Validation of NOAA's Multi-Sensor Precipitation Estimate in the Mid-Atlantic Region

Validation efforts of the NOAA's multi-sensor precipitation estimate (MPE) have been carried out at the mid-Atlantic region where fifteen dual or triple tipping bucket gauge sites were operated at 15 sites for two years. MPE is a gauge adjusted radar (WSR-88D) rainfall product. It is constructed on 4 x 4 km2 grid in an hourly basis and has been generated by the River Forecast Centers (RFCs). The Stage IV is a national mosaic of the MPE. In our study area, all gauge sites were within mid-Atlantic RFC domain and each MPE pixel corresponded to single gauge site except at NASA Wallops Flight Facility (WFF) where one pixel had four and another pixel had two gauge sites. These gauges were not in part of the MPE and therefore they were considered as an independent validation product. In the presence of dual or triple radar coverage, MPE employs the closest radar. In this study, there were three WSR-88Ds in the domain with no blockage. Among gauge networks, MPE employs Automated Surface Observing System (ASOS) and Cooperative Observing Program (COOP) gauges. We had 7 sites next to ASOS and 5 sites next to COOP gauges and the gauge sites next to ASOS gauges resulted in better agreement with MPE rainfall than the other sites. This finding demonstrated the strong dependency of MPE to the gauge bias adjustment. The correlation coefficient and percent absolute bias was employed to determine the agreement between the gauge and MPE for hourly, three-hourly, six-hourly, daily, monthly, and yearly rainfall. The gauges were considered as a reference and the rainfall statistics were derived when the gauge had at least 0.5 mm (two tips) hourly accumulation. Therefore, false MPE rainfall was not shown in this study. While the agreement improves with increasing time span, there were substantial differences in agreement between the two from season to season. This is attributed to differences in rain intensity and type. During passage of Hurricane Ernesto (2006), for instance, the agreement between the gauge and MPE was poor. The tropical cyclones exhibits more small and less large drops than extratropical cyclones at a given reflectivity and the National Weather Service default reflectivity rain rate relations severely underestimates rainfall. The rainfall statistics showed no dependency to the distance between the radar and gauge site. The rainfall statistics did not improve significantly when more than one gauge was within one MPE pixel. Considering rain threshold, the statistics showed improvement when the threshold is applied both gauge and MPE rainfall. This demonstrates the contribution of missed MPE rainfall in the statistics. The agreement between the MPE and gauges was poorer when the rain threshold was increased from 0.5 mm to 1 and 2 mm.