An Evaluation of the Consistency of Extremes in Gridded Precipitation Data Sets
We applied an advanced statistics method, called “tail dependence” to compare extreme precipitation from different gridded daily products. Analogous to centered correlation, this statistic removes biases and examines the co-occurrence of extremes.
This research revealed significant discrepancies extreme precipitation between satellite and station based products over the CONUS region. Hence, we conclude that satellite-based products are inappropriate for analysis of extreme precipitation, at least in the mid-latitudes. We also found discrepancies between gridded products based on the same station data, traceable to a different interpretation of the time of day of the measurement. This can be removed by considering pentadal rather than daily extreme precipitation.
Uncertainty in observational estimates of extreme precipitation is large. Satellite-based estimates disagree with station based estimates in the most well-observed part of the world, seriously questioning their reliability. Sampling issues may be at the root of satellites’ problem as any given point is not continuously observed as it is for a rain gauge. We recommend against the usage of TRMM or similar products as a benchmark. We further recommend, that except for NARR, which directly assimilates station precipitation, that reanalysis products also not be used as a proxy for observed extreme precipitation.