Severe Storms Prediction and Global Climate Impact on the Gulf Coast, Rice University, October 29‐31, 2008
Radar-based flood forecasting: Quantifying hydrologic prediction uncertainty
Baxter E. Vieux, Ph.D., P.E. J.M. Imgarten, Graduate Research Assistant 1University of Oklahoma, School of Civil Engineering and Environmental Science, National Weather Center, 120 David L. Boren Blvd., Suite 3600, Norman, OK 73072; PH (405) 325-3600; email: [email protected] Overview
Stormwater runoff significantly impacts flooding and water quality in urban catchments. Weather radar captures the variability of rainfall over watersheds from catchment to river basin scale What is the accuracy that can be achieved for catchments and river basins, and how does it scale with space/time? Hydrologic Prediction Problem
OU BRAYS BAY KEEG E AN L S B O AY H O R U TE WILLOW WA 1. Radar QPE
Action Level
2. Model 3. Stream Forecasts
Sensing Æ Predicting River Basin Hydrologic Modeling Vflo Continuous
Drainage network and hydraulics determine hydrologic response without presumptive unit hydrographs. Setup with geospatial data and physically realistic parameters
Saturation and infiltration rate excess is Rainfall modeled as a single layer with variation Runon
throughout the basin. Runon Kinematic wave grid-grid and in channels Runoff
defined by surveyed cross-section, rating Runon Infiltration curves or trapezoids ∂ h ∂ uh)( Channel hydraulics include cross-section, + −= IR rating curves, trapezoidal, modified Puls and ∂ t ∂ x ∂ A ∂ Q)( looped rating curves + = q ∂ t ∂ x Continuous soil moisture tracking by S 12/ u = O h23/ adjusting climatological ET according to n available soil moisture and radar rainfall in ⎡∂ h ∂ q ⎤ e)( = NR T + −− IR = 0)( each grid cell. ∫Ω ⎢ ⎥ ⎣ ∂t ∂ x ⎦ Vieux. B.E., 2004. Distributed Hydrologic Modeling = − Δ [ −θ )1( +θ ]+ Δ [ −θ )1( +θ FFtQQStCACA ] Using GIS, Kluwer Academic Publishers, new old old new old new Real-time Forecasting
Radar input Operational Distributed Model, Vflo™ Forecast stage in real- time for operational decisions Used in the Rice University/Texas Medical Center Flood Alert ? System (www.fas.org) Urban Catchment
Radar data used in this analysis are derived from both S- band (NEXRAD) with accuracy enhanced through bias correction. Model accuracy is assessed using radar QPE derived from the existing WSR-88D (KHGX) as input to a physics-based hydrologic model. The study catchment, Harris Gully, is a 10km2 subwatershed of the 260 km2 Brays Bayou located in Houston TX. Parts of the stormwater sewer draining Harris Gully runs through Rice University and the TMC. Evaluative Study Model prediction accuracy with NEXRAD input Radar Quantitative precipitation estimates (QPE) derived from radar and rain gauge.
OU BRAYS BAY K E EEG AN L S B O AY H OU R TE WILLOW WA Distributed hydrologic prediction
June 8, 2004, observed and Vflo™ simulated hydrographs1. at Harris the Harris Gully outlet 3. Scaling of Gully predictability 4.5 4.0 3.5 3.0 2.5
Stage (ft) Stage 2.0 1.5 1.0 0.5 0.0
7:12 9:36 12:00Observed Simulated 14:24 16:48 19:12 21:36 Time (CDT) 4. NEXRAD June 8, 2004, observed and Vflo™ simulated hydrographs at2. the HarrisImproved Gully outlet 1x1 km input
4.5 Hydraulics
4.0
3.5
3.0
2.5
Stage (ft) 2.0
1.5
1.0
0.5
0.0
7:12 9:36 12:00 14:24 16:48 19:12 21:36 Observed Simulated 2 Time (CDT) Vflo Harris Gully 40x40m 10 km Radar storm total for June 14, 2005 over Harris Gully.
June 8, 2004, observed and Vflo™ simulated hydrographs at the Harris Gully outlet June 8, 2004, observed and Vflo™ simulated hydrographs at the Harris Gully outlet 4.5
4.5 4.0 4.0 3.5 3.5 3.0 3.0 2.5 2.5 2.0 Stage (ft)
Stage (ft) 2.0 1.5 1.5 1.0 1.0
0.5 0.5 0.0 0.0 7:12 9:36 12:00 14:24 16:48 19:12 21:36 7:12 9:36 12:00 14:24 16:48 19:12 21:36 Time (CDT) ObservedTime (CDT)Simulated Observed Simulated Prediction Performance
10
9 Difference Observed Simulated Difference Observed Simulated in Peak Peak Peak in Peak Event Peak (ft) Peak (ft) (ft) Time Time Time (min) 8 8-Jun-04 4.2 4.2 0 12:25 12:35 10 13-Jun-04 6.1 11 4.9 18:55 18:55 0 7 23-Oct-04 3.8 5.1 1.3 11:00 11:05 5 11-Jul-05 2.3 3.5 1.2 14:45 14:55 10 12-Jul-05 4.6 4.5 -0.1 18:10 18:15 5 6 13-Jul-05 6.7 6.5 -0.2 15:05 14:55 -10 14-Jul-05 6.4 8.9 2.5 14:55 15:25 30 5 15-Jul-05 4.8 5 0.2 5:25 5:25 0 29-Apr-06 4.5 5.1 0.6 8:55 8:55 0 6-May-06 4.6 2.3 -2.3 6:50 7:25 35 4 17-Jun-06 4.8 5.1 0.3 13:15 12:55 -20
Simulated Stage (ft) 3 2 Stage prediction accuracy = 1 02468101.05 ft RMSE Measured Stage (ft) 0 Peak arrival time accuracy = 10.0 min. RMSE
Distributed Hydrologic Modeling
Vflo™ Distributed Runoff Model 500m - Blue River (1200km2) 200m – Connerville (420km2)
Flow Direction Effective Porosity Soil Depth Hydraulic Conductivity Slope Channel Parameterization Cross-sections Blue River Selected Hydrographs Model Calibration Objective Functions
RMSE Volume (in)
RMSE Peak (cfs)
Search parameter space to find minima Rainfall Products Gauge Only
30,000
SimulatedG = 0.6892(Observed) R2 = 0.1032
) 20,000
Simulated Peak (cfs 10,000
0 0 10,000 20,000 30,000 Observed Peak (cfs)
Density of Mesonet Stations Number of Stations = 13 Average distance = 40 Km ReRe--AnalysisAnalysis ofof NWSNWS StageStage III/MPEIII/MPE
MFB statistics
MFB AD MFB CAD LB CAD 99 GaugesGauges MIN 0.226 1.2 0.1 0.0 AVE 0.992 20.8 15.8 7.0 MAX 2.600 395.0 139.9 55.9 Rainfall Products Radar MFB
30,000
SimulatedMFB = 1.0202 (Observed) R2 = 0.4229 )
20,000
Simulated Peak (cfs 10,000
0 0 10,000 20,000 30,000 Observed Peak (cfs)
14yr Storm total=122,640 hourly maps Hydrograph Predictions
Observed Gauge Only Radar Local Bias Radar MFB Quantitative Precipitation Estimate Verification by Streamflow
1.00
Event Runoff Volume 0.90 Event Peak Discharge
0.80
0.70
0.60 Volume Improvement
0.50 over Gauge-Only
0.40
Coefficientof Determination (R2) 0.30 Peak Improvement over Gauge-Only 0.20
0.10
0.00 MFB ABRFC MPE Gauge Only Re-analysis Raw Stage III Mesonet Temporal Scaling of Hydrologic Prediction Accuracy
DailyDaily StreamflowStreamflow MonthlyMonthly StreamflowStreamflow
Blue River near Connerville (420 km2)