Grant agreement no. 776479 - COACCHD2.4 Impacts on Industry, Energy, Services, and Trade CO Page 2 Version 1.1 This project has received funding from the European Union’s Horizon

D2.4 Impacts on Industry, Energy, Services, and Trade

CO Page 1 Version 1.1

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Grant Agreement No 776479.

Grant agreement no. 776479

COACCH

CO‐designing the Assessment of Climate CHange costs

H2020‐SC5‐2016‐2017/H2020‐SC5‐2017‐OneStageB


Work Package: 2

Due date of deliverable: M22 (SEP/2019)

Actual submission date: 28/OCT/2019

Start date of project: 01/DEC/2017

Duration: 22 months

Lead beneficiary for this deliverable: Fondazione Centro Euro‐Mediterraneo sui Cambiamenti Climatici (CMCC)

Contributors: Jessie Ruth Schleypen (CA), Shouro Dasgupta (CMCC), Stefan Borsky (UNI GRAZ), Martin Jury (UNI GRAZ), Milan Ščasný (CUNI), Levan Bezhanishvili (CUNI)

Disclaimer

The content of this deliverable does not reflect the official opinion of the European Union. Responsibility for the information and views expressed herein lies entirely with the author(s).




Dissemination Level PU Public x CO Confidential, only for members of

the consortium (including the Commission Services)

CI Classified, as referred to in Commission Decision 2001/844/EC

Suggested citation

Schleypen, J.R., Dasgupta, S., Borsky, S., Jury, M., Ščasný, M., Bezhanishvili, L. (2019). D2.4 Impacts on Industry, Energy, Services, and Trade. Deliverable of the H2020 COACCH project.




Table of contents 1. Impacts on Labour Productivity.................................................................................... 6 1.1. Introduction ........................................................................................................... 6 1.2. Data ....................................................................................................................... 9 1.2.1. Socioeconomic Data ....................................................................................... 9 1.2.2. Climate Data ................................................................................................. 12

1.3. Methodology ....................................................................................................... 13 1.3.1. Historical response of sectoral productivity to climatic conditions ............. 13 1.3.2. Impact Projections ........................................................................................ 15

1.4. Results ................................................................................................................. 15 1.4.1. Historical response ....................................................................................... 15 1.4.2. Projected impacts ......................................................................................... 17

1.5. Conclusion ........................................................................................................... 18 1.6. References ........................................................................................................... 20

2. The role of global supply chains in the transmission of weather induced production shocks 23 2.1. Introduction ......................................................................................................... 23 2.2. Related literature and conceptual discussion ..................................................... 25 2.3. Empirical Implementation ................................................................................... 27 2.4. Data and summary statistics ............................................................................... 28 2.4.1. Supply chain connectivity ............................................................................. 30 2.4.2. Natural disaster data..................................................................................... 32 2.4.3. Sectoral supply chain shocks ........................................................................ 34 2.4.3.1. Supply chain shock index ....................................................................... 36

2.5. The results ........................................................................................................... 38 2.5.1. Sectoral decomposition ................................................................................ 40 2.5.2. Projections .................................................................................................... 40 2.5.2.1. Country‐specific predicted impacts ....................................................... 42 2.5.2.2. Sector‐specific projected SCS impacts ................................................... 48

2.6. Conclusions .......................................................................................................... 49 2.7. References ........................................................................................................... 51 2.8. Appendix .............................................................................................................. 55

3. Climate change and wind power in Europe ............................................................... 58 3.1. Introduction ......................................................................................................... 58 3.2. Data ..................................................................................................................... 59 3.4. Econometric framework ...................................................................................... 61 3.5. Results ................................................................................................................. 62 3.6. Impact of future climate change ......................................................................... 64 3.7. Discussion and conclusion ................................................................................... 66 3.8. References ........................................................................................................... 67

4. Vulnerability of Global Hydropower to Climate Change ............................................ 69 4.1. Introduction .......................................................................................................... 69 4.3. Hydropower and climatic data ............................................................................ 72




4.4. Econometric Framework ..................................................................................... 74 4.4.1. Non‐stationarity and cointegration .............................................................. 75

4.5. Empirical Results.................................................................................................. 75 4.6. Impacts of future climate change........................................................................ 77 4.6.1. The case of Europe ....................................................................................... 78

4.7. Discussion and conclusion ................................................................................... 79 5. Impact on energy demand in Europe ......................................................................... 82 5.1. Introduction .......................................................................................................... 82 5.2. Trends in Energy Demand in Europe .................................................................... 83 5.3. Data and Methodology ........................................................................................ 84 5.4. Results .................................................................................................................. 85 5.4.1. Spatial heterogeneity climate change impacts ............................................. 86

5.6. Discussion and conclusion .................................................................................... 89 5.7. References ............................................................................................................ 91

6. Impacts on Tourism .................................................................................................... 92 6.1. Theoretical Framework ....................................................................................... 92 6.2. Data ..................................................................................................................... 93 6.2.1. Tourism and Socioeconomic Data ................................................................ 93

6.3. Methodology and Results .................................................................................... 95 6.3.1. Temperature effect on tourism: monthly‐data analysis .............................. 95 6.3.2. Temperature and climate extremes effect on tourism: country‐pairs analysis 109

6.4. Conclusion ......................................................................................................... 113 Appendix: Effects of Climate Change on Tourism: A Review ........................................ 117




Version log

Version Date Released by Nature of Change

1.1 30/SEP/2019 CA First Draft

1.2 01/OCT/2019 CA Second Draft

1.3 07/OCT/2019 CA Third Draft

1.4 09/OCT/2019 CMCC Revised by the coordinator

1.5 28/OCT/2019 CA Final version

Deliverable Summary This task used econometric approaches to analyse the adverse impacts of climate changes including extreme events (specified in Task 1.3) on production and productivity for different sectors of the economy in Europe at higher spatial resolution (NUTS2) compared to previous studies. Future impacts (short‐term to long‐term) under varying socio‐economic scenarios was projected by combining the estimated responses with the RCP scenarios selected with stakeholders in WP1. Due to the absence of subnational projections in the SSPs, the authors present the results as a percent reduction in the dependent variable relative to the reference period. Data from Eurostat at NUTS2 level, as well as survey data such as the newly released EU‐LFS2, will be used to estimate the response of labour productivity and economic activity indicators across economic sectors – including tourism ‐ in Europe (CA, CMCC, CUNI). International supply chain risks under extreme events for European industrial sectors were assessed by determining inter‐sectoral linkages using input‐output data, as from the OECD STAN database (UNIGRAZ). These linkages are then used in a gravity model of international trade to determine the degree of cross sectorial and transboundary transmission of extreme weather impacts to the EU industrial sectors. The analysis focuses on selected impact chains (e.g. supply chain disruptions of raw materials, or decreased export demand due to climate change in other world regions) as identified in the stakeholder engagement process (Task 1.3). Impacts on energy supply (CMCC) were assessed using geo‐referenced data on the distribution of power plants in European countries3 and accounting for changes in the future energy mix. Energy needs in the built environment for cooling and heating across different sectors of the economy were analysed by combining empirically‐estimated response functions using data from ENERDATA with the set of chosen scenarios. The effect of climate change on tourist demand for European region over more than last 15 years (2000‐2016) has been analysed econometrically based on monthly and annual country‐level data provided by EUROSTAT. Two major regression models are discussed in the report. Approach 1 considers the tourism data recorded for the destination of tourists only, while in Approach 2 country of origin is additionally indicated. Results are the basis for projections that will feed the macroeconomic assessment next part of COACCH research.




1. Impacts on Labour Productivity

1.1. Introduction

The European Union (EU), made up of 28 countries, functions as a single market, and contributes about 23 percent to the world economy despite having only 7 percent of the world population. The industry and services sectors make up 22% and 66% of the total EU Gross Domestic Product (GDP), respectively. 1 Being a single market with a single currency, the countries within the EU have abolished all border controls and trade barriers, making mobility of goods and services, as well as people more fluid. Having a single currency also unifies efforts to maintain price stability2. More than 64% of the EU’s total trade take place between countries within the Union.3

Despite the EU being an economic powerhouse, its economy is still at risk of dangerous climate change. About 98% of the increase in temperature, in relation to Europe’s 2014 record, can be attributed to anthropogenic climate change (Kam et al., 2015; EURO4M, 2015; Füssel et al., 2016). According to the IPCC AR5 report on Europe (Kovats et al., 2014)4, observed trends and future projections show increases in mean temperature and high temperature extremes across Europe, particularly in Southern Europe; and there is minimal evidence that resilience to heat waves and fires has improved. Precipitation is expected to increase in Northern Europe but expected to decrease in the South. These changes will have large implications for human health through increases in heat waves, and can greatly affect sectors such as agriculture, forestry, energy, transport, and tourism5.

Stern (2007) wrote that "climate change will affect the basic elements of life for people around the world", such that global warming can directly affect water availability, glacial and polar ice melting, widespread of heat stress, sea‐level rise; which could lead to issues on food security, health, reduction in habitable land area, and environmental sustainability. Among the channels through which climate affects the day‐to‐day lives of individuals is the loss in labor productivity, and ultimately, the loss in income. Research on occupational health risk from increasing temperatures using changes in ambient temperature point to the reduction in labour productivity in higher temperatures as a result of natural human responses to avoid damages to health, e.g., workers slow down, take more breaks to re‐hydrate, and cool down (Kjellstrom et al., 2009; Parsons, 2014; Dell et al., 2014; UNDP, 2016); or in cases of severe temperature increases, excessive body temperature and

1 Cited information is based on 2018 GDP in constant 2010 USD and population data from the World Bank WDI. EU working population (ages 15‐64) constitutes 6.7% of the world population, while total EU population including children below 15 and elderly above 65 constitutes 6.8% of the world population. Data was extracted on 11 Sept 2019. 2 Source: https://www.ecb.europa.eu/mopo/html/index.en.html. Accessed on 12 Sept 2019. 3 Source: https://europa.eu/european‐union/about‐eu/figures/economy_en 4 Source: IPCC AR5 Ch 23 Europe. https://www.ipcc.ch/site/assets/uploads/2018/02/WGIIAR5‐Chap23_FINAL.pdf 5 Source website: Adaptation to Climate Change – How we will be affected. Link: https://ec.europa.eu/clima/policies/adaptation/how_en. Accessed on 09 September 2019.




dehydration can cause, not only slower worker but also workers to make more mistakes, have increased accidental injuries (Bouchama and Knochel, 2002; Schulte et al., 2016; Schulte and Chun, 2009). Temperature shocks that are unmitigated through adequate thermoregulatory infrastructure, such as air conditioning, cause poor countries to remain poor due to productivity losses from an already heat‐stressed workforce (Heal and Park, 2013).

Economic projections of climate change impacts for Europe amount to a total household welfare loss of €190 billion or 2% of EU GDP by 2080s if current climate conditions persist, and no public adaptation occurs (Ciscar et al., 2014). In 2015, the Conference of Parties brought global leaders into the Paris Agreement, which aim "to strengthen the global response to the threat of climate change by keeping a global temperature rise this century well below 2 degrees Celsius above pre‐industrial levels and to pursue efforts to limit the temperature increase even further to 1.5 degrees Celsius". According to the Climate Action Tracker,6 current policies in place can only bring the global temperature to 3.4°C above pre‐industrial levels. Considering the unconditional pledges of countries in their Nationally Determined Contributions (NDCs), global temperature increase will likely be limited to 3.2°C. Meeting the 2°C target alone significantly reduces the estimated impacts to the EU by €60 billion, or a reduction in GDP of 1.2% (Ciscar et al., 2014). Europe has experienced more severe heatwaves in most recent years. According to the Copernicus Climate Change Programme data, Europe in July 2019 experienced temperatures that is already 1.2°C warmer than pre‐industrial era7, with France, Germany, Netherlands, the UK and Belgium experiencing day temperatures of over 40°C. The Ministry of Health in France recorded 567 deaths due to the June 2019 heatwave, when temperature reached 46°C and additional 868 during the July 2019 heatwave, when temperature reached 42.6°C. The 2003 heatwave was reported to have caused 70,000 more deaths than previous years8. According to the recently launched PESETA III report (Gosling et al., 2018)9, if climate change remains unmitigated and no adaptation occurs, labor productivity in outdoor labor could decline by 10‐15% by the end of the century compared to present‐day in southern European countries such as Bulgaria, Greece, Italy, Macedonia, Portugal, Spain, and Turkey; while the northern countries such as Denmark, Estonia, Finland, Norway, and Sweden will have an estimated 2‐4% decline.

Among the three main economic sectors, the outputs of agriculture and industry sectors are most affected by rising temperatures, and only the services sector seems to be protected from weather and climatic shocks. (Dell et al., 2012; Burke et al., 2015; IMF, 2017). While

6 The Climate Action Tracker is produced by Climate Analytics, NewClimate Institute, and Ecofys. Link accessed on 12 December 2018. Source link: https://climateactiontracker.org/publications/warming‐projections‐global‐update‐dec‐2018/ 7 Source: https://public.wmo.int/en/media/news/july‐matched‐and‐maybe‐broke‐record‐hottest‐month‐analysis‐began. Accessed on 13 Sept 2019. 8 Source: https://www.bbc.com/news/world‐europe‐48756480. Accessed on 13 Sept 2019. 9 Source: Gosling S.N., Zaherpour J., Ibarreta D., PESETA III: Climate change impacts on labour productivity, EUR 29423 EN, Publications Office of the European Union, Luxembourg, 2018, ISBN 978‐92‐79‐ 96912‐6, doi:10.2760/07911, JRC113740. https://ec.europa.eu/jrc/en/publication/peseta‐iii‐climate‐change‐impacts‐labour‐productivity




rising temperatures are seen to be detrimental to most economic activity, there may still be positive impacts on some sectors. For instance, some crops that prosper in warmer conditions and summer tourism could benefit from this change (Bosello, et al., 2012; Grillakis, et al., 2016). Barrios and Ibañez (2015) estimate a modest annual increase of 0.32% of GDP for Northern Europe, while Southern Europe could experience a reduction by 0.45% of GDP per year given current climate conditions.

While economic systems are bound by political borders that define a country, state, or region, the atmosphere does not follow the same boundaries, and thus spatial entities close to each other will likely experience similar climates. Given a wide set of studies providing evidence of the non‐linear relationship between climate10 and economic productivity11, simultaneous impacts could likely happen. In a closely knitted regional economic union such as the EU, simultaneous impacts could spillover across regions. That is, negative impacts from climate shocks could have a ripple effect on the rest of the countries within the union, as well as, positive spillovers may occur such as knowledge sharing on adaptation practices, similar sociodemographic characteristics that point to higher levels of resilience (i.e., educational attainment of the population, gender equality, etc.)

Due to the EU’s strong economic integration, the lack of economic studies looking into indirect effects of climate change, and the explicit expression of interest from key stakeholders of the project, the authors of this research contribute to the existing economic studies by acknowledging the possibility and role of spillovers to and from countries within the EU. Furthermore, this research makes use of sub‐national, sub‐sectoral data to disaggregate further the differentiate impacts within a country. The results show that regions belonging to the same country could be at the losing and winning end of climate change, owing also to the difference in vulnerability of economic sectors to varying climate conditions. The results have strong policy relevance, because it provides scientific evidence to support informed decisions on how adaptation in a country should be prioritized (i.e., which sector and which region) and designed (i.e., adaptation action to reduce direct impacts would likely differ from action to prevent a negative spillover from an impact on a neighbouring spatial unit). For instance, if a country is highly sensitive to the impacts of another, then it would suggest strengthening sectors closely tied between the two to prevent the spreading of impacts. Furthermore, in view of global cooperation in fighting climate change, information on spillover effects incentivizes neighboring countries to ensure critical hubs do not experience severe and lasting impacts, should they also be at the receiving end of the negative spillovers; therefore, preventing a domino‐effect of climate impacts.

10 Mainly referring to temperature and precipitation. 11 See papers from Dell et al. (2014), and Tol (2018) for a review of economic impacts of climate change literature.




1.2. Data

1.2.1. Socioeconomic Data

We use EUROSTAT data on sectoral gross value‐added (GVA)12 in million Euros from 255 sub‐national regions (NUTS‐2) of 28 EU member states between 2000 ‐ 2015. The three sectors are defined as follows; (1) agriculture, which includes agriculture, forestry, and fishing; (2) industry, which includes mining, manufacturing, construction, and utilities (electricity, water, gas); and (3) services, which includes wholesale and retail trade; transport; accommodation and food service activities; information and communication, financial and insurance activities; real estate activities; professional, scientific and technical activities; administrative and support service activities, and public administration and defense; compulsory social security; education; human health and social work activities; arts, entertainment and recreation, repair of household goods and other services. Adhering to the focus of this task, we only present results for industry and services sectors.

12 The sectoral breakdown follows the European Classification of Economic Activities(NACE R2): [1] "Total ‐ all NACE activities"; [2] "Agriculture, forestry and fishing"; [3] "Industry (except construction)"; [4] "Manufacturing; [5] "Construction; [6] "Wholesale and retail trade, transport, accommodation and food service activities”; [7] "Wholesale and retail trade; transport; accommodation and food service activities; information and communication”; [8] "Information and communication”; [9] "Financial and insurance activities" ; [10] "Financial and insurance activities; real estate activities; professional, scientific and technical activities; administrative and support service activities; [11] "Real estate activities”; [12] "Professional, scientific and technical activities; administrative and support service activities"; [13] "Public administration, defense, education, human health and social work activities”; [14] "Public administration and defense; compulsory social security; education; human health and social work activities; arts, entertainment and recreation, repair of household goods and other services"; [15] "Arts, entertainment and recreation; other service activities; activities of household and extra‐territorial organizations and bodies"




Figure 1.1.1: Average Gross Value Added, 2000‐2015, EU NUTS‐2 regions

Industry GVA. The largest contributors to average industrial GVA (2000‐2015) are Germany (24.2%), France (12.8%), the United Kingdom (12.5%), Italy (12.1%), and Spain (9.3%).




Figure 1.1.2: Average Industry Gross Value Added, 2000‐2015, EU NUTS‐2 regions

Industry is mainly driven by manufacturing, of which the regions with the largest average industrial GVA and above 50% share of manufacturing are the regions of Lombardia which includes Milan in Northwest Italy (ITC4), Île de France which includes Paris (FR10) and Auvergne‐Rhône‐Alpes (FR71), Stuttgart (DE11), Oberbayern (DE21), Düsseldorf (DEA1) in Germany, Catalonia in Spain (ES51).

The region of Lombardia has a comparative advantage on manufacturing of metal products, production of base metals, wearing apparel, media printing and reproduction, and wooden products.13 The region of Île de France is the richest in the country, whose key industries include electronics and ICT, aeronautics, biotechnologies, finance, mobility, automobile, pharmaceuticals, and aerospace14; while the region of Auvergne‐Rhône‐Alpes, located in the eastern central part of France, is home to large chemical and plastics industry rubber production mechanical engineering industries and agro‐food industries. The region also

13 Source link: https://ec.europa.eu/growth/tools‐databases/regional‐innovation‐monitor/news/innovation‐policy‐lombardy‐top‐manufacturing‐region‐europe. Accessed on 10 Sept 2019. 14 Information taken from InvestParisRegion (2017) and cited in the EC website. Source link: https://ec.europa.eu/growth/tools‐databases/regional‐innovation‐monitor/base‐profile/ile‐de‐france. Accessed on 10 Sept 2019.




specializes in high‐tech industries such as pharmaceutics, nutrition and health, biotechnology and ICT.15

Services GVA. Similar to the industry sector, the main contributing countries in EU’s the average services sector GVA are Germany (19.7%), France (16.6%), the UK (16%, Italy (12.4%), and Spain (8.4%). The services sector is mainly driven by trade and transport16, public administration17, and finance18 subsectors.

Figure 1.1.3: Average Services Gross Value Added, 2000‐2015, EU NUTS‐2 regions

1.2.2. Climate Data

Our historical climatic data comes from the Global Land Assimilation System (GLDAS v2.1), this is a re‐analysed gridded climatic dataset, with 0.25° x 0.25° spatial and 3‐hourly

15 Source link: https://ec.europa.eu/growth/tools‐databases/regional‐innovation‐monitor/base‐profile/auvergne. Accessed on 10 Sept 2019. 16 NACE R2 sector ‐ Wholesale and retail trade; transport; accommodation and food service activities; information and communication 17 NACE R2 sector ‐ Public administration, defence, education, human health and social work activities 18 NACE R2 sector ‐ Financial and insurance activities; real estate activities; professional, scientific and technical activities; administrative and support service activities




temporal resolution. We begin with the gridded 3‐hourly data and compute the various aggregated indicators at the NUTS‐2 level. A Global Circulation Model (GCM) has a typical resolution of 200 km which is not suitable for the application at sub‐national scale; hence, for future projections, we opted to use data from four high‐resolution Regional Climate Models (RCM): KNMI RACMO22E, IPSL‐CM5A‐MR, MPI‐ESM‐LR, and CNRM‐CM5. Based on the stakeholder interests established in COACCH D1.5, we have focused our projections on the Representative Concentration Pathway (RCP) 4.5 as the likely scenario and closest to the proposed Nationally Determined Contributions (NDC) pathway, and an extreme scenario RCP 8.5, which represents the worst possible case. Table 1 below provides the descriptive statistics for some of the relevant variables. Table 1.1.1 Descriptive statistics of key variables (2000‐2015)

Variables Mean Min Max

Industrial GVA (€ million) 10367.9 90.5 92709.5 Services GVA (€ million) 29285.6 585.0 505670.6 Mean Temperature (°C) 10.2 ‐1.6 26.1 Max Temperature (°C) 14.2 1.6 31.9

Total Precipitation (mm) 2.5 0.4 7.1 WBGT (°C) 9.5 2.5 20.0

WSDI (number of days) 9.3 0.0 346.0

1.3. Methodology

1.3.1. Historical response of sectoral productivity to climatic conditions

Non‐spatial benchmark model Following Burke et al. (2015) and Newell (2018), we use a fixed‐effects panel regression, wherein our dependent variable 𝑙𝑛𝑦 is the log of labor productivity defined by GVA per hours worked, GVA per person employed, and GVA per working population in region 𝑖 in a given year 𝑡 for sector 𝑠. f(C) includes the non‐linear relationship between the temperature and productivity, controlled for by including both the linear and its squared‐term; an acclimatization variable ‘𝑡𝑑𝑒𝑣’ that we introduce in this study, computed as the difference of the mean temperature from a rolling average of four periods prior19; total annual precipitation (and its second‐degree polynomial), heat extremes represented by the Warm Spell Duration Index (WSDI)20. All our specifications include region 𝛼 and year 𝛾 specific fixed‐effects, and a quadratic time trend 𝜃 𝑡 𝜃 𝑡 to capture non‐linear technological changes.

19 The length of the rolling average is an arbitrary value and was guided by the observable length of a business cycle in EU regions 20 Annual count of days with at least 6 consecutive days when daily maximum is greater than 90th percentile. The 90th percentile was determined from months June‐August in years 1981‐2010 as the baseline years, and applied to summer months – May to October – for years 2000‐2015 to count the number of days exceeding this threshold.




𝑙𝑛𝑦 𝑓 𝐶 𝛼 𝛾 𝜃 𝑡 𝜃 𝑡

where 𝑓 𝐶 𝛽 𝑡𝑒𝑚𝑝 𝛽 𝑡𝑒𝑚𝑝 𝜑𝑡𝑑𝑒𝑣 𝛿 𝑝𝑟𝑒𝑐 𝛿 𝑝𝑟𝑒𝑐 𝜎𝑊𝑆𝐷𝐼

Given existing findings on the non‐linear and inverse U ‐shaped relationship of economic performance and in particular local temperatures, we expect for the set of coefficients of temperatures that 𝛽 0,𝛽 0. In this case, the results indicate a non‐linear relationship with an "optimal" value of the local temperature computed as 𝑡𝑒𝑚𝑝 | 𝛽 / 2 ∗ 𝛽 | for all economic sectors. Spatial Econometrics According to the first law of geography, “Everything is related to everything else, but near things are more related than distant things” (Tobler 1970). Using the Moran's I test to test for the presence of spatial correlation, the null hypothesis of spatial independence/randomization is rejected for all the climatic and productivity variables, suggesting that the climatic stressors and productivity in a given NUTS‐2 region is spatially dependent on those in the neighbouring NUTS‐2 regions. Thus, not controlling for the spatial dependence between NUTS‐2 regions will likely provide biased results. Hence, we utilize a binary spatial weight with the distance threshold set to 150 KM based on the greatest Euclidean distance measured between two places on a Cartesian plane. The spatial dependence between NUTS‐2 regions decays as distance between them increases. By incorporating the spatial weights, we control for spatially lagged climatic stressors and run a Spatial Durbin Model (SDM) model of the following form;

𝑙𝑛𝑦 𝜔𝑾𝑦 𝑓 𝐶 𝑾𝑓 𝐶 𝛼 𝛾 𝜃 𝑡 𝜃 𝑡

where W𝑓 𝐶 𝛽 𝑡𝑒𝑚𝑝 𝛽 𝑡𝑒𝑚𝑝 𝛽 𝑾𝑡𝑒𝑚𝑝 𝛽 𝑾𝑡𝑒𝑚𝑝 𝜑 𝑡𝑑𝑒𝑣 𝜑 𝑾𝑡𝑑𝑒𝑣

𝛿 𝑝𝑟𝑒𝑐 𝛿 𝑝𝑟𝑒𝑐 𝛿 𝑾𝑝𝑟𝑒𝑐 𝛿 𝑾𝑝𝑟𝑒𝑐 𝜎 𝑊𝑆𝐷𝐼 𝜎 𝑾𝑊𝑆𝐷𝐼 W is the spatial weights matrix21, 𝜔 is the spatial coefficient of the spatial autocorrelated dependent variable, 𝛽 ,𝛽 , 𝛿 , 𝛿 ,𝜑 ,𝜎 are the spatial coefficients of the spatially‐weighted independent variables, and the term W𝑓 𝐶 measures the potential spillovers effect that occurs in climatic stressors across NUTS‐2 regions (Baltagi 2003) while 𝑾𝑦 measures the possible spillover effect in productivity across the NUTS‐2 regions. The SDM specification is also completely specified and allows us to investigate the impact of both direct (contemporaneous) and indirect effects (spatial lags) of climatic stressors on labour productivity.

21 We use a binary spatial weight matrix with a distance threshold of 150 km.




1.3.2. Impact Projections

In order to estimate the impacts of future warming on sectoral, sub‐national productivity, we combine our spatial non‐linear econometric estimates with gridded climate data from four different regional climate models (KNMI RACMO22E, IPSL‐CM5A‐MR, MPI‐ESM‐LR, and CNRM‐CM5) under two different warming scenarios. We use the Delta method (Dasgupta, 2018) to combine the econometric estimates with future climate data with the resulting impacts being percentage change in productivity due to future climate change. We have opted not to use any of the SSP scenarios due to two main reasons: (1) the results already presented as percentage change in the labor productivity, given a unit change in the climate variables, which will remain the same for all SSP; and (2) our results are disagreggated to the NUTS‐2 level, in which case, weights need to be assigned in order to match the country‐level SSPs.

1.4. Results

1.4.1. Historical response

We have considered three dependent variables to represent labor productivity and have found GVA per working population the most satisfactory. An alternative to working population could have been the number of hours worked which, unfortunately, are not available for Belgium and Croatia. Another possibility could be offered by the number of employees at the sectoral level, which are not available for Lithuania and Ireland. Furthermore, exploratory regressions conducted both at the country‐level and NUTS‐2 level for the latter two labor productivity definitions produced results that were not statistically significant, or identifying optimal temperatures inconsistent with existing studies, or out‐of‐sample; thus, providing no basis for projections. In line with previous studies, (Burke et al. 2015; IMF 2017), we find non‐linear impacts of temperature and total precipitation on productivity for all the industry and services sectors considered, except for manufacturing. The derived optimal spatially lagged temperatures maximizing industry, and construction are 10.8°C, 10°C, respectively. For labour productivity in the services sector, the spatially lagged temperature is not statistically significant, however, the contemporaneous temperature is statistically significant with an optimum of 16.3°C. Our results suggest that, while productivity increases with initial increases in temperature, beyond a certain threshold, further increases in temperature results in a negative impact. As expected, the optimal temperature for the industry and construction sectors are comparatively lower than services, as the workers in this sector are more exposed to outside temperatures. The lower impacts experienced in the services sector is consistent with existing evidence (IMF, 2017). The derived optimal temperature for the services sector is above the sample's mean temperature, and therefore provides weaker evidence of the downward‐sloping productivity beyond the computed optimal temperature. The difference in estimated impacts for each sector suggests that economic structures play a significant role in determining the overall economic impact of climate change.




Figure 1.4 Nonlinear relationship between mean temperature and productivity

Non‐linear relationship between mean temperature and productivity (dark navy line) at the NUTS‐2 level with 95% confidence interval (light blue spikes). Left‐panel shows the impact on the industrial sector while right‐panel shows the impact on the construction sector. Specification controls for mean temperature (and its second‐degree polynomial), total precipitation (and its second‐degree polynomial), temperature shock, WSDI, income group‐maximum temperature interaction‐term, and year and region fixed‐effects.

We also find a non‐linear direct impact of temperature for construction with an optimum of 13°C, suggesting that temperature has both direct and indirect adverse beyond a certain threshold in this sector. Our estimates also suggest that there is significant negative direct impact of temperature shock and indirect impact of WSDI on both industrial and construction labour productivity. Table 1.2 Regression results

(1) (2) (3) (4) Industry Construction Manufacturing Services

Spatial lag of industrial GVA 0.789*** (0.000)

Spatial lag of construction GVA 0.899*** (0.000)

Spatial lag of manufacturing GVA 0.614*** (0.000)

Spatial lag of service GVA 0.597*** (0.000)

WSDI 0.001*** 0.001** 0.001** 0.000 (0.000) (0.036) (0.026) (0.584)

Temperature deviation ‐0.027** ‐0.054*** ‐0.007 ‐0.071*** (0.031) (0.001) (0.628) (0.000)

Total precipitation 0.116*** 0.153*** 0.087* 0.003 (0.004) (0.004) (0.062) (0.916)

Total precipitation‐squared ‐0.019*** ‐0.024*** ‐0.015* ‐0.000 (0.006) (0.008) (0.052) (0.938)

Mean temperature 0.056*** 0.063** 0.019 0.077*** (0.004) (0.014) (0.389) (0.000)

Mean temperature‐squared ‐0.001* ‐0.002*** ‐0.000 ‐0.002*** (0.084) (0.009) (0.945) (0.000)

Spatial lag of WSDI ‐0.003*** ‐0.003*** ‐0.003*** ‐0.002*** (0.000) (0.000) (0.000) (0.000)

Spatial lag of temperature deviation 0.004 0.016 ‐0.010 0.023* (0.812) (0.422) (0.547) (0.054)

Spatial lag of total precipitation 0.018 0.015 0.019 0.170*** (0.755) (0.850) (0.777) (0.000)

Spatial lag of total precipitation‐squared ‐0.006 ‐0.006 ‐0.004 ‐0.031*** (0.605) (0.684) (0.741) (0.000)

Spatial lag of mean temperature 0.049** 0.051* 0.050* 0.013 (0.034) (0.096) (0.060) (0.466)

Spatial lag of mean temperature‐squared ‐0.002** ‐0.003* ‐0.002 0.000 (0.021) (0.052) (0.174) (0.741)

Constant ‐5.453*** ‐6.875*** ‐3.961*** ‐3.761*** (0.000) (0.000) (0.000) (0.000)




Observations 3,067 3,067 2,875 3,067 R‐squared 0.680 0.737 0.518 0.744

Number of regions 256 256 240 256

robust p‐value in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Neither the direct or indirect effects of temperature are statistically significant for the manufacturing sector, however, spatially lagged WSDI has a negative impact on manufacturing labour productivity, suggesting that extreme climatic events are more critical in this sector. Thus, we have evidence that both gradual and extreme temperature events have significant negative impacts on labour productivity. Finally, the spatial lag of sectoral GVA in all the three sectors are positive and statistically significant. These provides evidence of adaptation based on increasing output taking place (Dasgupta and Bosello, 2016).

1.4.2. Projected impacts

Using a multi‐model mean of four RCMs, the results suggest that under an unmitigated warming scenario of RCP8.5, future climate change will result in a decline of industrial productivity by 4.3% by 2070. While labour productivity in the construction sector will decline by 6.6% by 2070. For both industrial and construction productivity, the highest declines will occur in Greece (Peloponnese, Thessaly, and Attica), Italy (Puglia), Spain (Region of Murcia and Andalusia), and Portugal (Algarve) while some regions in Austria, Estonia, Finland, Sweden, and the north‐eastern and north‐western Italian regions will gain. Figure 1.5 Impact projections

Future impact under RCP8.5 on industrial (left‐panel) and construction productivity (right‐panel) by 2070. The impacts are computed using the Delta method and a reference period of 1985 ‐ 2005.

It should be noted that these estimated future impacts are solely driven by changes in climatic stressors and do not consider any possible adaptation, although there is little evidence of adaptation taking place (Burke et al., 2015).




As a comparison we also compute projections for a more moderate warming scenario of RCP4.5. Our results suggest that industrial productivity will decline by 2.7% by 2070, while labour productivity in the construction sector will decline by 3.1%.

1.5. Conclusion

This paper investigated the impacts of both gradual and extreme climatic stressors on sectoral productivity in Europe using sub‐national (NUTS‐2) data combined with high‐resolution climatic data. Employing spatial econometric techniques, we find differentiated and non‐linear impacts of temperature across sectors and within countries. Econometric estimates suggest that a significant part of the total impacts to sectoral output are transmitted through losses in productivity from extreme heat and warming; which are experienced in large part by workers whose occupational environment are hard to control, e.g. construction. The estimates suggest that temperature that maximizes productivity in the industry, construction, and services sectors are 10.8°C, 10°C, and 16.3°C, respectively. The results are consistent with existing publications that the relationship between temperature and economic productivity is non‐linear; however, we find insufficient evidence to generalize the idea of a non‐linear relationship for all the economic sectors. The weak argument for the use of the quadratic form is drawn from the results of the services sector analysis, in which case, the derived optimal temperature is higher than any of the mean temperatures in the sample. The lack of evidence to support the presence of the downward‐sloping curve, therefore, suggests further inquiry in generalizing predefined forms of equation to represent the temperature‐productivity relationship. The paper also finds significant adverse effects of extreme heat events using WSDI and temperature shocks using the acclimatization variable, suggesting that both gradual and extreme climatic change affect productivity. To quantify the impacts of future climate change, we combine our non‐linear econometric estimates with future warming scenarios that consistent in the Paris Agreement. Assuming that the computed optimal temperature holds for years in the future, the results suggest that under an unmitigated warming scenario of RCP8.5, future climate change will result in a decline of industrial productivity by 4.3% and construction labour productivity by 6.6%. Due to future warming, regions in Greece (Peloponnese, Thessaly, and Attica), Italy (Basilicata and Puglia), Spain (Region of Murcia and Andalusia), and Portugal (Algarve) will suffer the highest declines while some of the colder regions in Estonia, Finland, northern Italy, and Sweden will experience gains. Under a more moderate warming scenario of RCP4.5, industrial productivity will decline by 2.7% by 2070, while labour productivity in the construction sector will decline by 3.1%. These declines in productivity due to climate change will likely be transmitted through overall economic activity and have negative multiplier effects. Our results provide evidence in support of strong mitigation action. Which show that under more moderate warming scenario and avoiding extreme climatic change, decline in labour productivity will be around 3%, significantly lower than the damages projected under RCP8.5. While the estimates of a scenario with the current NDCs are lower than estimated




impacts of current climate conditions, as Ciscar et al., (2014) have estimated; there is still room for significant reduction in impacts through mitigation action. Furthermore, in order to protect industries that have developed in regions with comparative advantage (e.g., from abundance in natural resources, technological advantage, etc.), adaptation action needs to be strengthened in light of anticipated climate changes and corresponding estimated impacts. The results of this study could, therefore, be further utilized to support cost‐benefit analyses of avoided climate change losses against the additional investment; as well as the estimation of adaptation costs, given avoided costs from mitigation.




1.6. References

Baccini, M., Biggeri, A., Accetta, G., Kosatsky, T., Katsouyanni, K., Analitis, A., Michelozzi, P. (2008). Heat Effects on Mortality in 15 European Cities. Epidemiology, 19(5), 711–719. https: //doi.org/10.1097/EDE.0b013e318176bfcd

Barrios, S., and Ibañez, J. N. (2015). Time is of the essence: adaptation of tourism demand to climate change in Europe. Climatic Change, 132(4), 645–660. https://doi.org/10.1007/ s10584‐015‐1431‐1

Bosello, F., Eboli, F., and Pierfederici, R. (2012). Assessing the Economic Impacts of Climate Change. Retrieved from http://papers.ssrn.com/abstract=2030223

Bouchama, A., and Knochel, J. (2002). Heat Stroke. The New England Journal of Medicine, 346, 1978–1988. https://doi.org/10.1056/NEJMra011089 Burke, M., Hsiang, S. M., and Miguel, E. (2015). Global non‐linear effect of temperature on economic production. Nature, (1), 1–16. https://doi.org/10.1038/nature15725

Burke, M., S. Hsiang, and E. Miguel. (2015). Global non‐linear effect of temperature on economic production. Nature; 527,235–239, doi:10.1038/nature15725.

Ciscar, J., Feyen, L., Soria, A., Lavalle, C., Raes, F., Perry, M., Ibarreta, D. (2014). Climate Impacts in Europe. The JRC PESETA II Project (JRC Scientific and Policy Reports No. EUR 26586EN). Seville, Spain. https://doi.org/10.2791/7409

Dell, M., Jones, B. F., and Olken, B. A. (2014). What Do We Learn from the Weather. The New Climate‐ Economy Literature. Journal of Economic Literature, 52(3), 740–798. https: //doi.org/10.3386/w19578

Grillakis, M. G., Koutroulis, A. G., Seiradakis, K. D., and Tsanis, I. K. (2016). Implications of 2°C global warming in European summer tourism. Climate Services, 1, 30–38. https://doi. org/10.1016/j.cliser.2016.01.002

Heal, G., and Park, J. (2013). Feeling the heat: temperature, physiology and the wealth of nations (No. 19725). NBER Working Paper Series (Vol. 12). https://doi.org/10.1007/ s13398‐014‐0173‐8.3

IMF. (2017). The effects of weather shocks on economic activity. How can low income countries cope (Chapter 3). Global Economic Outlook: Seeking sustainable growth. Retrieved from https: //www.imf.org/en/Publications/WEO/Issues/2017/09/19/world‐economic‐outlook‐october‐ 2017#Chapter3

Zivin, J.G., Hsiang, S.M., Neidell, M. (2018). Temperature and Human Capital in the Short and Long Run, Journal of the Association of Environmental and Resource Economists, University of Chicago Press, vol. 5(1), pages 77‐105.

Kjellstrom, T., Kovats, R. S., Lloyd, S. J., Holt, T., and Tol, R. S. J. (2009). The direct impact of climate change on regional labour productivity. Archives of Environmental and Occupational Health, 64(4), 217–227. https://doi.org/10.1080/19338240903352776




Kjellstrom, T., Otto, M., Lemke, B., Hyatt, O., Briggs, D., Freyberg, C., and Lines, L. (2016). Climate Change and Labour: Impacts of Heat in the Workplace. Retrieved from Climate Change and labour: Impacts of Heat in the Workplace

Millar, R. J., Fuglestvedt, J. S., Friedlingstein, P., Rogelj, J., Grubb, M. J., Matthews, H. D., Skeie, R.B., Forster, P.M., and Frame, D.J., Allen, M. R. (2017). Emission budgets and pathways consistent with limiting warming to 1.5C. Nature Geoscience, 10(10), 741–747. https: //doi.org/10.1038/ngeo3031

Newell, R.G., Prest, B.C. and Sexton, S.E. (2018). The GDP Temperature Relationship: Im‐ plications for Climate Change Damages. RFF Working Paper Series, WP 18‐17 REV. Available at http://www.rff.org/research/publications/gdp‐temperature‐relationship‐implications‐ climate‐change‐damages

Parsons, K. C. (2014). Human Thermal Environment. The Effects of Hot, Moderate, and Cold Environments on Human Health, Comfort, and Performance (3rd editio). New York: CRC Press. https://doi.org/10.4324/9780203302620

Rogelj, J., and Knutti, R. (2016). Geosciences after Paris. Nature Geoscience, 9, 187–189. https://doi.org/10.1038/ngeo2668

Rogelj, J., Mccollum, D. L., and Riahi, K. (2013). The UN’s ‘Sustainable Energy for All’ initiative is compatible with a warming limit of 2°C. Nature Publishing Group, 3(6), 545–551. https://doi.org/10.1038/nclimate1806

Sahu, S., Sett, M., and Kjellstrom, T. (2013). Heat exposure, cardiovascular stress and work

productivity in rice harvesters in India: implications for a climate change future. Industrial Health, 51(4), 424–31. https://doi.org/10.2486/indhealth.2013‐0006

Schleussner, C. F., Lissner, T. K., Fischer, E. M., Wohland, J., Perrette, M., Golly, A., and Schaeffer, M. (2016). Differential climate impacts for policy relevant limits to global warming: the case of 1.5°C and 2°C. Earth System Dynamics, 7(2), 327–351. https://doi.org/10.5194/esd‐ 7‐327‐2016

Schulte, P. a, and Chun, H. (2009). Climate change and occupational safety and health: es‐ tablishing a preliminary framework. Journal of Occupational and Environmental Hygiene, 6(April 2015), 542–554. https://doi.org/10.1080/15459620903066008

UNDP (2016). Climate Change and Labour: Impacts of heat in the workplace. Retrieved from

http://www.thecvf.org/wp‐content/uploads/2016/04/Climate‐and‐Labour‐Issue‐Paper_28‐ April‐2016_v1_lowres‐1.pdf







2. The role of global supply chains in the transmission of weather induced production shocks

2.1. Introduction

Every year numerous natural disasters happen worldwide. In 2018 only, 315 natural disaster occurred, which resulted in 11,804 deaths, 68 Million people affected and 131,7 billion USD in direct economic damages (CRED 2019).22 Besides direct social impacts and economic damages, natural disasters also have a widespread secondary or indirect impact on the economy. There is a fair amount of literature focusing on the sectoral and macroeconomic outcomes of natural disasters (e.g., Raddatz 2007, Noy 2009, Skidmore & Toya 2002, Dell, Jones & Olken 2012, Cavallo, Galiani, Noy & Pantano 2013, Hsiang & Jina 2014, Felbermayr, Gröschl, Sanders, Schippers & Steinwachs 2018).23 Additionally, literature has focused on weather and disaster effects on agriculture (e.g., Schlenker & Roberts 2009, Burke & Emerick 2016), health (e.g., Deschenes, Greenstone & Guryan 2009), labor (e.g., Graff Zivin & Neidell 2014), social conflict (e.g., Hsiang, Meng & Cane 2011, Hsiang, Burke & Miguel 2013), migration (e.g., Missirian & Schlenker 2017).

Few articles exist, which determine the impact of disasters on international trade.24 Jones & Olken (2010) analyze the impact of temperature and precipitation on a country's export growth. For an increase in temperature they find a reduction in exports for low income countries. Based on a gravity model of trade Oh & Reuveny (2010) and Gassebner, Keck & Teh (2010) analyze the impact of large natural and technological disasters on bilateral trade. Accounting for the size of the affected country and its political environment, the find a negative effect on a country's trade activity. Felbermayr & Gröschl (2013) use the impact of natural disasters on international trade in an instrumental variable procedure to consistently estimate the impact of international trade on per capita income. Their results suggest that depending on the degree of financial integration large natural disasters decrease exports and increase imports. Recent articles by Oh (2017) and EL‐Hadri, Mirza & Rabaud (2018) find that the potential negative impact of a disaster depends on the type of affected industry, country size and its level of development and the intensity of the disaster.

In addition, countries can be affected by large natural disasters abroad, which are propagated globally over economic network structures. Economies today are organized in fine interweaved networks of production units ‐ each commonly receiving input flows from their suppliers to produce products, which are then often used as inputs in other production units (Carvalho 2014). Idiosyncratic shocks, which are triggered for example by natural disasters, and affect only a specific production unit, can be widely dispersed in the economy

22 The year 2018 was below the 10‐year average, which is 348 natural disasters, 67,752 deaths, 198,8 million affected and 166,7 billion in direct economic damages. 23 The year 2018 was below the 10‐year average, which is 348 natural disasters, 67,752 deaths, 198,8 million affected and 166,7 billion in direct economic damages. 24 For a recent review of the literature on the impacts of natural disasters on international trade, see Osberghaus (2019).




through inter‐industry linkages. In course of this dispersion, the impact of a single shock is significantly multiplied. A prominent example of such an event is the 2011 flood in Thailand, which affected 14,500 companies for automobiles and computers situated in the inundated area around Bangkok. After the event, Nissan and Toyota had to suspend production worldwide, because of problems in obtaining parts from Thailand. Also, global prices of hard disks doubled, because almost half of the world's hard disk supply were produced in Thailand (Liverman 2016). Another event, which is thoroughly studied in the recent literature, is the 2011 Tōhoku Earthquake in Japan, which had large, significant impacts on the US manufacturing industry (Barrot & Sauvagnat 2016, Boehm, Flaaen & Pandalai‐Nayar 2019, Carvalho, Nirei, Saito & Tahbaz‐Salehi 2016). Recent studies, which study the role of production networks in the propagation of shocks across sectors (Acemoglu, Carvalho, Ozdaglar & Tahbaz‐Salehi 2012, Puzzello & Raschky 2014) or within sectors across firms (Barrot & Sauvagnat 2016, Boehm et al. 2019, Carvalho et al. 2016) give some empirical evidence on this aggregating relationship. The extend of this effect depends on the in‐ and outdegree distribution in the production network, i.e. the degree of connectivity between the production units.

In this paper, we provide empirical evidence of the effect of disruptions in the supply chain, which are caused by natural disasters, on a country's export activity. In particular, following Puzzello & Raschky (2014) we analyze how natural disasters propagate through sectors, which are strongly interconnected via input‐output linkages, from country to country. To analyze the role of the global production networks in the propagation of shocks, we first construct a measure, which captures the degree of input‐output connectivity between sectors and countries. We obtain information on input‐output linkages for a large set of countries from 1990‐2015 from the EORA global supply chain database (Lenzen, Kanemoto, Moran & Geschke 2012, Lenzen, Moran, Kanemoto & Geschke 2013). Second, we use a subset of extreme weather indices defined by the Expert Team on Climate Change Detection and Indices (ETCCDI, Karl, Nicholls & Ghazi 1999, Sillmann, Kharin, Zhang, Zwiers & Bronaugh 2013) based on daily temperature and precipitation data, to construct proxies for historical and projected future natural disasters. Finally, combining our variable of supply‐chain interlinkages and our proxy for natural disasters, gives us a measure of supply chain shocks to a sector and country, which is then used in a fixed‐effect model to estimate its impact on a sector's export performance. In a further step, as the frequency and intensity of natural disasters will increase in future due to climate change (e.g. Sillmann, Kharin, Zwiers, Zhang & Bronaugh 2013), we give insights in the future sectoral exposure to natural disaster shocks transmitted over the supply chain.

Our results highlight that supply chain disruptions, caused by large natural disasters abroad, significantly reduces a sector's export value. A one standard deviation increase in our supply chain shock measure reduces a sector's export value by around 11 percent. Further, we show that this negative effect is mainly driven by the manufacturing and agricultural sector. Finally, predicting future supply chain shocks we find a potentially strong impact of climate change on the extension of the negative effects of supply chain shocks on a sector's export value. Depending on the global circulation model and the representative emission pathway climate change reduces exports via supply chain shocks by about 8 percent to 26 percent.




The impact of climate change is heterogenous between countries and sectors and depends on the extend of a sectors global production network and the strength of increase in natural disaster in that region. Our results suggest, that it is countries in the tropics and subtropics, which will be particularly negative affected by these shocks in future.

This paper contributes to the literature on the role of production networks in the propagation of shocks across sectors. Our study is based on sector level information for a large number of countries over a long period of time. Although, we are less disaggregated than recent studies based on within sectors across firm’s variation (Barrot & Sauvagnat 2016, Boehm et al. 2019, Carvalho et al. 2016) we are able to consider a multitude of different natural disaster shocks happening all over the globe. This allows us to account for country and sectoral specificities to deal with supply chain disruptions. Further, to the best of our knowledge this is the first study to examine a country and sectoral exposure to supply chain shocks taking climate change induced changes in the occurrence of natural disasters into account. Finally, our study relates to the literature studying the macroeconomic impacts of natural disasters (e.g., Cavallo et al. 2013, Felbermayr et al. 2018, Mohan, Ouattara & Strobl 2018). We add one potential mechanism how disasters can affect macroeconomic outcomes even in regions outside of the disaster affected area.

The insights of this study have important policy implications. Adaptation policies to current natural disasters as well as potential future disaster exposure need to take the vulnerability of sectors to supply chain disruptions into account. Pro‐active measures to mitigate the impact of supply chain shocks can be based on information campaigns, regulation or firm‐level insurance. At individual level, firms, for example, can increase their level of geographical diversification in their global production network or intensify the use of storage facilities. Re‐active, post‐disturbance measures could be based on direct disaster relief aid to shorten the recovery period in the affected region and decrease the length of the supply chain disruption. However, the success of these post‐disturbance measures relies on several country level‐factors, like the quality of political and financial institutions.

The report is structured as follows. Section 2 gives a conceptional discussion on the mechanism how natural disasters propagate through the supply chain. Further, a short review of the recent literature, which analyse the role of input‐output linkages as a propagation mechanism of idiosyncratic productivity shocks, is given. Section 3 presents the empirical framework. In section 4 the data on supply chain vulnerability as well as the data on extreme weather events is introduced. Section 5 discusses the main results for the impact of supply chain vulnerability and present the climate change predictions. Finally, section 6 concludes.

2.2. Related literature and conceptual discussion

Natural disasters affect the productivity level of firms in a sector in different ways. They destroy tangible assets such as buildings and equipment as well as inventories, e.g., intermediate products and raw materials. Natural disasters can cause significant short‐term




population changes due to reallocation and by this directly affect the available labor supply (e.g., Belasen & Polachek 2008, Kirchberger 2017). Finally, firms can be affected by demand effects due to a reallocation and loss of existing customers.25 Depending on the vulnerability of firms in regard with these channels, natural disasters can be fatal to the firms and result in them being forced to close down and exit the market. Based on a European sample of firms Leiter, Oberhofer & Raschky (2009) find that firms' employment growth and accumulation of physical capital were significantly higher in regions that had experienced a major flood event. In a recent paper, Cole, Elliott, Okubo & Strobl (2019) use damage information of the 1995 Kobe earthquake and geo‐coded plant‐location data for Japan to show that a large idiosyncratic shock significantly reduces the number of firms in an affected market. Similarly, Basker & Miranda (2018) analysing the impact of Hurricane Katrina's damage to firms at the Mississippi coast in 2005, show that especially small and less productive firms exit the market in the post‐disaster period. In the case a firm is directly affected by a natural disaster, its productivity shock may be propagated through the production network to its customers as well as suppliers and, thereby, indirectly affect firms beyond the disaster region. To conceptualize this relationship, we build on the model of Carvalho et al. (2016), who examine the propagation of disaster shocks in the supply chain. In this model, a negative disaster shock to firm j can impact firms downstream in the production network, i.e., customers of firm j, and upstream firms, i.e., suppliers of firm j. Downstream firms are affected via two channels. First, a disaster shock to firm j decreases its productivity, which increases its price. Downstream firms buying its product have, therefore, to scale back production, which leads to a smaller output. Second, due to the price increase of firm j the downstream firm can substitute the affected input with labor, which, depending on the size and sign of the substitution elasticity leads to a further production decrease of the downstream firm. The effect on upstream firms ‐ the suppliers to firm j ‐ depends on whether labor and the affected inputs are gross substitutes or complements. The mechanism of the upstream propagation comes from the price change of firm j. As prices increases downstream firms buy less products, which leads the firm j to reduce its own input demand. Finally, the propagation effect decays in distance, i.e., the further away the downstream or upstream firm is from firm j in the supply chain, the smaller is the potential indirect disaster impact. This comes from the fact that the importance of a firm in an input‐output relationship decreases the more intermediate steps are between these two firms. Recently, a couple of studies came out, which analysed the role of input‐output linkages as a propagation mechanism of idiosyncratic productivity shocks.26 Carvalho et al. (2016), using an extensive dataset of supplier‐customer relationships of Japanese companies, show that the 2011 earthquake in Japan had significant negative impacts on the output of firms downstream as well as upstream of the affected firms, with a larger negative effect for the

25 A change in demand can change the productivity level of firms in case firms face increasing economies of scale. 26 This literature can in turn be placed in the larger strand of papers dealing with the microeconomic origins of macroeconomic fluctuations (e.g., Acemoglu et al. 2012, Acemoglu, Ozdaglar & Tahbaz‐Salehi 2017). For a recent overview see Carvalho (2014).




downstream firms. Barrot & Sauvagnat (2016) based their research on US data and find that a shock to suppliers propagate within the country and leads to substantial output losses at their direct customers. In an earlier cross‐country study Puzzello & Raschky (2014) find on a sectoral level a significant negative effect of supply chain shocks on a sector's export value in that year. More recently, Boehm et al. (2019) base their analysis on between‐country transmission of shocks using a database of American affiliates of Japanese multinationals. They show large output reductions in these companies compared to not‐affiliated companies in the US in the months following the 2011 earthquake in Japan. Finally, in a recent working paper Kashiwagi, Todo & Matous (2018) analyze the impact of hurricane sandy on the output of 110.000 major firms worldwide. They find large propagation effects between firms within the country, but do not find significant impacts on output between firms across countries. They argue, that internationalized firms can more easily substitute for their suppliers and customers and are, therefore, able to mitigate the propagation of shocks.27

2.3. Empirical Implementation

To answer our research questions, we specify a generic model that accounts for the impact of a natural disaster transmitted over the supply chain on a sector's exports as follows: 𝒀 𝒉𝒊𝒕 𝜷 𝟎 𝜷 𝟏 𝑺𝑪𝑺 𝒉𝒊𝒕

𝜷 𝟐 𝑿 𝒉𝒊𝒕 𝝀 𝒉𝒕 𝜽 𝒉𝒊 𝜻 𝒊𝒕 𝜺 𝒉𝒊𝒕 ( 1)

where 𝑌 is the export value of sector, ℎ, in country, 𝑖, and year, 𝑡. 𝑆𝐶𝑆 , is our parameter of interest and is the measure of the degree of a natural disaster shock transmitted over the supply chain to sector, ℎ, country, 𝑖, in year, 𝑡. Based on our discussion on supply chain propagation of disaster shocks in Section 2, we expect 𝛽 to be statistically significant. A non‐zero coefficient estimate of 𝑆𝐶𝑆 implies that a sector's export performance is affected by disasters happening abroad and being transmitted over the supply chain. We expect 𝛽 to be negative as supply chain shocks reduce the average productivity of the firms in an affected sector. 𝑋 is a vector of all sector characteristics in country 𝑖, which affect a sector's export intensity and which vary over time, e.g., disasters abroad, which serve as a proxy for the degree of foreign competition and could affect a sector's export performance but is not covered by our supply chain shock measure. 𝜆 , is a sector‐year dummy, which covers all factors that vary over sectors in a specific year and influence the export activity of a sector, e.g., business cycles. 𝜃 is a country sector dummy, which controls for all sector specific factors in a country, which are invariant over time, e.g., the capital intensity of specific sectors in a country, which makes them relatively inelastic to adopt to short‐term demand changes, or the degree of returns to scale. 𝜁 is a country‐year dummy, which captures all country specific factors, which change over time, and have an effect on a sector's export

27 Another strand of literature uses simulation analysis based on CGE or agent‐based modeling to asses the impact of supply chain shocks on a firm's output. See, for example, Otto, Willner, Wenz, Frieler & Levermann (2017), Inoue & Todo (2019a) and Inoue & Todo (2019b).




performance, e.g., the occurrence of a domestic disaster or the financial crisis in the year 2008. Finally, the error term, 𝜀 , is assumed to be i.i.d. and heteroscedasticity robust. Based on our fixed effect structure, we should be able to disentangle the impact of an exogenous short‐term supply chain shock on the exporters' productivity and the resulting export decision from other confounding factors. Our identification of an exogenous short‐term supply chain shock comes by comparing the export performance of different sectors, which are differently exposed to supply chain shocks as they are differently embedded in the global supply chain network, in the same country as well as with the export performance of the same sector in different countries. Finally, the coefficient estimate of 𝑆𝐶𝑆 may be biased, if important variables are omitted, which are correlated with our supply chain shock measure and influence the export performance of a sector. In general, we are able to control for most of confounding variation through our very strict fixed effect structure. However, for instance, the size and experience of exporters could influence the way how supply chain shocks affect the export performance of a sector. Large and experienced exporters may have a more efficient management of their supply chains, which makes them more able to better react to supply chain shocks. In an extension of our model as specified in equation (1) we are controlling for exporter size and experience.

2.4. Data and summary statistics

We utilize an unbalanced panel data set of 12 sectors from 174 countries around the world for the year 1990 to 2015.28 Information on a country's worldwide export flows stems from the World integrated Trade Solutions (WITS) data base, which itself relies on the UN's commodity trade statistic database.29 Data on a country's domestic and international input‐output structures is taken from the EORA global supply chain database. The EORA global supply chain database is based on the supply‐use tables from the full EORA multi‐regional input‐output tables, which have been converted to symmetric product‐by‐product input‐output tables using the industry technology assumption and aggregated to a common 26‐sector classification.30 To establish concordance between the 26 sectors in the EORA database and the UN's commodity trade statistic database we rely on Engel (2016). Table A2 in the appendix gives information on the sectors covered in our final dataset and their concordance with the ISIC Rev.3 classification in our export dataset. Finally, information to construct our proxy for disaster occurrence comes from five global climate models (GCMs) of the Coupled Model Intercomparison Project phase 5 ensemble (CMIP5, Taylor, Stouffer & Meehl 2012), which have been bias corrected within in the Inter‐Sectoral Impact Model Intercomparison Project 2A (ISIMIP2A, Hempel, Frieler, Warszawski, Schewe & Piontek 2013). Additionally, we use data from the emergency event database provided by the Center

28 For a detail list of countries and sectors covered see Table A1 and Table A2 in the appendix. 29 http://wits.worldbank.org/wits/ 30 Please see Lenzen et al. (2012), Lenzen et al. (2013) and https://worldmrio.com/eora26/ for a more detailed description.




of Research on the Epidemiology of Disasters at the University of Louvain.31 The emergency event database captures disaster events, for which at least one of the following criteria has been realized: (1) ten or more people died due to the disaster; (2) at least 100 people were affected; (3) a state of emergency has been declared; or (4) a call for international assistance has been made. For each disaster the type, information on the number of fatalities, the total number of people affected, and the total amount of estimated direct damages in US dollars is reported in the database. Table 2.1 Summary statistics

Table 2.1 depicts the summary statistics. On average the value of a country's sectoral exports amounts to 5,761,478 US$ per year and is led by China, which has an average export value above 1 billion US$ in the "Electrical and Machinery" sector for the year 2013 until 2015. Our variable of interest, a sectoral productivity shock in a country due to disruptions transmitted over the supply chain, lies between 0 and 1 with an average value of 0.399 for a country, sector and year. The proxy for foreign competition, measured as output weighted disasters abroad per sector, country and year, varies between 0.061 and 0.824, where the maximum of a foreign competition shock is in the "Electricity, Gas and Water" sector in the year 1997. Finally, a sector's size and export experience might affect its ability to cope with a productivity shock transmitted over the supply chain. The gross output of a given sector, which serves as a measure of sectoral size, is led by China's "Electrical and Machinery" sector. Our measure of export experience, a country's sector exports relative to the world exports in the previous year, is on average 1% and lead by France's "Electricity, Gas and Water" sector, which had an export share of around 54% in the year 1996. In Table 2.2 the number of observations per world region and sectoral group are shown. The manufacturing sector is in all world regions the sector where most countries are active each year, which is then followed by the agricultural sector. The energy sector is the least traded sector. All in all, Table 2.2 makes us confident that we have enough observations per

31 https://www.emdat.be/




country, sector and year to identify the impact of supply chain shocks on a country sector's export performance. Table 2.2 Observations by region and sector

2.4.1. Supply chain connectivity

From EORA's 26 sector multi‐regional input‐output tables, we use values for the intermediate good sales between each sector and country, which contain both inputs sourced domestically and inputs sourced abroad. We, then, divide the intermediate good sales matrix by the total output of each sector. This gives us a so‐called technical coefficient matrix, A, where each column of this matrix represents an industrial recipe used to produce a single industry's good. Finally, the total, i.e., direct and indirect, amount of inputs used in one sector's production from all other sectors, is given by the Leontief inverse, which is calculated as

𝐿 𝐼 𝐴 , and summarizes the network effects generated when final output changes. Each element of the Leontief inverse, 𝑙 , summarizes all direct and indirect effects created in sector 𝑖 to supply a single unit of final demand for sector 𝑗 in year 𝑡. Using this framework, we are able to classify each country's sector according to its degree of spatial connectivity. Thereby, production of sector 𝑗 can have two effects on the other sectors in an economy. If sector 𝑗 increases its output, i.e., the demand will be increased from sector 𝑗 for goods produced in other sectors 𝑗 used as inputs to production. The degree of interconnection of sector 𝑗 with those upstream sectors 𝑗 from which it derives its inputs is called "backward linkage". Formally, it is given as

𝐵𝐿1

1𝑛 ∗ 𝑚 1

,




where 𝑛 being the number of sectors and 𝑚 being the number of countries. Increased output of sector 𝑗 means that more goods produced by sector 𝑗 are available as inputs to production for all downstream sectors. The term "forward linkages" measures the degree of interconnection of a sector with those sectors to which it sells its outputs.32 It is given as

𝐹𝐿𝐺 1

1𝑛 ∗ 𝑚 1 𝐺 1

,

where 𝐺 stands for the Goshian inverse, which is the transposed Leontief inverse and pictures a supply‐side view of the input‐output relationships. Both measures, 𝐵𝐿 and

𝐹𝐿 captures direct and indirect effects as well as intra‐ and interregional linkages. The

larger 𝐵𝐿 and 𝐹𝐿 the stronger is a sectors degree of spatial interconnectivity. In

Figure 1, we plot these measures of spatial linkages for each sector in our sample for six different points in time. The y‐axis depicts the degree of backward linkages and the x‐axis the degree of forward linkages. Sectors, which are below one in both measures, are generally seen as independent and not strongly connected to other sectors. Sectors with a forward linkage measure, which is larger than one, can be classified as sectors, which are dependent on interindustry demand. Whereas, sectors with a backward linkage measure, which is larger than one can be classified as sectors, which are dependent on interindustry supply. Finally, sectors with both measures larger than one are seen as generally dependent and strongly connected to other sectors. Sectors in the European Union are marked red and the depicted number corresponds to the sector number as given in Table A2.

32 For an excellent introduction into input‐output analysis see Miller & Blair (2009).




Figure 2.1 Sectoral forward and backward linkages over time

Overall, it can be seen that the majority of sectors are not strongly interconnected with other sectors domestically as well as internationally. However, overtime the number of sectors, which are depended on interindustry demand, supply or both is increasing over time. In case a natural disaster is occurring, the spillover effects over the supply chain will be much stronger if these strongly interconnected sectors are affected.

2.4.2. Natural disaster data

To construct our natural disaster index, we use daily 2‐meter air temperature and precipitation rate measures of the WATCH Forcing Data ERA‐Interim (WFDEI, Weedon, Balsamo, Bellouin, Gomes, Best & Viterbo 2014) provided on a 0.5° x 0.5° regular latitude longitude grid. Climate extreme indices have been calculated using the ClimPACT2




package.33 Only indices, which provide continuous information, in contrast to absolute indices based on day counts, have been used. An overview of used indices per disaster type is given in Table 2.3, a description of the single indices can be found in the Appendix (Table A3 and Table A4). Table 2.3 Disaster types and associated indices

There is a large spatio‐temporal scale gap between our different data sources, where input‐output connectivity is measured at country‐sector‐year level and WFDEI data is based on daily temperature and precipitation values on a 0.5° x 0.5° regular latitude longitude grid. In the process constructing the natural disaster measure, we start using daily temperature and precipitation values of WFDEI to calculate monthly indices on the gridpoint scale. In a next step, all grid points within a country's borders have been aggregated in three ways, the unweighted mean, the minimum/maximum value34, and a weighted mean based on the within country spatial distribution of population (GPWv3 2005). In a next step, we calculated for every country, i, and every month, m, standardized anomalies from the long term monthly mean between the years, t, from 1990 to 2015, which is given as

γ𝑋 𝑋

σ,

where X corresponds to the climate extreme index of interest. For some climate extreme indices, we introduced additional conditions on the monthly values in order to prevent false detections in the subsequent analysis. This means that for all coldwave indices mean minimum temperature had to be below 0 °C and that for all heatwave indices mean 33 This is a freely available R software package (https://github.com/ARCCSS‐extremes/climpact2), which uses climdex.pcic and climdex.pcic.ncdf. It was developed by the Pacific Climate Impacts Consortium and its development was overseen by the World Meteorological Organisation's Expert Team on Sector‐specific Climate Indices (ET‐SCI). 34 This is depending on which distributions tail we were interested in. For instance, minima values of mean minimum temperature were considered for coldwaves, whereas maxima values of mean maximum temperature were used for heatwaves.




maximum temperature had to be above 30 °C. For all flooding indices the maximum 1‐day precipitation rate had to be at least 10 mm day‐1. And for all drought indices the respective SPI/SPEI had to be below ‐0.1. By this procedure we generated Nindices x Naggregations datasets for every disaster index. In a next step, we generated a disaster time‐series by selecting data values above a certain percentile threshold (p90, p95, p97.5 and p99). Finally, to select the indices, which best predict a potential disaster in a country, we calculated a score measuring the relative success rates in predicting disasters reported in the EM‐Dat database published by the Centre for Research on the Epidemiology of Disasters (EM‐DAT 2019). 35 Table 2.4 Summary statistics ‐ Disasters

Table 2.4 depicts for each disaster type the number of affected sectors for each country and year. On average 10 percent of all observations are either affected by a drought, heatwave, coldwave or flash‐flood event. Only, riverine‐floods with 1 percent on average happen less often. In total around 38 percent of our observations are on average affected by a disaster event. In the right panel of Table 2.4 the predicted natural disasters per disaster type and period are shown. The occurrence of nearly all disaster types increases in future, with the strongest increase in droughts, heatwaves and riverine floods.

2.4.3. Sectoral supply chain shocks

In a further step, we now combine the information of interindustry linkages and the occurrence of natural disasters (see Figure 2).

35 Table 3 in the appendix shows the best predictive indices and scores for the single disaster types. We do not use a disaster measure directly based on events reported in the EM‐Dat database out of following reasons. First, as the focus of this analysis is based to predict future ‐ climate change induced – impacts of disasters on a country’s export performance, we would not be able to project future disaster based on EM‐dat events, as the information of these events are mainly based on insurance claim reporting. Second, the EM‐dat database itself recently came under some critique as the probability and quality of reporting is not independent of a country's level of economic development and is, therefore, susceptible to potential endogeneity issues.




Figure 2.2 Sectoral forward and backward linkages and disaster shocks

As before, the y‐axis depicts the degree of backward linkages and the x‐axis the degree of forward linkages. Sectors, which are below one in both measures are generally seen as independent and not strongly connected to other sectors. Sectors, which are marked red, are sectors in countries a natural disaster, as determined by our natural disaster indices and as laid out in Section 4.2, has happened. In all years, sectors, which are strongly interdependent in the supply chain, i.e., sectors with a value above one in the forward‐ and backward linkage measures, are hit by natural disasters. These sectors, due to their spatial linkages, have a large potential to transmit natural disaster shocks to many other countries and sectors over the supply chain. Finally, the number of independent sectors hit by a natural disaster is increasing over time.




2.4.3.1. Supply chain shock index

To construct our measure of supply chain shock, we combine each element of our supply chain connectivity matrix, 𝐿, with our measure of natural disasters as laid out in Section 4.2. For each sector, ℎ, in country, 𝑖, the proportion of inputs potentially affected by natural disasters at year, 𝑡, is given as

𝑺𝑪𝐒𝐡𝐢𝐭 ∑ ∑𝐋𝐣𝐢𝐭 𝐠,𝐡

𝐋𝐢𝐭 𝐡𝐆𝐠 𝟏

𝐍𝐣 𝐢 𝐍𝐃𝐣𝐭, ( 2)

where 𝑔 1,2, … ,𝐺 is the domestic or imported input used in the production of country 𝑖's good ℎ and 𝐿 ℎ is the total per unit use of inputs in country 𝑖's sector ℎ at time 𝑡. Our disaster index, 𝑁𝐷 , is one, whenever country 𝑗 is hit by a natural disaster, as determined by

one of our disaster indices.36 Equation 2 shows that the level of shock a country's sector receives depends on two factors: First, how strongly the sector is connected to each other sector at home and abroad. The stronger the interdependence the larger the transmission of a disturbance. And second, how many of its trading partners are hit by a natural disaster. This means that a natural disaster, which hits only one sector, but this sector is strongly connected to sector ℎ in country 𝑖, can have the same impact on sector ℎ, as many sectors, which are hit by natural disasters, but which are not strongly connected to sector ℎ.

Figure 2.3 plots the distribution of the SCS index for different large regions in the world. Overall, the SCS measure of supply chain shocks features a bimodal distribution, with more of its density concentrated at the lower end of the support. The reason for this distribution lies in the general structure of intra‐ and intercountry supply chain linkage. In general, domestic inputs form the largest share in total inputs of a production process. Therefore, a disaster happening at home has not only a large direct impact on the sector itself, but also indirectly over the domestic supply chain (Kashiwagi et al. 2018). This leads to a larger value in the SCS index and explains part of the higher concentration of values at the higher end of the support. Whereas, inputs from abroad form a smaller share of total inputs in the production process and, therefore, disasters happening abroad will have a smaller value in the SCS measures and is part of the higher concentration of values at the lower end of the support. 37

36 Potential disaster are heatwave, coldwave, drought, springflood and riverineflood (see Section 4.2). 37 For more discussion on this distributional feature see Puzzello & Raschky (2014).




Figure 2.3 Distribution of supply chain shock index for different world regions

The bimodal distribution can be seen for all large regions in the world. However, in the European Union the mass of values in the lower and medium end of support is higher than for other regions. This can be explained that due to the single market and stronger export orientation countries in the European Union are more strongly interconnected, which means that these countries are more often receiving disasters happening abroad. Whereas, sectors in North America are receiving a large degree of inputs domestically, which explains the smaller mass of observation at the lower end of the support compared to the mass of observations at the higher end of the support.

In Figure 2.4 the exposure of 4 different sector groups for countries in the European Union compared to sectors in all other regions is depicted. The left panel of Figure 4 shows the distribution of the SCS index for sectors in the European Union. In the right panel of Figure 4 the SCS measure for sectors in all other regions is plotted. Interestingly, the manufacturing sector is the sector, which is particularly exposed to shocks in the supply chain happening abroad. This can be explained, that intermediate goods for the manufacturing sector are traded internationally and, in the last years in particular Asian countries have significantly gained in the market share of intermediate goods production, e.g., the production of inputs for consumer electronics in China. Regarding regional differences, it can be seen that the manufacturing sector in the European Union is stronger exposed to international supply chain shocks compared to the manufacturing sector in the other regions.




Figure 2.4 Distribution of supply chain shock index for different sectors for the EU and other regions

Finally, in Figure 2.5 the change in the distribution of the SCS index over time is plotted. Overall, the bimodal distribution of the SCS measure can be found over the whole time frame of our sample with an increase in the values at the higher end of the support, which is explained by the increasing degree of input outsourcing over time and the increase in the occurrence of natural disasters.

Figure 2.5 Distribution of supply chain shock index over time

2.5. The results

Table 5 presents the results, which are based on specification 1. Model (1) in Table 5 shows the outcome based on country‐year and sector‐year fixed effects. In all the other models (2‐6) in Table 5 the results are based on the full fixed effect structure, including country‐year dummies, country‐sector dummies and sector‐year dummies. Productivity shocks transmitted over the supply chain significantly reduce a sector's export performance. The country‐sector specific control variables are as expected. A decrease in foreign competition,

which is depicted with a large 𝑁𝐿𝐷 value, increases the export value. Larger sectors, measured by the gross output in that year, tend to export more. With including our full fixed‐effect structure, as shown in column (2) of Table 5, the impact of productivity shocks transmitted over the supply chain becomes slightly smaller but remains statistically




significant. A one standard deviation increase in the SCS measure decreases a sector's exports by around 11 percent. This is our preferred specification. Table 2.5 Estimation results ‐ Supply chain shocks and a country's exports

To ensure our estimated coefficients are not biased due to the omission of variables, which are correlated with the SCS measure and could influence the export performance of a country, in column (3) we include a sector's export share in the previous year as a proxy for experience. More experienced exporters may manage the supply chain more efficiently and, therefore, are able to switch to alternative suppliers in case the supply chain is hit by a natural disaster. Our parameter of interest, SCS$_{hit}$, remains robust in size and significance. In column (4) we use a stricter definition of the standard errors and cluster them at country‐sector level. In column (5) we re‐estimate specification 1 using a pseudo‐poisson maximum likelihood estimator as suggested by Silva & Tenreyro (2006), which is able to deal with 0 in the export variable and allows a more flexible treatment of the standard errors in regard with heteroscedasticity. Our estimate of the 𝑆𝐶𝑆 impact stays robust. Finally, in column (6) we use an alternative definition of natural disasters, which takes the intensity of the natural disaster events into account.38 Using a disaster intensity measure we find that a one standard deviation increase in the SCS measure decreases a sector's value of exports by around 13 percent.

38 This means the 𝑁𝐷 in equation 2 is the sum of the standard deviations for each index and country. In

contrast, in our baseline definition 𝑁𝐷 is just defined by dummy, which is 1 whenever a disaster occurred in a

country.




2.5.1. Sectoral decomposition

The effect of productivity shocks due to supply chain disruptions may be different for different sectors, e.g., the number and origin of inputs in the manufacturing sector may be different compared to the agricultural sector.

Table 2.6 Estimation results ‐ Sectoral decomposition

Further, as we have seen in Section 4.3.1 the distribution of the SCS measure varies for different sector groups, and points to the potential internationalization of input sourcing in the manufacturing sector. Table 2.6 presents the results, when we re‐estimate specification 1 for four large sector groups, i.e., the agricultural, manufacturing, energy and mining sector. Our estimates show that the negative effect of a supply chain shock on a sector's export value is mainly driven by the agriculture and, in particular, manufacturing sector. This is due to the different composition and number of inputs used in the production of the good being exported. The larger the number of inputs the higher the potential impact of supply chain shocks.

2.5.2. Projections

We next turn to the impact of supply chain shocks on a sector's export performance taking future exposure to natural disasters due to climate change into account. To this end, we combine the estimated negative relationship between supply chain shocks and a sector's export performance with predictions about future climate change.




We estimated future natural disasters, i.e., coldwaves, heatwaves, droughts as well as riverine flooding and flash floods, using temperature and precipitation data of five global circulation models (GFDL‐ESM2M, HadGEM2‐ES, NorESM1‐M, IPSL‐CM5A‐LR and MIROC‐ESM‐CHEM) of the CMIP5 ensemble (Taylor et al. 2012) bias corrected within the ISIMIP2A project (data denoted ISIe in Hempel et al. 2013) for the two emission scenarios39 following the approach described in Subsection 4.2. Populations dynamics on the regular ISIMIP grid (0.5° x 0.5, lat x lon) have been accounted for using GPWv3 (2005) for the observable past, and data from Jones & O'Neill (2016) for the future under the shared socioeconomic pathway 2 (SSP2). The percentile limits have been estimated over the period 1990‐‐2015, using data of the CMIP5 models historic experiment until 2005 and the respective RCPs from 2006. Subsequently, disasters for future time periods have been obtained for monthly country values exceeding the respective percentile threshold. The monthly country disasters have been collapsed for each year and country, which results in the final yearly time‐series. To predict supply chain impacts of climate change on a sector's export performance, we use the regression coefficient estimate based on specification 2 and the SCS measures based on natural disaster predictions for each global circulation models and the two emission pathways. We then calculate the predicted difference in the SCS measure for three future periods (2020‐2040; 2041‐2070; 2071‐2100) and the baseline period (1990‐2015) for each country and sector in our sample and multiply it by the estimated regression coefficient. It has to be noted that these calculations are based on strong assumptions. While, we allow for changes in the occurrence in natural disasters due to climate change and account for population dynamics, in all predictions we keep all other determinants, which could affect a sectors export performance fixed to mean values over our sample period from 1990 to 2015. The input‐output relationships to construct our SCS measure are based on the 2015 input‐output network. Therefore, the predictions can be seen as upper limits of climate change impacts, in particular for the time periods further in future, if only very limited adaptation process takes place.40 The predicted impacts of climate change for the three future time periods (2020‐2040; 2041‐2070; 2071‐2100) are shown in Table 7. All results have to be interpreted as mean annual change to the baseline period (1990‐2015). Depending on the global circulation model and the representative concentration pathway additional supply chain impacts of climate change on a sector's export performance will be in the range of ‐ 8% to ‐11% in the short‐term

39 The emission scenarios used in this study are the representative concentration pathways 2.6 (RCP2.6) and RCP4.5. This follows the scenario choices discussed in deliverable D 1.5 of the H2020 COACCH project (Hof et al. 2018), where RCP 2.6 – SSP2 and RCP 4.5 ‐ SSP2 are decided as core scenarios. These two scenarios are based on a low‐ and medium level emission development assumption. Other scenarios as discussed in deliverable D 1.5would also be possible and would lead to potentially stronger negative impacts of supply chain shocks due to an increase in natural disasters in future. As our projection exercise is based on strong assumptions regarding potential adaptation processes, we base this analysis only on these two representative concentration pathways to stay conservative in our predicted impacts. 40 To account for potential adaptation measures, long‐differences could be estimated. For a discussion on this issue see Dell, Jones & Olken (2014). However, as the time frame of our dataset is limited such an approach is unfortunately not feasible.




period (2020‐2040), ‐8% to ‐15% in the medium‐term period (2041‐2070) and ‐8% to ‐16% in the long‐term period (2071‐2100) with strong differences for the single country and sector, which can be seen at the minima and maxima values. The intuition behind the large differences for the single countries and sectors is the interdependence of a sector in the global production network as well as the exposure of its trading partners to natural disasters in the future. To account for not only the occurrence but also the intensity of future natural disaster, in Table 8 we present predicted impacts of climate change for the three future time periods (2020‐2040; 2041‐2070; 2071‐2100) using the intensity of natural disasters in the predictions. The supply chain impacts of climate change on a sector's export performance are large and will be in the range of ‐19% to ‐29% in the short‐term period (2020‐2040), ‐19% to ‐46% in the medium‐term period (2041‐2070) and ‐19% to ‐43% in the long‐term period (2071‐2100). Again, very large differences between the single country and sector can be observed.

2.5.2.1. Country‐specific predicted impacts

As indicated in Table 2.7 there is a considerable heterogeneity in the strength of a supply chain shock on a sector's export value. The value of the SCS measure is determined by two factors ‐ by the degree of connectivity between two sectors and if a sector is hit by a natural disaster. Therefore, the combination of a sector's trading partners in the supply chain and the change in the occurrence of natural disaster due to climate change determine the mean annual impact of our SCS measure for the future time periods considered in our analysis. Figure 2.6 and Figure 2.7 show the additional change in the average annual export value of a country for the RCP2.6 and the RCP4.5 scenarios, respectively. The predictions are based on the mean value of all five global circulation models and on the sector‐level predictions laid out before. Thus, the country‐specific predictions reported here represents the effect of country‐level heterogeneity in the




Table 2.7 Projected SCS impacts




Table 2.8 Projected SCS impacts ‐ Disaster Intensities




predicted changes in the natural disaster occurrence and its global production network. It does not account for possible heterogeneity in the historical relationship between a supply chain shock and average export value in a country. Plotting the average annual impacts over time gives an interesting picture for both RCPs. All countries' sectoral exports are negatively affected by climate change. However, the strongest effects can be found for countries in the tropics and sub‐tropics as this is the region, which will observe strong adverse weather changes due to climate change, which are then transmitted over interregional supply chain connections.

Table 2.9 Projected SCS impacts ‐ Country‐specific impact

Table 2.9 lists the five countries with the strongest predicted impact, and the five countries with the weakest predicted impact for the three periods and 2 representative concentration pathways considered in the study. In both representative concentration pathways countries less strongly hit by supply chain shocks in future, i.e., countries with a reduction in average export value between zero percent and 9 percent, are countries with a more localized production network, e.g., Afghanistan, or




Figure 2.6 Projected export change (mean over 5 GCMs) ‐ RCP 2.6




Figure 2.7 Projected export change (mean over all GCMs) ‐ RCP 4.5

countries with a regionally concentrated production network in regions less affected by climate change, e.g., Slovenia, Island and Denmark. Whereas countries, which are situated in regions more strongly affected by climate change also have a stronger SCS impact on mean annual exports in future, which is a reduction between seventeen and twenty‐four percent. These countries are, for example, Congo, Libya and Eritrea. Finally, Table 2.10 lists the impact of supply chain shocks on European Union countries for the two representative emission pathways and the three future time periods considered in




this study. All in all, countries in the European Union are differently affected by disaster induced shocks transmitted over the supply chain. Whereas, Austria, Slovenia and Hungary are less affected by climate induced shocks to the supply chain, countries like France, Spain, Netherlands and Portugal are strongly affected by these shocks. The reasons of these differences lie in the regional exposure to climate change induced disaster shocks and the composition of the global production network. Austria, for example, has an intensive supply chain connectivity with Germany, a country which will not be as strongly affect by climate change induced disaster shocks as other southern European countries in the near future. However, these difference in the supply chain shocks will decrease the further the projections are in future.

Table 2.10 Projected SCS impacts ‐ Country‐specific impact (EU countries)

2.5.2.2. Sector‐specific projected SCS impacts

The impacts of a climate change induced increase in supply chain shocks may differ across sectors because of different exposure to the global production network and the geographical distribution of climate change shocks. To explore heterogeneity across sectors Table 2.11 shows the mean, the minimum and the maximum value for each representative emission pathway for the twelve sectors considered in this study.




Table 2.10 Projected SCS impacts ‐ Sectors

It has to be noted that the outcomes in Table 2.11 are based on the regression coefficient of our 𝑆𝐶𝑆 measure for the whole sample, i.e., it represents a mean annual impact of a supply chain shock over all sectors in the baseline period. The combination of the input‐output connections and natural disaster predictions are based on country‐sector level. Alternatively, specification 1 could be estimated for each sector separately, which would allow to consider heterogenous responses of each sector to supply chain shocks. However, given our stringent fixed effect structure unfortunately this is not possible for the individual sectors in our sample. That is a limitation and in interpreting the results in Table 10 the uniform sectoral response to shocks has to be considered. Thus, the sector‐specific predictions represent the effect of country‐level heterogeneity in the predicted changes in the natural disaster occurrence and the sectors role in the global production network.

On average the annual export value per sector over the whole time period, i.e., from 1920 to 2100, is additionally reduced by around 11 percent compared to the baseline period for the RCP 2.6 and thirteen percent for RCP 4.5 scenario. Within each sector the heterogeneity is large and economic significant, meaning that, for example, in the "Wood and Paper" sector Eritrea will have an additional average reduction of around twenty‐two percent. Whereas, Afghanistan will have in the same sector no additional or even a slight decrease in the average negative impact for the same time period.

2.6. Conclusions

Today, the production of a final good in a country is based on numerous input‐output interlinkages domestically as well as increasingly internationally. Disturbances in one country




can be propagated over the supply chain leading indirectly to a change in other countries' macroeconomic outcomes. This report addresses the impact of one potential shock, which can be transmitted over the supply chain ‐ namely natural disasters. Combining a large dataset of input‐output connections with a natural disaster dataset, we find that a one standard deviation increase in supply chain shocks decreases a sector's export by 11 percent. Further, we show that this negative effect is mainly driven by the manufacturing and agricultural sector. Finally, predicting future supply chain shocks we find a potentially strong impact of climate change on the extend of the negative effects of supply chain shocks on a sector's export value. Although, the impact depends on the global circulation model in most of the cases it reduces exports and it is considerably large. Finally, the impact of climate change is heterogenous between the countries and depends on the extend of a sectors global production network and the strength of increase in natural disaster in that region. Our results suggest, that it is countries in the tropics and subtropics, which will be particularly negative affected by these shocks in future. These findings are economic important. We show that countries, which are regularly hit by natural disasters are also strongly interdependent in global production networks. Regarding domestic disasters, policy makers need to the take the prevalent risk of supply chain disruptions due to natural disasters into account. At national level, pro‐active measures, like zoning and building standards, could be implemented. Public information campaigns on disaster risk could incentivize private adaptation measures and insurance uptake. After a disaster has happened, financial disaster relief aid and solid financial institutions could speed up the disaster recovery period and decrease the length of the supply chain disruption. For disaster happening abroad, information campaigns could made companies aware of the potential risk of supply chain disruptions, which could incentivize companies, for example, to increase their level of geographical diversification in their global production network or intensify the use of storage facilities. At the global level one major contributor to natural disaster occurrence is climate change. International coordinated policy, which reduces the amount of greenhouse gases, will also reduce the future risk of natural disasters and therefore, will decrease the potential negative impact of future supply chain disruptions.




2.7. References

Acemoglu, D., Carvalho, V. M., Ozdaglar, A. & Tahbaz‐Salehi, A. (2012), The network origins of aggregate fluctuations, Econometrica 80(5): 1977‐2016.

Acemoglu, D., Ozdaglar, A. & Tahbaz‐Salehi, A. (2017), Microeconomic origins of macroeconomic tail risks, American Economic Review 107(1): 54‐108.

Barrot, J.‐N. & Sauvagnat, J. (2016), Input Specificity and the Propagation of Idiosyncratic Shocks in Production Networks, The Quarterly Journal of Economics 131(3): 1543‐1592.

Basker, E. & Miranda, J. (2018), Taken by storm: business financing and survival in the aftermath of hurricane katrina, Journal of Economic Geography 18(6): 1285‐1313.

Belasen, A. R. & Polachek, S. W. (2008), How hurricanes affect wages and employment in local labor markets, American Economic Review 98(2): 49‐53.

Boehm, C. E., Flaaen, A. & Pandalai‐Nayar, N. (2019), Input linkages and the transmission of shocks: Firm‐level evidence from the 2011 tohoku earthquake, The Review of Economics and Statistics 101(1): 60‐75.

Botzen, W. J. W., Deschenes, O. & Sanders, M. (2019), The economic impacts of natural disasters: A review of models and empirical studies, Review of Environmental Economics and Policy 13(2): 167‐188.

Burke, M. & Emerick, K. (2016), Adaptation to climate change: Evidence from US agriculture, American Economic Journal: Economic Policy 8(3): 106‐40.

Carvalho, V. M. (2014), From micro to macro via production networks, Journal of Economic Perspectives 28(4): 23‐48.

Carvalho, V. M., Nirei, M., Saito, Y. & Tahbaz‐Salehi, A. (2016), Supply chain disruptions: Evidence from the great east japan earthquake, Becker Friedman Institute for Research in Economics Working Paper No. 2017‐01(1): 1‐46, URL: http://dx.doi.org/10.2139/ssrn.2893221

Cavallo, E., Galiani, S., Noy, I. & Pantano, J. (2013), Catastrophic natural disasters and economic growth, The Review of Economics and Statistics 95(5): 1549‐1561.

Cole, M. A., Elliott, R. J. R., Okubo, T. & Strobl, E. (2019), Natural disasters and spatial heterogeneity in damages: the birth, life and death of manufacturing plants, Journal of Economic Geography 19(2): 373‐408.

CRED (2019), Natural Disasters 2018, Brussels:CRED . Dell, M., Jones, B. F. & Olken, B. A. (2012), Temperature shocks and economic growth:

Evidence from the last half century, American Economic Journal: Macroeconomics 4(3): 66‐95.

Dell, M., Jones, B. F. & Olken, B. A. (2014), What do we learn from the weather? The new climate‐economy literature, Journal of Economic Literature 52(3): 740‐98.

Deschenes, O., Greenstone, M. & Guryan, J. (2009), Climate change and birth weight, American Economic Review 99(2): 211‐17.

EL‐Hadri, H., Mirza, D. & Rabaud, I. (2018), Natural disasters and exports: New insights from a new (and an old) database, The World Economy, forthcoming.

EM‐DAT (2019), The Emergency Events Database', Universite catholique de Louvain (UCL) ‐ CRED, D. Guha‐Sapir ‐ www.emdat.be, Brussels, Belgium.




Engel, J. (2016), Analysis of value‐added trade and intermediates exports of commonwealth countries and comparator groups, International Trade Working Paper 2016/11, 1 ‐ 14, Commonwealth Secretariat, London.

Felbermayr, G. & Gröschl, J. (2013), Natural disasters and the effect of trade on income: A new panel IV approach, European Economic Review 58, 18 ‐ 30.

Felbermayr, G., Gröschl, J., Sanders, M., Schippers, V. & Steinwachs, T. (2018), Shedding light on the spatial diffusion of disasters, CESifo Working Papers (7146), 1‐71.

Gassebner, M., Keck, A. & Teh, R. (2010), Shaken, not stirred: The impact of disasters on international trade, Review of International Economics 18(2), 351‐368.

GPWv3 (2005), Gridded population of the world, version 3 (gpwv3): Population count grid. URL: http://sedac.ciesin.columbia.edu/data/set/gpw‐v3‐population‐count Graff‐Zivin, J. & Neidell, M. (2014), Temperature and the allocation of time: Implications for

climate change, Journal of Labor Economics 32(1), 1‐26. Hempel, S., Frieler, K., Warszawski, L., Schewe, J. & Piontek, F. (2013), A trend preserving

bias correction – the ISI‐MIP approach, Earth System Dynamics 4(2): 219‐236.

Hof, A., van Vuuren, D., Watkiss, P., Hunt, A. (2018), D1.5 Impact and policy scenarios co‐designed with stakeholders, Deliverable of the H2020 COACCH project.

Hsiang, S. & Jina, A. (2014), The causal effect of environmental catastrophe on long‐run economic growth: Evidence from 6,700 cyclones', NBER Working Paper (20352), 1 ‐ 69.

Hsiang, S. M., Burke, M. & Miguel, E. (2013), Quantifying the influence of climate on human conflict, Science 341(6151).

Hsiang, S. M., Meng, K. C. & Cane, M. A. (2011), Civil conflicts are associated with the global climate, Nature 476: 438‐441. Inoue, H. & Todo, Y. (2019a), Firm‐level propagation of shocks through supply‐chain

networks, Nature Sustainability 2(9): 841‐847. Inoue, H. & Todo, Y. (2019b), Propagation of negative shocks across nation‐wide firm networks, PLOS ONE 14(3): 1‐17. Jones, B. F. & Olken, B. A. (2010), Climate shocks and exports, American Economic Review

100(2), 454‐59. Jones, B. & O'Neill, B. C. (2016), Spatially explicit global population scenarios consistent with

the Shared Socioeconomic Pathways, Environmental Research Letters 11(8): 084003. Karl, T. R., Nicholls, N. & Ghazi, A. (1999), CLIVAR/GCOS/WMO Workshop on Indices and

Indicators for Climate Extremes Workshop Summary, Springer Netherlands, Dordrecht, pp. 3‐7. Kashiwagi, Y., Todo, Y. & Matous, P. (2018), International propagation of economic shocks

through global supply chains, Technical Report No.E1810, WINPEC Working Paper. Kirchberger, M. (2017), Natural disasters and labor markets, Journal of Development

Economics 125: 40 ‐ 58. Leiter, A. M., Oberhofer, H. & Raschky, P. A. (2009), Creative disasters? Flooding effects on

capital, labour and productivity within European firms, Environmental and Resource Economics 43(3): 333‐350.

Lenzen, M., Kanemoto, K., Moran, D. & Geschke, A. (2012), Mapping the structure of the world economy, Environmental Science & Technology 46(15): 8374‐8381.




Lenzen, M., Moran, D., Kanemoto, K. & Geschke, A. (2013), Building EORA: A global multi‐region input‐output database at high country and sector resolution, Economic Systems Research 25(1): 20‐49.

Liverman, D. (2016), U.S. national climate assessment gaps and research needs: Overview, the economy and the international context, Climatic Change 135(1): 173‐186.

Miller, R. E. & Blair, P. D. (2009), Input‐Output Analysis: Foundations and Extensions, 2 edn, Cambridge University Press.

Missirian, A. & Schlenker, W. (2017), Asylum applications respond to temperature fluctuations, Science 358(6370): 1610‐1614.

Mohan, P. S., Ouattara, B. & Strobl, E. (2018), Decomposing the macroeconomic effects of

natural disasters: A national income accounting perspective, Ecological Economics 146: 1 ‐ 9.

Noy, I. (2009), The macroeconomic consequences of disasters, Journal of Development Economics 88(2): 221 ‐ 231.

Oh, C. H. (2017), `How do natural and man‐made disasters affect international trade? A country‐level and industry‐level analysis', Journal of Risk Research 20(2), 195‐217. Oh, C. H. & Reuveny, R. (2010), Climatic natural disasters, political risk, and international

trade', Global Environmental Change 20(2), 243 ‐ 254. Osberghaus, D. (2019), `The effects of natural disasters and weather variations on

international trade and financial flows: a review of the empirical literature, Economics of Disasters and Climate Change, forthcoming.

Otto, C., Willner, S., Wenz, L., Frieler, K. & Levermann, A. (2017), Modeling loss propagation in the global supply network: The dynamic agent‐based model acclimate, Journal of Economic Dynamics and Control 83, 232 ‐ 269.

Puzzello, L. & Raschky, P. (2014), Global supply chains and natural disasters: implications for international trade, in Asia and Global Production Networks, Chapters, Edward Elgar Publishing, chapter 4, pp. 112‐147.

Raddatz, C. (2007), Are external shocks responsible for the instability of output in low income countries?, Journal of Development Economics 84(1), 155 ‐ 187.

Schlenker, W. & Roberts, M. J. (2009), Nonlinear temperature effects indicate severe damages to U.S. crop yields under climate change, Proceedings of the National Academy of Sciences 106(37), 15594‐15598.

Sillmann, J., Kharin, V. V., Zhang, X., Zwiers, F. W. & Bronaugh, D. (2013), Climate extremes indices in the CMIP5 multimodel ensemble: Part 1. Model evaluation in the present climate', Journal of Geophysical Research: Atmospheres 118(4), 1716‐1733.

Sillmann, J., Kharin, V. V., Zwiers, F. W., Zhang, X. & Bronaugh, D. (2013), Climate extremes indices in the CMIP5 multimodel ensemble: Part 2. Future climate projections, Journal of Geophysical Research: Atmospheres 118(6), 2473‐2493.

Silva, J. S. & Tenreyro, S. (2006), The log of gravity, The Review of Economics and Statistics 88(4), 641‐658.

Skidmore, M. & Toya, H. (2002), Do natural disasters promote long‐run growth?, Economic Inquiry 40(4), 664‐687.

Taylor, K. E., Stouffer, R. J. & Meehl, G. A. (2012), An overview of CMIP5 and the experiment design, Bulletin of the American Meteorological Society 93(4), 485‐498.




Weedon, G. P., Balsamo, G., Bellouin, N., Gomes, S., Best, M. J. & Viterbo, P. (2014), The WFDEI meteorological forcing data set: WATCH Forcing Data methodology applied to ERA‐Interim reanalysis data, Water Resources Research 50(9), 7505‐7514.




2.8. Appendix










3. Climate change and wind power in Europe

3.1. Introduction

Renewable energy sources for electricity generation relates to three of the five pillars of the Energy Union (EU) and wind energy currently accounts almost one‐fifth of the EU’s total installed power generation capacity (Eurostat, 2018) and covered 14% of EU’s electricity demand in 2018, while globally, the share of wind power mix increased to 7.5%. In order to meet the EU’s binding renewable energy target of at least 32% of final energy consumption by 203041 will require continued focus on power generation from wind. Wind energy is the second highest contributor of power generation capacity in the EU with a total installed capacity of 178.8 GW in the EU. However, wind power generation is both dependent on and vulnerable to changes in wind speed. Wind speeds and patterns across Europe are expected to change in the future with regional differences (Christensen et al., 2013).

Figure 3.1 Total power generation capacity in the EU, 2008‐2018

Future climate change will likely alter both mean and extreme wind speeds over Europe (Pryor and Barthelmie, 2010; Fischer‐Bruns et al., 2005; Stone et al., 2001), however, the changes will be heterogenous in nature (Karnauskas et al., 2017; Pryor et al., 2012; Brayshaw et al., 2011) Extreme wind speeds can have a negative impact on the longevity of wind turbines (Manwell et al., 2009) and gale‐force wind speeds reduces turbine integrity and energy output.

Near‐surface wind conditions are affected by changes in wind circulation, land cover, and changes in storm intensity (Tobin et al., 2016; Pryor and Barthelmie, 2013; Hueging et al., 2013). Studies that have investigated the impact of climate change on regional wind speeds and the wind power sector have tended to focus on wind energy potentials (Reyers et al., 2016; Tobin et al., 2016; Hueging et al., 2013) or have mostly compare historical and future wind speeds to estimate changes in energy potential. Moemken et al. (2018) use nine different global and regional climate models and find relatively minor increase in annual and winter energy outputs in Europe but significant decreases during the winter. Weber et al. (2018) find a change in wind speed distribution over Europe which may lead to reduced

41 https://ec.europa.eu/energy/en/topics/energy‐strategy‐and‐energy‐union/2030‐energy‐strategy




energy output in the future. Onba (2019) uses downscaled regional climate models for Japan and concludes that wind energy outputs will decline over southern and central Japan due to climate change.

Changes in wind patterns due to climate change can affect wind power generation through increased variability in generation, damages to wind turbines due to extreme weather events, intermittency in generation leading to increased firm backup capacity, and icing on wind turbines (Chandramowli and Felder, 2014; Pryor and Barthelmie, 2010). Results are highly uncertain and characterized by strong seasonality. Carvalho et al. (2017) is one of the few studies that examines the effect of climate change on European wind power resources systematically with different climate models. Tobin et al. (2014) assessed the potential impacts of climate change on wind generation, finding that mean energy yields will reduce by less than 5% by 2050 (2°C scenario). Other studies have focused predominantly on Northern Europe or the UK (Pryor et al., 2005; Hdidouan and Staffel, 2017).

However, the literature lacks analysis combining high‐temporal and spatial resolution data to investigate the impacts of climate change on winder energy output. This paper seeks to fill this gap by empirically analyzing the impact of changes in wind speed on wind power production using hourly data at the NUTS‐2 level in Europe. In a second step, we combine our econometric response functions with future Regional Climate Models (RCMs) under various warming scenarios to estimate the impact of future climate change on energy output in Europe. The rest of the paper is organized as follows; section 2 provides a description of the wind power generation and the climatic data, section 3 introduces the methodological and econometric frameworks, the empirical results are described in section 4, while section 5 provides the future vulnerability assessment of wind power due to climate change, and section 6 concludes.

3.2. Data

We merge high frequency (hourly) wind speed, air density, and wind energy output data at the NUTS‐2 level for Europe for 30 years to estimate historical response functions and compute impacts of future climate change.

3.3. European Meteorological derived high‐resolution renewable energy source (EMHIRES)

Our wind energy output data comes from EMHIRES, which provides renewable energy generation data for the EU‐28, Norway, Switzerland, and the non‐EU the Western Balkans countries42. The wind power data are available at the hourly level for 1986‐2015 at the country‐level along with at various levels of sub‐national aggregation (NUTS‐1 and NUTS‐2). The full load hours across Europe is rather heterogenous (Figure 2), with Sweden being the best performer by a significant margin (0.69), followed by Ireland (0.31) and Belgium being the worst performer in terms of load hours (0.12).

42 https://setis.ec.europa.eu/publications/relevant‐reports/emhires‐dataset‐part‐i‐wind‐power‐generation




Figure 3.2 Full load hours

Full load hours ‐ the ratio between the sums of the energy produced (GWh) and the maximum possible generation (installed capacity (GW)*8760h (GWh)) per country

EMHIRES combines wind power generation data from more than 21,000 wind farms (reconstructed by gap filling and statistical techniques) with meteorological data from NASA’s Modern Era Retrospective‐Analysis for Research and Applications (MERRA) and uses power to generate electrical power output. The wind power generation data are validated compared against the actual wind power generation outputs provided by the Transmission System Operators (TSOs) for 2015 as a validation measure and to correct any systematic errors. The reconstructed database contains 16,171 wind farms located in European countries, of which 85 are offshore. Finally, the database is aggregated to the country and sub‐national levels43.

Figure 3.3 Wind farms locations across Europe (left) and wind average wind load factor capacity (right)

3.3.1. Climatic data

43 See Gonzalez Aparicio (2016) for a detailed description of the methodology used to construct EMHRES wind database.




Our climatic data comes from ERA544 hourly data on single levels, a fifth‐generation atmospheric reanalysis dataset from the European Centre for Medium‐Range Weather Forecasts (ECMWF). ERA5 provides data at hourly intervals from 1979 onwards at 0.25°×0.25° spatial resolution. We extract instantaneous 10‐meter wind gust for 255 NUTS‐2 regions, mean temperature and surface pressure data are also extracted to compute air density. The final merged dataset has more than 67 million observations. The descriptive statistics of the main variables in the analysis is provided in Table 1 below.

Figure 3.4 Wind speed distribution (left) and air density distribution (right)

Table 3.1 Descriptive statistics of the main variables Variable Mean Min Max

Wind speed (m/s) 6.93 0.08 45.89 Air density (kg/m3) 1.21 0.97 1.51

Mean temperature (°C) 10.31 ‐41.74 45.72 Load factor capacity (%) 25 0.00 100

3.4. Econometric framework

We run a panel regression with location (NUTS‐2) and multiple time (year, month, and hour) fixed‐effects. The location fixed‐effects allow us to control for time‐invariant NUTS‐2‐specific heterogeneity (e.g. geographic factors) while time fixed‐effects control for unspecified exogenous influences that affect all the regions. Wind turbines require optimal condition to operate produce the highest energy output, to compute the optimal wind speed, we control for 10‐m instant wind gust and its second‐degree polynomial (Fan and Miao, 2015; Pieralli et al., 2015; Carta, 2012). We also include air density as relatively denser air exerts more pressure on the rotors resulting in higher power output Fan and Miao (2015).

𝑦 𝑓 𝐶𝑙𝑖𝑚𝑎𝑡𝑒 𝛿𝑋 𝛼 𝜖 (1)

𝑦 : log of hourly wind factor capacity in NUTS‐2 region i

𝑓 𝑐𝑙𝑖𝑚𝑎𝑡𝑒 : 10‐m instant wind gust (and its second‐degree polynomial) and air density at time t

44 https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis‐era5‐single‐levels?tab=overview




𝛼 : time‐invariant NUTS‐2 fixed‐effects

𝛾 : year, month, and hourly fixed‐effects

As our dependent variable, wind power factor capacity, positive, continuous, and right‐skewed – we utilize a gamma Generalized Linear Model with a log‐link. Along with being more flexible compared to an Ordinary Least Squares type specification, the gamma regression also has a better fit to our data.

We compute the potential impacts of future climate change on wind energy production in Europe by combining the estimated parameters from Equation (1) with two Representative Concentration Pathway (RCP4.5 and RCP8.5) trajectories simulated using multiple RCMs to obtain the ratio of wind power generation with climate change relative to wind power generation under the current climate. Future climatic data in our analysis are a from four high‐resolution Regional Climate Models (RCM): KNMI RACMO22E, IPSL‐CM5A‐MR, MPI‐ESM‐LR, and CNRM‐CM5. Based on the stakeholder interests established in COACCH D1.5, we have focused our projections on the Representative Concentration Pathway (RCP) 4.5 as the likely scenario and closest to the proposed Nationally Determined Contributions (NDC) pathway, and an extreme scenario RCP 8.5, which represents the worst possible case.

3.5. Results

We find that wind load factor capacity in Europe is maximized at 10.1 m/s (≈36.4 KM/hour) when considering the full panel. This optimal condition is on the lower‐bound of 10–15 m/s suggested by the technical literature on wind speed and wind power generation (Li and Zhi, 2016; Dixon and Hall, 2014; Le Gouriérès, 1982). This result suggests that wind load factor capacity over Europe peaks at 10.1 m/s, beyond which the generation declines. As expected, air density has a positive impact on load factor capacity, as increased air density exerts added pressure on the turbines, thereby increasing power generation.

Figure 3.5 Non‐linear relationship between hourly wind speed and log of load factor capacity




Figure 3.5: Non‐linear relationship between hourly wind speed and log of load factor capacity (dark navy line, relative to the optimum condition) during 1986‐2015 with 95% confidence interval (blue, with robust standard errors, N= 66,531,935).

Specification includes air density, and NUTS‐2, year, month, and hour fixed‐effects. Histogram shows hourly distribution of wind speed. Marginal‐effects of wind speed impacts on wind load factor capacity

We also segregate the data by seasons45 and find varying optimal conditions by season. The optimal wind speed maximizing wind load factor capacity during Summer in Europe is 9.7 m/s while in the Winter, the optimal speed increases to 16.6 m/s.

45 Spring: March, April, May; Summer: June, July, August; Fall/Autumn: September, October, November, Winter: December, January; February.




Table 3.2 Regression results (1) (2) (3)

Dependent variable: Growth of wind power generation

Full‐sample Summer Winter

Hourly wind speed 0.047 0.131 0.090

(0.000) (0.000) (0.000) Hourly wind speed squared ‐0.002 ‐0.007 ‐0.003

(0.000) (0.000) (0.000) Air density 0.364 0.055 2.819

(0.000) (0.000) (0.000) Constant ‐1.378 ‐2.014 ‐2.037 (0.000) (0.000) (0.000)

Observations 66,531,935 16,632,984 16,632,984

Number of NUTS‐2 regions 234 234 234

Robust p‐value in parentheses

*** p<0.01, ** p<0.05, * p<0.10, + p<0.15

All specifications include NUTS‐2, year, month, and hour fixed‐effects

3.6. Impact of future climate change

We compute the impacts of future climate change by combining our econometric estimates with various warming scenarios (equation 2) under Representative Concentration Pathway (RCP4.5 and RCP8.5) simulated using four different regional climate models (KNMI RACMO22E, IPSL‐CM5A‐MR, MPI‐ESM‐LR, and CNRM‐CM5). We define current and future climate as the mean of the climatic variables between 1986‐2005 (historical), 2030‐2050, and 2050‐2070, respectively. We combine the daily mean of instant wind speed with the fitted response from equation (1) to obtain the ratio of future to current wind power generation;

∈

∈𝑒𝑥𝑝 ∑ 𝛽 ∆𝑊𝑆 ∈

(2)

Our results show that, under RCP4.5 load factor capacity from wind power will decline by 5.6% by 2050 compared the reference period of 1986‐2005, while and by 2070 the reduction will be 7.3%. The biggest declines in load factor capacity due to changing wind patterns will be in northern Austria, northeast Italy, and eastern Switzerland, while wind power generation will increase in parts of the United Kingdom and Ireland.




Figure 3.6 Impact of future changes in wind speed on load factor capacity from wind power under RCP4.5

Figure 3.7 Impact of future changes in wind speed on load factor capacity from wind power under RCP8.5

Under an unmitigated climate change scenario of RCP8.5, load factor capacity is projected to decline by 6.9% by 2050 while by 2070, the declines are projected to be 9.7%. These declines are higher than the impacts suggested by Tobin et al. (2014) who found that mean energy yields will reduce by less than 5% by 2050 (2°C scenario). Under this climate change scenario, the highest declines in wind power generation will be in eastern and western Sweden, and in Andalusia, Spain. Weber et al. (2018) found a high probability for low wind power generation in Sweden in the future. Our methodology to compute the projected changes to future




changes in wind pattern has certain limitations; we only consider the changes in wind speed and patterns and not any changes in technology.

3.7. Discussion and conclusion

Wind energy currently accounts for 20% of the EU’s total installed power generation capacity while covering for 14% of EU’s electricity demand in 2018. Meeting the EU’s binding renewable energy target will require continued focus on power generation from wind, however, wind power generation is both dependent on and vulnerable to changes in wind speed and patterns. Investigating and understanding the impacts of future climate change on wind speed and wind pattern and the consequent impact on wind power generation is critical for the future renewable energy management in Europe. Most of the literature on the impact of changing wind patterns due to climate change on wind power generation focus mainly on energetic resources and not on energy system impacts and is there are very few European‐wide studies. Using hourly load factor capacity and climatic data at the NUTS‐2 level, this paper seeks to fill this gap. Our analysis from a GLM gamma regressions shows that wind power generation in Europe is maximized at around 10 m/s and that there are significant seasonal (Summer and Winter) differences in optimal climatic conditions.

In a second step, we combine our non‐linear econometric estimates with future wind speed patterns from a multi‐model mean under two warming scenarios (RCP4.5 and RCP8.5). The results suggest that under RCP4.5, wind power generation is projected to decline by 5.6% by 2050 and by7.3% by 2070. Under an unmitigated climate change scenario of RCP8.5, load factor capacity of wind power is projected to decline by 6.9% by 2050 while by 2070 the declines are projected to be 9.7% in Europe. The future impacts on wind power generation will be highly heterogenous with the northern regions of Austria and Italy, eastern regions of Switzerland and Sweden, and the Andalusian regions of Spain projected to suffer the highest decline in wind power generation. However, similar to most of climate change impacts, there will also be regions that will gain in‐terms of increased wind power generation. The findings from this paper are essential for the future planning of wind energy networks in Europe, especially to prepare for a reduction in wind energy generation under future warming.




3.8. References

Brayshaw, D.J., Troccoli, A., Fordham, R., Methven, J. (2011). The impact of large‐scale atmospheric circulation patterns on wind power generation and its potential predictability: A case study over the UK. Renew Energy, 36, 2087–2096.

Carta, J.A. (2012). Wind Power Integration in Sayigh, A. (ed.) Comprehensive Renewable Energy; pp. 569‐622.

Carvalho, D., Rocha, A., Gómez‐Gesteira, M., Silva Santos, C. (2017). Potential impacts of climate change on European wind energy resource under the CMIP5 future climate projections. Renewable Energy, 101, 29‐40.

Chandramowli, S.N. & Felder, F.A. (2014). Impact of climate change on electricity systems and markets ‐ A review of models and forecasts. Sustainable Energy Technologies and Assessments, 5, pp. 62–74.

Christensen J. H., Kanikicharla K. K., Marshall G., and Turner J. (2013). Climate phenomena and their relevance for future regional climate change. Cambridge University Press, Cambridge, UK.

Dixon, S.L., and Hall, C.A. (2014). Wind Turbines in Dixon, S.L., and Hall, C.A. (ed.) in Fluid Mechanics and Thermodynamics of Turbomachinery, pp. 419‐485.

Fan, L., and Miao, Z. (2015). AC Machine Modeling in Fan, L., and Miao, Z. (ed.). Modeling and Analysis of Doubly Fed Induction Generator Wind Energy Systems; pp. 8‐33.

Fischer‐Bruns, I., Von Storch, H., González‐Rouco, J. F., Zorita, E. (2005) Modelling the variability of midlatitude storm activity on decadal to century time scales. Climate Dynamics, 25(5), pp. 461‐476.

Gonzalez Aparicio, I., Zucker, A., Careri, F., Monforti, F., Huld, T., Badger, J. (2016). EMHIRES dataset. Part I: Wind power generation European Meteorological derived High‐resolution RES generation time series for present and future scenarios; EUR 28171 EN; 10.2790/831549.

Hdidouan, D., Staffell, I. (2017). The impact of climate change on the levelised cost of wind energy. Renewable Energy, 101, 575‐592.

Hueging, H., Born, K., Haas, R., Jacob, D., and Pinto, J. G. (2013). Regional changes in wind energy potential over Europe using regional climate model ensemble projections. Journal of Applied Meteorology and Climatology, 52(4), 903–917. https://doi.org/10.1175/JAMC‐D‐12‐086.1.

Karnauskas, K.B., Lundquist, J.K., Zhang, L. Southward shift of the global wind energy resource under high carbon dioxide emissions. Nat. Geosci., 11, 38–43.

Le Gouriérès, D. (1982). Wind power plants: Theory and design.

Li, G., and Zhi, J. (2016). Analysis of Wind Power Characteristics in Wang, N., Kang, C., and Ren, D. (ed.) Large‐Scale Wind Power Grid Integration, pp. 19‐51.




Manwell, J. F., McGowan, J. G., and Rogers, A. L. (2009). Wind Energy Explained: Theory, Design and Application (2nd ed.). Chichester, UK: John Wiley. https://doi.org/10.1002/9781119994367.index

Moemken, J., Reyers, M., Feldmann, H., and Pinto, J. G. (2018). Future changes of wind speed and wind energy potentials in EURO‐CORDEX ensemble simulations. Journal of Geophysical Research: Atmospheres, 123, 6373–6389. https://doi.org/10.1029/2018JD028473.

Ohba, M. (2019). The Impact of Global Warming on Wind Energy Resources and Ramp Events in Japan. Atmosphere, 10(5), 265; https://doi.org/10.3390/atmos10050265.

Pieralli, S., Ritter, M., and Odening, M. (2015). Efficiency of wind power production and its determinants. Energy, 90; 429‐438.

Pryor, S.C., Barthelmie, R.J., Kjellström. E. (2005). Potential climate change impact on wind energy resources in northern Europe: analyses using a regional climate model. Climate Dynamics, 25; 815‐835.

Pryor, S. C. and Barthelmie, R. J. (2010). Climate change impacts on wind energy: A review. Renewable and sustainable energy reviews, 14(1), pp. 430‐437.

Pryor, S., Barthelmie, R.J., Clausen, N.E., Drews, M., MacKellar, N., Kjellström, E. (2012). Analyses of possible changes in intense and extreme wind speeds over northern Europe under climate change scenarios. Clim. Dyn, 38, 189–208.

Pryor, S.C., Barthelmie, R.J. (2010). Climate change impacts on wind energy: a review. Renewable and Sustainable Energy Reviews, 14:430‐7.

Pryor, S. C., and Barthelmie, R. J. (2013). Assessing the vulnerability of wind energy to climate change and extreme events. Climatic Change, 121(1), 79–91. https://doi.org/10.1007/s10584‐013‐0889‐y

Reyers, M., Moemken, J., and Pinto, J. G. (2016). Future changes of wind energy potentials over Europe in a large CMIP5 multi‐model ensemble. International Journal of Climatology, 36(2), 783–796; https://doi.org/10.1002/joc.4382.

Stone, D. A., Weaver, A. J., and Stouffer, R. J. (2001). Projection of climate change onto modes of atmospheric variability. Journal of Climate, 14(17), 3551‐3565.

Tobin, I., Vautard, V., Balog, I. et al. (2014). Assessing climate change impacts on European wind energy from ENSEMBLES high‐resolution climate projections. Climatic Change. https://doi.org 10.1007/s10584‐014‐12910.

Tobin, I., Jerez, S., Vautard, R., Thais, F., van Meijgaard, E., Prein, A., et al. (2016). Climate change impacts on the power generation potential of a European mid‐century wind farms scenario. Environmental Research Letters, 11(3), 034013; https://doi.org/10.1088/1748‐9326/11/3/034013.

Weber, J., Gotzens, F., and Witthaut, D. (2018). Impact of strong climate change on the statistics of wind power generation in Europe. Energy Procedia, 153 pp. 22‐28; https://doi.org/10.1016/j.egypro.2018.10.004.




4. Vulnerability of Global Hydropower to Climate Change

4.1. Introduction

Hydropower represents the main renewable source of electricity production for several countries in the world. In 2017, electricity generation from hydropower accounted for 16.3% of the total world gross electricity production46. The international debate about greenhouse gases emission has put an increasing pressure on countries to reduce their dependency on energy derived from fossil fuels. Renewable energy sources play a critical role in this process and hydropower remains the largest contributor among the renewables; 68.8% of electricity generated from renewables in 2016 and 48% of installed capacity of total renewable installed electricity capacity. The installation of new hydropower plants has registered a substantial increase in the past few years (REN21 2013). This power generation technology is one of the main pillars of the energy strategy of countries as China and Brazil for the next future (IEA 2013). Notwithstanding the large mitigation potential hydropower has, this option strongly depends on meteorological variability and trends. Rising concerns have stirred the debate about the vulnerability of hydropower technologies to climate change and the possibility for this important renewable source to sustain its future development (Mukheibir 2013). Moreover, the focus to minimize environmental and social impacts are driving towards the construction of construction of reservoirs characterized by a relatively small size, a trend which is expected to significantly increase the vulnerability of the hydropower sector to climate change (IEA 2013). Only a few countries have studied the impact of climatic stressors on the vulnerability hydropower supply on global scale (Blackshear et al. 2010; Hamududu and Killingtveit 2012), while most assessments have focus on specific regions (Ecuador, Hasan and Wyseureb 2018; China, Fan et al. 2018; Taiwan, Chiang et al. 2013; Sicily, Aronica and Bonaccorso 2012; Northern Europe, Bye 2008; Nordic regions, Beldring et al. 2006) or basins (Barnett et al. 2004; Van Rheenen et al. 2003). Hamududu and Killingtveit (2012) develop a GIS‐based analysis aimed at exploring the linkage between changes in runoff and hydropower supply. The relationship between these two variables is not empirically‐based and, other factors, such temperature, drought or flood, are not considered. Moreover, the analysis does not consider seasonal variability and focuses on the national geographical scale, characterizing the median country changes. Blackshear et al. (2010) present a very useful framework that can be used to identify the main factors affecting hydropower generation, discussing also

46 https://www.iea.org/statistics/electricity/




how the response and sensitivity varies with the type of facility (run‐of‐the river, pumped systems, reservoir dams). However, the paper does not provide actual estimates of the sensitivity of different types of dams. Basin‐specific studies generally rely on hydrogeological models (Van Rheenen et al 2003), and only a few regional analyses have adopted statistical approaches (Blasing et al. 2013). Blackshear et al. (2011) use the existing literature and geographic databases to develop a framework aimed at assessing hydropower vulnerability at global scale. Their approach relies on the spatial comparison between climate data and the geographic location of hydropower facilities. The study identifies and discusses the main mechanisms through which climate change could affect hydropower generation, indicating how different types of hydropower facilities (reservoir‐based, run‐of‐the‐river, pumped‐storage) would be differently affected by the various mechanisms. This paper explores the vulnerability of global hydropower to the variability temperature as well as changes in extreme conditions of precipitation, and temperature. A statistical model is used to estimate the elasticity of hydroelectricity generation to the historical variations (1971‐2016) in temperature and drought indicators47, while controlling for potential confounding factors at the global scale. We then combine our estimated response functions with future warming scenarios to assess the future vulnerability of hydropower generation in 140 countries. We combine the estimated elasticities with future changes in exposure to drought and temperature around 2050 and 2070 under two warming scenarios (Representative Concentration Pathways 4.5 and 8.5, van Vuuren et al. 2011) simulated by five different Global Circulation Models (GCMs). Future climatic data in our analysis are a from four high‐resolution Regional Climate Models (RCM): KNMI RACMO22E, IPSL‐CM5A‐MR, MPI‐ESM‐LR, and CNRM‐CM5. Based on the stakeholder interests established in COACCH D1.5, we have focused our projections on the Representative Concentration Pathway (RCP) 4.5 as the likely scenario and closest to the proposed Nationally Determined Contributions (NDC) pathway, and an extreme scenario RCP 8.5, which represents the worst possible case. The rest of the paper is organized as follows. Section 2 describes the framework and the approach. Section 3 introduces the hydropower and climate data. Section 4 provides the empirical model; Section 5 discusses the empirical results, while Section 6 provides future vulnerability assessment of hydropower, and Section 7 concludes.

4.2. A framework of meteorological effects on hydropower Hydropower generation is determined by a number of climatic and non‐climatic factors. Water is the primary input for hydro generation and its availability is a complex function of water sources (ground water, snowpack, streamflow, and reservoir storage) and competitive uses (e.g. water use for irrigation, cooling and heating demand). The impact of climate change on hydropower generation depends on the changes in precipitation, temperature, their interaction, their implications in terms of ground water and snowpack accumulation, as well as on the facility characteristics. Table 1 describes through which main mechanisms meteorological factors can affect electricity supply from hydropower.

47 Standard Precipitation Index (SPI)




Table 4.1 Climatic stressors and hydropower generation – a framework (Blacksheare et al. 2011)

Supply‐side Demand‐side

Temperature

Evaporation Snowpack/ice melting

Runoff

Demand for heating/cooling Demand for irrigation

Precipitation

River discharge Snowpack Runoff

Demand for heating/cooling Demand for irrigation

Runoff

River discharge Water availability

Temporal variability (Dry/wet periods)

Water availability Demand for heating/cooling

Demand for irrigation

Temperature can affect the electricity potential from hydropower plants by affecting the processes of evapotranspiration and snow melting, and therefore the speed at which water flows into the reservoirs and the duration of snowpack as additional water storage capacity (supply side effect). Temperature can also affect the demand for electricity, especially for cooling and heating (demand side effect). Annual precipitation captures the flow effect of water availability for electricity production in a given year, however, in the case of large dams, it does not inform about the impacts on the stock due to variations in the volumes stored in the reservoirs. Thus, it is important to consider also explicit indicators of temporal variability that can capture the effects of persistent periods of water scarcity or abundance. As reviewed in the latest IPCC 5th assessment report, econometric or statistical approach is one of the possible types of economic analyses used in the literature to estimate climate change effects and evaluate adaptation options (Chambwera et al. 2014). These approaches have been mostly applied to analyze climate change impacts in agriculture (Lobell and Burke 2010; Schlenker and Roberts 2009) and energy demand (De Cian and Wing 2017; Auffhammer Mansur 2012; Barreca 2012; De Cian et al. 2012; Deschenes and Greenston 2011), and only a few applications exist in the context of energy supply (Blasing et al 2012). Most of the studies on energy supply impacts have applied simulation approaches, such as hydro‐climatic simulation approaches (Hamududu and Killingtveit 2012) or process‐based models describing the water‐cycle processes (Prudhomme et al. 2013). Simulation methods are very detailed with respect to the analysis of exposure but the modelling of sensitivity, specifically the response of power generation to changes in average climatic variables as well as in extreme events, require extensive data inputs and calibration. Units with different sizes can exhibit different degrees of sensitivity that can vary with the type of dams (run‐of‐river versus dams with large storage reservoirs), with the storage capacity, as well as with the alternative (and competitive) uses for water (such as agriculture or the presence of waterways), a factor extremely limiting for the operation of the reservoirs. For this reason, simulation approaches are generally more data‐intensive than econometric ones, and the data requirement that often constrains the applicability of models to limited geographic areas. Econometric approaches do not describe all mechanisms but rather attempt to identify robust relationships between climate stressors and the endpoint of interest.




4.3. Hydropower and climatic data

We merge hydropower generation data from Global Energy & CO2 Data (Enerdata 2018) with high‐resolution climatic data from the Global Land Data Assimilation System (GLDAS v2.1) dataset (Rodell et al. 2004) for 1971 – 2016. Enerdata provides comprehensive statistics on electricity generation at the country‐level by energy source. We constructed an updated country‐level dataset using GLDAS (spatial resolution of 0.25°×0.25°; 3‐hourly temporal resolution). We first extract climatic data for all the land‐masked grid‐cells in the world and match it to individual countries and in the second step, the gridded data were aggregated to the country‐level. We tested several variables characterizing changes in gradual climatic and variability, such as extreme wetness and aridity including temperature (minimum, maximum, and mean), total precipitation, Standard Precipitation Index (SPI), and Warm Spell Duration Index (WSDI). Table 2 below provides descriptive statistics for the main variables used in the analysis. We model changes in inter‐annual variability by using the Standard Precipitation Index (SPI) (McKee et al. 1993). The SPI is a widely used drought indicator (McKee et al., 1993; Núñez et al. 2014; Orlowsky et al., 2013). The index represents the number of standard deviations that the cumulative precipitation over a desired time scale deviates from the long‐term median (Guttman 1994). A long‐term record of precipitation for the desired periods, in our case obtained from the GLDAS historical data (1960‐2010), is fitted to a Pearson type III distribution (Guttman 1999; Kumar et al. 2009; WMO 2012; Núñez et al. 2014) and then transformed into a normal standardized distribution so that the mean SPI for the location and desired period is zero (Edwards and McKee, 1997). Different duration periods can be used to analyse the effect of precipitation anomalies of different persistency. Depending on the problem at stake, the SPI can be defined for durations between 3 and 24 months. For example, in the context of agriculture, a key indicator is soil moisture, which is sensitive to precipitation anomalies over relatively short time scales, between 1 and 6 months (agricultural drought) (Beguería et al, 2014). Groundwater and large reservoirs tend to be more resilient and therefore are sensitive to longer time scale anomalies, between 6 and 24 months (hydrological drought) (WMO 2012). Since the SPI is a normalized value, it is a valuable indicator both for wet and dry periods (WMO 2012). Positive SPI values indicate greater than median precipitation, and negative values indicate less than median precipitation. The intensity to be chosen as representative has been widely discussed in literature (McKee et al. 1993; WMO 2012; Guttman 1999; Kumar et al. 2009) with the conclusion that values of plus and minus 1.5 represent reasonable thresholds to identify very wet and dry periods. To examine the impact of hydrological drought, we use the 6, 12, and 24‐month SPI. Table 4.2 Descriptive statistics

Variable Mean Std. Dev. Min Max

Hydropower generation 14,746.21 54,484.19 0.00 1,193,650 Installed hydroelectricity capacity 4,230.65 14,900.33 0.00 332,110




Electricity final consumption 67,427.84 297,694.40 2.10 5,131,368 Log of GDP per capita 8.18 1.54 4.75 11.64 Hydro power share 0.34 0.35 0.00 1.00 Nuclear power share 0.04 0.12 0.00 0.87

Coal share 0.13 0.25 0.00 1.00 Gas share 0.15 0.26 0.00 1.00 Oil share 0.32 0.35 0.00 1.00

Mean temperature 18.3 8.2 ‐9.2 30.1 Total precipitation 8,456 6,410 124.3 40,817.3

Count of negative SPI6 0.077 0.115 0 1 Count of negative SPI12 0.077 0.143 0 1 Count of negative SPI24 0.077 0.155 0 1




Figure 4.1 Trends in electricity production and capacity

It is essential to clarify that statistical and process models are characterized by two distinct approaches. Statistical models take into consideration the historical records to study the response of energy production to various inputs. The second approach studies the physical conditions of the hydrological system to estimate the potential ability to generate power. Although process models are probably more accurate in assessing the potential generation at dam level given water availability, they have a limited geographic coverage and very often do not account for operating strategies. The few global‐scale studies have mostly focused on the direct impacts of changes in future runoff on the physical capability of power generation (McKee et al. 2013) Hydropower production is not only function of water availability and dam technical characteristics but also depends on a country’s energy system, its strategic management, other supply sources, water uses, and electricity demand. By using historical data, our approach partially manages to take into consideration socio‐economic factors influencing the peculiarity of the operating rules used to manage the country energy system and the specific reservoir.

4.4. Econometric Framework

Most studies (Hamududu and Killingtveit 2012; Prudhomme et al. 2012) on climate change impacts and hydropower rely on process‐based models or simulation approaches. Only a few studies have adopted statistical approaches in the context of energy supply (Blasing et al. 2013), an alternative method (Chambwera et al. 2014) that has been used extensively to analyse climate change impacts in other sectors (De Cian et al. 2013; Deschenes and Greenstone 2011; Lobell and Burke 2010). We use a panel regression model to estimate the parameters characterizing a reduced‐form relationship to investigate the impact of both gradual and extreme climatic stressors on hydropower generation at the country‐level. We control for a set of climatic variables and number of other covariates controlling for time‐invariant country‐specific heterogeneity (country fixed‐effects), unspecified exogenous influences affecting all countries and units (year fixed‐effects), and confounding factors such as installed power generation capacity, total electricity consumption, and electricity generation mix.




A panel regression with both country and year fixed‐effects to estimate the impact of both gradual and extreme climatic stressors on country‐level hydropower generation;

ln 𝑦 𝑓 𝐶𝑙𝑖𝑚𝑎𝑡𝑒 𝛿𝑋 𝛼 𝛾 𝜖 (1)

𝑙𝑛 𝑦 : log of annual hydropower generation

𝑓 𝐶𝑙𝑖𝑚𝑎𝑡𝑒 : mean temperature/temperature growth, total precipitation, SPI (6/12/24 months)

SPI: number of months SPI was below ‐1.5 in country i in year t

𝑋 : vector of control variables controlling for installed hydroelectric capacity, final electricity consumption, share of hydropower capacity, and electricity production mix (gas, oil, coal, and nuclear).

𝛼 : time‐invariant country fixed‐effects

𝛾 : linear and quadratic time trends Potential impacts of future climate change are computed by combining the estimated parameters from Equation (1) with two Representative Concentration Pathway (RCP4.5 and RCP8.5) trajectories simulated using five GCM models described in the Supplementary

Information to obtain the ratio of hydropower generation with climate change 𝐺 ∈ relative

to hydropower generation with current climate (𝐺 ∈ ) supply.

4.4.1. Non‐stationarity and cointegration

Although some of the variables are non‐stationary (I(1)), they are cointegrated; thus, we run a first‐difference specification along with a level specification.

4.5. Empirical Results

Table 3 provides results from our first‐differenced specifications. Columns (1) – (3) in Table 3 controls for temperature, SPI (6/12/24 months), hydroelectric capacity, final electricity consumption, share of hydropower capacity, and electricity production mix while specifications (4) – (6) adds total annual precipitation and its second‐degree polynomial. Based on the Root Mean Square Error (RMSE), the specification controlling for SPI‐12 (Newell et al. 2018) is our preferred specification. We find that a 1°C increase in mean temperature results in between 0.025 and 0.029 percentage points decline in hydropower generation (Table 3). SPI informs about the stock effect due to potential variations in volumes stored in reservoirs over long time periods due to prolonged wet or dry periods. All the SPI coefficients are negative and statistically significant, indicating that persistent droughts reduce average annual hydropower generation. The occurrence of one additional month classified in a long‐term dry event (24‐month SPI lower than the standard deviation threshold defining severe dry events, ‐1.5) reduces hydroelectricity generation by between 0.16 ‐ 0.17 percentage points, while the




occurrence of one additional medium‐term dry event (12‐months SPI < ‐1.5) reduces generation from small units by 0.32 ‐ 0.36 percentage points; while short‐term (6‐months SPI < ‐1.5) drought events reduces hydroelectricity generation by between 0.35 ‐ 0.41 percentage points. The practices of imposing water release from reservoirs for minimizing drought impacts in agriculture are widely used especially in areas where agriculture has a higher value added. Finally, precipitation has an inverted U‐shaped relationship with hydropower generation, however, the coefficients are rather small. Drought events also have larger impact on the largest producers of hydropower (</>10TWh/year), we run separate regressions by segregating our dataset according to this criterion. According to our main specification, an additional month classified in a medium‐term dry event (12‐month SPI < ‐1.5) reduces hydropower generation by 0.48 percentage points among the largest producers of hydropower. While among the countries producing less than this threshold, this impact is 0.34 percentage points (Table A1). Table 4.3 First‐differenced regression estimates (1) (2) (3) (4) (5) (6)

Dependent variable: Growth of hydropower generation

Lagged & first‐differenced installed hydro capacity 0.132*** 0.127*** 0.129*** 0.132*** 0.128*** 0.130***

(0.003) (0.004) (0.003) (0.003) (0.004) (0.003)

Lagged electricity consumption growth 0.070 0.065 0.049 0.069 0.066 0.054

(0.309) (0.344) (0.479) (0.318) (0.339) (0.443)

Nuclear share ‐0.719*** ‐0.714*** ‐0.733*** ‐0.720*** ‐0.711*** ‐0.723***

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

Coal share ‐0.892*** ‐0.877*** ‐0.924*** ‐0.892*** ‐0.872*** ‐0.906***

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

Gas share ‐0.777*** ‐0.767*** ‐0.792*** ‐0.777*** ‐0.765*** ‐0.784***

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

Oil share ‐0.785*** ‐0.772*** ‐0.800*** ‐0.788*** ‐0.775*** ‐0.796***

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

ΔMean temperature ‐0.028*** ‐0.029*** ‐0.028*** ‐0.027*** ‐0.027*** ‐0.025***

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

ΔTotal precipitation 8.18×106*** 1.29×10‐5*** 1.9×10‐5***

(0.001) (0.000) (0.000)

ΔTotal precipitation squared −4.76×10‐10 −7.63×10‐10** −8.23×10‐10**

(0.180) (0.033) (0.038)

ΔSPI6 <1.5 ‐0.407*** ‐0.352***

(0.000) (0.000) ΔSPI12 <1.5 ‐0.359*** ‐0.320***

(0.000) (0.000) ΔSPI24 <1.5 ‐0.170*** ‐0.158***

(0.000) (0.000)

Constant 0.251*** 0.247*** 0.255*** 0.252*** 0.247*** 0.253***

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

Observations 5,491 5,491 5,491 5,491 5,491 5,491

R‐squared 0.169 0.173 0.141 0.171 0.180 0.157

Adj. R‐squared 0.0976 0.102 0.0668 0.0999 0.110 0.0846

RMSE 0.2472 0.2466 0.2514 0.2469 0.2455 0.2490


*** p<0.01, ** p<0.05, * p<0.10, + p<0.15

All specifications include country and linear and quadratic year fixed‐effects




4.6. Impacts of future climate change

We compute the impacts of future climate change by combining our econometric estimates with various warming scenarios under Representative Concentration Pathway (RCP4.5 and RCP8.5) simulated using five different climate models (CCSM4, GFDL‐CM3, INM‐CM4, IPSL‐CM5A‐MR, and MIROC5). We define current and future climate as the mean of the climatic variables between 1986‐2005 (historical), 2030‐2050, and 2050‐2070, respectively. We combine the annual mean of the climatic variables with the fitted response from model (5) in Table 1 to obtain the ratio of future to current hydroelectricity;

𝐺 ∈

𝐺 ∈𝑒𝑥𝑝 𝛽 ∆𝑆𝑃𝐼 ∈ 𝑒𝑥𝑝 𝛽 𝑙𝑛

𝑇 ∈ ,

𝑇 ∈,

Under RCP4.5, global hydropower generation will decline by 3.6% by 2050 and by 5.3% by 2070 due to future climate change. While under unmitigated climate change (RCP8.5), global hydropower generation will decline by 4.2% (2050) and by 7.3% (2070), respectively. There will be significant losses among the largest producers of hydropower; Brazil (4% under RCP4.5 by 2070 and 6.2% under RCP8.5 by 2070), Canada (6.8% under RCP4.5 by 2070 and 10.6% under RCP8.5 by 2070), China (5.6% under RCP4.5 by 2070 and 7.4% under RCP8.5 by 2070), Russia (7.6% under RCP4.5 by 2070 and 10.3% under RCP8.5 by 2070), and USA (4.8% under RCP4.5 by 2070 and 7.5% under RCP8.5 by 2070)

Figure 4.2 Climate change impacts on hydropower (% change) under RCP 4.5 and RCP 8.5 simulated by 5 GCMs




4.6.1. The case of Europe

The European countries are also projected to suffer reductions in hydropower production under both the climatic scenarios (Figure 4). Under a moderate warming scenario of RCP4.5, the highest declines will be in Finland (6.3%; hydropower share of electricity production was 25% in 2016), Estonia (6.2%; hydropower share of electricity production was 0.3% in 2016), and Serbia (5.9%; hydropower share of electricity production was 30% in 2016) by 2050, while by 2070 – the highest declines are projected in Slovenia (10.5%; hydropower share of electricity production was 25% in 2016), Croatia (9.8%; hydropower share of electricity production was 54% in 2016), and Austria (9.6%; hydropower share of electricity production was 63% in 2016). Under the unmitigated climate change scenario of RCP8.5, the highest declines in hydropower generation by 2050 will be in Finland (7.7%; hydropower share of electricity production was 25% in 2016), Sweden (6.6%; hydropower share of electricity production was 40% in 2016), and Estonia (6.4%; hydropower share of electricity production was 0.3% in 2016). By 2070, the highest declines in hydropower generation due to climate change are projected in Serbia (13.3%; hydropower share of electricity production was 30% in 2016), Romania (12.7%; hydropower share of electricity production was 30% in 2016), Hungary (12.7%; hydropower share of electricity production was 1% in 2016), and Sweden (12.3%; hydropower share of electricity production was 40% in 2016).




Figure 4.3 Climate change impacts on hydropower (% change) in Europe under RCP 4.5 and RCP 8.5


Rising concerns have stirred the debate about the future vulnerability of hydropower to climate change and of its sustainable development as renewable source of energy. Although storage hydropower could help to mitigate climate change and cope with water scarcity and flood events, climate change is expected to modify the future conditions in which the hydropower operators are called to manage the storage capacity. Our econometric analysis suggests that both gradual and extreme climatic change adversely affects hydropower generation. We also find that large producers of hydropower such as Brazil, Canada, China, Russia, and USA are more vulnerable to increased drought related extreme events compared to the smaller producers.




Combining our response‐functions with a multi‐model mean of climate models, we find that under unmitigated climate change (RCP8.5), global hydropower generation will decline by 4.2% (2050) and by 7.3% (2070), respectively, with all the largest producers of hydropower suffering significant declines of up to 10%. These declines are smaller under lower warming scenario of RCP4.5.




Annex

(1) (2)

Dependent variable: Growth of hydropower generation

Lagged log of installed hydroelectricity capacity 0.301* 0.120**

(0.077) (0.014) Lagged log of electricity consumption 0.085 0.069

(0.234) (0.369) Nuclear share ‐1.094*** ‐0.586***

(0.000) (0.001) Coal share ‐1.107*** ‐0.809***

(0.004) (0.000) Gas share ‐0.917*** ‐0.732***

(0.001) (0.000) Oil share ‐1.093*** ‐0.741***

(0.000) (0.000) ΔMean temperature ‐0.020*** ‐0.030***

(0.005) (0.001) ΔTotal precipitation 0.000*** 0.000**

(0.001) (0.018) ΔTotal precipitation squared 0.000 ‐0.000+

(0.992) (0.138) SPI6 <1.5 ‐0.411*** ‐0.308***

(0.000) (0.000) Constant 0.577*** 0.401*** (0.000) (0.000)

Observations 1,479 4,034 R‐squared 0.328 0.153 Adj. R‐squared 0.288 0.103 Number of countries 36 109


*** p<0.01, ** p<0.05, * p<0.10, + p<0.15

All specifications include country and linear and quadratic year fixed‐effects

Table A1: Large vs. small hydropower producers




5. Impact on energy demand in Europe

5.1. Introduction

Energy demand is increasing globally, causing greenhouse gas emissions from the energy sector also to increase. In the EEA countries, final energy consumption increased by 2.1% in the EU‐28 and by 7.5% between 1990 and 2016 (Eurostat, 2018). The energy sector is also heavily affected by climatic stressors and future climatic conditions are likely to increased demands for energy required for cooling services through increased number of extreme temperature events, however, demand for cooling services might decrease due to the fewer low temperature extremes (De Cian and Wing, 2017; Mideksa and Kallbekken, 2010). Combined with changes in economic growth and the rising population, the mix of fuel in energy demand by various sectors is likely change as well. It is also important to investigate the future impacts of climate change on energy demand to develop adaptation and mitigation policies (Damm et al., 2017 and Eskeland and Mideksa, 2010).

Temperature is one of the major drivers of energy demand in Europe, affecting summer cooling and winter heating for households, industry, and service sectors. Higher temperatures are expected to raise electricity demand for cooling, decrease demand for heating, and to reduce electricity production from thermal power plants (Mideksa and Kallbekken, 2010). These responses are largely autonomous and can therefore be considered as an impact or an adaptation. However, cooling is predominantly powered by electricity (which is more expensive), while heating uses a wider mix of energy sources, this particular distinction needs to be controlled for.

The impact of climatic stressors on energy demand have been rather extensively researched (De Cian and Wing, 2017; De Cian et al., 2013; Howell and Rogner, 2014; Schaeffer, 2012; Bazilian et al., 2011). However, sub‐national estimates of future climate change on energy demand in Europe is lacking in the existing literature. Kitous and Després (2018)48 provide aggregated results for EU‐28 with a focus on selected regions. The authors find that heating needs decline compared by 27% by the end of the century but cooling needs increase significantly. According to EC (2018), final energy consumption in the EU is expected to decrease by 26% by 2050, with energy demand declining in the residential, industrial, transport, and the tertiary sectors. However, these results are also at the aggregate level with no spatial disaggregation. Pilli‐Sihvola et al. (2010), using an econometric methodology found that demand for heating will likely decline in Central and Northern Europe due to future warming. However, due to increasing temperature, cooling demand is likely to increase in Southern Europe. Eskeland and Mideksa (2010) estimated a decrease in electricity consumption in the northern European countries but an increase in demand the southern due to increased warming.

The current literature also provides limited information on the combinations of sectors and fuels with much of the literature focusing on electricity and the residential sector (Schaeffer,

48 As part of PESETE III.




2012). We combine econometric estimates with high‐resolution climatic data from Regional Climate Models (RCMs) to estimate the impact of future climate change at the NUTS‐2 level in the EU under various warming scenarios. Projections are computed for electricity, petroleum products, and natural gas, in four economic sectors (agriculture, industry, residential, and commercial).

5.2. Trends in Energy Demand in Europe

Figure X.1 shows the trends in final energy consumption across the industrial, transport, residential, and the service sectors since 2000. The data shows that the final energy consumption has increased in the transport and service sectors while decreasing in the industrial and residential sectors. However, these aggregated trends in the EU may not be representative as Germany, France, the United Kingdom, and Italy consumed 55.4% of the final energy consumption while fourteen member states consumed less than 10% in 2016.

Figure X.1: Trends in final energy consumption by sector (right) in the EU (EC, 2018)

Figure X.2 provides both the historical and projected breakdown of the aggregated final energy consumption in the EU. The projections are computed by the PRIMES model ‐ an EU energy system model which simulates energy consumption and the energy supply system49. The projected changes in energy mix (left‐panel) suggests that the demand for fossil fuel will decline while the use of electricity will increase. As for the sectoral demand breakdown, the final energy consumption from transport and residential are likely to decline due to efficiency gains (EC, 2018). While aggregated projections are available from the PRIMES model, designing low‐carbon, energy‐efficient energy systems and high mitigation energy policies require projections of sub‐national level climate change impacts for Europe.

49 https://ec.europa.eu/clima/sites/clima/files/strategies/analysis/models/docs/primes_model_2013‐2014_en.pdf




Figure X.2: Final energy consumption by fuel type (left) and sector (right) in the EU (Eurostat, 2018)

5.3. Data and Methodology

We use econometric estimates from De Cian and Sue Wing (2017); the authors analyse per capita demand for three different final energy carriers associated with heating and cooling (electricity, petroleum products, and natural gas, in four economic sectors (agriculture, industry, residential, and commercial) for tropical and temperate countries as a function of per capita gross domestic product (GDP) and exposure to hot (>27.5 °C) and cold (<12.5 °C) days. We use the long‐run temperature elasticities in combination with future changes in temperatures under two warming scenarios for the EU. We utilize population projections from the Shared Socioeconomic Pathways‐2 (SSP2) to construct two baselines for global energy demand in 2050 and 2070 and compare them to a scenario without climate‐change impacts. De Cian and Sue Wing (2019), using data from 204 countries for 1970–2014, estimates elasticities and temperature semi‐elasticities of sectoral energy demand. The relationship between energy demand, weather, income, and prices as a dynamic adjustment process using an Error‐Correction Model (ECM)50. The authors find that temperature change impacts energy demand in a majority of energy carrier, sector, region combinations. Demand for energy increases heterogeneously with hot days across energy carriers and sectors. The paper also shows that extreme cold weather could reduce energy demand, especially in industry and agriculture.

50 Please see De Cian and Sue Wing (2019) for a detailed description of the econometric methodology.




We combine these estimates with future projections of temperature. Our temperature projections are simulations of two representative concentration pathway scenarios (RCPs; van Vuuren et al., 2011) indicative of a high‐warming scenario (RCP8.5) in which climate change is unabated and moderate‐warming (RCP4.5)51 scenario in which mitigation policies are pursued. We use four different regional climate models (KNMI RACMO22E, IPSL‐CM5A‐MR, MPI‐ESM‐LR, and CNRM‐CM5) to compute a multi‐model mean. These bias‐corrected and downscaled RCMs are part of the EURO‐CORDEX52 climate simulations and are available at a spatial resolution of 12 KM. We define current and future climate as the mean of temperature between 1986‐2005 (historical), 2030‐2050, and 2050‐2070, respectively. The climate data used as the input is the difference in the hot (>27.5 °C) and cold (<12.5 °C) bins between the historical and future periods.

5.4. Results

Climate projections show that the number of hot days in EU (a relatively cold region) is expected to increase on average under both the warming scenarios, with some regions in Greece and Cyprus expected to have an additional 30 hot days by 2070. The number of cold days is set to decline under both RCP4.5 and RCP8.5 and some regions in Italy, Portugal, and Spain are expected to have 30 fewer cold days by 2070.

We find heterogenous changes in energy demand by energy carrier and sector in the EU due to future climate change (Figure 5.3 and Table 5.1). The most significant increases in energy demand are expected in the industrial sector (11.4%) from natural gas and the service sector from electricity (40.1%) under RCP8.5 by 2070. These increases are projected to be 3.8% and 15.9%, respectively by 2050. Given the expanding service sector and substantial share of energy demand in this sector being met from electricity, this increase could potentially exert additional pressure on the electricity infrastructure in Europe. Energy demand in the residential sector is projected to decline significantly from natural gas (‐27.5%) and petroleum (‐41.5%) with only demand from electricity increasing (3.8%) under RCP8.5 by 2070. In the case of the agricultural sector, energy demand from electricity is projected to increase by 2% 2070. These declines in energy demand are likely driven by reduced energy consumption for heating, fuel switching (e.g., from petroleum products to electricity in the commercial sector), or as a result of increased energy efficiency in the residential sector.

51 Our choice of scenarios is based on the stakeholder interests established in COACCH D1.5. We have focused our projections on the RCP4.5 as the likely scenario and closest to the proposed Nationally Determined Contributions (NDC) pathway and an extreme scenario RCP8.5, which represents the worst possible case. 52 EURO‐CORDEX ‐ Coordinated Downscaling Experiment ‐ European Domain (https://euro‐cordex.net).




Figure 5.3: Change in climate‐related final energy demand by 2050 and 2070 under RCP4.5 and RCP8.5.

RCP 4.5 RCP 8.5

Sector 2030 ‐ 2050 2050 ‐ 2070 2030 ‐ 2050 2050 ‐ 2070

Agriculture

Electricity 0.6 1.0 0.8 2.0

Natural gas ‐ ‐ ‐ ‐

Petroleum ‐ ‐ ‐ ‐

Industry


Natural gas 2.7 4.9 3.8 11.4

Petroleum ‐ ‐ ‐ ‐

Residential


Natural gas ‐25.9 ‐33.5 ‐27.5 ‐41.5

Petroleum ‐23.7 ‐30.8 ‐25.1 ‐38.3

Commercial

Electricity 13.1 20.6 15.9 40.1

Natural gas ‐ ‐ ‐ ‐

Petroleum ‐14.4 ‐19.1 ‐15.4 ‐24.4

Table 5.1: Percentage change in climate‐related final energy demand by 2050 and 2070 under RCP4.5 and RCP8.5.

Note: Some energy carrier – sector combinations could not be computed as the elasticities computed by De Cian and Sue Wing (2019) were not statistically significant.

5.4.1. Spatial heterogeneity climate change impacts

One of our major contributions is the projection of change in energy demand due to future climate change at the sub‐national (NUTS‐2) level for all the energy carrier – sector combinations. The summary of the results for RCP4.5 and RCP8.5 are provided in Table X.1 while the distribution of impacts across the NUTS‐2 regions are provided for RCP8.5 in Figure X.4. The highest increase in energy demand across energy carriers, driven by rise in cooling




demand, are projected in the NUTS‐2 regions of Thessaly (Greece), Central Macedonia (Greece), Andalusia (Spain), and Yugoiztochen (Southeastern Belgium). Driven by decline in heating demand, the regions with the highest declines in the residential sectors are projected to be South Aegean (Greece), Algarve (Portugal), and Ceuta (Spain).




Electricity

Natural Gas

Petroleum




Figure 5.4: Distribution of climate change impacts by energy carrier, sector, and NUTS‐2 region under RCP8.5 by

2070 computed using a multi‐model mean of four RCMs. Note: White indicates missing data.


Our results suggest that the impact of future climate change on energy demand in Europe will be heterogeneous, across energy carriers, sectors, and NUTS‐2 regions. The most significant increases in energy demand are expected in the industrial sector (11.4%) from natural gas and the service sector from electricity (40.1%) under RCP8.5 by 2070. Under a more moderate warming scenario of RCP4.5, these increases are projected to be 4.9% and 20.6%, respectively. Given the expanding service sector and substantial share of energy demand in this sector being met from electricity, this increase could potentially exert additional pressure on the electricity infrastructure in Europe.

Energy demand in the residential sector is projected to decline significantly from natural gas (‐27.5%) and petroleum (‐41.5%) with only demand from electricity increasing (3.8%) under RCP8.5 by 2070. In the case of the agricultural sector, energy demand from electricity is projected to increase by 2% 2070. Comparable estimates at the sub‐national level are difficult to find for Europe; Damm et al. (2017) found that a warming of 2°C reduces electricity consumption in most European countries. With the highest decrease expected in Norway, Sweden, Estonia, Finland, and France.

Our estimates also suggest that electricity is the most dominant energy carrier in terms of increased energy demand due to climate change. In the case of sectors, the increases in demand will be largely driven by the industrial (4.5%) and commercial sectors (5.2%), while the demand in the agricultural sector projected to increase slightly by 0.7%. Final total energy demand in the residential sector is expected to decline by 25.3% under an




unmitigated climate change scenario. These results are in line with van Ruijven et al. (2019), who find a median 2% net reduction in the total final energy consumption for Europe.

Our results could be important not only for sector and region‐specific mitigation policies but also for network and infrastructure operators. As our analysis provides impacts based on hot and cold days, supply side decision on future peak demand requirements can be tailored by region.




5.7. References

Bazilian, M. et al. (2011). “Considering the energy, water and food nexus: towards an integrated modeling approach”. Energy Policy 39 (12):7896–7906, https://doi.org/10.1016/j.enpol.2011.09.039.

Damm, A., Köberl, J., Prettenthaler, F., Rogler, N., and Töglhofer, C. (2017). “Impacts of +2°C Global Warming on Electricity Demand in Europe”, Climate Services, 7, 12–30; https://doi.org/10.1016/j.cliser.2016.07.001.

De Cian, E., Lanzi, E., Roson, R. (2013). “Seasonal temperature variations and energy demand”. Climatic Change, 116, 805–825. http://dx.doi.org/10.1007/s10584‐012‐0514‐5.

De Cian, E. and Wing, I. S. (2017). Global energy consumption in a warming climate. Environ Resource Econ. 1–46. https://doi.org/10.1007/s10640‐017‐0198‐4.

Eskeland, G. S., Mideksa, T.K. (2010). “Electricity demand in a changing climate”. Mitigation Adaptation Strategies Global Change, 15, 877–897. http://dx.doi.org/10.1007/s11027‐010‐9246‐x.

European Commission (2018). In‐depth analysis in support of the commission communication com:(2018) 773.

Eurostat (2018). Eurostat Database on Energy Statistics. Available: https://ec.europa.eu/eurostat/web/energy/data. [Accessed September 2019].

Howell M., Rogner H.H. (2014). “Assessing integrated systems”. Nature Climate Change, 4 (4), pp. 246‐247, https://doi.org/10.1038/nclimate2180.

Kitous A., Després J., Assessment of the impact of climate change on residential energy demand for heating and cooling, EUR 29084 EN, Publications Office of the European Union, Luxembourg, 2018, ISBN 978‐92‐79‐77861‐2, https://doi.org/10.2760/96778, JRC108692.

Mideksa, T.K. and Kalbekken, S. (2010). “The impact of climate change on the electricity market: A review”, Energy Policy, 38(7); https://doi.org/10.1016/j.enpol.2010.02.035.

Pilli‐Sihvola, K., Aatola, P., Ollikainen, M., Tuomenvirta, H. (2010). “Climate change and electricity consumption—Witnessing increasing or decreasing use and costs?” Energy Policy, 38, 2409–2419. http://dx.doi.org/10.1016/j.enpol.2009.12.033.

Schaeffer, R. (2012). “Energy sector vulnerability to climate change: a review”. Energy 38:1–12.




6. Impacts on Tourism53

6.1. Theoretical Framework

Since the mid‐1990s, climate change and its effects on different sectors of the economy, in particular the tourism industry, has been one of the main topics in the economic literature. The impact of climate change on tourism is addressed either by its direct effects, such as increasing temperature, or the result caused by its secondary effects, such as rising sea levels (Rosselló‐Nadal, 2014). In order to study the impacts of climate change on tourism demand, there are different methods used in the literature. A large number of papers studies this impact by using quantitative methods, while a relatively limited part of the literature addresses the issue by using surveys and qualitative approaches (Steiger et al., 2017). Applying econometric methods, the former mostly uses climate indices and tourism demand models to investigate the impact of climate change on tourism. On the other hand, the latter relies on surveys and also experts’ opinions and predictions on the interaction of climate effects and the tourism industry to evaluate the relationship between climate change and tourism. To start a comprehensive research, one can primarily consult a significant part of the literature that critically reviews and summarizes works done in the field. Most of these studies the literature on general tourism activities using different climate change measures, such as the Tourism Climate Index (TCI) (Weaver 2011; Scott 2011; Scott et al. 2012; Becken 2013; Pang et al. 2013; Rosselló‐Nadal 2014), while some review studies concentrate only on one segment of the industry, e.g. winter tourism (Elsasser and Bürki 2002; Yang andWan 2010; Gilaberte‐Búrdalo et al. 2014; Steiger et al. 2017). In line with other review studies, Scott et al. (2012) suggest that since climate change will alter the competitiveness of tourist destinations, all destinations need to adapt to climate change by capitalizing on new opportunities in order to minimize the projected risks associated with local and global impacts of climate change. The latter group consistently concludes that as a reduction in the number of winters is projected due to climate change, a comprehensive multidisciplinary approach is needed to investigate the impacts of climate change on winter tourism. In addition, there are some literature reviews that mainly focus on surveys, qualitative data, decision making and perceptions regarding the projected effects of climate change on the tourism industry (Gössling and Hall 2006; Gössling et al. 2012). Reviewing the existing literature, Gössling and Hall (2006) claim that current models do not capture a wide range of aspects of climate change effects on tourism, which are necessary for achieving the understanding of tourist response in order to have accurate projections.

53 Milan Ščasný, Levan Bezhanishvili, Shouro Dasgupta are grateful for very valuable comments and advice on econometric modelling to Professor Anna Alberini, University of Maryland and Charles University in Prague.




We review over 50 papers in order to better understand how the weather‐related effects on tourism flows have been analysed in empirical studies. These results is presented in Appendix. This study contributes to that literature by analysing the effect of climate change on tourist demand for European countries, applying quantitative econometric analyses and using both monthly and annual data from EUROSTAT. We specifically aim at the effect of temperature and climate extremity on number of arrivals and nights spent. We analyse the effect of present temperature and extremes and a one‐year lag of these variables to examine the effect of past experience. We use monthly and annual country‐level data. While the former data are used to specifically analyse the effect on tourism in Europe during summer months (June to September), the latter approach allow us to analyse the aggregate effect during the entire year. In order to examine different patterns in visiting different countries, we analyse the effect of temperature on tourism for five regions in Europe, differing in climate and income.

This section is organized as follows. In the next section, the data used for analyses is discussed. In section 4.3, the descriptive statistics, methodology and results are presented, finally, in the last section conclusion is presented.

6.2. Data

6.2.1. Tourism and Socioeconomic Data

We use EUROSTAT country‐level data on tourism and socio‐economic characteristics. To measure the impact of climate change on tourism demand, we gathered monthly, quarterly, and annual tourism‐related data. Analyses based on monthly‐data is relying on tourism data recorded in “Arrivals at tourist accommodation establishments” and “Nights spent at tourist accommodation establishments” EUROSTAT’s databases, which depict number of arrivals and nights spent by residents and non‐residents in hotels; holiday and other short‐stay accommodation; camping grounds, recreational vehicle parks and trailer parks monthly in 38 European countries for the period 1990M01‐2019M07. Quarterly data records number of trips and number of nights spent by tourists (for the population aged 15 years and over) of which the main purpose is holidays or business, and which involve at least one or more consecutive nights spent away from the usual place of residence for the period 1996Q1 – 2011Q4 for 31 European countries.




Two types of annual data are available. The first type of database contains tourism data for European countries where tourists are either local residents or foreigners. The (country) origin of tourists is also known in the second database. Apart number of arrivals and number of nights spent by tourists, number of establishments and bed places are provided at NUTS 2 level for the period 1990‐1917. Besides tourism data, our econometric models are relying on socioeconomic data, including GDP (Gross domestic product at market prices, Chain linked volumes (2010), million euro), GDP in EURO PPS, and Population density (Number of inhabitants per square kilometre). Both databases are available on EUROSTAT website at NUTS 2 level for 2000‐2017. Socioeconomic data as well as data on tourism are unbalanced.

6.2.2. Climate Data

Our historical climatic data comes from the Global Land Assimilation System (GLDAS v2.1), this is a re‐analysed gridded climatic dataset, with 0.25° x 0.25° spatial and 3‐hourly temporal resolution. We begin with the daily temperature, precipitation, humidity data and compute the various aggregated indicators at the country and NUTS‐2 level.

ptotal ‐ total amount of precipitation over given period;

psummer ‐ total amount of precipitation for summer season;

tmax ‐ maximum temperature for a region (or country) over given period;

tmin ‐ minimum temperature for a region (or country) over given period;

tmean ‐ average temperature for a region (or country) over given period. Based on daily climate data we construct different climate extremity indices in order to check the effect of extremity events on tourism demand.54

HImax ‐ the maximum of daily Heat Index over the period.

THImax ‐ the maximum of daily Temperature Humidity Index (Sometimes they call it Discomfort Index (DI)) over the period.

THImin ‐ the minimum of daily Temperature Humidity Index over the period. (Sometimes they call it Discomfort Index (DI).

WSDI ‐ Warm spell duration index is defined as annual or seasonal count of days with at least 6 consecutive days when the daily maximum T exceeds the 90th percentile in the calendar 5‐day window for the base period 1979‐2009. (Data is provided only for NUTS2 annual level).

HI_Caution ‐ number of days where 27℃ < HI < 32.5℃.

HI_Ext_Caution ‐ number of days where 32.5℃ <= HI < 39.5℃.

54 For detailed description of the indexes, see, for instance, https://en.wikipedia.org/wiki/Heat_index, https://www.slideshare.net/omkarjoshi31521/why‐so‐discomfort‐discomfort‐index‐47964323, https://www.slideshare.net/omkarjoshi31521/why‐so‐discomfort‐discomfort‐index‐47964323.




Other weather‐related indexes tested in our modelling are: trange ‐ mean of daily (Tmax ‐ Tmin) over the period; summerdays ‐ number of days where Tmax > 25℃ over the period; tropnights ‐ number of days where Tmin > 20℃ over the period; stdmax ‐ std of Tmax over the period; stdmean ‐ std of Tmin over the period; HI_Danger ‐ number of days where 39.5℃ <= HI < 51.5℃; HI_Ext_Danger ‐ number of days where HI >= 51.5℃; DI_uncomf_exist ‐ number of days where THI <=14.9℃ or THI > 26.5℃; DI_uncomf_prop ‐ number of days where THI <=14.9℃ or THI > 26.5℃; DI_uncomf_exist_neg ‐ number of days where THI <=14.9℃; DI_uncomf_exist_posit ‐ number of days where THI >=26.5℃; DI_uncomf_prop_posit ‐ number of days where THI >=30.1℃.

6.3. Methodology and Results

6.3.1. Temperature effect on tourism: monthly‐data analysis

We use monthly data for summer, covering June to September 2000‐2016, for 29 European countries. Since we use in our analysis monthly data average temperature is 16.9 °C, maximum monthly (average) temperature is 30.8 °C. Average annual GDP is about 25,000 Euro, with a range at 6,200 to 77,300 PPS Euro. Countries vary in climate, geography, and area – on average there are 125 people per km2, see Table 6.1. Table 6.1 Sample descriptive statistics, monthly data, June‐Sept, 2000‐2016

Variable N Mean Std Min Max

Dependent variables nights [1000/month] 1820 11 859 17 783 101 87 171

arrivals [1000/month] 1820 3 327 4 725 43 23 389

Socio‐econ controls

GDP PPS pc 1536 24 804 11 430 6 200 77 300

GDP EUR2010 pc 1700 25 899 16 183 3 250 84 547

Population density 1812 125.71 105.86 2.80 503.10

Weather controls tmean 1700 16.86 4.66 ‐2.00 30.23

tmax 1700 30.78 6.76 4.37 46.07

precipitation 1700 74.97 40.31 0.00 247.79

HI_max 1700 24.09 5.77 1.91 38.37

THI_max 1700 21.58 4.04 2.78 28.53

THI_min 1700 10.03 4.47 ‐7.16 23.88

WSDI 1816 9.50 12.34 0.00 115.00

Regions NORTH 1820 0.21 0.41 0 1

WEST 1820 0.23 0.42 0 1

EAST 1820 0.22 0.41 0 1

SOUTH 1820 0.19 0.39 0 1




BALKAN 1820 0.15 0.35 0 1

The dataset records the number of arrivals and the number of nights spent in country of destination, without knowing from which country tourists are coming (i.e. country of origin). In order to examine different patterns in visiting countries we group all countries in five groups that differ in weather (warm, cold) and income. These five groups include NORTH, WEST, EAST, SOUTH, and BALKAN.55 As shown in table 2, countries grouped in the five groups differ with respect to economic performance (GDP per capita), population density, outdoor temperature, and weather in general (as measured by heat‐related indexes). Moreover, arrival flows and nights spent also vary considerably, see Table 6.2. Table 6.2 Descriptive country statistics by regions, summer months 2000‐2016, averages

Variable ALL NORTH WEST EAST SOUTH BALKAN

nights [1000/month] 11 936 8 452 18 497 2 689 25 628 3 956

arrivals [1000/month] 3 421 2 948 6 091 902 5 702 859

GDP PPS pc 24 482 32 611 38 731 17 114 22 914 13 110

GDP 2010 pc 26 482 38 998 41 640 10 553 21 558 7 758

Population, million 12.84 13.46 27.59 9.96 25.26 8.04

Population density, p/km2 219.60 71.88 238.11 84.38 113.32 84.14

Weather controls

tmean 16.65 11.99 15.71 16.79 21.83 19.08

tmax 30.52 23.36 31.03 30.14 37.80 33.07

max(tmean) 30.23 19.81 22.40 23.90 30.23 24.12

max(tmax) 46.07 33.96 41.43 40.32 46.07 43.15

Himax 23.87 18.00 24.33 23.50 29.83 25.98

THImax 21.44 17.29 22.15 21.43 24.86 22.85

THImin 9.89 6.56 8.79 10.82 12.64 12.04

WSDI 9.18 6.96 9.14 7.94 12.65 10.57

Attractiveness of country to visit depend on many factors, including cultural and natural characteristics and tourist infrastructure quality. Country area also determines its capacity to accept tourists. Larger countries have also more tourists’ opportunities and facilities, we disregard domestic, within‐country, tourists flow in presented analysis. As a result, number of tourist’s arrivals and nights spent vary considerably across countries. While only about 50,000 people a month arrived at some countries, more than 20 million people a month visited another one, giving average at 3–4 million arrivals a month in various years, see

55 NORTH include the Northern European country (Denmark, Island, Northern Ireland, Finland, Norway, Sweden, and the United Kingdom). WEST include countries in the Western Europe and close to the Alps (Austria, Belgium, France, Germany, Luxemburg, Netherlands, and Switzerland). EAST includes Baltic and Visegrad countries, specifically, the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, and Slovakia. SOUTH includes Cyprus, Greece, Italy, Portugal, and Spain. BALKAN includes Bulgaria, Croatia, North Macedonia, Romania, Slovenia, and Turkey.




Figure 6.1. The same tendency can be observed for nights pent – while the average is 11–14 million nights spend a month in various years, there are countries with 100,000 nights spent a month and other ones that record almost 90 million nights spent a month. Figure 6.1 Nights spent and arrivals per month and annual GDP, 2000‐2018

Arrivals and nights spent also vary across regions visited by tourists. Southern countries and countries in Western Europe were visited more frequently than countries in Eastern Europe and Balkan. Tourist flow is stronger in July and August, gets a bit lower in June, and is considerable smaller in September, see Figure 6.2. Figure 6.2: Arrivals and nights spent by visited regions and by summer months.

0

5.000

10.000

15.000

20.000

25.000

30.000

35.000

0

2.000

4.000

6.000

8.000

10.000

12.000

14.000

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

2014

2015

2016

2017

2018

Annual GDP, EUR2010 per capita, by year

nights and arrivals in 1000 per m

onth, by year

nights arrivals GDP EUR 2010 pc




Estimation results Using the monthly data on tourist’s destination, we estimate three groups of models with fixed effect on country of destination and time effects on year and summer months. Table X3 reports the results when we control for average monthly temperature, while Table X4 displays the estimation results when controlling for maximal monthly temperature. Table X5 then provides the results when weather is defined by heat‐measured indexes, such as THI, HI, or WSDI. In all models we measure tourist’s performance by either number of arrivals, or number of nights spent, both transformed in natural logarithm. Since we do not know from which country visiting tourists originate, we can’t control for socio‐economic characteristics of country of origin. GDP, therefore measures economic performance of country of destination that may also reflect how visited country is expensive. Better economic performance is also likely associated with better tourist infrastructure and more facilities that may hence make its places more attractive to visit. In all models we found that the economic performance of country of destination, as measured by GDP per capita, is positively associated with number of arrivals, however it is negatively associated with number of nights spent there. Following our prior intuition, the former effect may be due to better infrastructure that attract more people to visit a country, while the latter effect is due to higher price level in more wealthy countries. In both cases, absolute value of elasticity is about 0.10, indicating relatively small effect of economic performance of destination country on tourists flows. More populated countries are less attractive to be visited. Population density increased by 1 person per square‐kilometer reduces number of arrivals and nights spent by about 0.8–0.9 %. Arrivals and nights spent are the largest in July and August and tourism is weaker in September than in June. With respect to weather controls, we do not find statistically significant effect of precipitation. Effect of temperature on tourism has in most models an inverted U shape and it is similarly associated with both dependent variables. Figure 6.4. displays the effect of average and maximal monthly temperature on arrivals per month. In the case of average temperature, the turning‐point is a few degrees of Celsius after sample average (grey circle), implying that the curve is steeply declining in the whole area around current sample




maximum in average monthly temperature (red diamond). The effect of maximum in daily temperature over a month on arrivals is also negative after the average value, although its pace is not as steep as in the case of average monthly temperature, see Figure 6.3. Association between temperature and arrivals is not the same across the regions. In countries that are relatively colder (NORTH), the effect of increasing temperature is always positive, increasing attractiveness of their places (blue lines in Figure 6.4). In WEST, increase in maximum monthly temperature has stronger effect than an increase in average temperature (orange lines). We can observe similar same association between temperature and arrivals, as in WEST, also in EAST (grey lines). Tourists coming to countries grouped in EAST are in particular sensitive to changes in average temperature beyond 35 °C. Both yellow curve characterising visits in countries from SOUTH follow a declining trend and the effect of increasing average monthly temperature is higher than the effect of increasing its maximum. It seems temperature is not so important factor for travelling to countries in BALKAN group.




Table 6.3: Results, controlling for average monthly temperature, summer 2000‐2016

Dependent variable ln(arrivals) ln(arrivals) ln(arrivals) ln(nights) ln(nights)

Coeff Coeff Coeff Coeff Coeff

ln(GDP) 0.1017 * 0.0875 * 0.0874 * -0.1099 * -0.1265 **

population density -0.0079 *** -0.0081 *** -0.0081 *** -0.0089 *** -0.0088 ***

temp 0.0803 *** 0.0870 ***

temp-sq -0.0018 *** -0.0021 ***

temp*NORTH 0.0564 * 0.0523 *** 0.0032

temp-sq*NORTH -0.0001 0.0014

temp*WEST 0.0154 0.0152 0.0166

temp-sq*WEST -0.0006 -0.0006 -0.0003

temp*EAST 0.0757 *** 0.0755 *** -0.0099

temp-sq*EAST -0.0017 ** -0.0017 ** 0.0005

temp*SOUTH 0.0732 ** 0.0731 ** 0.2098 ***

temp-sq*SOUTH -0.0017 ** -0.0017 ** -0.0051 ***

temp*BALKAN 0.0218 0.0216 0.0002

temp-sq*BALKAN 0.0001 0.0001 0.0007

precipitation 1.12E-05 6.95E-06 7.11E-06 1.75E-05 3.23E-05

precipitation-sq -4.86E-09 -4.27E-09 -4.29E-09 -6.17E-09 -7.93E-09 **

July 0.2344 *** 0.2265 *** 0.2263 *** 0.4069 *** 0.3947 ***

August 0.2469 *** 0.2411 *** 0.2411 *** 0.4044 *** 0.3985 ***

September -0.0386 *** -0.0479 *** -0.0482 *** -0.0635 *** -0.0751 ***

constant 12.7486 *** 13.1715 *** 13.1790 *** 16.1219 *** 16.4921 ***

Fixed effects

Country Y Y Y Y Y

Years Y Y Y Y Y

No. obs. 1416 1416 1416 1416 1416

No. groups 29 29 29 29 29

F(24,1363) 139.28 114.23 118 154.27 124.34

F test (all ui=0) 1793.57 1231.61 1233.68 1582.32 1036.74

corr(u_i, Xb) -0.6048 -0.6342 -0.635 -0.629 -0.511




Table 6.4: Estimation results, controlling for maximum monthly temperature, summer 2000‐2016, FE model

Dependent var ln(arrivals) ln(arrivals) ln(arrivals) ln(nights) ln(nights)

Coeff Coeff Coeff Coeff Coeff

ln(GDP) 0.0992 * 0.0689 0.0692 -0.1094 * -0.1447 **

population density -0.0085 *** -0.0079 *** -0.0079 *** -0.0095 *** -0.0090 ***

max(t) 0.0598 *** 0.0613 ***

max(t)-sq -0.0009 *** -0.0009 ***

max(t)*NORTH 0.0633 ** 0.0236 *** 0.0373

max(t)-sq*NORTH -0.0008 -0.0003

max(t)*WEST 0.0664 *** 0.0655 *** 0.1301 ***

max(t)-sq*WEST -0.0012 *** -0.0012 *** -0.0022 ***

max(t)*EAST 0.1042 *** 0.1034 *** 0.0396

max(t)-sq*EAST -0.0016 *** -0.0016 *** -0.0006

max(t)*SOUTH -0.0810 *** -0.0810 *** -0.0701 **

max(t)-sq*SOUTH 0.0010 ** 0.0010 ** 0.0008 *

max(t)*BALKAN 0.0007 0.0000 -0.0082

max(t)-sq*BALKAN 0.0002 0.0002 0.0004

precipitation -7.55E-06 -7.19E-06 -7.49E-06 1.10E-05 1.31E-05

precipitation-sq -2.88E-09 -3.09E-09 -2.94E-09 -5.55E-09 -5.82E-09

July 0.2699 *** 0.2630 *** 0.2619 *** 0.4302 *** 0.4237 ***

August 0.2761 *** 0.2707 *** 0.2705 *** 0.4233 *** 0.4184 ***

September -0.0620 *** -0.0590 *** -0.0611 *** -0.0816 *** -0.0806 ***

constant 12.7029 *** 13.4898 *** 13.5978 *** 16.0430 *** 16.9911 ***

Fixed effects

Country Y Y Y Y Y

Years Y Y Y Y Y

No. obs. 1416 1416 1416 1416 1416

No. groups 29 29 29 29 29

F(24,1363) 135.74 110.45 113.92 151.71 121.86

F test (all ui=0) 1312.89 1123.03 1123.44 1183.9 895.27

corr(u_i, Xb) -0.6331 -0.8264 -0.8325 -0.6522 -0.8021




Figure 6.3: Effect of temperature change on arrivals, in 1000 per month.

Note: Red diamonds indicate maximum in maximum monthly temperature and average temperature, respectively, while grey circles represent average value of maximum or average temperature across countries.

0,0

0,5

1,0

1,5

2,0

2,5

3,0

10 15 20 25 30 35 40 45 50

arrivals (in 1000 a m

onth)

average/max temperature, summer [°C]

arrivals (temp) nights (temp) arrivals (tmax) nights (tmax)




Figure 6.4: Association between arrivals and temperature, by regions of destination

Panel A – average monthly temperature

Panel B – maximal monthly temperature

Note: Red diamonds indicate maximum in maximum monthly temperature and average temperature, respectively, while grey circles represent average value of maximum or average temperature across countries.

Elasticity measures how sensitive tourists flows (arrivals, nights spent) are with respect to changes in temperature. For instance, elasticity at ‐0.30 indicates that 10% change in temperature result in decrease in tourism by 3 percentage points. Due to quadratic specification of temperature‐related variables, elasticity of demand for arrivals, and nights spent, respectively, is a function of temperature. It means that, for instance, elasticity of ‐0.80 at current temperature maximum (say 30 °C) indicates that increasing this maximum by 5 % (by 1.5 °C) may result in a decrease in tourism flows by 4.0 %.

0

1

2

3

4

5

6

7

10 15 20 25 30 35


onth)

average temperature, summer [°C]NORTH WEST EAST SOUTH

0

1

2

3

4

5

6

20 25 30 35 40 45 50

arrivals (1000 a m

onth)

max temperature, summer [°C]

NORTH WEST EAST SOUTH




Using the estimates reported in tables 2‐3, we can derive the elasticity estimates. We found the association follows an inverted U‐shape, and this form is similar for both temperature variables and both dependent tourism variables. Specifically, in the case of maximum in monthly temperature (upper curves), elasticity is positive and declining to zero around level of 35 °C (that is current maximum in NORTH countries), from that elasticity is negative and declining even more steep. For instance, around the maximum level at 41 °C, that is the maximum observed in the past in countries in WEST and EAST, elasticity is ‐0.54 for arrivals, and ‐0.59 for nights spent, respectively, see Figure 6.5. Tourists flows, and in particular nights spent, are even more sensitive on changes in average monthly temperature. Elasticity at the level of maximums for NORTH, around 20 °C, is still positive, approximately +0.18 for arrivals, and +0.05 for night spent, respectively. For higher average temperatures, elasticity is getting negative and steeply declines. For 22–24 °C, that is the maximum of average temperature in WEST, EAST, and BALKAN, and country average for average temperature in SOUTH, the elasticity is small but already negative (up to ‐0.13, and ‐0.35, respectively). In the area around 30 °C elasticity gets larger value, around ‐0.80 for arrivals, and ‐1.20 for nights, respectively. We note that 30 °C is current maximum in average temperatures in SOUTH. Figure 6.5: Elasticity of arrivals and nights with respect to changes in temperature

Note: For semi‐log model specification implies the elasticity as (a*T) for linear specification, lnY=a*T, and (a+2*b*T)*T for a quadratic form, lnY=a*T+b*T^2, where T is weather control, and a and b are coefficients to be estimated.

Lagged effect Table 6.5. reports the results for fixed effect models when controlling for the weather variables lagged by one year. Panel A displays the results for number of arrivals and panel B shows the results for nights spent.

‐2.2

‐1.7

‐1.2

‐0.7

‐0.2

0.3

0.8

10 15 20 25 30 35 40 45 50

temperature, summer °C

elasticity ‐ arrivals(tmax) elasticity ‐ nights_spent(tmax)

elasticity ‐ arrivals(temp) elasticity ‐ nights_spent(temp)




The results are qualitatively similar as the results presented in the models without lags. Effect of GDP in a country of destination on arrivals is positive and significant, with implied elasticity at 0.14–0.15, while coefficient for GDP in models with nights spent is negative but not significant at any convenient level. Population density makes places less attractive with respect to arrivals and nights spent. Additionally, to maximum in daily temperature and average temperature over a month, we also control for weather indices, such as HI max, THI, HI Caution and HI Extreme Caution. In all models we support a quadratic form (an inverted U‐shape) of association between weather measures and tourism variables, with exemption for HI Caution. Association between temperature lagged by one year and arrivals follows similar trend as temperature controls without lags, as shown in Figure 6.6 (with lag) and Figure 6.3 (no lag), respectively. Figure 6.6: Temperature lagged by 12 months and arrivals

0,0

0,5

1,0

1,5

2,0

2,5

3,0

10 15 20 25 30 35 40 45 50


onth)

average/max temperature, summer [°C]

arrivals (temp) tmean arrivals (tmax) tmax




Table 6.5: Estimation results for lagged weather‐controls, FE model, summer months 2000‐2016. Panel A ‐ number of arrivals

arrivals tmax temp HI_max THI HI_Caution HI_ExtremCaution

Coeff. Coeff. Coeff. Coeff. Coeff. Coeff.

lngdp 0.1507 *** 0.1449 *** 0.1355 *** 0.1437 *** 0.1405 *** 0.1453 ***

popdenst ‐0.0083 *** ‐0.0078 *** ‐0.0081 *** ‐0.0081 *** ‐0.0081 *** ‐0.0082 ***

L12_weather 0.0536 *** 0.0770 *** 0.0629 *** 0.1365 *** ‐0.0036 ‐0.0689 **

L12_weather‐sq ‐0.0008 *** ‐0.0018 *** ‐0.0012 *** ‐0.0031 *** ‐0.0001 0.0073 *

month

July 0.2732 *** 0.2429 *** 0.2745 *** 0.2764 *** 0.2793 *** 0.2749 ***

August 0.2775 *** 0.2530 *** 0.2782 *** 0.2777 *** 0.2815 *** 0.2777 ***

September ‐0.0635 *** ‐0.0419 *** ‐0.0646 *** ‐0.0687 *** ‐0.0897 *** ‐0.0888 ***

constant 12.2796 *** 12.3681 *** 12.4856 *** 11.7109 *** 13.2215 *** 13.1891 ***

Fixed effects

Country Y Y Y Y Y Y

Year Y Y Y Y Y Y

No. obs. 1472 1472 1472 1472 1472 1472

No. groups 29 29 29 29 29 29

F(24,1363) 171.07 175.62 172.34 172.28 165.74 164.97

F test (all ui=0) 3798.57 4540.63 3941.35 3865.24 4257.55 4389.96

corr(u_i, Xb) ‐0.6105 ‐0.59 ‐0.6064 ‐0.6049 ‐0.6044 ‐0.61

R‐sq

within 0.7311 0.7274 0.7274 0.7273 0.7196 0.7186

between 0.0771 0.0791 0.0791 0.0808 0.0749 0.0765




overall 0.0545 0.0581 0.0581 0.0587 0.0563 0.0579

Panel B –nights spent

nights tmax temp HI_max THI HI_Caution HI_ExtremCaution

Coeff. Coeff. Coeff. Coeff. Coeff. Coeff.

lngdp ‐0.0488 ‐0.0566 ‐0.0640 ‐0.0565 ‐0.0724 ‐0.0578

popdenst ‐0.0080 *** ‐0.0074 *** ‐0.0078 *** ‐0.0077 *** ‐0.0076 *** ‐0.0078 ***

L12_weather 0.0517 *** 0.0833 *** 0.0577 *** 0.1216 *** ‐0.0054 ‐0.1049 ***

L12_weather‐sq ‐0.0008 *** ‐0.0020 *** ‐0.0011 *** ‐0.0029 *** ‐0.0003 * 0.0099 **

month

July 0.4302 *** 0.4089 *** 0.4341 *** 0.4386 *** 0.4435 *** 0.4337 ***

August 0.4229 *** 0.4061 *** 0.4253 *** 0.4267 *** 0.4351 *** 0.4257 ***

September ‐0.0818 *** ‐0.0634 *** ‐0.0875 *** ‐0.0961 *** ‐0.1082 *** ‐0.1068 ***

constant 15.4495 *** 15.5027 *** 15.7082 *** 15.0661 *** 16.4707 *** 16.3632 ***

Fixed effects

Country Y Y Y Y Y Y

Year Y Y Y Y Y Y

No. obs. 1472 1472 1472 1472 1472 1472

No. groups 29 29 29 29 29 29

F(24,1363) 182.29 186.79 182.71 182.13 185.24 179.46

F test (all ui=0) 2974.76 3815.36 3071.84 3005.65 3766.77 3868.37

corr(u_i, Xb) ‐0.5971 ‐0.5743 ‐0.5955 ‐0.5963 ‐0.5868 ‐0.5973

R‐sq

within 0.7384 0 0 0.7382 0 0

between 0.1059 0 0 0.1158 0 0




overall 0.0574 0 0 0.0615 0 0



6.3.2. Temperature and climate extremes effect on tourism: country‐pairs analysis

We analyse annual data for 2000‐2016 period, for 37 European countries. This dataset records the number of nights spent in country of destination, while the country of origin of tourists is also reported. For this specific database, we run a classical regression model for temperature and precipitation as well as a model where the covariate main of interest is a climate extreme index. Our goal is to check whether and how extreme weather influences on tourism demand. The basic econometric model is as follows:

ln 𝑛𝑖𝑔ℎ𝑡𝑠 𝛼 ∙ ln 𝐺𝐷𝑃𝑑 𝛽 ∙ ln 𝐺𝐷𝑃𝑜 𝛾 ∙ ln 𝐷𝑒𝑛𝑠𝑖𝑡𝑦 𝛿 ∙ TEMPθ ∙ 𝑇𝐸𝑀𝑃 𝜗 ∙ prec μ ∙ prec 𝑢 𝑌𝐸𝐴𝑅 𝜀

where i denotes country of destination, while j is country of origin. Subscript t is time (year). The term 𝑢 is country origin‐destination fixed effects (defined by country‐

pairs) and 𝑌𝐸𝐴𝑅 is time effect (years). Natural logarithm of the dependent variable – nights spent – is used. GDPs for origin as well as for destination countries are included in the model. It is expected that GDPs ought to be positively correlated to number of nights tourist spent. In addition, we control the regression by population of density per square kilometre of country of destination. The two main climate variables (TEMP) are temperature (tmax, or tmean) and precipitation (prec). We assume a quadratic relationship between weather variables and tourism dependent variable. We estimate panel data model with Fixed Effects on country‐pairs and years (FE panel model fits our data better than RE model, supported by Hausman test). The below Table 6 displays the results for model analysing number of nights, using weather controls defined by average temperature and maximum temperature, respectively. Figures 1 and 2 then displays corresponding effect of temperature change on absolute number of nights and related elasticity that is temperature‐specific. In brief, we find that max temperature and total precipitation significantly affect amount of nights spent by tourists. For both weather variables, association follows an inverted U‐shaped form, indicating on a turning‐point in the temperature‐tourist nights relationship. The results are convincing since the max temperature and total precipitation affect positively tourists’ decisions to spend more nights at tourist destinations only until certain point, while, beyond 40℃ of annual maximum temperature and 27,000 mm of total precipitation, hotter and wetter conditions have negative effect on nights spent. In other words, extreme hot and abundant precipitation are disliked by tourists,



resulting in shortening length of vacation time. Contrary to maximum temperature, tourists flow is not sensitive with respect to mean temperature, see Table 6.6. GDP per capita in country of origin positively affects length of vacation, with implicit income elasticity at 1.15. Economic performance of country of destination, as measured by its GDP per capita, does not affect tourists’ decisions. Tourists are also sensitive to population density; increasing density by 10 % would reduce length of tourist stay, ceteris paribus, by about 18.5 %. Table 6.6: Results – number of nights, country‐pairs FE model Dependent variable ln(nights) ln(nights) Coeff Coeff

ln(GDP_destin) ‐0.0202 ‐0.0072 ln(GDP_origin) 1.1473 *** 1.1463 *** population density ‐1.8994 *** ‐1.8711 *** Tmax 0.0413 ** tmax‐sq ‐0.0005 ** Tmean ‐0.0078 tmean‐sq 0.0007 Precipitation ‐8.82E‐05 ‐0.0001 precipitation‐sq 1.13E‐7 1.18E‐07 Constant 7.7779 *** 8. 3428 *** Fixed effects Country‐pairs Y Y Years Y Y

No. obs. 7765 7765 No. Groups 870 870 F(22,6873) 100.41 99.86 F test (all ui=0) 335.09 372.38 corr(u_i, Xb) ‐0.6845 ‐0.6812 R‐sq Within 0.2432 0.2414 Between 0.0471 0.0465 Overall 0.0443 0.0436

Significance level *** <1%, ** <5%, * <10%



Figure 6.7: Results – nights spent and maxim temperature, country destination‐origin FE, year‐time effect Panel A – Effect on temperature on number of nights (in thousands)

Panel B – elasticity of number of nights with respect to temperature change



Further, we analyse the effect of climate extremity on nights spent. The underlying model is similar as the one above, using country pairs fixed effects and year time effect:

ln 𝑛𝑖𝑔ℎ𝑡𝑠 𝛼 ∙ ln 𝐺𝐷𝑃𝑑 𝛽 ∙ ln 𝐺𝐷𝑃𝑜 𝛾 ∙ ln 𝐷𝑒𝑛𝑠𝑖𝑡𝑦 𝛿∙ 𝐸𝑋𝑇𝑅𝐸𝑀 𝑢 𝑌𝐸𝐴𝑅 𝜀

Similar to the above model the natural logarithm of the number of tourist arrivals is used. GDPs for origin and destination countries are used in the model. Moreover, we control the regression model by population of density per square kilometre of country of destination. Climate extremity (EXTREM) is measured by two indexes. First one is WSDI that has been widely used as a weather extremity index in different econometric methods in order to analyse the effect of weather extremes on various dependent variables. WSDI is Warm Spell Duration Index and it is defined as annual (or seasonal) count of days with at least six consecutive days when the daily maximum T exceeds the 90th percentile in the calendar 5‐day window for the base period 1979‐2009. Second one is Heat Index. This index is defined as the number of days where the index is between 27℃ and 32.5℃. Since, this temperature may cause fatigue with prolonged exposure and/or physical activity, this weather state is classified as “Caution”, therefore HI_Caution. Table 6.7 displays the results for Fixed Effect model. Again, GDP per capita in destination country does not affect number of nights spent by tourists, while GDP in country of origin is positively and significantly associated with tourists stay, with income elasticity at around the unity. WSDI has negative impact on tourists’ decision with respect to how many nights to spend in tourist destinations. It means the higher WSDI the less nights tourists spent on visiting places. The effect of second extremity index – HI_Caution – is even stronger affecting nights spent negatively. In addition to analysing the current weather condition on tourists’ behaviour, we investigate whether weather condition in past influences current tourists’ decision. We therefore estimate the panel FE model, controlling for WSDI, and HI_Caution, respectively, lagged by one year. Table 6.7 reports these results. Similarly, the results we obtained for WSDI and HI in the year when tourist visit was made show that lagged extremity indices are negatively correlated to future tourist decisions and their lagged effect is almost identical to the current one. Both these models clearly indicate the negative impact of extreme weather conditions on the decision of tourists.



Table 6.7: Result for country‐pair FE model, climate extremity indexes

dependent variable ln(nights) HI no lag

ln(nights) WSDI no lag

ln(nights) HI lagged

ln(nights) WSDI lagged

Coeff Coeff Coeff Coeff

ln(GDP) ‐0.0246 ‐0.0623 ‐0.1255 ‐0.1043 *

ln(oGDP) 1.1414 *** 1.0970 *** 0.9591 *** 1.0930 ***

population density ‐1.7366 *** ‐1.8856 *** ‐1.4201 *** ‐1.5300 ***

extremity index ‐0.0039 *** ‐0.0012 ** ‐0.0036 *** ‐0.0013 ***

constant 7.9867 *** 9.6009 *** 9.3816 *** 8.5291 ***

Fixed effects

Country pairs Y Y Y Y

Years Y Y Y Y

No. obs. 7765 8884 8051 9200

No. groups 870 899 870 899

F stat 116.06 136.19 117.18 152.02

F test (all ui=0) 390.83 413.56 427.92 456.87

corr(u_i, Xb) ‐0.664 ‐0.7097 ‐0.617 ‐0.6524

R‐sq

within 0.2428 0.2548 0.2371 0.2783

between 0.0464 0.0313 0.049 0.0297

overall 0.0443 0.0301 0.045 0.0275

6.4. Conclusion

The impacts of climate change on the tourism industry has been a subject of research for more than two decades. The need of the tourism industry to forecast the possible outcomes of a climate change scenario (which is almost inevitable), has led to a considerable bulk of literature, that we review, see Appendix. The presented research aims to summarize the current knowledge about the impacts of the climate change on the tourism industry by exploring the results coming from the empirical literature when various modelling approaches were used. In this particular study, we analyse econometrically the effect of climate change on tourist demand for European region over more than last 15 years (2000‐2016) based on monthly and annual country‐level data provided by EUROSTAT. Based on the data type and structure two major regression models are discussed in the report. Approach 1 considers the tourism data recorded for the destination of tourists only, while in Approach 2 country of origin is additionally indicated. Approach 2 can however be performed on annual basis, since the data are available for years and country. Data are richer for Approach 1; this analysis can be performed on monthly,



quarterly, or annual basis, analysing aggregated the data for countries or, for selected variables, also for NUTS2 regions. In Approach 1, in order to examine different patterns in visiting countries, we group European countries in five different groups that differ in weather (warm, or cold) and income. These five groups include NORTH, WEST, EAST, SOUTH, and BALKAN. This analyses specifically aims at the effect of temperature and climate extremes on tourisms in Europe during summer months, June to September. Applying econometrical analyses, we do not find statistically significant effect of precipitation on tourism flows, either measured by number of arrivals or nights spent. Effect of temperature has in most models an inverted U‐shape form, indicating the turning point in the temperature‐tourism relationship. We found that economic performance in visiting country, as measured by GDP per capita in country of destination, is weakly but positively associated with number of arrivals, however, places with higher income make the lengths of tourists visit shorter, most likely due to higher price level. Countries with higher population density are also less attractive to be visited. We do not find statistically significant effect of precipitation. However, we find an inverted U‐shape association between temperature and tourism variables in almost all of our models. This shape is also found for the weather controls measuring weather extremity. In the case of average monthly temperature, the turning‐point is a few degrees of Celsius beyond sample temperature average, implying that the effect of increasing average temperature is negative and steeply declining in the whole area around present sample temperature maximum. The effect of maximum temperature is also negative, although its pace is not as steep as in the case of average monthly temperature. Elasticity measures how sensitive tourists flows (arrivals, nights spent) are with respect to changes in temperature. Due to the model specification used, magnitude of our elasticity estimate depends on the magnitude of temperature. In the case of maximum temperature, we get the elasticity that begins to be positive and then declining to zero around level of 35 °C (present maximum in NORTH), from that point the elasticity becomes negative and declining quiet steep. Around the maximum level at 41 °C (the maximum observed in the past in WEST and EAST), the elasticity is ‐0.54 for arrivals, and ‐0.59 for nights spent, respectively. Elasticity has similar shape for average temperature. It gets small but positive values up to 20 °C (the maximum levels in NORTH), it becomes negative for higher magnitudes of average temperatures, and then steeply declines as average temperature increases. For instance, in the area around 30 °C, the elasticity is around ‐0.80 for arrivals, and ‐1.20 for nights, respectively. Since current maximum in average temperatures in SOUTH is close to 30 °C so it is reasonable to expect larger decreases in tourism flows in particular in countries located in the Southern Europe as the effect of climate change.



The association between temperature and tourist arrivals is not, however, same across the five regions. In countries that are relatively colder (NORTH), the effect of increasing temperature is always positive, increasing attractiveness of their places. In WEST region, an increase in maximum temperature has stronger effect than an increase in average temperature. And we can observe similar association between temperature and arrivals in both WEST as well as EAST European region. Tourists coming to countries grouped in EAST are in particular sensitive to changes in average temperature beyond 35 °C. Visits in countries from SOUTH are declining with respect to increasing temperature, and the effect is higher for increasing average monthly temperature than for increasing temperature maximum. It seems temperature is not important factor for travelling to countries in BALKAN. The econometrical analyses in Approach 2, i.e. controlling for country‐pairs effects, similarly indicates an inverse U‐shaped dependence of the number of nights spent by tourists on maximum annual temperature. These findings also hold for total annual rainfall and number of nights. This explains that, increasing maximum temperature and precipitation leads to an increase in the number of nights spent at tourist destination to a certain threshold point. Hitting the threshold causes decrease in the duration of tourists visits. Based on the results the threshold for maximum temperature is 40℃ while for total annual rainfall it is 27,000 mm. Apart observing the sensitivity of tourists’ behaviour towards temperature and precipitation, we have also analysed the influence of weather extremes on the duration of visits. It appears that some of the extreme indices have significant negative effect on the amount of holiday nights.



References:

Becken, S. (2013). A review of tourism and climate change as an evolving knowledge domain. Tourism Management Perspectives, 6, 53‐62.

Dubois, G., & Ceron, J. P. (2006). Tourism and climate change: Proposals for a research agenda. Journal of Sustainable Tourism, 14(4), 399‐415.

Elsasser, H., & Bürki, R. (2002). Climate change as a threat to tourism in the Alps. Climate research, 20(3), 253‐257.

Gilaberte‐Búrdalo, M., López‐Martín, F., Pino‐Otín, M. R., & López‐Moreno, J. I. (2014). Impacts of climate change on ski industry. Environmental Science & Policy, 44, 51‐61.

Gössling, S., & Hall, C. M. (2006). Uncertainties in predicting tourist flows under scenarios of climate change. Climatic change, 79(3‐4), 163‐173.

Gössling, S., Scott, D., Hall, C. M., Ceron, J. P., & Dubois, G. (2012). Consumer behaviour and demand response of tourists to climate change. Annals of tourism research, 39(1), 36‐58.

Pang, S. F., McKercher, B., & Prideaux, B. (2013). Climate change and tourism: An overview. Asia Pacific Journal of Tourism Research, 18(1‐2), 4‐20.

Rosselló‐Nadal, J. (2014). How to evaluate the effects of climate change on tourism. Tourism Management, 42, 334‐340.

Scott, D. (2011). Why sustainable tourism must address climate change. Journal of Sustainable Tourism, 19(1), 17‐34.

Scott, D., Gössling, S., & Hall, C. M. (2012). International tourism and climate change. Wiley Interdisciplinary Reviews: Climate Change, 3(3), 213‐232.

Steiger, R., Scott, D., Abegg, B., Pons, M., & Aall, C. (2019). A critical review of climate change risk for ski tourism. Current Issues in Tourism, 22(11), 1343‐1379.

Weaver, D. (2011). Can sustainable tourism survive climate change?. Journal of sustainable Tourism, 19(1), 5‐15.

Yang, J., & Wan, C. (2010). Progress in research on the impacts of global climate change on winter ski tourism. Advances in climate change research, 1(2), 55‐62.



Appendix: Effects of Climate Change on Tourism: A Review

118

Appendix

Effects of Climate Change on Tourism:

A Review

1 Weather measures induced by Climate Change

1.1 Single Measures

Climate is a broad term and there are various measures used to present for it. The most

common one is the temperature-related measure. Some authors used annual average tem-

perature (Berrittella et al. 2006; Hamilton et al. 2005a,b; Bigano et al. 2005, 2006; Hamilton

and Tol 2007). while others preferred more specific details of temperature including: av-

erage temperature of the warmest month(Lise and Tol 2002), mean temperature in August

CET (Agnew and Palutikof 2006), minimum and maximum temperatures (Hein et al. 2009;

Rosselló-Nadal 2014; Priego et al. 2015; Maddison 2001), global average temperature (Braun

et al. 1999). Additionally, some papers did not use temperature to measure climate change,

they also used precipitation (Michailidou et al. 2016; Lépy et al. 2014; Steiger 2010; Den-

stadli et al. 2011; Nyaupane and Chhetri 2009), rainfall (Eugenio-Martin and Campos-Soria

2010; Rosselló-Nadal 2014) or sea-level rise (Bigano et al. 2008; Scott et al. 2012).

From other perspective, some authors investigated the importance of climate change on

winter tourism therefore they used either snow-related measures (Hoffmann et al., 2009;

Falk, 2010) or combined two important indicators including temperature and snow depth

(Yang and Wan, 2010; Tervo, 2008; Steiger et al., 2017; Falk, 2010; Scott et al., 2006; Elsasser

and Bürki, 2002).

It is clear that temperature and other measures can somehow reflex climate situations

and its impact on tourism industry. Several papers used temperature-related measures

found out that temperature does not have a reverse impact on tourism but it might cause

119

a gradual shift of tourism destination (Hamilton et al., 2005a; Berrittella et al., 2006; Mad-

dison, 2001). It might imply that an increase in temperature can bring negative impacts to

some regions but can generate positive impacts to other regions. More interestingly, an in-

crease in temperature might lead to a decrease in snow depth, but it seems that ski tourism

is slightly affected by this (Scott et al., 2006; Falk, 2010). This can be partly explained that

services suppliers aware about possible climate change and they can apply some measures

to prevent potential losses (Hoffmann et al., 2009; Elsasser and Bürki, 2002). Generally,

usage of single index or combination of some single indexes might be appropriate in some

specific cases but it is unlikely to generalize it.

1.2 Climate Composite Index

Tourism Climate Index Among several indices developed over the last decades to investigate the suitability of cli-

mate for tourism, the most used one in the literature is Mieczkowski (1985)’s Tourism Cli-

mate Index (TCI). This index systematically assesses the most important climatic factors

regarding the quality of the tourism experience. It uses the widely available monthly data

relevant to the tourist destinations. The index consists of 7 climate variables (sub-indices):

monthly mean for maximum daily temperature and minimum of daily relative humidity

(CID); mean daily temperature and mean daily relative humidity (CIA), total precipitation

(P), total hours of sunshine (S), and average wind speed (W), where the relative weightings

of sub-indices are 40%, 10%, 20%, 20%, and 10%, respectively. Hence, the index has the

following expression:

TCI = 2[(4xCID) + CIA + (2xP) + (2xS) + W ]

Moreover, Mieczkowski (1985) suggests a standardized rating system ranging from impossi-

ble (-30) to ideal (100), to provide a common basis for the better interpretation of the index

(Table 1). Despite its limitations, TCI is still used as the most common index in the litera-

ture relevant to tourism, and scoping analysis of possible changes in the climate resources

for tourism under climate change.

Using TCI, Scott et al. (2004) study the spatial and temporal distribution of climate re-

sources for tourism in North America under baseline conditions from 1961 to 1990. The

authors also project two climate change scenarios for 2050s and 2080s. The climate change

120

TCI Scors Category -30 to 0 Impossible 10 to 19 Extremely unfavorable 20 to 29 Very unfavorable 30 to 39 Unfavorable 40 to 49 Marginal 50 to 59 Acceptable 60 to 69 Good 70 to 71 Very good 80 to 89 Excellent 90 to 100 Ideal

Table 1: Rating categories for TCI

scenarios are obtained from the Canadian Climate Impact Scenarios Project (CCIS 2002)

based on climate change experiments conducted at international climate modeling centers.

These scenarios are constructed in line with the recommendations of the Intergovernmental

Panel on Climate Change (IPCC). The authors conclude that based on analysis a substantive

redistribution of climate resources for tourism is possible as a result of the climate change,

particularly for the warmer scenario. They also show that Canada and northern USA would

benefit form the projected climate change, which means a longer warm-weather tourism

season would develop the tourism industry in these regions. On the other hand, shorter

and warmer winters based on the climate change projection would reduce the impetus for

Canadians to travel to the warmer destinations. Therefore, there will be an increase in the

number of destinations for short-term sun holidays in these regions.

Amelung and Viner (2006) investigate the suitability of tourism in the Mediterranean

region using TCI and future climate change scenarios in 2020s, 2050s, and 2080s. They use

two different data sets containing monthly climatic data (1961-1990) to do the scenarios

and analyze the results. The authors study four scenarios, which are based on the Special

Report on Emissions Scenarios (SRES) done by IPCC. These scenarios indicate an estimation

of increasing global temperature by 1.5 ◦C to 5.8 ◦C during 21st century. However, the

authors use a different classification of TCI distributions adopted from Scott and McBoyle

(2001), rather than what is presented in Table 1, for analyzing seasonal TCI patterns. For

example, if all monthly ratings are over 80, the distribution qualifies as optimal. They find

a potentially large impact of climate change on tourism in both positive and negative sense.

According to the projections, in spring and autumn most of the Mediterranean region will

face an improvement in TCI rating, particularly Spain, Greece, and Turkey. In addition,

TCI score in the Mediterranean region will deteriorate, whereas there will be an improving

121

conditions in western and northern Europe.

Hein et al. (2009) review some advances in modeling the effect of climate change on

the tourism industry with a particular focus on the case study of Spain. First, the authors

use TCI in order to analyze the current and next 50 years of climate suitability of Spain,

based on existing climate change models and scenarios. Second, Hein et al. (2009) model

the potential change in the number of tourists in Spain as a function of climate change.

Furthermore, they use monthly tourism data from 2004 to establish the baseline model in

order to forecast tourist flows in 2060, based on climate change projections for the period

2051-2080. The authors use relative attractiveness of a region and climatic changes (TCI) to

model tourist flows addressed as visitor nights spent by foreign visitors. The paper shows

that there is an insignificant difference between present situation and the projected winter

TCI in 2060 in Europe, while Summer will face a significant change in climatic scores. On

the basis of these projected climatic changes, Hein et al. (2009) forecast a decrease of 5% to

14% in the number of tourists visiting Spain in 2060.

In another paper, Amelung and Nicholls (2014) assess the effect of projected climate

change on the Australia’s tourism industry by using TCI. The authors use the same datasets

and scenario analysis approach for 2020s, 2050s, and 2080s, as other relevant studies. Based

on analysis, Northern regions of Australia are projected to face deterioration in their climatic

suitability in large periods of the year, Australia’s summer months (Dec-Feb) in particular.

In contrast, southern parts are projected to have substantial improvements in their climatic

conditions, which assure suitable weather for general tourism activities. Similarly, using the

Tourism Climatic Index, Nicholls and Amelung (2015) study the effects of climate change

on non-winter rural tourism in the Nordic region. They conclude that there will be a consid-

erable increase in the climatic attractiveness of the southern and eastern parts of the region.

Grillakis et al. (2016a), based on the 2 ◦C global warming scenario, conclude that the

climate is expected to be more favorable for outdoor activities for the majority of the Eu-

ropean countries. Mediterranean countries are projected to encounter the largest decrease

in TCI score, unlike the other parts of Europe. However, they are still projected to exhibit

higher TCI scores compared to other European countries. Therefore, it means that under

2 ◦C global warming, there will be an increase in the climate competitiveness of the other

European destinations.

TCI and its related results exhibit some drawbacks. Monthly timescale, for instance,

can be a drawback of the index since it cannot capture the extreme climatic events which

can affect the tourism industry for an entire tourist season, even though they are smoothed

122

through the monthly averaging. Another shortcoming is the subjective scoring of each sub-

index that might be different over time and also among different regions. Eugenio-Martin

and Campos-Soria (2010) point out three main drawbacks of Mieczkowski’s TCI. First, since

it is designed with respect to an average tourist and general tourist activities, it can not

distinguish tourists/destinations. Second, the assessment of sub-indices depends on under-

standing of thermal comfort that requires further study. Third, the subjective weights are

used for aggregation of the index.

There are two main data sources to compile TCI: the CRU CL 1.0 and the Hadley Cen-

tre’s HadCM3 circulation model (GCM). The first one, CRU CL 1.0, includes mean monthly

surface climate over all global areas from 1961 to 1990 which are in grid-based form. The

following variables are available: precipitation and frequency of wet-day, mean tempera-

ture and foundation to calculate minimum and maximum temperature, vapour pressure,

sunshine, cloud, frequency of frost and wind speed. This data set is considered the best

available source Amelung and Nicholls (2014). It only covers the period 1961-1990, how-

ever. Therefore, it is necessary to combine with Hadley Centre’s HadCM3 general circulation

model which were used to predict some changes in climate during three periods: the 2020s,

the 2050s and the 2080s. There are four scenario families A1, A2, B1 and B2 which include

forty individual scenarios.

Other Composite Indices

Addressing subjectivity issue in TCI method, Scott et al. (2008) introduce Climate Index for

Tourism (CIT). This index integrates all features of TCI with other important sub-indices

relevant to beach tourism, such as heavy rain and strong winds. CIT is also calibrated

against questionnaires to resolve the subjectivity issue in the rating scales of TCI.

Yu et al. (2009) introduce and develop a Modified Climate Index for Tourism (MCIT) to

overcome limitations of previous TCI and CIT. The modified index uses hourly data unlike

the previous indices. Moreover, two elements of visibility and significant weather are in-

cluded in the index. The final aggregated index and also sub-indices are categorized in three

levels, ideal, marginal, and unsuitable. The authors use historical observation data from

Alaska (1943-2005) and Florida (1953-2005) for the purpose of constructing the index and

statistical analysis. The results show that high-latitude regions, such as Alaska, exhibit a po-

tential positive impact on climate resources from warming temperatures. In contrast, there

are negative impacts on Florida. Further, Alaska encounters an improvement of weather

123

conditions in the spring and summer, while weather conditions in Florida deteriorate in the

summer but improve in the winter. These results are in line with those of Perch-Nielsen

(2010) showing the less vulnerability of high-latitude countries against the impact of cli-

mate change on beach tourism. However, findings in Perch-Nielsen (2010) indicate that the

level of vulnerability might differ between developed and developing countries.

Perch-Nielsen et al. (2010) utilize the adjusted TCI to investigate the future climate com-

fort for tourism in Europe. The authors make three adjustments to the original index in

order to make the index reflect the current state of knowledge. First, using daily data in-

stead of monthly data. Second, using apparent temperature to measure the thermal com-

fort (Steadman, 1984). Third, in order to measure wind speed, the authors use wind chill

equivalent temperature (Osczevski and Bluestein, 2005), instead of originally used wind chill

index (Siple and Passel, 1945). They use daily data from five regional climate models to

compare the reference period 1961-1990 to the projected scenario in 2071-2100.

Analyzing the results, Perch-Nielsen et al. (2010) conclude that by redistributing cli-

mate resources for tourism in an effective way, climate change would introduce winners and

losers in different locations a and seasons. Most parts of Central and Northern Europe can

be considered winners, whereas Southern Europe regions are regarded as losers with the

decreasing number of good days on average. Although southern countries encounter the

sharpest drop in the summer months, they are still winners in other months of the year as

the number of good days increases. It is projected that southern countries have 15 good

days per month in the winter months, while there are less than 5 good days per month in

the rest of Europe. In addition, based on the projection, there are more than 20 days of good

conditions per month in four to five months of the year in the southern regions. Therefore,

Southern Europe generally displays more favorable climate for tourism compared to the rest

of Europe.

Grillakis et al. (2016b) examine the impact of a 2 ◦C increase in the global temperature on

the summer European tourism, based on the projections for periods 2016-2045, 2037-2060,

and a reference period 1971-2000. The authors thus use TCI and CIT to analyze the impact.

The paper identifies dramatic changes in the projected climate comfort levels estimated by

both TCI and CIT. Some Europeans regions, particularly northern regions will experience a

remarkable increase in summertime climate comfort, while the Mediterranean regions will

benefit the least.

The TCI indices are widely used in the literature for assessing effect of climate change

on the demand-supply of tourism. The research can be divided into Four main blocks based

124

on the methodology used. Four main methods are the econometric analysis, Travel Cost

Models (TCM), gravity model, and qualitative approaches.

1.3 Extreme Events

Clearly, extreme events are relevant to tourism activities and these can directly bring neg-

ative impacts on tourism by destroying infrastructures and causing human loss. However,

Gössling and Hall (2006) believe that weather extreme events only bring short-term impacts

on tourism preferences while its long term effects are fuzzy. The authors claim that the rela-

tionship between weather extremes and tourism activities is not simply linear and it needs

further research on this issue. Moreover, Gössling and Hall (2006) indicate that tourism

is highly susceptible to other extreme events such as terrorism, war, epidemics. However,

these types of event do not happen frequently and they can be considered low-probability

events. Therefore, indicators for such rare extreme events are unlikely to provide the statis-

tical robustness for empirical analysis ?. Consequently, it needs a trade-off between statisti-

cal robustness and the selection of less extreme events. For example, Perch-Nielsen (2010)

used only two indicators to present for flood including “the relative change in the maximum

5-day precipitation total of the year and the absolute change in the fraction of total precipi-

tation due to events exceeding the 95th percentile of the climatological distribution for wet

day amounts” ((Perch-Nielsen, 2010), p.589).

2 Conceptual Approach

Conceptually, the effects of weather or climate change on tourism may be analyzed either

econometrically by regressing the tourism outcome variable on weather control or by means

of a macro-structural model like a CGE model that is appropriately enriched. The core

of this paper is devoted to describe the former approach (see Section 4), while the latter

approach is briefly summarized below in this section.

2.1 Simulation Models

The econometric estimates may be also used to simulate or predict the effects on tourism

outcome due to climate change (Bigano et al., 2008; Hamilton and Tol, 2007). One of the

most common simulation model is Hamburg Tourism Model (HTM) which is used to study

and predict the effect of climate change on tourism in present and different future sce-

125

narios. Hamburg Tourism Model (HTM) is an economic simulation that models the travel

choices of tourists from 207 countries choosing destination out of other 206 countries. This

model helps to analyze the patterns of global tourism (Hamilton et al., 2005b). Hamilton

et al. (2005a) offer 1.1. versions of the HTM; they estimate CO2 emission for various sce-

narios from international tourism, allow for higher income elasticity and study the effect

of changed in international tourism on domestic tourism. The HTM is widely criticized by

Bigano et al. (2005), Bigano et al. (2006), Hamilton et al. (2005a), Hamilton and Tol (2007).

These papers point out main drawbacks of HTM to be as follows. HTM allows for the quali-

tative but not quantitative analyses, and it is better at testing sensitivity rather than making

prediction. Even though the model is successful in reproducing the current picture of the

international tourism, its long-term prediction power is reasonably weak. Moreover, the

data used in these studies is too general for the purpose and only measure used for climate

change is temperature. Further drawbacks are that the model does not distinguish tourism

by seasons, purpose and tourists by age. Hamilton and Tol (2007) offer 1.2 of the Ham-

burg Tourism Model to make it more precise by introducing regions. The authors follow

the econometric methodology of Lise and Tol (2002), and use the same data used by Bigano

et al. (2004).

2.2 Economy-Wide Effects

Computable General Equilibrium (CGE) is another method used in the literature to address

the effect of climate change on the tourism industry. This framework is derived from tradi-

tional Input-Output (IO) models in order to estimate the effects of changes in a part of the

economy on the remainder.

Using a world CGE model, Berrittella et al. (2006) study the implications of climatic

changes on tourism demand. The CGE model includes parameters indicating the struc-

ture of the economy. However, since the authors intend to forecast changes at future dates,

the model is re-calibrated to obtain datasets for certain future years, 2010, 2030, and 2050.

Then, authors run a series of simulations scenarios to capture the tourism-related impacts of

climate change. Simulation results indicate that economic impacts increase by time as tem-

perature gradually increases. Moreover, the global effect of the climate change on tourism

is not significant, and roughly zero in 2010. In contrast, assuming global mean warming

as 1.03 ◦C relative to 1997, climate change is a more important issue for the tourism in-

dustry in 2050. North America, Australasia, Japan, Eastern Europe, and the former Soviet

126

Union states will positively be affected by climate change, while Mediterranean and tropical

countries will become less attractive destinations.

Bigano et al. (2008) use an extended version of a multi-country CGE model, the Global

Trade Analysis Project (GTAP) model (Hertel, 1997) to analyze the effects of sea level rise

on tourism. The benchmark datasets are derived at 2010, 2030, and 2050 to entail inserting

forecast values into the model calibration data. In addition, the Hamburg tourism model

(HTM) is used to assess the impacts of climate change on tourism. Calibrated for 1995, HTM

as an econometric simulation tool estimates tourism flows between countries, the share of

international tourists in total tourists, and the number tourist by country. The results show

a significant reduction of tourism demand in warmer countries, −19%, −8%, −7% in tropical

islands, Middle East and South East Asia, respectively. The results also indicate an increase

in tourism demand by 1.3% and 8% for Western Europe or Japan and Korea, which are

regions at the higher latitudes.

Pham et al. (2010) investigate the induced effects of climate change on five Australian

tourism destinations. For this purpose, the authors use a comparative static CGE eco-

nomic model and a set of simulations under scenarios for periods 2005-2020, 2005-2050,

and 2005-2070. Based on the simulation results, the authors find that as the tourism in-

dustry is labor-intensive, a decrease in tourism demand will reduce the age income, which

leads to higher inequality in the society. Moreover, climate change-induced effects are not

distributed evenly across the country. These results support the importance of investment

in climate change adaptation and environmental preservation measures.

2.3 Qualitative Research

There are survey based alternative approaches addressing the impact of climate change on

tourism, which are used fewer times in the literature. Assessing different types of survey

and questionnaire, papers based on these methods usually tend to be on a qualitative basis

(Dodds and Kelman, 2008; Moreno, 2010; Lépy et al., 2014; Michailidou et al., 2016).

For example, Moreno (2010) analyzes data of tourist’s views on the role of climate change,

collected at one Dutch and one Belgian airport in 2007. The travelers are asked for their

views on factors making a destination (un)favorable, and also the role of climate change on

their destination choice. In line with a majority of the relevant literature, Mediterranean

regions are found to be the most vulnerable to climate change impact. The pilot research

shows that increasing temperature in the Mediterranean countries together with improve-

127

ment in weather conditions in respondents’ home countries are the most important factors

threatening the Mediterranean tourism industry. Moreover, the author claims that high tem-

peratures are not necessarily associated with the concept of bad weather for the respondents.

Even heat-wave term does not affect very negatively the level of satisfaction of respondents.

Moreover, Michailidou et al. (2016) study climate change mitigation and adaptation mea-

sures in the case of Greece’s tourism industry since the average temperature in Greece is

projected to increase at least 2.4 ◦C under the modest scenario accompanied with a decline

in precipitation until the end of 21st century (Bank of Greece, 2001 as cited in Dodds and

Kelman, 2008). Providing a methodological framework based on Multi-Criteria Decision

Analysis (MCDA), authors obtain the decisions of a panel of experts upon different issues

in the tourism context. Then, authors provide an optimal ranking for the basic scenario.

The results demonstrate the ability the feasibility of integrating MCDA into the tourism

management by the proposed framework and methodology.

3 Econometric studies

3.1 Theoretical Model

Travel Cost Model

Hotelling (1949) originally introduces Travel Cost Method (TCM). However, it has been

widely refined during the years (Clawson and Knetsch, 2013). The method mainly aims to

estimate economic costs or benefits of changes regarding (i) access to a recreation site, (ii)

adding or elimination of a recreation site, and (iii) environmental quality at a recreation site.

There are three main approaches within the TCM method.

The original and the simplest one, Zonal TCM, defines geographical divisions surround-

ing the site in order to collect the number of visitors from each division in the last year. Then

one can regress the number of visits on travel costs from each zone, which might include

demographic and other data in model to construct a demand function. While this approach

assumes homogeneity in population and travel costs among different zones, the two other

approaches, Individual TCM (ITCM) and Random Utility TCM (RUTCM), address these as-

sumptions as well as other neglected issues in the first approach, which make them more

complicated but more precise. RUTCM, for instance, assumes that individuals make trade-

offs between site quality and price of travel based on their preferences. However, there are

128

some limitations in TCM studies, such as the value of time, and the type of traveller. More-

over, Freeman (1992) argues that the absence of substitute site qualities and prices in travel

cost models make it impossible to examine the effect of changes in destination characteris-

tics at more than one site.

Introducing the Pooled Travel Cost Model (PTCM), Maddison (2001) investigates the

welfare impact of climate change on British tourists’ holiday destinations and the number

of trips to particular destinations. The author uses quarterly data on international travel

by British residents taken in 1994 for the dependent variable. Other variables, such as

population, GDP per capita, beach length, quarterly averaged maximum temperature, and

quarterly precipitation in the capital city are included in the travel cost dataset. Overall,

305 observations from 87 countries are available in the dataset. The results of the semi-log

regression analysis indicate that British tourists are attracted to climates which deviate little

from an averaged daytime maximum of 30.7 ◦C. Furthermore, due to the climate change the

nearby destinations are likely to become more popular, which might lead to a non-negligible

welfare gain to British tourists.

A model with unknown travel distances and costs can show decisive factors which make

destinations more popular than other ones. Lise and Tol (2002) used the model as following:

LnArrivals = β0 + β1Year + β2Area + β3Popden + β4Coast + β5GDPPC + β7TW + β8TW 2+

+β9PS + β10PS2 + error

Where the two main independent climate variables are temperature (TW) and precipi-

tation (PS) and the dependent variable is number of tourist arrivals under the natural loga-

rithm form. Note that, the authors include both temperature-squared and precipitation-

square which implies that there could be an optimal temperature and precipitation for

tourism. If temperature and precipitation are more than these optimal points, it might bring

some negative effects. Additionally, other control variables including destination price lev-

els (GDPPC), land surface of are per country (Area), total length of the coast (Coast) and

population density (Popden) are used in the estimated model. Finally, Year variable is used

to clear out all yearly trend from the model. The model is estimated by Original Least

Square (OLS). The main purpose of the model is to estimate the sensitivity of tourist choice

to climate.

However, this model with unknown travel distances and costs can not reflect the differ-

129

ent tastes of people from different countries. For testing these differences, another database

with travel origins need using and further model modification is necessary. Lise and Tol

(2002) add distance between capitals of travel origins and travel destinations into the model

for each origin country as follow:

LnArrivals = β0 + β1Year + β2Area + β3Popden + β4Coast + β5GDPPC + β7TW + β8TW 2+

+β9PS + β10PS2 + β11Dist + error

Dist variable can show the difference in the option of tourists from different nations, but this

model can be suffered from shortage of data and then the analysis at cross-country level can

be crude. Consequently, Lise and Tol (2002) suggest taking a deeper analysis by conducting

a pooled travel-cost model (PTCM) for a single country (Dutch) as following:

LnVisits = β0 + β1Fare + β2GDP + β3Pop + β4Popden + β5Coast + β6Pday + β7Dist + β8TQ+

+β9TQ2 + β10PQ + β11Q1 + β12Q2 + β13Q3 + +error

This is the model for Dutch tourist demand only and LnVisits is logarithm of number of

Dutch tourists visiting destination countries and each destination has different climate and

non-climate characteristics which are included in the model. Two climate variables are TQ

and PQ which are average day and night temperature and total precipitation respectively.

This demand equation helps understand which factors determine the preferences of Dutch

tourists. However, it is limited by a rigid assumption that included variables do not change

over tourist destinations.

Gravity Model

Gravity model is inspired by the Newton’s universal law of gravitation. Based on mass and

distance between two objects Newton’s law of gravity measures attraction of these objects.

Similarly, Gravity Models aim to measure the "attraction" between two economic factors,

(i.e demand for goods or labor) in different country of origin, assuming that the distance

between the countries decreases the correlation between these factors (Anderson, 2011).

To asses effect of climate change on tourism Priego et al. (2015) investigate the effect of

increasing temperature on domestic destination choice in Spain.

130

Using a gravity model, that includes a temperature parameter, Priego et al. (2015) find

that international flows are expected to increase with a country’s economic size. They define

a gravity equation in the domestic context to capture tourism within Spanish regions. Using

Pooled Ordinary Least Squares (POLS) that includes years fixed effects, the authors estimate

the model. In addition, the domestic tourism data are obtained from survey conducted in

a period from 2005 to 2007. The estimated results show that the southern regions in Spain

will negatively be affected by an expected increasing temperature, in particular at the south

of Madrid, as well as the Spanish Mediterranean provinces. On the other hand, northern

provinces will be among those regions that benefit from an increase in temperature. Hence,

according to Priego et al. (2015) domestic tourism has similar patterns to the international

one, where northern and southern regions are negatively correlated in terms of tourism

attractiveness.

3.2 Tourism Outcome Variable

In literature, the tourism outcome is typically measured through variables either linked to

demand or supply.

Tourism Demand

With respect demand-linked tourism, the dependent variable is measuring either the num-

ber of overnight stays (Falk (2010); Agnew and Palutikof (2006); Grillakis et al. (2016a)) or

the number of arrivals and/or departures (Hamilton et al. (2005a); Goh (2012); Lise and

Tol (2002); Hein et al. (2009); Bigano et al. (2008)). The specific variables used in partic-

ular paper differ and some of them are recorded in our database, see Appendix. If they

are country-based ones, majority of papers make use of national statistics. For example,

Falk (2010) collects data on number of overnight stays from Statistic Austria or Hein et al.

(2009) uses data on number of visitor per night from annual Spanish inbound tourism sur-

vey while Bigano et al. (2005) and Agnew and Palutikof (2006) take advantage of Italy na-

tional statistics and the UK statistical office when conducting research on Italy and the UK

respectively. Apart from that, some papers examine the effect of climate on tourism at the

international comparative level. These papers either collect data from each country Goh

(2012) or take advantage of available international data sources. More specifically, Hamil-

ton et al. (2005a) collect data from World Resource Institute, Grillakis et al. (2016b) use

Eurostat database while Lise and Tol (2002) take data from World Development Indicators

131

CD-ROM and Maddison (2001) make use of number of return trips from International Pas-

senger Survey. Additionally, Bigano et al. (2006) pool various sources including Euromon-

itor, Institutional Resources and TWO to have data on number of arrivals and departures

and number of nights. Note that, apart from Hamilton et al. (2005a) and Lise and Tol (2002)

who use yearly database, the rest make use of monthly database.

In terms of qualitative research, demand of tourism can be measured by decision on holiday

(Hares et al. (2010); Braun et al. (1999)), willingness to travel domestically or internation-

ally Eugenio-Martin and Campos-Soria (2010) or travel behaviour Gössling and Hall (2006).

These indicators are constructed by conducting an interview with travellers.

Tourism Supply

Tourism supply is measured by activities of tourism service providers and they typically re-

flect providers market behaviours. These data are collected mainly from interviewing these

providers or from business statistics collected by national or regional statistical offices. For

instance, Elsasser and Bürki (2002) examine the mitigation and adaptation strategies of ski

resort in the Alps towards climate change. Similarly, Hoffmann et al. (2009) study the strate-

gic directions for ski lift operators in Switzerland under the scenario that the snow depth

could be less in the future. This strategies of operators are collected from Swiss nation-wide

survey.

3.3 Review of Econometric Studies

Summer Tourism

A part of literature focuses on the impact of climate change on the tourism industry in

warmer seasons, in particular summertime. The main finding of the literature is that since

climate change is positively incorporated with hotter and drier summer conditions, it will

affect the increasing trend of domestic tourists in Northern European countries. On the

other hand, there will be a decline in tourism deman in warmer regions, particularly Mediter-

ranean destinations (see Bigano et al., 2005).

Subak et al. (2000), Agnew and Palutikof (2006) investigate the effect of climate change

on tourism in the case of the United Kingdom. They study the influence of weather con-

ditions on the international and domestic tourism within the scope of time series analysis.

132

Despite that Subak et al. (2000) uses only three time series, every series were significantly

sensitive to the weather variability. Agnew and Palutikof (2006) investigate the issue us-

ing annual data for international tourism and monthly data for domestic tourism. They

find that international tourism is sensitive to preceding year weather fluctuations, whereas

weather fluctuation in the period of travel is important for domestic tourism. After analyz-

ing the unusually warm 1995 year in the UK, the authors conclude that warm and dryness

of the climate encouraged increase in the domestic tourism, Whereas preceding year with

damp and rainy weather motivated international travel. However, there are several studies

showing that tourists make travel decisions few month prior to their holidays, those mostly

related to summer activities (Money and Crotts, 2003; Perez and Juaneda, 2000).

A series of related studies investigate the effect of climate change on global tourism

(Hamilton et al., 2005a,b; Hamilton and Tol, 2007). In both papers, the authors use data

on the flow of tourists between 207 countries in 1995 (Bigano et al., 2004). They build a

simulation model that aims to present flows of tourists between 2000-2075. Firstly, Hamil-

ton et al. (2005a) find that climate change has a smaller contribution to increasing trend of

tourism over time, compared to population and income changes. In the following study, the

authors find that even though the tourism is found to have increasing trend, the speed of

increase may slow down later on as the most of the demand for travel is met (Hamilton et al.,

2005b). Together with increasing emission of carbon dioxide and climate change, Hamilton

et al. (2005b) state that the preferred tourist destinations will shift towards higher latitude

and altitude locations. The authors claim that this would mean that tourists from the tepid

climates, those that supply the majority of tourist market, would prefer to spend their va-

cation mainly in the home country. Hamilton et al. (2005b) argues that whereas climate

change decreases international tourism, it has smaller effect compared to population and

economic growth, which is consistent with Hamilton et al. (2005a).

Hamilton and Tol (2007) describe the HTM according to 1.2 version of Bigano et al.

(2005). In the model, the size of the population and average income are the main determi-

nants of the number of tourists generated by the given country. Distribution of domestic

and international tourism depend on the climate and per capita income in the origin coun-

try. The proxy of the climate measure is only temperature. The distance between origin and

destination, climate, per capita income in the destination countries, and a general attractive-

ness index are observed as the determinants of the allocation of international tourists. Other

variables are included for the efficiency reasons but are held constant in the simulation.

Since Hamilton and Tol (2007) only have the data of tourist destinations, but not their

133

origin, the model is marginally improved by including the regions of the destination coun-

try. The model shows the significant influence of climate change on the tourism in the

particular region. The authors point out the need of the regional climate change scenario,

which can be non-homogeneous over regions. For instance, the continental interior, com-

pared to the ocean board, becomes warm faster. The results show that the regional effect of

the climate change is relatively small, however it can be translated into large absolute terms

(0.5% change in German tourism trips yields in change of number of tourists by 400000 in-

dividuals). Otherwise, the model results are inline with those of Hamilton et al. (2005a,b),

and show that while climate change has negative effect, population and economic growth

have stronger influence on the tourism industry.

Eugenio-Martin and Campos-Soria (2010) study the relation between origin country

weather and destination choice in outbound tourism demand. The purpose of the analy-

ses is to test the hypothesis that a good climate is demotivator for the residents to travel

abroad.The authors consider household rather than aggregated level travel decision. This

approach allows for analyzing regional and socioeconomic characteristics as destination

choice determinants. Therefore the paper uses household data from survey of 16183 house-

holds in 15 European Union countries in 1997 that was used in the Eurobarometer 48 (Euro-

pean Commission, 1998 as cited in Eugenio-Martin and Campos-Soria, 2010). According to

the survey, Europeans from Scandinavian and North-European countries travel most often.

Even though one of the main drivers of this difference appears to be the financial state of

the countries, there are other reasons that have significant impact on the frequency of travel.

However, financial constraints are also subject to interpretation. Portuguese, For instance,

might not be willing to pay as much as Danish residents given different preference, which

depend on various other aspects, e.g. climate and attractiveness of the own country.

Defining the climate in econometric model framework is complicated, therefore the

authors choose to use the regional climate index for tourism purposes. Eugenio-Martin

and Campos-Soria (2010) choose to use a bivariate probit since domestic and international

tourism are related. They analyze the probabilistic results with the help of GSI and non-

parametric approaches and estimate the importance of the origin climate condition for

travel destination choice. The results show that family size is negatively correlated, whereas

income and education encourage traveling; women are more kin on traveling compared to

men. Moreover, age has positive effect on international and not effect on domestic traveling

habits. According to research, marital status has no effect on travel frequencies, whereas

families with more children prefer to spend their vacations domestically. Household living

134

at the coast are seen to be more willing to travel domestically rather than internationally,

whereas resident of the large community is likely to travel to any destination. Finally, the

model shows high significance of the climate index. The good local climate condition is

positively correlated with probability of traveling domestically and causes low probability

of travelling abroad.

Using a dynamic panel approach, Moore (2010) evaluates the potential effects of climate

change on Caribbean tourism. Since TCI captures the attractiveness of the destinations un-

der climate change scenarios and does not provide a quantitative assessment of its impact

on tourism demand, the author provides a standard demand model incorporated with TCI

for each Caribbean island and its competitors. The author uses annual data from the pe-

riod 1980-2004 for 18 Caribbean islands. Using data from 1980-2000, the tourism demand

model is estimated by, and also used to project the tourists arrivals between 2001 and 2004

under four different scenarios. Overall, the results show a decline in the number of tourists

in the projected scenarios. In addition, the results suggest that under the worst climate

change scenario, arrivals to the Caribbean will decrease by 1% per year. However, the mag-

nitude of the decline is not homogeneous and will differ from one island to another one.

Rosselló-Nadal et al. (2011) focus on the climate change as a factor that motivates tourism.

The authors study the effect of origin country weather on the travel decisions of residents.

They base their research on the United Kingdom, as it represents the third biggest tourism

spender country in the world and provides high quality data on international tourism.

Rosselló-Nadal et al. (2011) take monthly data for the UK outbound flows (1980 - 2009),

as a measure of international tourism, from the International Passenger Survey. The au-

thors use a transfer function model in order to understand the weather sensitivity of British

outbound flows. The scope of the analysis of Transfer function models is broader com-

pared to that of a classical ARIMA to multiple time series. Moreover, The transfer function

reduces random component by incorporating the explanatory variables. Therefore, it is pre-

fared of the other methodologies. Overall, Rosselló-Nadal et al. (2011) investigate effect of

six weather variables on outbound tourist flows: mean maximum daily temperature, mean

minimum daily temperature, days of air frost (AF), total rainfall, total sunshine duration

(SD), and monthly average temperature (AT). They assume optimal holiday temperature to

be 16.3 ◦C and construct variables for high temperature (=1 if higher than 16.3 ◦C, zero

otherwise) and low temperature (=1 if lower than 16.3 ◦C, zero otherwise) around this es-

timate. In order to analyze the impact of global warming on the tourism flows, the authors

simulate the warmer climate according to UKCIP (2002) climate change scenarios for the

135

United Kingdom (the hypothesized increase of average temperatures by 1 ◦C, 2 ◦C, and 3 ◦ C). The simulation shows that global warming has negative effect on outbound flows of the

United Kingdom. The effect of the warmer climate is the strongest during summer, whereas

it affects the springtime tourism the least.

Denstadli et al. (2011) study how the summer tourists perceive the weather conditions

in Scandinavia. The authors study the links between weather expectations and perceptions,

and adaptive behavior of vacationer. Specifically, the study aims at understanding the ef-

fects of weather variability on vacationers decisions. Denstadli et al. (2011) study whether

they prolong (terminate) the vacation in more (less) favorable weather conditions; they are

farther interested how weather variability affects tourists overall satisfaction from the desti-

nation and on the probability of return to the same location. The authors focus on the sum-

mer vacationers to archipelago of Vesteralen in Northern Norway and address three specific

issues: how tourists perceive the weather (in comparison to their prior expectations), and

adapt their activities and willingness to return to the weather conditions in Vesteralen.

The research is based on the survey of the Vesteralen population of tourists and other

leisure travelers living outside the discussed location. The survey includes questions cov-

ering the frequency of encountered weather variability, the thermal comfort, aesthetic and

physical sensation. Moreover, the survey asks to rate the overall perception of weather and

to specify if visitors adjust plans due to weather variability during the vacation period. Ad-

ditionally, the visitors are asked if they would return to Vesterlan during the summertime

in the next three years. The weather perceptions and disconfirmation (comparison of the

expectation and reality) is compared using a MANCOVA between two subgroups of the in-

terviewees - (i) first time and repeat and (ii) domestic and international travelers. In general,

the study shows that one third of visitor’s plans are altered by the weather conditions. Most

frequent adjustment is to prolong the stay due to nice weather. Furthermore, the research

shows no significant difference between the first time and repeat visitors of Vesteralen. In

contrast, in the case of domestic and international visitors the difference is found to be sig-

nificant. The domestic visitors are more resilient to changes in their plans; around 75%

of domestic and 62% of international visitors choose not to adjust their plans according to

weather. The international visitors are more likely to alter their plans (17%) due to weather

conditions, than to shorten the period of stay (5%).

The objective of Goh (2012) is to build the tourism demand model by including climate

factor into the Economic and socio-psychological framework. Including the climate vari-

able into the classical tourism demand model, the author studies importance of climate.

136

Goh (2012) studies long and short haul tourism to Hong Kong from four major origin coun-

tries, i.e. the United States and the United Kingdom (long haul), and China and Japan (short

haul). In addition to the discrete and distinct seasons, number of events affect demand on

tourism in Hong Kong (Goh, 2012). The latter is crucial in analyzing the effect of seasonal-

ity and interventions, as well as for examining the robustness of the study. Using the error

correction model (ECM), the author incorporates climate into the classical tourism demand

framework. The author measures effect of climate by TCI adjusted according to Hong Kong

climate conditions. TCI is found to be significant and positive for all four origin countries.

The index is significant at 1% level for the US and 5% level for the UK, China and Japan.

Thus, the author finds stronger influence of climate for the US compared to the China and

Japan. Goh (2012) further explains that the possible reason is that unlike the US, China and

Japan have very similar weather to Hong Kong. This makes travelers from short-haul coun-

tries less sensitive to the weather changes in Hong Kong, compared to long-haul countries.

Winter Tourism

Besides the literature on beach and summertime tourism, a broad literature focuses on win-

ter tourism. The winter tourism, particularly skiing tourism, similar to other sectors within

the tourism industry has been showing vulnerability to global climate change. The vul-

nerability of the winter tourism industry has been a central topic not only for academic

researchers, but also for entrepreneurs. Most of the entrepreneurs in the Finnish tourism in-

dustry consider a 90–120-day-long winter season to be adequate for making a profit. More-

over, potential weather conditions causing cancellations are above 2 ◦C, for a period of days,

extremely low temperatures (below -25 ◦C), high wind, and rain (Tervo, 2008). Consider-

ing expert’s views and scientific methods, researchers have been investigating the impacts

of climate change on the winter tourism industry for more than two decades. Scott et al.

(2006), Steiger (2010), and Falk (2010) are among the most cited works in this field.

Scott et al. (2006) study the impacts of climate change on the ski-based tourism in east-

ern North America, particularly Quebec, Michigan and Vermont. The authors use a panel of

climate record from 1961 to 1990 as the baseline period. Then, they investigate the projec-

tions of climate change induced effects on snow conditions at each ski area, and also changes

in ski season length. The findings suggest that in the 2020s, even the high impact climate

change scenario poses only a minor risk to ski areas at each of the study areas. Consistent

with Steiger (2010), another major finding is that the projected length of ski season in 2050s

137

under different climate change, is not as severe as projected in earlier studies that did not

adequately incorporate snowmaking. However, the average reduction of season length is

still projected to be significant in 2050s, at least under the high impact scenario.

The object of study for Hoffmann et al. (2009) is how corporate sector adopts to the cli-

mate change with the help of survey on Swiss ski lift operators. The data on Swiss ski lift

operators is informative due to the high dependence of the ski lift operators on the natu-

ral snow level, i.e. on the climate conditions. The authors develop corporate adaptation

strategies to capture the measures implemented by companies in the reaction to the cli-

mate change, e.g. adopt, expand beyond the affected business, share the risk. The authors

have formulated four group of hypothesis. They hypothesize that climate change awareness,

firm’s business vulnerability to the climate change, and firm’s dependence on the affected

business are positively correlated with the adaptation measures that firm undertakes. Con-

versely, they test that the more the firm is uncertain over the climate change the less it takes

measures to adopt, expand beyond the affected business, and tries to hedge uncertain future

outcomes by increasingly sharing risk of financial impacts.

Hoffmann et al. (2009) develop 26 adaptation measures along the three categories of the

adaptation goals - protect the affected business, expand beyond the affected business, and

share risks of financial impacts. Around 50% (124) of all Swiss ski lift operators participate

in the survey, and the data gives no signs of sample selection. The authors choose to estimate

the model using simple OLS approach and make robustness checks using count model - the

negative binomial model (NBM) - estimated by Maximum Likelihood. The results show no

significant evidence for the claim that firm’s business vulnerability positively affects adapta-

tion measure. The study further rejects the hypothesis for all the various directions through

which perceived uncertainty affects adaptation measure. In contrast, the results show signifi-

cant positive influence of climate change awareness on the corporate adaptation. Moreover,

the study show that the more the firm depends on the affected business, the more it takes

adaptation measures in order to protect the business.

Falk (2010) studies the relationship between the snow depth and the number of overnight

stay in the 28 Austrian ski resorts during 1986-2005. The study has some distinct features:

it is first to use dynamic, heterogeneous panel data technique, its dataset is significantly

more detailed compared to previous studies (nearly 50% of the winter season overnight

stays in western Austria), and the author tests for the parameter stability of the snow depth

and tourism relationship. The author looks at the effect of snow accumulation, GDP per

capita, price index of accommodation prices, and early Easter holiday effect (captured by

138

the dummy variable) on the output(measured as number of overnight stays-winter tourism

demand) over time in each of 28 ski resorts. It is likely that the relation between snow

depth and overnight stays is different for different ski resorts, as the ski resorts differ by

size, elevation and snow-making capacity.

Furthermore, the author uses technique adopted from Pesaran et al. (1999) for the dy-

namic heterogeneous panel data. Applying the mean group (MG) and the pooled mean

group (PMG) estimators, the author solves the heterogeneity problem. The MG estimator

estimates the coefficients separately for each ski resort and takes averages of these coeffi-

cients. The PMG estimator requires long-term coefficients to be equal across ski resorts;

in this case PMG is both efficient and consistent (can be tested by Hausman test), whereas

MG is only consistent. The results show that the overnight stays elasticity of snow depth

is 0.10. However, the number of overnight stays is shown to be independent of the snow

accumulation in the high-elevation resorts. Additionally, long-run overnight stay elasticity

of GDP per capita in high elevation resorts accedes that of low-elevation resorts. Finally, the

study shows that the relationship between dummy capturing the early Easter holidays and

demand on winter tourism is significant and positive.

4 Research Gap and Further Research

The impacts of climate change on the tourism industry has been a main subject of research

for more than two decades. The need of the tourism industry to forecast the possible out-

comes of a climate change scenario (which is almost inevitable), has led to a considerable

literature. This paper aims to summarize the current knowledge about the impacts of the

climate change on the tourism industry by exploring some methodologies and results in the

literature.

One of the main results in the literature, regardless of the methodology used, is a neg-

ative effect of global warming for tourism, since the most important motivation for tourist

flows is favorable climate condition. In other words, climate change will negatively affect

both current winter and summer tourist destinations. On the other hand, by a decline in the

number of preferred destinations abroad, it seems that climate change will be an important

factor in increasing the number of domestic trips, as well as new tourist destinations.

Although almost all studies using either quantitative or non-empirical methodologies

indicate the same result, because of the complexity of the topic, there is still a research

gap providing opportunities for further research. More developed indices capturing differ-

139

ent climatic variables, tourist perceptions and reactions to climate change, as well as the

responses of supply-side are required to be considers for carrying out similar studies.

Moreover, as for the further research agenda, transportation and its effects on climate

change and subsequently tourism is one of the neglected issues in the literature. In addition,

further research might be developed based on regional climate models, since at the moment,

there is not enough local/regional scenarios for climate change to obtain the reliable forecast

of its local impacts on tourism.

Finally, it is worth mentioning that there is almost no known study on tourism and cli-

mate change in Eastern Europe, South America, Africa, and Middle East. These regions are

underrepresented in both winter tourism (excluding Africa and middle East) and summer

tourism literature. This research gap can be addressed in further studies to achieve a better

understanding of the impact of climate change on tourism.

140

References

Agnew, M. and Palutikof, J. (2006). Impacts of short-term climate variability in the uk on

demand for domestic and international tourism. Climate Research, 31(1):109–120.

Amelung, B. and Nicholls, S. (2014). Implications of climate change for tourism in australia.

Tourism Management, 41:228–244. Amelung, B. and Viner, D. (2006). Mediterranean tourism: exploring the future with the

tourism climatic index. Journal of sustainable tourism, 14(4):349–366.

Anderson, J. E. (2011). The gravity model. Annu. Rev. Econ., 3(1):133–160. Berrittella, M., Bigano, A., Roson, R., and Tol, R. S. (2006). A general equilibrium analysis

of climate change impacts on tourism. Tourism management, 27(5):913–924.

Bigano, A., Bosello, F., Roson, R., and Tol, R. S. (2008). Economy-wide impacts of climate

change: a joint analysis for sea level rise and tourism. Mitigation and Adaptation Strategies

for Global Change, 13(8):765–791.

Bigano, A., Goria, A., Hamilton, J., and Tol, R. (2005). The effect of climate change and

extreme weather events on tourism.

Bigano, A., Hamilton, J., Lau, M., Tol, R., and Zhou, Y. (2004). A global database of domestic

and international tourist numbers at national and subnational level. fondazione eni e.

Technical report, Mattei Working Papers 3.05, Milano, Italy.

Bigano, A., Hamilton, J., and Tol, R. (2006). The impact of climate change on domestic and

international tourism: a simulation study.

Braun, O. L., Lohmann, M., Maksimovic, O., Meyer, M., Merkovic, A., Messerschmidt, E.,

Riedel, A., and Turner, M. (1999). Potential impact of climate change effects on prefer-

ences for tourism destinations. a psychological pilot study. Climate Research, 11(3):247–

254.

Clawson, M. and Knetsch, J. L. (2013). Economics of outdoor recreation. RFF Press. Denstadli, J. M., Jacobsen, J. K. S., and Lohmann, M. (2011). Tourist perceptions of summer

weather in scandinavia. Annals of Tourism Research, 38(3):920–940.

141

Dodds, R. and Kelman, I. (2008). How climate change is considered in sustainable tourism

policies: a case of the mediterranean islands of malta and mallorca. Tourism Review Inter-

national, 12(1):57–70.

Elsasser, H. and Bürki, R. (2002). Climate change as a threat to tourism in the alps. Climate

research, 20(3):253–257.

Eugenio-Martin, J. L. and Campos-Soria, J. A. (2010). Climate in the region of origin and

destination choice in outbound tourism demand. Tourism Management, 31(6):744–753.

Falk, M. (2010). A dynamic panel data analysis of snow depth and winter tourism. Tourism

Management, 31(6):912–924.

Freeman, A. M. (1992). The measurement of environmental and resource values: theory and

methods. Technical report, Resources for the Future.

Goh, C. (2012). Exploring impact of climate on tourism demand. Annals of Tourism Research,

39(4):1859–1883.

Gössling, S. and Hall, C. M. (2006). Uncertainties in predicting tourist flows under scenarios

of climate change. Climatic Change, 79(3):163–173.

Grillakis, M. G., Koutroulis, A. G., Seiradakis, K. D., and Tsanis, I. K. (2016a). Implications

of 2 c global warming in european summer tourism. Climate Services, 1:30–38.

Grillakis, M. G., Koutroulis, A. G., and Tsanis, I. K. (2016b). The 2 c global warming effect

on summer european tourism through different indices. International journal of biometeo-

rology, 60(8):1205–1215.

Hamilton, J. M., Maddison, D. J., and Tol, R. S. (2005a). Climate change and international

tourism: a simulation study. Global environmental change, 15(3):253–266.

Hamilton, J. M., Maddison, D. J., and Tol, R. S. (2005b). Effects of climate change on inter-

national tourism. Climate research, 29(3):245–254.

Hamilton, J. M. and Tol, R. S. (2007). The impact of climate change on tourism in germany,

the uk and ireland: a simulation study. Regional Environmental Change, 7(3):161–172.

Hares, A., Dickinson, J., and Wilkes, K. (2010). Climate change and the air travel decisions

of uk tourists. Journal of transport geography, 18(3):466–473.

142

Hein, L., Metzger, M. J., and Moreno, A. (2009). Potential impacts of climate change on

tourism; a case study for spain. Current Opinion in Environmental Sustainability, 1(2):170–

178.

Hertel, T. W. (1997). Global trade analysis: modeling and applications. Cambridge university

press.

Hoffmann, V. H., Sprengel, D. C., Ziegler, A., Kolb, M., and Abegg, B. (2009). Determinants

of corporate adaptation to climate change in winter tourism: An econometric analysis.

Global Environmental Change, 19(2):256–264.

Hotelling, H. (1949). Letter of june 18, 1947, to newton b. Drury. Included in the report The

Economics of Public Recreation: An Economic Study of the Monetary Evaluation of Recreation

in the National Parks, pages 1947–15.

Lépy, É., Heikkinen, H. I., Karjalainen, T. P., Tervo-Kankare, K., Kauppila, P., Suopajärvi,

T., Ponnikas, J., Siikamäki, P., and Rautio, A. (2014). Multidisciplinary and participatory

approach for assessing local vulnerability of tourism industry to climate change. Scandi-

navian Journal of Hospitality and Tourism, 14(1):41–59.

Lise, W. and Tol, R. S. (2002). Impact of climate on tourist demand. Climatic change,

55(4):429–449.

Maddison, D. (2001). In search of warmer climates? the impact of climate change on flows

of british tourists. Climatic change, 49(1-2):193–208.

Michailidou, A. V., Vlachokostas, C., and Moussiopoulos, N. (2016). Interactions between

climate change and the tourism sector: Multiple-criteria decision analysis to assess miti-

gation and adaptation options in tourism areas. Tourism Management, 55:1–12.

Mieczkowski, Z. (1985). The tourism climatic index: a method of evaluating world climates

for tourism. Canadian Geographer/Le Géographe Canadien, 29(3):220–233.

Money, R. and Crotts, J. C. (2003). The effect of uncertainty avoidance on information

search, planning, and purchases of international travel vacations. Tourism Management,

24(2):191 – 202.

Moore, W. R. (2010). The impact of climate change on caribbean tourism demand. Current

Issues in Tourism, 13(5):495–505.

143

Moreno, A. (2010). Mediterranean tourism and climate (change): A survey-based study.

Tourism and Hospitality Planning & Development, 7(3):253–265. Nicholls, S. and Amelung, B. (2015). Implications of climate change for rural tourism in the

nordic region. Scandinavian Journal of Hospitality and Tourism, 15(1-2):48–72.

Nyaupane, G. P. and Chhetri, N. (2009). Vulnerability to climate change of nature-based

tourism in the nepalese himalayas. Tourism Geographies, 11(1):95–119.

Osczevski, R. and Bluestein, M. (2005). The new wind chill equivalent temperature chart.

Bulletin of the American Meteorological Society, 86(10):1453–1458. Perch-Nielsen, S. L. (2010). The vulnerability of beach tourism to climate change—an index

approach. Climatic change, 100(3-4):579–606.

Perch-Nielsen, S. L., Amelung, B., and Knutti, R. (2010). Future climate resources for

tourism in europe based on the daily tourism climatic index. Climatic change, 103(3-

4):363–381.

Perez, E. A. and Juaneda, S. C. (2000). Tourist expenditure for mass tourism markets. Annals

of Tourism Research, 27(3):624 – 637.

Pesaran, M. H., Shin, Y., and Smith, R. P. (1999). Pooled mean group estimation of dynamic

heterogeneous panels. Journal of the American Statistical Association, 94(446):621–634.

Pham, T. D., Simmons, D. G., and Spurr, R. (2010). Climate change-induced economic

impacts on tourism destinations: the case of australia. Journal of Sustainable Tourism,

18(3):449–473.

Priego, F. J., Rosselló, J., and Santana-Gallego, M. (2015). The impact of climate change on

domestic tourism: a gravity model for spain. Regional environmental change, 15(2):291–

300.

Rosselló-Nadal, J. (2014). How to evaluate the effects of climate change on tourism. Tourism

Management, 42:334–340.

Rosselló-Nadal, J., Riera-Font, A., and Cárdenas, V. (2011). The impact of weather variabil-

ity on british outbound flows. Climatic change, 105(1-2):281–292.

Scott, D., Gössling, S., and de Freitas, C. R. (2008). Preferred climates for tourism: case

studies from canada, new zealand and sweden. Climate Research, 38(1):61–73.

144

Scott, D., Gössling, S., and Hall, C. M. (2012). International tourism and climate change.

Wiley Interdisciplinary Reviews: Climate Change, 3(3):213–232. Scott, D. and McBoyle, G. (2001). Using a ‘tourism climate index’to examine the impli-

cations of climate change for climate as a tourism resource. In Proceedings of the first

international workshop on climate, tourism and recreation, pages 69–88.

Scott, D., McBoyle, G., Minogue, A., and Mills, B. (2006). Climate change and the sustain-

ability of ski-based tourism in eastern north america: A reassessment. Journal of sustain-

able tourism, 14(4):376–398.

Scott, D., McBoyle, G., and Schwartzentruber, M. (2004). Climate change and the distribu-

tion of climatic resources for tourism in north america. Climate research, 27(2):105–117.

Siple, P. A. and Passel, C. F. (1945). Measurements of dry atmospheric cooling in subfreezing

temperatures. Proceedings of the American Philosophical Society, 89(1):177–199.

Steadman, R. G. (1984). A universal scale of apparent temperature. Journal of Climate and

Applied Meteorology, 23(12):1674–1687.

Steiger, R. (2010). The impact of climate change on ski season length and snowmaking

requirements in tyrol, austria. Climate Research, 43(3):251–262.

Steiger, R., Scott, D., Abegg, B., Pons, M., and Aall, C. (2017). A critical review of climate

change risk for ski tourism. Current Issues in Tourism, pages 1–37.

Subak, S., Palutikof, J. P., Agnew, M. D., Watson, S. J., Bentham, C. G., Cannell, M. G. R.,

Hulme, M., McNally, S., Thornes, J. E., Waughray, D., and Woods, J. C. (2000). The impact

of the anomalous weather of 1995 on the u.k. economy. Climatic Change, 44(1):1–26.

Tervo, K. (2008). The operational and regional vulnerability of winter tourism to climate

variability and change: The case of the finnish nature-based tourism entrepreneurs. Scan-

dinavian Journal of Hospitality and Tourism, 8(4):317–332.

UKCIP (2002). Climate change scenarios for the United Kingdom: the UKCIP02 scientific report.

Tyndall Centre for Climate Mental Sciences University. Yang, J. and Wan, C. (2010). Progress in research on the impacts of global climate change

on winter ski tourism. Advances in climate change research, 1(2):55–62.

145

Yu, G., Schwartz, Z., and Walsh, J. E. (2009). A weather-resolving index for assessing the

impact of climate change on tourism related climate resources. Climatic Change, 95(3-

4):551–573.

Documents

Grant agreement no. 776479 - COACCHD2.4 Impacts on Industry, Energy, Services, and Trade CO Page 2 Version 1.1 This project has received funding from the European Union’s Horizon