Rainfall Input for Distributed Hydrologic Modeling: The Case for Radar

Physically-based distributed parameter hydrologic models are very sensitive to rainfall input. Inaccuracies in the space-time description of rainfall are typically amplified through runoff predictions. Rain gages do not represent ground truth for rainfall at the watershed scale because they are truly point samples while watersheds are more sensitive to the spatial distribution of rainfall. The use of rain gage data necessitates spatial interpolation of the rainfall field. Weather radars provide an indication of areally-averaged precipitation within the radar pulse-volume. Unfortunately, there are no unique relations between the radar observables and the rainfall rate. Furthermore, radar-rainfall estimates are contaminated by a host of random and systematic error sources. In this study, we examine the impact of each separate source of radar-rainfall estimation error on runoff predictions from the hydrologic model CASC2D. Two different approaches are used. First, S-band single- and dual-polarization radar observations of the extreme warm-rain convective storm that impacted Fort Collins, Colorado, on July 28, 1997, are converted to rainfall rates using published algorithms. Secondly, a physically based atmospheric model of convective rainfall is coupled with an active microwave radiative transfer model to simulate radar observations of a thunderstorm. This methodology allows direct assignment of radar-rainfall errors. Results address the relative importance of random and systematic errors, including range effects and vertical profile of reflectivity, on runoff predictions. Additionally, simulations driven by rain gage and radar-rainfall input are compared to illustrate the impact of the different sampling characteristics on model performance.