The impact of precipitation dataset choices on forecast verification during the HMT

Over the continental United States, a variety of different sources for precipitation observations and estimates are available for use as verification datasets for numerical forecast models. These range from radar and satellite estimates generally available in continuous spatial grids, to gage networks of varying spatial and temporal resolution. Under difficult conditions (eg., extreme terrain and heavy rainfall) the performance characteristics of different precipitation observations can vary widely. As a result, model verification scores based on these different datasets will not generally be identical. Understanding how the choice of data impacts verification results thus becomes an important aspect of the verification process.

In the present work, we address this issue by using several observation datasets (primarily independent gage datasets and radar-derived estimates) to evaluate QPF fields produced by the time-lagged and multi-model ensemble system built from 3 km WRF model runs during the HMT-west-2006 field experiment in the American River Basin in Northern California during December-March 2005/6 and again in 2006-2007. Precipitation forecasts from this experiment allow interesting additions to QPF evaluation, including as they do strongly orographic precipitation over complex mountainous terrain during several episodes of very heavy precipitation. We compare scores computed from the so-called Stage IV gridded radar/gage analyses with scores computed from independent hourly and daily gage sites in the Western United States. We also segregate the verification datasets by the rigor of quality screening applied to them, and by other aspects of the datasets including observation density and accumulation period.