Characterizing Precipitation Product Errors across the United States using Triple Collocation

Since all precipitation products contain errors, the true precipitation is unknown. This makes validation of precipitation estimates a challenging problem. However, with the increase in availability of data products from a wide range of instruments (satellite, ground-based radar, and gauge), it is now possible to apply the Triple Collocation (TC) technique to characterize the uncertainties in each of the products with respect to the true. Classical TC takes advantage of three collocated data products of the same variable and estimates the mean squared error of each, without requiring knowledge of a perfect (zero error) truth. In this study, triplets of NEXRAD-IV, TRMM 3B42 V7, GPCP and GPI products are used to quantify the associated spatial error characteristics across a central part of the continental US. This is the first study of its kind to explore precipitation estimation errors using TC across the US. A logarithmic (multiplicative) error model is used to relate the precipitation estimates to the hypothetical truth. For precipitation, this is more realistic than the additive error model used in the original TC derivations, which is generally appropriate for wind comparisons. This study provides realistic error estimates of the precipitation products that can be incorporated into hydrological and meteorological models especially those used in data assimilation. Physical interpretations of the error fields (related to topography, climate, etc) are also explored. The methodology presented in this study can be used to quantify the associated uncertainties with precipitation estimates from each of the constellation of GPM satellites. Such quantification is a vital step towards optimally merging these estimates.