A new methodology for weighting high-resolution model simulations to project future rainfall in the Middle East

A divergence metric was used to combine 4 high-resolution climate models to generate more reliable simulations of future rainfall. The approach is based on the assumption that the use of multiple models (an ensemble) is superior to the use of a single model, even if one of the models is shown to better capture past trends. Such an approach is especially useful in areas with steep climatic gradients, where large-scale climate models are not effective in capturing orographic and local effects. We applied the methodology to the Middle East, and specifically to Israel, where climate shifts from arid to humid temperate occur over a distance of around 400 km. Model weights were determined by calculating the similarity between the probability distributions of the models and those of the historical data using the Jenson-Shannon divergence metric. These weights were then applied to future model projections. Annual amounts of rainfall, numbers of wet days and numbers of 3 d wet spells were analyzed. Compared with observed data, the weighted ensemble outperformed the equal weights ensemble, which outperformed the best model. For the northern and central stations, average annual amounts of rainfall decreased in both near- and far-future periods, with most of the change occurring at the peak and in the left-hand tail and less change in the right-hand tail of the probability distribution. This, combined with the change in the right-hand tail of the distribution in numbers of wet spells in the near future, suggests that the decline in overall rainfall will be higher than the corresponding decline in extreme events; or in other words even though there will be less rainfall, the extreme events will remain, and even possibly increase. In the south, a mixed trend of slightly increasing median amounts of rainfall and slightly decreasing extreme events is projected.