Operational Flood Forecasting for Bangladesh: Tom Hopson, NCAR Peter Webster, GT A. R. Subbiah and...

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











Calibrate model




Generate hindcasts






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

PDF

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











Calibrate model




Generate hindcasts






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]

Operational Flood Forecasting for Bangladesh: Tom Hopson, NCAR Peter Webster, GT A. R. Subbiah and...

Documents

Transcript of Operational Flood Forecasting for Bangladesh: Tom Hopson, NCAR Peter Webster, GT A. R. Subbiah and...