Statistical methods for understanding hydrologic change

The climate change issues are becoming everyday more and more important, not only for scientists and specialists but also for large part of public opinion. A significant part of climate change science is related to its implications on the hydrologic cycle. Among all hydrologic processes, rainfall is a very important variable as it is a fundamental component of flood risk mitigation and drought assessment, as well as water resources availability and management. For these issues, a local assessment is required, and specific investigations tools are necessary. We use a hierarchical modeling approach to investigate a collection of spatially referenced time series of rainfall extreme values. We assume that the observations follow a generalized extreme value (GEV) distribution whose locations are spatially and temporally dependent where the dependence is captured using a geoadditive model. Geoadditive models analyze the spatial distribution of the studied variable while accounting for the explicit consideration of linear and nonlinear relations with relevant explanatory variables, as well as the spatial correlation described by a standard spatial autocorrelation function. Under the additivity assumption they can handle the covariate effects by combining the ideas of additive models and kriging, both represented as linear mixed model. This approach, based on the generalized mixed model/splines paradigm, has achieved a valuable success during the last decade as useful tool with which to study the spatial distribution of climate variables as well as in other contexts. The study area is the Tuscany Region, in Central Italy. The rainfall dataset is composed by the time series of about 700 recording rain gauges, spatially distributed over an area of about 23.000 km2. The record period covers mainly the second half of 20th century.