HR: 0800h
AN: H51B-0361    [Abstracts]
TI: Flood Estimation at Ungaged Locations: a Hybrid Geographic and Predictor-Variable Region-of-Influence Regression Model
AU: * Eng, K
EM: keng@usgs.gov
AF: U.S. Geological Survey, National Research Program, 12201 Sunrise Valley Drive, MS 430, Reston, VA 20192 United States
AU: Milly, P C
EM: cmilly@usgs.gov
AF: U.S. Geological Survey, National Research Program, Geophysical Fluid Dynamics Laboratory/NOAA, Route 1, Forrestal Campus, Princeton, NJ 08542 United States
AU: Tasker, G D
EM: gdtasker@usgs.gov
AF: U.S. Geological Survey, National Research Program, 12201 Sunrise Valley Drive, MS 430, Reston, VA 20192 United States
AB: Establishment of hydrologically similar regions based on a set of basin characteristics for region-of-influence regression is often insufficient for estimating floods, because this set is often missing aspects of basin properties and climate, such as profiles of hydraulic conductivity. The goal of this study was to improve flood estimates at ungaged locations by incorporating missing aspects using proximity in geographic space in addition to predictor-variable space to form hydrologic regions. Three types of region-of-influence (RoI) regression approaches were used to estimate values of the 50-year-return peak discharge. The analysis was performed with streamflow records from 1,091 gages in 10 states in the southeastern United States. Parameter values of regression models were estimated from gaged sites contained within a RoI. The RoI was established in three ways: the geographic RoI (GRoI) contained the n geographically closest sites; the predictor-variable RoI (PRoI) contained the n closest sites in (normalized) predictor-variable space; and the hybrid RoI (HRoI) contained the n closest sites in predictor-variable space that were less than a geographic distance, D, from the site of the regression. Parameter estimates were calculated by a generalized-least-squares (GLS) procedure, which (in comparison with ordinary least squares) minimized biases induced by cross-correlation of the flow records at spatially close sites. A split-sampling framework was employed, with the RoI parameters n and D optimized over one set of gages and performance evaluated over the remaining gages. The HRoI approach yielded lower estimation errors than either the PRoI or GRoI approaches. It is concluded, for the 50-year peak flow in the study region, the basin characteristics considered (area, slope, and annual precipitation) provide important, but incomplete information that is helpful to improving the estimation of discharge at ungaged location. Consideration of geographic proximity of stations boosts the ability to accurately estimate discharge by providing a useful surrogate for basin characteristics, such as storm arrival rates, that determine discharge that are not included in the analysis.
DE: 1816 Estimation and forecasting
DE: 1821 Floods
DE: 1860 Streamflow
DE: 1869 Stochastic hydrology
DE: 1874 Ungaged basins
SC: Hydrology [H]
MN: Fall Meeting 2005