Can Radar Quantitative Precipitation Estimates Reproduce Extreme Precipitation Statistics in Central Arizona?

In this study, we assess the ability of 4-km, 1-h Quantitative Precipitation Estimates (QPEs) from the Stage IV analysis of the NEXRAD radar network to reproduce the statistics of extreme precipitation (P) in central Arizona, USA. As reference, we use 19 years of records from a dense network of 257 rain gages. For each radar pixel and gage record, we fit the generalized extreme value (GEV) distribution to the series of annual maximum P at durations, τ, from 1 to 24 hours. We found that the GEV scale and shape parameters estimated from the radar QPEs are slightly negatively biased when compared to estimates from gage records at τ = 1 h; this bias tends to 0 for τ ≥ 6 h. As a result, the radar GEV quantiles for return period, T_R, from 2 to 50 years exhibit negative bias at τ = 1 h (median between -23% and -12% for different T_R’s), but the bias is gradually reduced as τ increases (average of +4% for τ = 24 h). The relative root-mean-square-error (RRMSE) ranges from 17% to 44% across all τ’s and T_R’s and these values are similar to those computed between gages and operational design storms from NOAA Atlas 14. Lastly, we found that radar QPEs reproduce fairly well observed scaling relationships between the GEV location and scale parameters and P duration, τ. Results of our work provide confidence in the utility of Stage IV QPEs to characterize the spatiotemporal statistical properties of extreme P and, in turn, to improve the generation of design storm values.