Estimation of Radar-Rainfall Error Spatial Covariance

Characterization of radar-rainfall error spatial covariance requires high-density rain gauge networks and high quality data. The authors use data from two rain gauge networks: IIHR ground validation network at Iowa City, Iowa consisting of 25 platforms, with average intergauge distance of about 5 km, and the Iowa City Municipal Airport network consisting of 9 platforms and covering approximately 1 km2. In both networks, each platform is equipped with dual tipping-bucket rain gauges for redundancy and improved data quality. Using two years of data, the authors obtain the correlation structure of point rainfall at daily, hourly and 15-min temporal scales. The radar-rainfall product is based on Level II reflectivity data from the Davenport, Iowa WSR-88D radar. Radar rainfall estimates obtained at different spatial and temporal resolutions are compared with the rainfall from the gauge network. The radar grid resolutions used are HRAP grid, 2×2 km2 grid and polar grid (10 ×1 km) and the temporal resolutions are daily, hourly and 15-min intervals. The authors investigate several approaches to estimating radar-rainfall error covariance. These include a method analogous to the error variance separation that uses only radar data at the rain gauge locations and accounts for gauge representativeness error, an extension of this approach that uses radar data at neighboring locations, and a non-parametric approach based on using parabolic kernel. The authors discuss development of the methods, their assumptions and differences, and early results.