Sampling-Related Variability in Raingage Network Products

The representation of precipitation using sparse gage measurements presents formidable challenges due to incomplete and nonhomogeneous sampling. With feasible density and frequency of observations, gage networks cannot capture the full spectrum of space and time scales at which precipitation may fall. If unrecognized or unadjusted, this sampling inadequacy can lead to a biased or distorted characterization of precipitation. We examine the magnitude of this sampling effect on two products: gridded fields of precipitation analyzed from point observations, and verification scores for model-generated precipitation computed using these observed fields. To do so, we first directly compare analyses and frequency statistics produced using several independent U.S. raingage datasets, including measurements made by the operational array of near-real-time daily and hourly reporting rain gages acquired from the National Centers for Environmental Prediction, and two sets of high-quality raingage measurements made by volunteer observers and archived at the National Climatic Data Center. Next, to assess intra-network sampling differences, we employ resampling techniques (including bootstrapping) to estimate the variability of gridpoint precipitation values and model verification scores that must be assumed when the analyses or scores are used. We also discuss the implications of combining precipitation observations from platforms with vastly different sampling characteristics (e.g., radar and gages).