Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
A global perspective on developments in
operational flood forecasting systems
Dr Jan Verkade, Deltares
BHS Conference on Flood Forecasting, London, October 1, 2019
Brief introduction: Jan Verkade @ Deltares
Deltares:
• Dutch national R&D institute for water management
• Developers of Delft-FEWS forecast production system – backbone of systems
such as those of EA, SEPA, NRW, BoM, NWS, RWS and many more
Jan:
• Hydrologist; forecasting system developer
(England, Ireland, Québec, Netherlands, Belgium, Sudan)
• R&D coordinator @ Deltares, “Enabling Early Warning” programme
• Fluvial forecaster at Rijkswaterstaat River Forecasting Service
(Dutch rivers Rhine and Meuse; EFAS)
This presentation: “perspective on developments” → my perspective
Forecast – decision – response chain
Source: Verkade, J. S. ‘Estimating Real-Time Predictive Hydrological Uncertainty’. Ph.D. dissertation, Delft University of Technology, 2015. http://dx.doi.org/10.4233/uuid:a7e8ac36-4bdb-4231-a11e-
d46778b2ae4a
Development ‘themes’
1. Towards better forecasts
2. Towards better decisions
3. Organizations and cooperation
Theme 1: Towards better forecasts
“And very little ever gets done…”???
Are we there yet? → where’s “there”?
• Estimates of predictive uncertainty that can justifiably be interpreted as probabilistic forecasts
• Forecasters are able to effectively communicate our forecasts to downstream users
• End users are able to effectively ingest, use the forecasts
• …
• Many (most even) systems: ensemble NWP → model → hydro ensemble
• But few include other, non future weather related, uncertainties
• Few agencies communicate uncertainty estimates to end users
• Few end users are able to effectively use estimates of uncertainty
in their forecast informed decision making
(Scratching the surface)
Data assimilation: using near real-time data in your forecast
• Various types: error correction, parameter updating, state updating, input updating
• DA can have a considerable “black box” feeling to it
• Mathematically complex – little if any physics
• Not always clear where ‘goodness’ originates: model chain, or data assimilation
• Increased availability of (notably, remotely sensed) data adds to appeal of DA:
• Soil moisture estimates, for updating states of a hydrological model
• River ice, for updating/changing rating curves, changing parameters of hydrodynamic
models
• These appear to be making their way to operational systems
(Rijkswaterstaat, Province of Québec)
• Various agencies are trying to reduce the black-boxiness of
ARMA (EA, NRW, Wales)
Impact Forecasting
• Generally defined as “forecasting not only hazards, but also impacts” → towards “risk”
• Considerable attention in meteorology, scientific realm; lots of pilots
• Few, if any, operational systems include forecasts of consequences of flooding
• Often, however, impacts are addressed in forecast informed decision making
• More of an uptake in forecasting systems operated by / used for NGOs
(e.g., Red Cross Forecast based financing)
# people affected (ensemble member X) expected # people affected
Nowcasting
• “Nowcasting concerns the accurate description of the current weather situation together with
very short-term forecasts obtained by extrapolating the real-time observations” (Foresti e.a.,’16)
• Active field of research
• Propagation of radar observed precipitation
• Temporal and spatial development of convective precipitation
• Use of satellite observed precip
• Blending of NWP and nowcasts
• Some uptake in operational flood forecasting systems
• Visualization of radar estimated precip fairly widely done
• Some meteorological agencies make available
nowcasting products
• Some hydro agencies even route these through their
models
Foresti, L., Reyniers, M., Seed, A. and Delobbe, L.: Development and verification of a real-time stochastic precipitation nowcasting system for urban
hydrology in Belgium, Hydrol. Earth Syst. Sci., 20(1), 505–527, doi:10.5194/hess-20-505-2016, 2016.
Reservoir Optimization
• Multi-purpose reservoirs: often conflicting objectives
• Hydropower production: maintain highest possible
water level
• Flood protection: keep reservoir as empty as possible
• Additional ‘constraints’: recreational use, ecological
objectives, energy prices
• Development:
simple reservoir rules → feedback control → predictive
control
• Uptake of either of these varies – depending on nature of
problem owner
(hydropower producer v public safety authorities)
Black Box models
• “Any model that does not explicitly simulate physical processes”
• Neural networks, Artifical Intelligence, Machine Learning
• Fueled by “Big Data”
• Are they ‘better’? Jury’s still out…
• Promising applications: AI models of processes where human behavior has significant role
• Reservoir ops
• Weirs
• Pumps
Forecast Verification – How good is my forecast?
• Various reasons: administrative, diagnostic, operational
• Not a given – many agencies do not routinely verify past forecasts
• the Why and How of forecast verification aren’t always
immediately clear
• expertise or capacity may be unavailable
• feedback mechanism not always well understood
• sometimes, reluctancy to face the facts
• On the upside
• incidental verification is done
• some agencies are considering, even developing operational
verification systems
• most existing systems are operated by hydropower producers
(BPA, TVA, CEATI)
Open data
• Increasingly often, organizations make available their data
• NWP providers – often, observations also (e.g., DWD, NCEP, ECCC, KNMI, Met Éireann)
• Remote sensing operators (satellites)
• Hydrological forecasting agencies: forecasts, observations
• Rationale includes:
• Innovation
• New knowledge from combined data
sources and patterns in large data
volumes
• Quite a few forecasting systems make use
of multiple NWP products originating from
abroad – if anything, for robustness
Flood forecasting on continental, global scales
• Forecasting mandates often reside with national,
sub-national authorities
• Increasing number of continental, global
forecasting systems
• Europe: EFAS (JRC)
• Global: GLOFAS (JRC), E-Hype (SMHI), gloffis
(Deltares), Google, …
• Rationales include: catering for different target
audience, scientific development
• R&D at global scale feeds back to ‘local’
systems
• Forecast quality is ever-increasing. Will it ever
be comparable to that of ‘local systems’?
• Differences with ‘local’ systems: role of a human
forecaster
Theme 2: Towards better decisions
4 oktober 2019
On the assumption of rationality: “prospect theory”
• Expected values:
• A: 50% x 1,000 + 50% x 0 = 500
• B: 100% x 450 = 450
• Yet more people choose B over A…
• Observed decision making is inconsistent with
“risk based decision making”
“Conventional wisdom is that false alarms
reduce the public's willingness to respond
to future events. This paper questions this
conventional wisdom.”
http://depts.washington.edu/forecast/wordpress/wp-content/uploads/2013/06/NDM-poster_5.9.13.pdf
Unacceptably high rates of non-compliance with weather warnings
(e.g., only 46% compliance among the most vulnerable residents in
Sandy’s path, Baker & Downs, 2013) may be due in part to a “cry
wolf” effect (Breznitz, 1985): ignoring warnings due to the past
experience of false alarms.
Rebecca Pliske, a psychologist, found that veteran meteorologists would make weather
forecasts first by looking at the data and forming an expert judgment; only then would
they look at the computerised forecast to see if the computer had spotted anything that
they had missed. (Typically, the answer was no.) By making their manual forecast first,
these veterans kept their skills sharp, unlike the pilots on the Airbus 330. However, the
younger generation of meteorologists are happier to trust the computers. Once the
veterans retire, the human expertise to intuit when the computer has screwed up could be
lost.
https://www.theguardian.com/technology/2016/oct/11/crash-how-computers-are-setting-us-up-disaster
Should we improve our forecasts, or our decisions?
Source: Verkade, J. S. ‘Estimating Real-Time Predictive Hydrological Uncertainty’. Ph.D. dissertation, Delft University of Technology, 2015. http://dx.doi.org/10.4233/uuid:a7e8ac36-4bdb-4231-a11e-
d46778b2ae4a
Increasingly often addressed by stakeholders and scientists – emphasis on uncertainty
mgt:
•University of Washington’s Decision Making With Uncertainty Lab
https://depts.washington.edu/forecast/
•Rand Corporation’s Decision Making Under Uncertainty Lab
•Deltares’ Decision Making in Uncertainty Lab
•Society for Decision Making Under Deep Uncertainty (www.deepuncertainty.org)
Forecast informed decision making
Some related background materials
• US Storm Prediction Center, “Forecast Decision
Making”: https://youtu.be/ayadlJlQM9k
• Susan Joslyn, “Uncertainty Forecasts and the General
Public”: https://youtu.be/SfXlt40StpA
• Hurricane Dorian New York Times article:
https://www.nytimes.com/interactive/2019/08/29/opinio
n/hurricane-dorian-forecast-map.html
Theme 3: Organizations and co-operation
4 oktober 2019
Communities of Practice
• Organizations actively seek other organizations for inspiration,
help, joint R&D
• Quite a few ‘mechanisms’ around, including
• HEPEX Hydrometeorological Community of Practice
(www.hepex.org)
• Delft-FEWS Community for Real-time Hydro Forecasting
• UK “Four Agencies” initiative: NRW, SEPA, EA, DfI
• Bilateral agreements, such as MoU between EA and BoM
• Provinces of Québec and Alberta – exploration of joint R&D
• Increasingly often, these are used to exchange expertise and
experience
• Training programs
• Forecasting algorithms
• …
Summarizing…
• Uncertainty estimation: on our way, lots to do still
• Big Data, AI, et cetara have yet to find their way to operations
• Some developments in operational systems take place in other
realms:
• Impact forecasting: largest development in systems operated
by/for NGOs
• Verification systems: largest development in systems operated
by hydropower agencies
• Lots of attention to ‘improving forecasts’, less so to ‘improving
decisions’
https://www.deltares.nl/en/blog/the-future-of-realtime-hydrological-forecasting/