Increasing resolution of climate assessment and projection of temperature and precipitation in an alpine area.

In mountain regions, important differences in the time trends of climate series can be detected even within relatively small areas, leading to uncertainty when assessing climate change. The paper deals with a structured algorithm for high-resolution downscaling of climate characterisation in a region (precipitation and temperature), leading to a twofold application: increasing spatial resolution of past climate definition for the area and attaining high-resolution downscaling for climate projections. In the first stage, multi-variate analysis ('partial least squares' regression) was applied to a number of time series (10) in order to obtain climate averages for a larger number of sites. Predictions made with single-site values (such as seasonal means) can in some cases be improved by applying 'random perturbation' of the value and averaging single predictions in the ensemble. This analysis laid the foundation for implementing the same technique to the output of statistical downscaling of multi-model climate projections. Climate shift in the study area (Trentino), located in the north-eastern Italian Alps, was simulated for two 30-year time windows: 2021-2050 and 2071-2099. Progressive warming is predicted, being stronger in the summer, along with a mixed, seasonally differentiated trend for precipitation. [ABSTRACT FROM AUTHOR]