Welcome To
This Presentation
Downscaling and Modeling the Climate of Blue Nile River Basin-Ethiopia
By: Netsanet Zelalem
Supervisors: 1. Prof. Dr. rer.nat.Manfred Koch, Kassel University2. Dr. Solomon Seyoum, IWMI, Ethiopia
Nov9/2012Kassel University, Germany
Statement of the Problem
• High population pressure, poor water and land management and climate change are inducing declining agricultural productivity and vulnerability to climate impact [Haileslassie et al., 2008].
• In order to alleviate poverty and food insecurity, it is widely recognized to utilize water resources such as Blue Nile.
• So, assessment of the impact of climate change on future water resource may provide substantial information to the area where more than 85% of the basin depends entirely on rain-fed agriculture.
Objective
• Evaluate the possible relationships between large scale variables with local meteorological variables.
• Evaluate the most common statistical downscaling methods, SDSM and LARSWG, for the assessment of the hydrological conditions of the basin.
• Generate climate change scenarios for the basin using different emission scenarios and AOGCMs (Atm.and Ocean).
• Investigate the possiblity of climate change on hydrology in UBRB based on the downscaled meteorological scenario data.
• Provide streamflow predictions of the basin for current and downscaled future climate conditions.
Contents
• Background on Climate System• Study Area• Data collection, analysis and results• Climate Modeling• Results of Climate Modeling• Conclusions
Background (Climate system)Climate is a statistical description of weatherincluding averages and variability.The earth climate system is an interaction of various
components of climate system: Ocean Land surface Atmosphere Cryospher Biosphere Anthropogenic
---Background (Climate system)• Climate Change: refers to a statistical significant variations
that persist for an extended period, typically decades or longer.
• The mea annual global temperature has increased by about 0.3-0.60C since the late 19 century.
---Background (Climate change Impact )• Today, the impact of climate change become the
biggest concern of mankind.
---Background (Climate Change Impact)• This will impact the hydrology of the watershed systems
and hence it exhibits long-term changes.
---Background (Climate Change Impact)• This impact needs integrated modeling to evaluate
alternate future watershed scenarios.• IPCC findings indicate that developing countries, such as
Ethiopia, will be more vulnerable to climate change
Higher Relative Risks
Lower Relative Risks
---Background (Climate Model)• Climate Models try to simulate the likely responses of
climate system to a change in any of the parameter interactions between them mathematically.
• Generally refers as GCMs (Global Circulation Models)• The 3-D model formulation is based on the fundamental
laws of physics:Conservation of energyConservation of momentumConservation of mass andThe “Ideal Gas Law”
---Background (Emission Scenarios)• Emission scenarios are
important components and tools for the modeling of climate change (Werner and Gerstengarbe, 1997)
Emissions 2011-2030 2046-2065 2080-2099
A2 0.64 1.65 3.13
A1B 0.69 1.75 2.65
B1 0.66 1.29 1.79
---Background (Downscaling GCM)• In climate change impact
studies, hydrological modeling:Are usually required to
simulate sub-grid scale phenomenon.
Require input data (such as pcp, temp) at similar sub-grid scale.
• Downscaling is a means of relating the large scale atmospheric predictor variables to local scale so as to use for hydrological model inputs.
---Background (Downscaling Methods)1. Dynamic downscaling
Extract local-scale information by developing and using regional climate models (RCMs) with the coarse GCM data used as boundary conditions.
2. Statistical downscalingDrive the local scale information from the larger scale
through inference from the cross-scale relationship.
It Can be categorized in to three typesRegression downscalingStochastic weather generatorsWeather typing schemes
---Background (Statistical downscaling)
1. Regression downscaling techniques: Predicted=f(Predictors). The function f could be. Linear or non-linear regression.2.Stochastic weather generators: The relationships between daily weather generator
parameters and climatic average can be used to characterize the nature of future daily statistics (wilby, 1999).
---Background (Statistical downscaling)3. Weather typing schemes Involve grouping local, meteorological variables in
relation to different classes of atmospheric circulation. Future regional climate scenarios are constructed by:
Resembling from observed variable distribution Climate change is then estimated by determining the
change of the frequency of weather classes.
Study area
---Study AreaFeatures of Upper Blue Nile watershed
The total area=176,000 km2
Latitude: 7° 45’ and 12° 45’N and longitude: 34° 05’ and 39° 45’E
Altitude: Min. 485m to Max. 4,257m aslUBNB has 14 sub-basinsIt contributes 40% of Ethiopia surface water resources
[World Bank 2006]87% of the Nile flow at Aswan dam is from Ethiopia
from this UBNB contributes 60% and the Atbara (13%) and the Sobat (14%)
Data sources
Data Name Sources
PrecipitationMaximum TemperatureMinimum Temperature
NMA
NCEP www.ncep.noaa.gov
GCMs
WCRP CMIP3 Multi-Modal data sethttp://esg.llnl.gov:8080/index.jsp
World Climate Data Center http://www.mad.zmaw.de/wdc-for-climate/cera-data-model/index.html
Data Collection and Quality Checking• After collection of precipitation data from 53 stations and
temperature from 33 stations for 1970-2000 period at daily time scale, data quality( Such as, filling missing data and consistency check) control has been conducted.
• Areal precipitation and temperature based on ThiessenPolygon method: Stn.Results:
Sub-Basin Results of Observed Data
Large Scale Data
Criterion to chose GCMs
1. Based on outputs of
MAGICC-SCENGEN
2. Based on data availability
3. Based on their participation
IPCC-AR4
4. Allowable number of GCMs
ECHAM-5, GFDLCM21 and
SCIRO-MK3
Data of selected GCMs
• A1b and A2 emission scenarios are considered to account the worst (A2) and the middle(A1B).
• Re-griding has been done using Xconv package.
GCM EmissionScenario of A1B and A2
Current ConditionScenario
65 yearsInto Future Scenario
100 years Into FutureScenario
AtmosphericResolutions(Deg)
Echam5 1970-2000 2046-2065 2081-2100 1.9x1.9
GFDLCM2.1 1970-2000 2046-2065 2081-2100 2.0x2.5
CSIRO-MK3 1970-2000 2046-2065 2081-2100 1.9x1.9
NCEP 1970-2000 2.5X2.5
Large-scale Predictor VariablesS
No Predictor variablesDesign
ation
S
No Predictor variablesDesignat
ion
1 Air pressure at sea level mslp 11 Northward wind @850mpa p8_v
2 Precipitation flux prat 12 Northward wind @500mpa p5_v
3 Minimum air temperature tmin 13 Meridional surface wind speed p_v
4 Maximum air temperature tmax 14 Specific humidity @850mpa s850
5 Surface air tempratur@2m temp 15 Specific humidity @500mpa s500
6 Air temperature @850mpa t850 16 Geopotential height @850mpa p850
7 Air temperature@500mpa t500 17 Geopotential height @500mpa p500
8 Eastward wind@850mpa p8_u 18 Relative humidity @500mpa r500
9 Eastward wind@500mpa p5_u 19 Relative humidity @850mpa r850
10 Zonal surface wind speed p_u
Large Scale Data
Re-analysis grid lines covering the study areaName of
Subbasin
Grid box
considered
Name of
sub basin
Grid box
considered
Tana 22 and 23 Anger 12and 22
Belles 12,13,
22 and 23
Wonbera 12 and 22
Dabus 12 Muger 22 and 32
D idessa 11,12,
21and 22
Beshilo 22,23,
32 and 33
Guder 22 Wolaka 22 and 32
Fincha 22 N/Gojam 22 and 23
S/Gojam 22 Jimma 22 and 32
Statistical Downscaling Tools
• Two statistical downscaling tools:• *SDSM: A regression based statistical downscaling
model (wilby, et al., 2002)
• *LARS-WG: Long Ashton Research Station Stochastic Weather Generators (Semenov et al, 1998).
SDSM: A regression based Statistical Downscaling models
• Identify predictand relationships using multiple linear regression techniques.
• The predictor variables provide daily information concerning the large-scale state of the atmosphere,
• The predictand describes condition at the site scale.
LARS-WG• Generate precipitation, min and
max temperature.• Semi-empirical distributions are
used to state a day as wet/dry series.
• Semi-empirical distributions are used for precipitation amounts, dry/wet series.
• Semi-empirical distributions are used for Temperature. It is conditioned on wet/dry status of a day.
Cases considered• Three cases are employed in climate modeling
• All the cases are applied for each of 14 sub-basins in UBNB.
Type GCMs Emission Period ToolsCase-1 echam5 a1b, a2 2050s, 2090s SDSMCase-2 echam5 a1, a2 2050s, 2090s LARS-WGCase-3 Echam5, gfdl21
& csiro-mk3a1b, a2 2050s, 2090 LARS-WG
Climate modeling-Case1 SDSM reduces the task into a number
of discrete processes as follows: 1. Quality control of data and
transformation. 2. Selection of appropriate predictor
variables for model calibration. 3. Calibrate Model. 4. Generate the daily data. 5. Analyze the outputs. 6. Scenario generation: Then analysis
of climate change scenarios
Selecting predictor variables• Predictor is selected based on correlation analysis off-line of
SDSM and using SDSM screening methods in the software.
SDSM Calibration ApproachModel calibration is performed in two approaches:Unconditional: It assumes a direct link between the
regional-scale predictors and the local predictand. • Maximum and minimum temperatureConditional: depend on an intermediate variable such as
the probability of wet-day occurrence, intensity, amount etc.
• PrecipitationThe performance of calibration result for each sub basin
Results-Case1
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
2050s_A1B
56.5398570682714
50.6178592203631
18.7517964941868
20.680995641808
17.6311267120
3
30.3049786971396
50.0999916256843
53.0836467398137
9.47728277495148
18.6656023946175
45.4398817586599
71.6198618835854
2050s_A2
58.037791609252
40.7018163799141
16.152452080642
18.3858223906661
16.4661748979571
28.9431768157827
42.7483752505986
47.8417676360748
8.90037634372148
21.4068305948191
38.3561599392011
59.4481031407495
2090s_A1B
108.789001475902
73.2560000113955
42.0655849483734
35.4505823281153
41.5313957283356
82.0454731115789
102.226095199723
100.220246311381
15.4098335664761
32.4556475754166
66.9899801297406
109.605317992617
2090s_A2
111.14322060747
84.0650531908927
45.861560978674
39.4375328118225
45.1181453291102
90.6020198906623
108.758896122186
107.088399085689
20.9833029139384
32.2608226714777
72.0939149812107
116.656888395815
1030507090
110130
Simulated RF change from observed at Muger
Re
la
tive
ch
an
ge
(%
)
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2047
2049
2051
2053
2055
2057
2059
2061
2063
2065
2082
2084
2086
2088
2090
2092
2094
2096
2098
2100
0500
10001500200025003000 Trend line of simulated RF at Muger
Observed Control 2050s_A1B2050s_A2 2090s_A1B 2090s_A2
RF
(m
m)
Results-case1Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
2050s_A1B
-11.9930058267654
-32.0980301738977
-38.547057330613
-36.2256394665523
-48.2886452950688
-46.186351170232
-41.957561295554
-23.4202136313885
-11.0395047910016
-31.7250646563016
-21.1008447573487
-9.82121216341527
2050s_A2
-4.46968300268275
-34.1996522532874
-38.155983099162
-28.0665395772585
-42.176674085796
-42.6980184567346
-34.4911947966843
-25.1120590860266
-17.0114756425644
-24.3961467042082
-21.3427597564296
-9.97921040628668
2090s_A1B
-20.3668397911135
-52.6938176530704
-63.1316600966
6
-65.6534137933796
-66.1744957152001
-64.7676491513639
-58.1420110835669
-51.806218694887
-55.4310153262774
-38.0095288946405
-31.7302803329002
-23.5413602886664
2090s_A2
-31.3900435522635
-54.8524440900349
-65.7767566643183
-65.6271651253081
-68.4318014870236
-67.9810644647242
-61.1894460632419
-49.9965652795155
-49.0428098605348
-42.6868309053216
-44.7927641388414
-24.3913811212018
-70
-50
-30
-10
Simulated RF change from observed at Wonbera
Re
la
tiv
e ch
an
ge
(%
)19
7019
7319
7619
7919
8219
8519
8819
9119
9419
9720
0020
4820
5120
5420
5720
6020
6320
8120
8420
8720
9020
9320
9620
99
0
500
1000
1500
2000
2500Trend line of simulated RF at Wonbera
Observed Control2050s_A1B 2050s_A2
RF
(m
m)
Climate Modeling –Case2 The weather generator consists of three main sections: Model calibrationAnalysis of observed station data in order to calculate the weather
generators. Model validationQtest is used for determining how well the model is simulating
observed conditions. The statistical characteristics of the observed data are compared
with those of the synthetic data. Model useGenerating the synthetic weather based on the available data
parameter generated during model calibration or by combining scenario file with the generated parameter to account climate change.
Incorporating Climate Scenario• Climate changes derived from GCMs can be incorporated
in stochastic weather generator by applying climate change scenarios expressed on a monthly basis in the relevant climate variable.
e5ab_2050 e5a2_2090
month m.rain wet dry min max tsd rad m.rain wet dry min max tsd radJan 1.66 1.04 0.97 2.31 1.85 1.31 1.00 2.94 1.01 0.98 2.54 1.51 1.06 1.00Feb 2.20 0.97 1.02 2.60 1.89 1.13 1.00 1.20 1.01 1.00 2.08 1.99 1.04 1.00Mar 0.91 0.98 1.01 2.39 2.60 1.53 1.00 1.05 0.99 0.99 2.00 2.36 1.26 1.00Apr 1.11 1.05 1.00 2.19 1.95 1.10 1.00 1.12 1.03 1.01 1.85 1.49 1.06 1.00May 0.85 1.30 1.17 2.73 3.06 1.27 1.00 0.87 0.74 1.03 2.33 2.43 1.17 1.00Jun 0.80 0.98 0.84 2.97 4.06 1.23 1.00 0.87 0.89 1.04 2.70 3.63 1.12 1.00Jul 1.00 1.44 1.18 2.96 3.59 1.18 1.00 1.02 1.48 1.06 2.56 2.91 1.24 1.00Aug 1.24 1.54 1.80 2.27 1.71 1.22 1.00 1.20 1.76 1.28 2.18 1.80 1.13 1.00Sep 1.17 1.62 1.98 2.15 1.63 1.52 1.00 1.10 1.52 1.34 1.90 1.56 1.19 1.00Oct 1.01 1.24 1.31 2.56 2.23 1.32 1.00 1.05 1.03 0.84 2.26 1.79 1.16 1.00Nov 1.33 0.98 0.98 2.92 2.00 1.33 1.00 1.16 1.02 1.03 2.49 1.85 1.07 1.00Dec 2.93 1.02 1.01 3.17 2.21 1.54 1.00 2.33 1.00 1.00 2.55 1.87 1.04 1.00
Results-Case2
Winter Spring Summer Autumn
pcpa1b_2050s
125.937342978295
0.222532507815593
1.47799219153169
-1.4035038016
4694
pcpa2_2050s
74.8228589525257
1.12651227796044
1.14450375511057
-4.3898727272
6738
pcpa1b_2090s
4.90951271560589
-11.463109518
4856
-0.4338644706
63076
10.0827704673863
pcpa2_2090s
-17.502366327
1532
-4.3769455586
3903
-1.5848989934
8576
10.886970651294
-25
25
75
125
UBNB Seasonal pcp
Re
lati
ve
ch
an
ge
(%
)Tm
xa1b_
2050s
Tmxa
2_2050
sTm
na1b_
2050
sTm
na2_2
050s
Tmxa
1b_209
0sTm
xa2_2
090s
Tmna
1b_20
90s
Tmna
2_209
0s
1.0
2.0
3.0
4.0
5.0 WinterSpringSummerAutumn
Tem
per
atu
re C
han
ge
(0C
)
UBNB Seasonal Temprature Change
Climate Modeling: Case-3• The methodology is same as case-2. • The climate change scenario is constructed from 3GCMs.
Winter Spring Summer Autumn
pc-pa1b_2050s
0.154857092426357
-3.116675556
97433
-5.199142559
43182
-6.146566987
79637
pcpa2_2050s
3.90832887314738
-2.063423865
45666
-4.239632875
90028
-6.815913772
98443
pc-pa1b_2090s
-14.62056405
03331
-8.542518929
40405
-5.739208881
47881
4.22672097466272
pcpa2_2090s
-11.46112809
87629
-8.079263240
83682
-6.188251033
69393
5.16059747666773
-18
-13
-8
-3
3
8
UBNB Seasonal pcp
Rela
tive
chan
ge (%
)
Comparison of Mono-Modal and Multi-Modal Approaches
• Multi-modal approach under estimated pcp prediction and this is more apparent in 2050s than 2090s.
• Annual relative % change in pcp increases due to relatively high increase in dry periods.
• Tmx and Tmn change has no significant difference between two approaches in 2050s.
• Multi-modal approach underestimates both Tmx and Tmn during 2090s
• Summer season in the case of mono-modal is warmer while spring season is warmer in multimodal approach.
Comparison of Mono-Modal and Multi-Modal Approaches
Mono/Multi-modal Comparisons
Comparisons of SDSM and LARS-WG outputs
• Generally, downscaled precipitation results from SDSM and LARS-WG show marked difference.
• Both downscaling tools illustrate an increase in maximum and minimum temperature in both 2050s and 2090s time compare with the base line period.
SDSM and LARS-WG Comparison
SDSM and LARS-WG Comparison
Conclusions
• LARS-WG performs better in precipitation prediction than SDSM.
• simulation of future precipitation using SDM has significant spatial variation than LARS-WG.
• LARS-WG illustrate similar trend across each sub-basins in the simulation of precipitation, maximum and minimum temperature.
• LARS-WG shows better performance over the study area than SDSM.
THANK YOU.