Statistical downscaling of seasonal precipitation in Israel

Careful planning of the use of water resources is critical in the semi-arid eastern Mediterranean region. The relevant areas are characterized by complex terrain and coastlines, and exhibit large spatial variability in seasonal precipitation. Global seasonal forecasts provide only partial information of the precipitation as a result of their coarse spatial resolution. We present two statistical downscaling methods of global forecasts, both identifying past-analogue synoptic-weather patterns and their connection to precipitation at specific stations. The first method utilizes a classification of the large-scale weather patterns into regimes, and the other identifies the closest past analogues directly without grouping the weather events. The validation of the algorithms using NCEP/NCAR reanalyses and past precipitation observations at 18 stations shows that both methods provide good skill in predicting mean precipitation amounts and quantiles of the precipitation distribution, and in reproducing the observed inter-annual and spatial variability. Both methods show good correlations between predicted and observed precipitation amounts (∼0.8), and the downscaled precipitation reproduces the observed differences between the stations, which are not available in the coarse global models. Based on these results, we downscaled the operational global-seasonal forecasts issued by the NCEP CFS1.0 ensemble. This approach could also have utility in climate change scenario downscaling.