Estimates of Point Rainfall Extremes from Satellite Precipitation Products: Testing in Northeastern Italy

Reliable extreme precipitation estimates are essential for understanding, predicting, and mitigating natural disasters. However, their inference at a global scale is limited by the sparse and uneven distribution of direct rainfall observations. Satellite-based estimates provide a promising source of information for extreme value analysis, but their high uncertainty and low spatial resolution hinder their applicability. Additionally, grid sizes ranging from 10 to 600 km² prevent direct comparisons with point-scale extreme value estimates, as point- and area-averaged statistics inherently differ in their construction.
This study addresses this limitation by assessing the sensitivity of a downscaling method for extreme value statistics, based on random field theory and the Metastatistical Extreme Value Distribution (MEVD). We use a comprehensive dataset of approximately 140 rain gauges in northeastern Italy, along with multiple satellite precipitation products, including IMERG, CMORPH, CHIRPS, SM2RAIN, MSWEP, and PERSIANN. The downscaling process, based on the autocorrelation structure of precipitation fields, is applied individually to each product at the grid cells corresponding to the available rain gauges.
To perform the downscaling process, two key variables must be obtained: the variation in the wet fraction (Beta) and the variation in the rainfall spatial correlation (lambda), both between satellite pixel and point scale. However, this requires defining a neighborhood centered on the point of interest, as well as a function that represents the decay of spatial correlation within this neighborhood. Therefore, we assess the method's sensitivity by considering three different neighborhood sizes (3, 5, and 7 pixels) and two functions to represent the spatial correlation (Exponential kernel with Power law tail and Stretched exponential).
Finally, downscaling results for extreme daily event magnitudes with a 50-year return period at the point scale are obtained from the above methodology and are validations against those derived from rain gauge time series.