Estimating the probability of rare or yet unobserved rainfall extremes is essential for hazard quantification, especially in the context of an intensifying water cycle and of short observational records, common even in countries with long hydro-meteorological monitoring tradition. Data limitations can be mitigated by using regionalization techniques, which augment observational information, and by employing effective statistical models , such as the Metastatistical Extreme Value Distribution (MEVD), which maximizes the use of available observations. In this work, we develop the MEVD Regionalized framework (MEVD-R) to reduce uncertainty in estimating the probability of rare events that are not present in the training sample. Extensive testing using a global dataset of 40,000 rain gauges across Europe, North America, and Australia demonstrates that MEVD-R yields negligible systematic error and greatly reduces uncertainty compared to traditional approaches. MEVD-R can even provide estimates when available data are too sparse for conventional methodologies to offer reliable results.

Droughts have pervasive societal impacts and remain difficult to characterize observationally, due to the limited number of droughts sampled in instrumental records. One approach to improving the statistical basis of drought occurrence probability estimation is to extend the observational record using proxy climatic archives, such as those based on tree-ring information. Additionally, since droughts are rare and characterized by multiannual durations and inter-arrival times, it is important to devise and apply statistical techniques that make full use of all of the available information so as to improve our ability to quantify the rarest droughts. We extract data from a publicly available tree-ring based Palmer Drought Severity Index (PDSI) dataset, the Old World Drought Atlas, for two sites in Italy where long rainfall and temperature observational time series are leveraged for a meaningful comparison. Drought events are defined in terms of drought deficit volumes below a threshold PDSI value, and are studied through the Metastatistical Extreme Value Distribution (MEVD) to quantify the occurrence probability of extreme drought events. The estimation uncertainty associated with a variety of possible assumptions in MEVD analysis is studied, in specific comparison with the performance obtained using the traditional Generalized Extreme Value distribution, through a cross-validation methodology. Results suggest that MEVD-based formulations are more robust and flexible with respect to traditional ones. The combination of paleoclimatic data and methodologies capable of using most of the existing information provide more reliable estimates of drought recurrence times, which may be used to design more effective drought risk management plans.