On the parametric and NEXRAD-radar sensitivities of a distributed hydrologic model suitable for operational use

Abstract

The recent deployment of the NEXRAD (WSR-88D) radars in the United States has made spatially distributed rainfall data available in an operational environment. In addition, national databases now exist with high-resolution digital terrain-elevation, soils, and land-use and land-cover data. The availability of such spatially distributed data brings into focus the question of the utility of hydrologic models with spatially distributed parameters and input in an operational environment. Are the data and models accurate and reliable enough to allow their use in an operational environment? The present study discusses a methodology for addressing this issue based on a sensitivity analysis of event streamflow to parameter and radar input within a Monte Carlo framework for the Illinois River basin in Oklahoma and Arkansas. This framework allows the incorporation of parametric and radar-rainfall uncertainties in the analysis and illustrates their effects on the results of the sensitivity. The particular distributed model used incorporates components that are adaptations of operational models used often in a spatially-lumped manner to produce estimates of soil water, runoff and streamflow in watersheds with an area of order 103km2. A GIS is used to sub-divide the study watershed into small sub-catchment units with areas of a few hundred square kilometers. Soil water accounting and channel routing models are used to generate runoff and streamflow over the network of sub-catchments and streams. Spatially-distributed soil databases are used to condition relevant soil–water model parameters to reflect the soil field capacity. The main conclusions are that: (a) the distributed model forced by NEXRAD data produces results comparable to those produced by the operational spatially-lumped models using raingauge data; (b) sensitivity of flow statistics on various spatial scales to parameters and radar-rainfall input that are uncertain is scale dependent; and (c) in view of parametric and input uncertainty, in several cases, a spatially lumped model has response that cannot be statistically distinguished from that of the distributed model. This study is an initial step toward understanding the issues surrounding the use of distributed models in an operational environment. Repeating the sensitivity analysis for other areas with different geomorphologic, soils and climate characteristics is necessary to investigate aspects pertaining to higher variability in soils and terrain elevation, and the presence of snow accumulation and melt. Sensitivity analysis of soil water estimates to parameters and radar rainfall input is also a natural extension of this work.