GPM Precipitation Estimates over the Walnut Gulch Experimental Watershed/LTAR site in Southeastern Arizona

Precipitation profiles from the GPM Core Observatory Dual-frequency Precipitation Radar (DPR) form part of the a priori database used in GPM Goddard Profiling (GPROF) algorithm passive microwave radiometer retrievals of rainfall. The GPROF retrievals are in turn used as high quality precipitation estimates in gridded products such as IMERG. Due to the variability in and high surface emissivity of land surfaces, GPROF performs precipitation retrievals as a function of surface classes. As such, different surface types may possess different error characteristics, especially over arid regions where high quality ground measurements are often lacking. Importantly, the emissive properties of land also result in GPROF rainfall estimates being driven primarily by the higher frequency radiometer channels (e.g., > 89 GHz) where precipitation signals are most sensitive to coupling between the ice-phase and rainfall production. In this study, we evaluate the rainfall estimates from the Ku channel of the DPR as well as GPROF estimates from various passive microwave sensors. Our evaluation is conducted at the level of individual satellite pixels (5 to 15 km in diameter), against a dense network of weighing rain gauges (90 in 150 km2) in the USDA-ARS Walnut Gulch Experimental Watershed and Long-Term Agroecosystem Research (LTAR) site in southeastern Arizona. The multiple gauges in each satellite pixel and precise accumulation about the overpass time allow a spatially and temporally representative comparison between the satellite estimates and ground reference. Over Walnut Gulch, both the Ku and GPROF estimates are challenged to delineate between rain and no-rain. Probabilities of detection are relatively high, but false alarm ratios are also high. The rain intensities possess a negative bias across nearly all sensors. It is likely that storm types, arid conditions and the highly variable precipitation regime present a challenge to both rainfall retrieval algorithms. An array of ground-based sensors is being deployed during the 2017 monsoon season to better understand possible reasons for this discrepancy.