By Audrey Ishii¹, Terry Ortel¹, Thomas Over¹,and Elizabeth Murphy¹

A near real-time streamflow simulation system for Salt Creek in northeastern Illinois is being tested and enhanced by the U.S. Geological Survey in cooperation with Du Page County Department of Development and Environmental Concerns.  Various precipitation inputs using gaged and National Weather Service NEXt generation RADar (NEXRAD) data are being tested and compared for use in near real-time simulation. Rainfall runoff is simulated for 62 square miles of the watershed and routed through 23 miles of mainstem and various tributaries in order to evaluate and optimize operations at various flood-control reservoirs.  The rainfall-runoff time series are generated from the Hydrologic Simulation Program--Fortran (HSPF) utilizing inputs of precipitation, air and dewpoint temperatures, solar radiation, wind, and computed potential evapotranspiration, and are input to the dynamic-wave model Full EQuations (FEQ) for routing through channel and control structures.   The GENeration and analysis of model simulation SCeNarios (GENSCN) program is utilized for input, simulation, and display of the input meteorologic and hydrologic data, and for display and analysis of the routed streamflow stage and discharge hydrographs, reach profiles, and storage volumes.

The watershed is divided into two major subwatersheds for the purpose of near real-time simulation:  Upper and Lower Salt Creek. Six U.S. Geological Survey telemetered streamgage stations are located along Salt Creek. For both the Upper and Lower Salt Creek subwatersheds, observed streamgage station data are used as the upstream boundary conditions.  The two subwatersheds respond differently for single rainfall events because the Upper Salt contains reservoirs that result in a much longer time to peak following a rainfall event than the time to peak following a rainfall event in the Lower Salt.   The observed stages at the streamgages at river miles 31.8, 20.1, and 14.9 are utilized as checks in near real-time on simulation accuracy, which provides a qualitative measure of confidence in the predicted stages.

Seven telemetered rainfall gages are located within the Salt Creek subwatershed boundaries, but a set of three and a separate pair are located within 2.5 miles of each other, effectively providing backups during operations, when some may have gage malfunctions or data-transmission failures.  Therefore, four rainfall gages were selected as representative for precipitation coverage—two rainfall gages to represent the Upper Salt Creek subwatershed, and two gages for the Lower Salt Creek subwatershed.  The assignment of watershed area to rainfall gages for the Upper Salt was determined by reference to Thiessen polygons based on gage locations and subwatershed boundaries.  For the Lower Salt, the division between subwatersheds for rainfall gage assignment is determined by the hydraulic characteristics of the subwatershed—the narrow neck down which the floodwave travels, and the wider downstream area from which local flooding is generated.  Missing data are filled with the nearest apparently functioning rainfall gage.  This situation may result in effectively reducing the number of rainfall gages for each subwatershed to one gage; therefore, the one-gage case also was simulated for each subwatershed and compared to the two-gage case for quality of simulated peak stages.   All gage data was adjusted by factor of 1.14 based on comparison with weighing gage network data.  The NEXRAD Stage III/Multisensor Precipitation Estimates data are obtained from the North Central River Forecasting Center (NCRFC) for grid cells with areas of 6.18-square miles (16-square kilometers), and processed as hourly precipitation inputs.  Synthetic precipitation inputs are an average of the areally weighted grid-cell values covering the subwatershed areas.

In order to evaluate the expected quality of forecasts, a suite of eight events (recurrence intervals ranged from less than 2 years to more than 25 years at the various locations) since October 1999, were simulated with three precipitation input scenarios for each of the subwatersheds—one and two rainfall-gage and the weighted-sum average NEXRAD precipitation inputs.   The peak-elevation differences in all cases were generally less than 0.5 foot, but were significantly larger using the NEXRAD inputs.  In practice, the more input simulation data that are required in near real-time, the slower and more labor intensive the simulation process will be.  Both data sets from rainfall gages and NEXRAD are subject to outages both in collection and transmittal of the data that require trade-off in time spent estimating the missing data for presumed increases in accuracy.

Ishii, A.L., Ortel, T.W., Over, T.M., Murphy, E.A., 2003, Nexrad and rainfall-gage precipitation inputs for near real-time flood simulation of Salt Creek in Du Page County, Illinois (abstract) in 2003 National Hydrologic Warning Council/Southwestern Association of ALERT Systems Abstracts, Dallas, p. 43.