Evaluation of reanalysis, spatially interpolated and satellite remotely sensed precipitation data sets in central Asia

摘要

The accuracy of any gridded climatic data sets is as important as their availability for regional climate and ecological studies. In this study, the accuracy of estimated precipitation in central Asia from three recently developed reanalysis data sets, Modern-Era Retrospective Analysis for Research and Applications (MERRA), ECMWF Interim Re-Analysis (ERA-Interim), and Climate Forecast System Reanalysis(CFSR), is evaluated through comparisons with observations from 399 stations during 1979–2010. An interpolated precipitation data set from station observations and a satellite remotely sensed data set, Tropical Rainfall Measuring Mission (TRMM) 3B42, are included in the evaluation. Major results show that MERRA data have higher accuracy than ERA-Interim and CFSR, although they all overestimate the observed precipitation especially in late spring and early summer months, suggesting errors in their ways of representing convective precipitation in that region. In comparison, the interpolated and satellite-sensed data, which provide no upper air information/data, have higher accuracy. While all these data sets have difficulty in describing stations’ precipitation in mountainous areas, the reanalysis data sets have particularly large discrepancies. In examining the discrepancy in the reanalysis data, a Precipitation-Topography Partial Least Squares method is proposed to incorporate certain terrain/geographic effects on precipitation in mapping the gridded data to the station locations for comparison with the observation. The outcome suggests that the estimated station precipitation by this new method is closer to the observed than the method without considering those factors. The improvement by this method and by possible other methods taking into account different details/aspects of the influences indicates that it is only meaningful to compare the accuracy or relevance of gridded data sets to station observations in a relative sense among various data sets.