Evaluation of MPE data for the Midwestern United States

A comparison of multi-sensor (radar and gage) and gage precipitation estimates at a monthly temporal resolution and a county spatial resolution was undertaken for the midwestern United States. Precipitation data were collected from February 2002 through October 2006 from two sources: 1) multi-sensor precipitation estimates (MPE) based on the Stage III/IV algorithm developed by the National Weather Service (NWS) Office of Hydrology and NWS River Forecast Centers, and 2) quality-controlled NWS cooperative raingage (QC_Coop) data from the National Climatic Data Center (NCDC). The gage data were employed as the reference standard.

The monthly median of the percent differences in county-averaged monthly precipitation estimated by MPE and QC_Coop in the midwestern U.S., for ~750 counties, were mainly within +/- 12.5%, with a median percent difference of + 6%. The positive difference indicates that overall, the MPE values tend to be smaller than the QC_Coop values. The correspondence between gage and MPE data was found to vary across the region. Precipitation amount, range to the nearest WSR-88D radar, latitude zone and season were examined to attempt to account for the variation in differences between QC_Coop and MPE precipitation estimates.

Percent differences between QC_Coop and MPE values vary with precipitation amount, with QC_Coop values less than MPE values at low precipitation amounts and QC_Coop values greater than MPE values at higher precipitation amounts. This was found for all seasons and regions, except for winter months (November-February) in the most northern (>44° N) latitude zone. During the cold months in the far northern latitudes, QC_Coop precipitation generally was larger than MPE amounts for all categories of precipitation amount. This underestimation of winter precipitation at northern latitudes by MPE is likely the result of use of the HADS data set to adjust the radar amounts in the MPE algorithm. The HADS network is comprised primarily of tipping bucket gages which are known to significantly underestimate frozen precipitation.

At ranges within 30 km of the nearest radar in the summer and at ranges greater than 150 km from the radar in the winter, the MPE algorithm appears to fail to correct underestimations that are likely the result of radar error. MPE values appear less subject to error at all latitudes in the summer months when convective precipitation is dominant, and in all seasons for latitudes where non-frozen precipitation is most prevalent.