Flood Forecasting for Bangladesh Tom Hopson NCAR Journalism Fellowship June 14-18, 2010.
Operational Flood Forecasting for Bangladesh: Tom Hopson, NCAR Peter Webster, GT
description
Transcript of Operational Flood Forecasting for Bangladesh: Tom Hopson, NCAR Peter Webster, GT
Operational Flood Forecasting for Bangladesh:Operational Flood Forecasting for Bangladesh:
Tom Hopson, NCARTom Hopson, NCARPeter Webster, GTPeter Webster, GT
A. R. Subbiah and R. Selvaraju, ADPCA. R. Subbiah and R. Selvaraju, ADPC
Climate Forecast Applications for Bangladesh (CFAB): Climate Forecast Applications for Bangladesh (CFAB): USAID/CARE/ECMWF/NASA/NOAAUSAID/CARE/ECMWF/NASA/NOAA
Bangladesh StakeholdersBangladesh Stakeholders: Bangladesh Meteorological Department, Flood Forecasting and Warning Center, : Bangladesh Meteorological Department, Flood Forecasting and Warning Center, Bangladesh Water Development Board, Department of Agriculture Extension, Disaster Management Bureau, Bangladesh Water Development Board, Department of Agriculture Extension, Disaster Management Bureau, Institute of Water Modeling, Center for Environmental and Geographic Information Services, CARE-Institute of Water Modeling, Center for Environmental and Geographic Information Services, CARE-BangladeshBangladesh
Contact: [email protected]: [email protected]
Overview:Bangladesh flood forecasting
I. CFAB forecasting context
II. Forecasting techniques -- using Quantile Regression for:1. precipitation forecast calibration2. Post-processing to account for all errors
III. 2007 Floods and Warning System Pilot Areas
(World Food Program)
Damaging Floods:large peak or extended durationAffect agriculture: early floods in May, late floods in September
Recent severe flooding: 1974, 1987, 1988, 1997, 1998, 2000, 2004, and 20071998: 60% of country inundated for 3 months, 1000 killed, 40 million homeless, 10-20% total food production2004: Brahmaputra floods killed 500 people, displaced 30 million, 40% of capitol city Dhaka under water2007: Brahmaputra floods displaced over 20 million
River Flooding
CFAB Project: Improve flood warning lead time
Problems:1.Limited warning of upstream river discharges (Dhaka: ~24-48 hr warning, only)2.Precip forecasting in tropics difficult
Skillful CFAB forecasts benefit from:1. Large catchments => river discharge results from “integrated” inputs over large spatial and temporal scales2. Skillful data inputs: ECMWF, TRMM, CMORPH, CPC-rain gauge3. Partnership with Bangladesh’s Flood Forecasting Warning Centre (FFWC)
=> daily border river stage readings useful for data assimilation
Daily Operational Flood Forecasting Sequence
Forecast Trigger: ECMWF forecast files
Updated TRMM-CMORPH-CPC precipitation estimates
Updated distributed model parameters
Updated outlet discharge estimates
Above-critical-level forecast probabilities transferred to Bangladesh
Lumped Model Hindcast/Forecast Discharge Generation
Distributed Model Hindcast/Forecast Discharge Generation
Multi-Model Hindcast/Forecast Discharge Generation
Discharge Forecast PDF Generation
Calibrate model
Statistically corrected downscaled forecasts
Generate forecasts Generate hindcasts Generate forecasts Generate hindcasts
Update soil moisture states and in-stream flows
Generate hindcasts
Calibrate AR error model
Calibrate multi-model
Generate forecasts Generate hindcasts
Generate forecasted model error PDF
Convolve multi-model forecast PDF with model error PDF
Generate forecasts
7
Example of Quantile Regression (QR)
Our application
Fitting precipitation quantiles using QR conditioned on:
1) Reforecast ens
2) ensemble mean
3) ensemble median
4) ensemble stdev
5) Persistence
8
Calibration Procedure
For each quantile:
1) Perform a “climatological” fit to the data
2) Starting with full regressor set, iteratively select best subset using forward step-wise cross-validation
– Fitting done using QR– Selection done by:
a) Minimizing QR cost functionb) Satisfying the binomial distribution
3) 2nd pass: segregate forecasts into differing ranges of ensemble dispersion, and refit models.
=> have different calibration for different atmospheric stability regimes
=> ensures ensemble skill-spread has utility
Pro
babi
lity
Precipitation
obs ForecastPDF
Pre
cip
TimeForecastsobserved
Regressors for each quantile: 1) corresponding ensemble 2) ens mean 3) ens median 4) ens stdev 5) persistence
Significance of Weather Forecast Uncertainty on Discharge Forecasts
3 day 4 day
Calibrated Precipitation Forecasts Discharge Forecasts
1 day 4 day
7 day 10 day
1 day 4 day
7 day 10 day
Daily Operational Flood Forecasting Sequence
Forecast Trigger: ECMWF forecast files
Updated TRMM-CMORPH-CPC precipitation estimates
Updated distributed model parameters
Updated outlet discharge estimates
Above-critical-level forecast probabilities transferred to Bangladesh
Lumped Model Hindcast/Forecast Discharge Generation
Distributed Model Hindcast/Forecast Discharge Generation
Multi-Model Hindcast/Forecast Discharge Generation
Discharge Forecast PDF Generation
Calibrate model
Statistically corrected downscaled forecasts
Generate forecasts Generate hindcasts Generate forecasts Generate hindcasts
Update soil moisture states and in-stream flows
Generate hindcasts
Calibrate AR error model
Calibrate multi-model
Generate forecasts Generate hindcasts
Generate forecasted model error PDF
Convolve multi-model forecast PDF with model error PDF
Generate forecasts
Producing a Reliable Probabilistic Discharge Forecast
Step 1: generate discharge ensembles from precipitation forecast ensembles (Qp):
1/51
1
Qp [m3/s]
Prob
abil
ity
Step 3: combine both uncertainty PDF’s to generate a “new-and-improved” more complete PDF for forecasting (Qf):
Qf [m3/s]
1Pr
obab
ilit
y
Step 2: a) generate multi-model hindcast error time-series using precip estimates;b) conditionally sample and weight to produce empirical forecasted error PDF:
1000
-1000
forecasthorizon
time
PDF 1
-1000 1000Residual [m3/s]
[m3/s]
Residuals
=>
a) b)
2004 Brahmaputra Ensemble Forecasts and Danger Level Probabilities
3 day 4 day
5 day
3 day 4 day
5 day
7 day 8 day
9 day 10 day
7-10 day Ensemble Forecasts
7 day 8 day
9 day 10 day
7-10 day Danger Levels
Brahmaputra Discharge Forecast VerificationBrahmaputra Discharge Forecast VerificationRank Histograms
Brier Skill Scores
CRPS Scores
Daily Operational Flood Forecasting Sequence
Forecast Trigger: ECMWF forecast files
Updated TRMM-CMORPH-CPC precipitation estimates
Updated distributed model parameters
Updated outlet discharge estimates
Above-critical-level forecast probabilities transferred to Bangladesh
Lumped Model Hindcast/Forecast Discharge Generation
Distributed Model Hindcast/Forecast Discharge Generation
Multi-Model Hindcast/Forecast Discharge Generation
Discharge Forecast PDF Generation
Calibrate model
Statistically corrected downscaled forecasts
Generate forecasts Generate hindcasts Generate forecasts Generate hindcasts
Update soil moisture states and in-stream flows
Generate hindcasts
Calibrate AR error model
Calibrate multi-model
Generate forecasts Generate hindcasts
Generate forecasted model error PDF
Convolve multi-model forecast PDF with model error PDF
Generate forecasts
Five Pilot Sites chosen in 2006 consultation workshops based on biophysical, social criteria:
Rajpur Union -- 16 sq km-- 16,000 pop.
Uria Union-- 23 sq km-- 14,000 pop.
Kaijuri Union-- 45 sq km-- 53,000 pop.
Gazirtek Union-- 32 sq km-- 23,000 pop.
Bhekra Union-- 11 sq km-- 9,000 pop.
A v e r a g e D a m a g e ( T k . ) p e r H o u s e h o l d i n P i l o t U n i o n
7 , 2 5 5
2 8 , 7 4 5
6 0 , 9 9 3
6 4 , 0 0 0
4 0 5 8
0
1 0 , 0 0 0
2 0 , 0 0 0
3 0 , 0 0 0
4 0 , 0 0 0
5 0 , 0 0 0
6 0 , 0 0 0
7 0 , 0 0 0
U r i a G a z i r t e k K a i j u r i R a j p u r B e k r a
U n i o n
Average Damage (Tk) per
Household
2007 Brahmaputra Ensemble Forecasts and Danger Level Probabilities
7-10 day Ensemble Forecasts 7-10 day Danger Levels
7 day 8 day
9 day 10 day
7 day 8 day
9 day 10 day
Community level responses to 2007 flood forecastsCommunity level responses to 2007 flood forecasts
• Planned evacuations to identified high grounds with adequate communication and sanitation facilities
Economically, they were also able to:• Move livestock to high lands
with additional dry fodder.
• Early harvesting of rice and jute anticipating floods.
• Protected fisheries by putting nets in advance
Selvaraju (ADPC-UNFAO)
ConclusionsConclusions 2003: Daily operational probabilistic discharge forecasts “experimentally” disseminated based
on lumped model and 51-member ECMWF ensemble
2004: -- Multi-model and post-processing approach operational-- initializing watersheds using TRMM / CMORPH-- Forecasts automated-- CFAB became Bangladesh federal government entity-- forecast the severe Brahmaputra floods
2005: CFAB became HEPEX test bed
2006: -- Forecasts incorporated into national flood warning program and hydraulic model-- 5 vulnerable pilot areas designated and trained on using 1-10day probabilistic forecasts.
2007: 5 pilot areas warned many days in-advance during two severe flooding events
2008-2009: Ongoing expansion of the warning system thoughout Bangladesh
Further technological improvements through HEPEX test bed collaborations
2003: Daily operational probabilistic discharge forecasts “experimentally” disseminated based on lumped model and 51-member ECMWF ensemble
2004: -- Multi-model and post-processing approach operational-- initializing watersheds using TRMM / CMORPH-- Forecasts automated-- CFAB became Bangladesh federal government entity-- forecast the severe Brahmaputra floods
2005: CFAB became HEPEX test bed
2006: -- Forecasts incorporated into national flood warning program and hydraulic model-- 5 vulnerable pilot areas designated and trained on using 1-10day probabilistic forecasts.
2007: 5 pilot areas warned many days in-advance during two severe flooding events
2008-2009: Ongoing expansion of the warning system thoughout Bangladesh
Further technological improvements through HEPEX test bed collaborations
Thank [email protected]