Sensitivity analysis of variability in reflectivity-rainfall relationships on runoff prediction
Chakradhar Goud Malakpet, Univ. of Louisiana, Lafayette, LA; and E. Habib, E. A. Meselhe, and A. Tokay
Radar-rainfall information presents a significant potential for improving our ability to provide accurate and timely flood predictions. However, it is recognized that radar-rainfall estimates are subject to several sources of uncertainty. The current study focuses on uncertainty caused by natural variability in radar reflectivity-rainfall relationships and how it impacts runoff predictions. An intensive analysis is performed on disdrometric rain drop size distribution data for the year 2004, which are obtained from a 35-km2 experimental watershed in Lafayette, southwest Louisiana. Runoff response at the watershed outlet is simulated using a well calibrated physically-based distributed hydrologic model, GSSHA (Gridded Surface Subsurface Hydrologic Analysis). Rainfall intensity, R, is estimated from one minute disdrometric reflectivity, Z, using different Z-R relationships. Z-R relationships are derived at different time scales: climatological, daily, event-based, and physical-process based). The effect of using different estimation methods, such as least square fitting, bias-corrected, fixed-exponent, and default operational relation, are also examined. Z-R relationships for physical processes are derived by classifying and separating rain records into convective, transitional and stratiform phases based on vertical profile of radar reflectivity, developed from volume-scan WSR-88D radar data. A filtering technique was adapted to eliminate uncertainties like, drop sorting, small sampling volume and instrumental noise in one minute DSDs of disdrometer.
The quantitative and qualitative assessments of errors in runoff prediction at the watershed outlet are carried out for three periods in 2004, which have distinct physical rainfall characteristics. The first period (June 21-28, 2004), is mostly convective with high intensity squall line storms crossing Lafayette, on 24 and 25 June, 2004. These two events have produced about seven inches of rainfall. The second period (October 04-12, 2004), includes Tropical Storm Matthew which made landfall near Houma, southeast Louisiana, on the morning of October 10. Tropical Storm Matthew, observed in Lafayette, measured rainfall amounts as high as 10 inches on October 8-9, 2004. The third period (November 17-30, 2004), consists of scattered convective squall line storms with high intensities and short duration. All the events in this period recorded about one inch of rainfall each and five inches in total. Variations in the simulated runoff hydrographs due to variability in Z-R relationships is quantified and compared to reference runoff hydrographs. The reference runoff simulations are based on rainfall intensities estimated from the drop size distribution measurements. The results showed that more accurate runoff predictions could be obtained when a Z-R relationship, which corresponds to the physical process of the event, is used for estimating rainfall intensity. There are significant errors in runoff predictions when climatological and daily Z-R relationships are used for rainfall estimation. Event averaged and physical process based Z-R relationships produce results which are in good agreement with the reference values. However, the physical process based Z-R relationship being the most accurate. Initially this analysis is performed on reflectivity values obtained from disdrometer to check the validity of the analysis. The analysis will be repeated using the WSR-88D radar reflectivity data to see whether similar conclusions can be made.
Extended Abstract (588K)
Session 3, Coupled Hydrological Modeling
Wednesday, 17 January 2007, 8:30 AM-11:15 AM, 213A
Previous paper Next paper
Browse or search entire meeting
AMS Home Page