Change points in predictors-predictand relationships within the scope of statistical downscaling

A statistical downscaling approach allowing for change points in the relationships between atmospheric predictors and local precipitation is introduced. Change point analysis within generalized linear models, and change points in the predictor characteristics were used to develop a change point statistical downscaling approach. The approach is illustrated for station-based winter precipitation in the Mediterranean area. In this study, 94 stations were considered. The change point analysis yielded 37 stations with robust change points in the predictors-precipitation relationships. An added value regarding statistical model performance of the change point approach compared to the use of statistical models without change points was observed for 15 out of the 37 stations. In the projections under enhanced greenhouse gas forcing the application of the change point approach for the 15 stations affects the downscaled precipitation amounts, with significant differences compared to the application of downscaling models without change points for about one third of these stations. Under RCP4.5 and RCP8.5 scenario conditions, mainly increases of winter precipitation are projected until the end of the 21st century for parts of the western and northern Mediterranean area, whereas the north-eastern and eastern Mediterranean areas are affected by decreases.