Daily to Seasonal Operational Flood Forecasting Tom Hopson, NCAR and ADPC

1. Daily to Seasonal Operational Flood Forecasting Tom Hopson, NCAR and ADPC Peter Webster, CFAB and Georgia Tech A. R. Subbiah, ADPC

2. Overview: Bangladesh flood forecasting I. Overview of daily to seasonal weather forecast products II. Seasonal forecasting: Bangladesh CFAB example III. Short-term forecasting: Bangladesh CFAB example 1. Where does good predictability derive? 1. precipitation forecast bias removal 2. multi-model river forecasting 3. accounting for all error: weather and hydrologic errors IV. Future Work: Dartmouth Flood Observatory

3. Utility of a Three-Tier Forecast System SEASONAL OUTLOOK: Long term planning of agriculture, water resource management & disaster mitigation especially if high probability of anomalous season (e.g., flood/drought) 30 DAY FORECAST: Broad-scale planning schedules for planting, harvesting, pesticide & fertilizer application and water resource management (e.g., irrigation/hydro-power determination). Major disaster mitigation resource allocation. 1-10 DAY FORECAST: Detailed agriculture, water resource and disaster planning. E.g., fine tuning of reservoir level, planting and harvesting.

4. forecast products for hydrologic applications • Seasonal -- ECMWF System 3 - based on: 1) long predictability of ocean circulation, 2) variability in tropical SSTs impacts global atmospheric circulation - coupled atmosphere-ocean model integrations - out to 7 month lead-times, integrated 1Xmonth - 41 member ensembles, 1.125X1.125 degrees (TL159L62), 130km • Monthly forecasts -- ECMWF - “fills in the gaps” -- atmosphere retains some memory with ocean variability impacting atmospheric circulation - coupled ocean-atmospheric modeling after 10 days - 15 to 32 day lead-times, integrated 1Xweek - 51 member ensemble, 1.125X1.125 degrees (TL159L62), 130km • Medium-range -- ECMWF EPS - atmospheric initial value problem, SST’s persisted - 6hr - 15 day lead-time forecasts, integrated 2Xdaily - 51 member ensembles, 0.5X0.5 deg (TL255L40), 80km • Short-range -- RIMES - 26-member Country Regional Integrated Multi-hazard Early Warning System (RIMES) WRF Precipitation Forecasts - 3hr - 5 day lead-time, integrated 2X daily - 9km resolution

5. Motivation for Generating Ensemble Discharge Forecasts (from ensemble weather forecasts) • Greater accuracy of ensemble mean forecast (half the error variance of single forecast) • Likelihood of extremes • Non-Gaussian forecast PDF’s • Ensemble spread as a representation of forecast uncertainty

6. Overview: Bangladesh flood forecasting I. Overview of daily to seasonal forecast products II. Seasonal forecasting: Bangladesh CFAB example III. Short-term forecasting: Bangladesh CFAB example 1. Where does good predictability derive? 1. precipitation forecast bias removal 2. multi-model river forecasting 3. accounting for all error: weather and hydrologic errors IV. Future Work: Dartmouth Flood Observatory

8. Seasonal rainfall prediction for 2006 An example of seasonal predictions of precipitation issued in JFMA 2006 (left) and MJJA 2006 (right), to be compared with the observed rainfall (dotted line) and climatology (dashed line). The seasonal forecasts correctly indicate months in advance ‘higher than normal’ rainfall.

9. Overview: Bangladesh flood forecasting I. Overview of daily to seasonal forecast products II. Seasonal forecasting: Bangladesh CFAB example III. Short-term forecasting: Bangladesh CFAB example 1. Where does good predictability derive? 1. precipitation forecast bias removal 2. multi-model river forecasting 3. accounting for all error: weather and hydrologic errors IV. Future Work: Dartmouth Flood Observatory

10. CFAB Project: Improve flood warning lead time • Problems: • Limited warning of upstream river discharges • Precipitation forecasting in tropics difficult Good forecasting skill derived from: 1. good data inputs: ECMWF weather forecasts, satellite rainfall 2. Large catchments => weather forecasting skill “integrates” over large spatial and temporal scales 3. Partnership with Bangladesh’s Flood Forecasting Warning Centre (FFWC) => daily border river readings used in data assimilation scheme

11. 1) Rainfall Inputs Rain gauge estimates: NOAA CPC and WMO GTS 0.5 X 0.5 spatial resolution; 24h temporal resolution approximately 100 gauges reporting over combined catchment 24hr reporting delay Satellite-derived estimates: NASA TRMM 0.25X0.25 spatial resolution; 3hr temporal resolution 6hr reporting delay geostationary infrared “cold cloud top” estimates calibrated from SSM/I and TMI microwave instruments 3) Satellite-derived estimates: NOAA CPC “CMORPH” 0.25X0.25 spatial resolution; 3hr temporal resolution 18hr reporting delay precipitation rain rates derived from microwave instruments (SSM/I, TMI, AMSU-B), but “cloud tracking” done using infrared satellites 4) Weather forecasts: ECMWF GCM 51-member ensemble weather forecasts at 1-day to 15-day forecast lead-times (nominal resolution about 0.5degree)

12. Comparison of Precipitation Products: Rain gauge, GPCP, CMORPH, ECMWF

13. 2) Spatial Scale -- Increase in forecast skill (RMS error) with increasing spatial scale -- Logarithmic increase

14. Merged FFWC-CFAB Hydraulic Model Schematic Primary forecast boundary conditions shown in gold: Ganges at Hardinge Bridge Brahmaputra at Bahadurabad 3) Benefit: FFWC daily river discharge observations used in forecast data assimilation scheme (Auto-Regressive Integrated Moving Average model [ARIMA] approach)

15. Daily Operational Flood Forecasting Sequence

16. Transforming (Ensemble) Rainfall into (Probabilistic) River Flow Forecasts Rainfall Probability Discharge Probability Rainfall [mm] Discharge [m3/s] Above danger level probability 36% Greater than climatological seasonal risk?

17. ECMWF 51-member Ensemble Precipitation Forecasts 5 Day Lead-time Forecasts => Lots of variability • 2004 Brahmaputra Catchment-averaged Forecasts • black line satellite observations • colored lines ensemble forecasts • Basic structure of catchment rainfall similar for both forecasts and observations • But large relative over-bias in forecasts

18. Forecast Bias Adjustment • done independently for each forecast grid • (bias-correct the whole PDF, not just the median) Model Climatology CDF “Observed” Climatology CDF Pmax Pmax Precipitation Pfcst Padj 25th 50th 75th 100th 25th 50th 75th 100th Quantile Quantile In practical terms … ranked forecasts ranked observations 0 1m 0 1m Precipitation Precipitation

19. Bias-corrected Precipitation Forecasts Original Forecast Brahmaputra Corrected Forecasts Corrected Forecast => Now observed precipitation within the “ensemble bundle”

20. Daily Operational Flood Forecasting Sequence

21. Discharge Multi-Model Forecast • Multi-Model-Ensemble Approach: • Rank models based on historic residual error using current model calibration and “observed” precipitation • Regress models’ historic discharges to minimize historic residuals with observed discharge • To avoid over-calibration, evaluate resultant residuals using Akaike Information Criteria (AIC) • If AIC minimized, use regression coefficients to generate “multi-model” forecast; otherwise use highest-ranked model => “win-win” situation!

22. 2003 Model Comparisons for the Ganges (4-day lead-time) hydrologic lumped model hydrologic distributed model Resultant Hydrologic multi-model

23. Multi-Model Forecast Regression Coefficients - Lumped model (red) - Distributed model (blue) • Significant catchment variation • Coefficients vary with the forecast lead-time • Representative of the each basin’s hydrology -- Ganges slower time-scale response -- Brahmaputra “flashier”

24. Daily Operational Flood Forecasting Sequence

25. Significance of Weather Forecast Uncertainty Discharge Forecasts Precipitation Forecasts Discharge Forecasts 1 day 4 day 1 day 4 day 3 day 4 day 10 day 7 day 7 day 10 day

26. What do we mean by “calibration” or “post-processing”? “bias” obs Forecast PDF Probability Probability Forecast PDF obs “spread” or “dispersion” calibration Basin Rainfall [mm] Basin Rainfall [mm] • Post-processing has corrected: • the “on average” bias • as well as under-representation of the 2nd moment of the empirical forecast PDF (i.e. corrected its “dispersion” or “spread”)

27. 1 PDF Step 1: generate discharge ensembles from precipitation forecast ensembles (Qp): Probability 1/51 Qp [m3/s] Step 2: a) generate multi-model hindcast error time-series using precip estimates; b) conditionally sample and weight to produce empirical forecasted error PDF: a) 1000 forecast horizon b) 1 Residuals PDF [m3/s] time => Residual [m3/s] -1000 1000 -1000 1 Step 3: combine both uncertainty PDF’s to generate a “new-and-improved” more complete PDF for forecasting (Qf): Probability Qf [m3/s] Producing a Reliable Probabilistic Discharge Forecast

28. Significance of Weather Forecast Uncertainty Discharge Forecasts 2004 Brahmaputra Discharge Forecast Ensembles Corrected Forecast Ensembles 7 day 8 day 7 day 8 day 3 day 4 day 5 day 9 day 10 day 9 day 10 day

29. 2004 Brahmaputra Forecast Results 2 day Above-Critical-Level Cumulative Probability Confidence Intervals Critical Q black dash 50% 95% 7 day 8 day 7 day 8 day 3 day 4 day 5 day 9 day 10 day 9 day 10 day

30. 2007 Brahmaputra Ensemble Forecasts and Danger Level Probabilities 7-10 day Ensemble Forecasts 7-10 day Danger Levels 7 day 8 day 7 day 8 day 9 day 10 day 9 day 10 day

31. Overview: Bangladesh flood forecasting I. Overview of daily to seasonal forecast products II. Seasonal forecasting: Bangladesh example III. Short-term forecasting: Bangladesh example 1. Where does good predictability derive? 1. precipitation forecast bias removal 2. multi-model river forecasting 3. accounting for all error: weather and hydrologic errors IV. Future Work: Dartmouth Flood Observatory

32. Satellite-based River Discharge Estimation Bob Brakenridge, Dartmouth Flood Observatory, Dartmouth College

33. http://www.dartmouth.edu/~floods/ • River Watch • Day/Night Flood detection on a near-daily basis regardless of cloud cover. • Measurement of river discharge changes; current flood magnitude assessments • Immediate map-based prediction of what is under water • Access to rapid response detailed mapping as new maps are made • Access to map data base of previous flooding and associated recurrence intervals.

34. Application to the Ganges River Basin

35. MODIS sequence of 2006 Winter Flooding 2/24/2006 C/M: 1.004 3/15/2006 C/M: 1.029 3/22/2006 C/M: 1.095

36. Objective Monitoring of River Status: The Microwave Solution The Advanced Microwave Scanning Radiometer - Earth Observing System (AMSR-E) is a twelve-channel, six-frequency, passive-microwave radiometer system. It measures horizontally and vertically polarized brightness temperatures at 6.9 GHz, 10.7 GHz, 18.7 GHz, 23.8 GHz, 36.5 GHz, and 89.0 GHz. Spatial resolution of the individual measurements varies from 5.4 km at 89 GHz to 56 km at 6.9 GHz. AMSR-E was developed by the Japan Aerospace Exploration Agency (JAXA) and launched by the U.S. aboard Aqua in mid-2002.

37. One day of data collection (high latitudes revisited most frequently)

38. Example: Wabash River near Mount Carmel, Indiana, USA Black square shows Measurement pixel. White square is calibration pixel.

39. Site 98, Wabash River at New Harmony, Indiana, USA

40. Guide to Predicting Inundation Irrawaddy River, Burma The current hydrologic status and discharge or C/M ratio can be used to determine present inundation extent. 2/17/2003 1.18 9/1/2002 1.82 7/24/2004 2.17

41. Conclusions • 2003: CFAB Brahmaputra/Ganges forecasts went operational • 2004: -- Forecasts fully-automated -- forecasted severe Brahmaputra flooding event • 2007: 5 pilot areas warned many days in-advance during two severe Brahmaputra flooding events Future Work • Dartmouth Flood Observatory river discharge estimates assimilated for improved skillful long-lead forecasts • Fully-automated forecasting scheme relying on global inputs (ECMWF forecasts, satellite rainfall) rapidly and cost-effectively applied to other river basins with in-country capacity building

42. Thank You!