Component analysis of errors in satellite‐based precipitation estimates

Satellite‐based precipitation estimates have great potential for a wide range of critical applications, but their error characteristics need to be examined and understood. In this study, six (6) high‐resolution, satellite‐based precipitation data sets are evaluated over the contiguous United States against a gauge‐based product. An error decomposition scheme is devised to separate the errors into three independent components, hit bias, missed precipitation, and false precipitation, to better track the error sources associated with the satellite retrieval processes. Our analysis reveals the following. (1) The three components for each product are all substantial, with large spatial and temporal variations. (2) The amplitude of individual components sometimes is larger than that of the total errors. In such cases, the smaller total errors are resulting from the three components canceling one another. (3) All the products detected strong precipitation (>40 mm/d) well, but with various biases. They tend to overestimate in summer and underestimate in winter, by as much as 50% in either season, and they all miss a significant amount of light precipitation (<10 mm/d), up to 40%. (4) Hit bias and missed precipitation are the two leading error sources. In summer, positive hit bias, up to 50%, dominates the total errors for most products. (5) In winter, missed precipitation over mountainous regions and the northeast, presumably snowfall, poses a common challenge to all the data sets. On the basis of the findings, we recommend that future efforts focus on reducing hit bias, adding snowfall retrievals, and improving methods for combining gauge and satellite data. Strategies for future studies to establish better links between the errors in the end products and the upstream data sources are also proposed.