Modeling Range-Dependent Biases in Long-Term Radar-Based Precipitation Estimates

A statistical model is presented that minimizes mean-field and range-dependent biases in long-term radar precipitation estimates from radar networks. The model uses collocated radar-gauge pairs as input data to compute bias as a function of range for individual effective radar coverage areas. Additionally, this study introduces percent of normal precipitation as an indicator of range-dependent biases, using gauges to calibrate the magnitude of the mean-field bias. The algorithm can be applied to a single radar or a network of radars and is robust enough to adapt to input radar precipitation estimates with varying levels of quality control, as long as an appropriate precipitation climatology is available. The bias adjustment model is tested on daily Stage IV precipitation data that were aggregated to monthly and multi-month time scales across 104 WSR-88Ds in the central and eastern United States. The reduction of the RMS Stage IV precipitation estimate errors across the study domain between 2009-2012 was about 40% to 50% across all time scales.