An Error Model for Uncertainty Quantification in High-Time-Resolution Precipitation Products

Abstract

his study proposes a new framework, Precipitation Uncertainties for Satellite Hydrology (PUSH), to provide time-varying, global estimates of errors for high-time-resolution, multisatellite precipitation products using a technique calibrated with high-quality validation data. Errors are estimated for the widely used Tropical Rainfall Measuring Mission (TRMM) Multisatellite Precipitation Analysis (TMPA) 3B42 product at daily/0.25° resolution, using the NOAA Climate Prediction Center (CPC) Unified gauge dataset as the benchmark. PUSH estimates the probability distribution of reference precipitation given the satellite observation, from which the error can be computed as the difference (or ratio) between the satellite product and the estimated reference. The framework proposes different modeling approaches for each combination of rain and no-rain cases: correct no-precipitation detection (both satellite and gauges measure no precipitation), missed precipitation (satellite records a zero, but the gauges detect precipitation), false alarm (satellite detects precipitation, but the reference is zero), and hit (both satellite and gauges detect precipitation). Each case is explored and explicitly modeled to create a unified approach that combines all four scenarios. Results show that the estimated probability distributions are able to reproduce the probability density functions of the benchmark precipitation, in terms of both expected values and quantiles of the distribution. The spatial pattern of the error is also adequately reproduced by PUSH, and good agreement between observed and estimated errors is observed. The model is also able to capture missed precipitation and false detection uncertainties, whose contribution to the total error can be significant. The resulting error estimates could be attached to the corresponding high-resolution satellite precipitation products.