资源环境科技发展态势分析平台

The estimation of the frequency of intense rainfall events is a crucial step for quantifying their impact on human societies and on the environment. This process is hindered by large gaps in ground observational networks at the global scale, such that extensive areas remain ungauged. The increasing availability of satellite-based rainfall estimates, while providing data with unprecedented resolution and global coverage, also introduces new challenges: the scale disparity between gridded and rain-gauge precipitation products on the one hand, and the short length of the available satellite records on the other. Here we propose a statistical framework for the estimation of rainfall extremes that is specifically designed to simultaneously address these two key issues, providing a new way of estimating extreme rainfall magnitudes from space. A downscaling procedure is here introduced to recover the spatial correlation and the probability density function of daily rainfall at the point (gauge) scale from coarse-scale satellite estimates. The results are then combined with a recent statistical model of extremes (the Metastatistical Extreme Value distribution), which optimizes the use of the information obtained from relatively short satellite observational time series. The methodology is tested using data from the Tropical Rainfall Measuring Mission Multisatellite Precipitation Analysis over the Little Washita River, Oklahoma. We find that our approach satisfactorily reproduces downscaled daily rainfall probability density functions and can significantly improve the Tropical Rainfall Measuring Mission Multisatellite Precipitation Analysis-based estimation of quantiles with return times larger than the length of the available data set (19years here), which are especially important for several water-related applications.