A statistical, spatial, and hydrologic comparison of gauge-based and MPE-based rainfall measurements

Abstract

Gauge-based and multi-sensor precipitation estimation (MPE) data are compared on hourly, daily, monthly and event time scales at site locations over a 12-year period. Gauge data is collected at 16 sites within a 950 km2 portion of the Upper Iowa River in northeast Iowa. Average relative MPE bias is positive for all but the event time scale, and has a magnitude of less than 0.10 for all scales. Gauge and MPE average correlation coefficients range from 0.73 on the hourly scale to 0.92 on the event and monthly scales. The MPE relative bias standard deviation decreases from 1.70 mm on the hourly scale to 0.27 mm on the monthly scale. Decomposition of hourly bias reveals that the false positive portion is the most significant component. Seventy percent of MPE accumulation have a relative bias of 0.5 or less when hourly accumulations are 7 mm or greater. Pearson product-moment coefficient analysis reveals strong similarities in spatial correlations as a function of site separation. Rainfall time series for the basin are constructed from the two data sources and used as input to a Blocked Topmodel rainfall runoff scheme to provide another means of comparison on a basin-wide spatial scale. Five goodness-of-fit measures are used for quantifying the viability of simulated flows. No statistically significant difference in annual means using the difference sources is found for any of the measures.