Inferring the uncertainty of satellite precipitation estimates in data‐sparse regions over land

The global distribution of precipitation is essential to understanding earth's water and energy budgets. While developed countries often have reliable precipitation observation networks, our understanding of the distribution of precipitation in data‐sparse regions relies on sporadic rain gauges and information gathered by spaceborne sensors. Several multisensor data sets attempt to represent the global distribution of precipitation on subdaily time scales by combining multiple satellite and ground‐based observations. Due to limited validation sources and highly variable nature of precipitation, it is difficult to assess the performance of multisensor precipitation products globally. Here, we introduce a methodology to infer the uncertainty of satellite precipitation measurements globally based on similarities between precipitation characteristics in data‐sparse and data‐rich regions. Five generalized global rainfall regimes are determined based on the probability distribution of 3‐hourly accumulated rainfall in 0.25° grid boxes using the Tropical Rainfall Measurement Mission 3B42 product. Uncertainty characteristics for each regime are determined over the United States using the high‐quality National Centers for Environmental Prediction Stage IV radar product. The results indicate that the frequency of occurrence of zero and little accumulated rainfall is the key difference between the regimes and that differences in error characteristics are most prevalent at accumulations below ~4 mm/h. At higher accumulations, uncertainty in 3‐hourly accumulation converges to ~80%. Using the self‐similarity in the five rainfall regimes along with the error characteristics observed for each regime, the uncertainty in 3‐hourly precipitation estimates can be inferred in regions that lack quality ground validation sources.