Combined Use of Radar and Gauge Measurements for Flood Forecasting - PowerPoint PPT Presentation

Combined Use of Radar and Gauge Measurements for Flood Forecasting Using a Physics-based Distributed Hydrologic Model National Hydrologic Warning Council Dallas Texas October 23, 2003 Baxter E. Vieux, Ph.D., P.E. Vieux & Associates, Inc. Norman, Oklahoma USA www.vieuxinc.com And Professor of Civil Engineering and Environmental Science University of Oklahoma

Technological Advances in Rainfall Measurement • Advances in rainfall measurement technology have made new approaches to hydrologic prediction possible, and with more accuracy than ever before. • Technological advances in precipitation measurement (radar/satellite/gauge) and hydrologic modeling allow us to better plan, design, and forecast performance of drainage infrastructure in preparation for the next flood.

Distributed Radar Input NEXRAD 10 cm Doppler Radar — • 160+ installed • ~130 in US • Elsewhere internationally

Twin Lakes, Oklahoma • The first operational WSR-88D • Installed in May 1990 at Twin Lakes, Oklahoma • Prototyped at National Severe Storms Laboratory (NSSL), Norman, OK • Movie ‘Twister’

Radar measures reflectivity

Reflectivity and rainfall rate Radar rainfall— • Reflectivity depends on Z=300 R 1.4 drop size Z=250 R 1.2 distribution Z Rainfall , y t i v i t c e • Rainfall rate Rate, R l f e R depends on drop size distribution

Combining Systems Radar Rain Gauge Rain Gauge Better Rainfall Estimates than either system alone

Physics -based distributed modeling • “Physics-based” means that conservation laws of mass momentum and energy are used to make hydrologic predictions • Hydrodynamics are used to generate both flow rates and flood stage • Represents spatial variability of parameters and inputs • Distributed modeling is accomplished by subdividing the domain of interest • Fully distributed models use computational elements such as grid cells

Adapted from— Rhodda and Rhodda, Proceedings of the Royal Society, 1999. Classifying hydrologic models Deterministic Stochastic Black Box Hydrodynamics Conceptual (Neural Nets) Models that benefit from using radar inputs and Distributed Lumped geospatial data Fully Distributed Semi-Distributed Statistical Grid/Unit Subareas Distribution

Distributed Hydrologic Modeling ∂ ∂ h ( uh ) + = − R I ∂ ∂ t x Factors controlling runoff: Rainfall 1. Rainfall/Snowmelt Input Runon 2. Channel/overland Hydraulics Runon 3. Drainage network 4. Soil Infiltration/Impervious Runoff 5. Land Cover 6. Antecedent Moisture Runon 7. Water Control Structures Infiltration

V flo ™ Distributed Hydrologic Analysis and Prediction www.vieuxinc.com

Blue River— Importance of channel hydraulics • Basin located in south central Oklahoma. • Subject of longstanding research and the National Weather Service experiment to compare distributed models (DMIP) • 1200 km 2 modeled with 270 m resolution • NWS gauge-adjusted radar (NEXRAD Stage3) • Model simulations for 23 events (18 calibration and 5 verification) • Event based simulation initialized by simple soil moisture scheme.

Achievable Accuracy Case Studies • Within a distributed modeling framework, an important question is: How accurately can hydrographs be simulated using physics-based hydrologic models and gauge-adjusted radar?

Model setup Blue River

Blue River October 21, 1996 Discharge - Blue River Basin Uncalibrated, No Rating Curves, No Mod Puls Routing Initial Saturation of 30% 300 200 Discharge (cms) 100 0 10/21/1996 0:00 10/22/1996 0:00 10/23/1996 0:00 10/24/1996 0:00 10/25/1996 0:00 Date (UTC) Observed Simulated

Blue River March 25, 1997 Discharge - Blue River Basin Uncalibrated, No Rating Curves, No Mod Puls Routing Initial Saturation of 50% 100 Discharge (cfs) 50 0 3/25/1997 0:00 3/26/1997 0:00 3/27/1997 0:00 3/28/1997 0:00 Date (UTC) Observed Simulated

Blue river volume and peak V flo ™ Blue Volume RMSE= 9.8 mm 100.00 α =1.0 and β =1.0. y = 0.9464x 80.00 R 2 = 0.7412 Simulated (mm) 60.00 40.00 Peak Blue 20.00 600.00 0.00 0.00 20.00 40.00 60.00 80.00 100.00 Observed (mm) 400.00 3 /s) Simulated (m V flo ™ 200.00 RMSE= 52.0 m 3 s α =0.75 and β =1.0. 0.00 0.00 200.00 400.00 600.00 Observed (m 3 /s)

Texas Medical Center/Rice University Flood Alert System Urban real-time Main flood Street forecasting— • Texas Medical Center relies on Texas an operational Medical distributed model Center flood forecasting • Radar + V flo ™ Brays Bayou www.floodalert.org

Response Real-time prediction Information Flood Observations

V flo ™ Brays Bayou Gessner Main Street N W E Main Street S Z $ Brays Bayou Roughness 0.01 - 0.015 Z $ 0.015 - 0.018 0.018 - 0.025 0.025 - 0.05 > 0.05 No Data 0 20 Kilometers Drainage area 260 km 2 Model resolution 120 x120 m

Testing reliability • Optimizing the rising limb— Select a threshold and measure observed and simulated time to cross the threshold called time to flood (TTF). • Adjust parameters to optimize TTF, peak and time to peak for three calibration storms • Validate performance

Forecasts based on Hydrograph rising limb 18000.00 16000.00 Only optimizing for peak and 14000.00 Discharge (cfs) 12000.00 time to peak does not necessarily 10000.00 8000.00 match the rising limb making 6000.00 4000.00 forecast thresholds accurate 2000.00 0.00 6/5/2001 6/5/2001 6/5/2001 6/5/2001 6/5/2001 6/5/2001 6/5/2001 6/5/2001 6/5/2001 6/6/2001 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00 0:00 18000.00 Date (CDT) 16000.00 14000.00 Discharge (cfs) 12000.00 Optimizing for TTF improves 10000.00 8000.00 rate of rise that will be used in 6000.00 4000.00 a real-time flood alert system 2000.00 0.00 6/5/2001 6/5/2001 6/5/2001 6/5/2001 6/5/2001 6/5/2001 6/5/2001 6/5/2001 6/5/2001 6/6/2001 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00 0:00 Date (CDT)

1 st wave August 15 Verification

2 nd wave August 15 Verification

No model adjustment Gauge adjusted radar Verification— 15 August 2002 Verification event Main Street Discharge m 3 /s 8/15/2002 7:00 8/15/2002 12:00 8/15/2002 17:00 8/15/2002 22:00 8/16/2002 3:00 Date (EST) 8/16/2002 8:00 8/16/2002 13:00 8/16/2002 18:00 8/16/2002 23:00 8/17/2002 4:00 8/17/2002 9:00

Historic event performance Radar to Stream Gauge Volume •Verification 200 of QPE using Adjusted 180 y = 1.076x stream gauge 160 R 2 = 0.9646 volumes Radar Rainfall Volume (mm) 140 •Radar 120 100 adjustment 80 improves Unadjusted (+) 60 y = 1.1003x efficiency R 2 = 0.2129 40 from 20 R 2 =0.2129 to 0 0 50 100 150 200 R 2 =0.9646 Stream Gauge Volume (mm)

Rainfall Runoff Prediction in Real- Time • Rainfall-runoff prediction is particularly important for a variety of applications such as water resources management, flood prediction, emergency management.

Hydrographs Greenville Louisberg Measured Simulated TS Allison

V flo ™ Predicted Inundation Web Display

Hurricane Floyd Transportation Impacts Pitt-Greenville Airport (PGV), Pitt County Photo Courtesy of North Carolina Emergency Management

Stage Sensitivity Summary Calibration sensitivity 90 78.9 80 70 60 % difference 50 40 30.7 30 15.0 20 9.1 8.3 7.6 10 1.3 0 rainfall channel channel overland infiltration channel hydraulic width side slope slope slope roughness

Summary 1. Physics-based distributed modeling can produce accurate predictions in real-time at any location in a drainage network. 2. Made possible by technological advances in radar rainfall measurement 3. Consistent performance across storm sizes/type 4. Physically realistic parameters from geospatial data 5. High achievable accuracy in peak and rising limb predictions given good channel hydraulic data 6. Event reconstruction tests reliability of operational flood forecasting systems

Further information Vieux B.E. 2002. “Predictability of Flash Floods Using Distributed Parameter Physics-Based Models.” Report of a Workshop on Predictability & Limits-To-Prediction in Hydrologic Systems , Committee on Hydrologic Science, Water Science and Technology Board, Board on Atmospheric Sciences and Climate, National Research Council, ISBN 0-309-08347-8. pp. 77-82. Vieux, B.E., and F.G. Moreda, (2003). Ordered Physics-Based Parameter Adjustment of a Distributed Model. Chapter 20 in Advances in Calibration of Watershed Models , Edited by Q. Duan, S. Sorooshian, H.V. Gupta, A.N. Rousseau, R. Turcotte, Water Science and Application Series, 6 , American Geophysical Union, ISBN 0-87590- 355-X pp. 267-281. Vieux. B.E., (2001) Distributed Hydrologic Modeling Using GIS , ISBN 0- 7923-7002-3, Kluwer Academic Publishers, Norwell, Massachusetts, Water Science Technology Series, Vol. 38. p. 293. Second Edition expected 2004 English and Chinese

--Ganges River Distributary, Bangladesh www.vieuxinc.com Questions?