Downscaling of Temperatures Over the Eastern Mediterranean for the 21st Century

ABSTRACT The Eastern Mediterranean is a climate‐change hotspot, yet global climate models (GCMs) cannot resolve the sharp coastal, mountain, valley, and desert gradients that shape local temperature variability in Israel. We develop an analogue‐based statistical downscaling method, using a K‐nearest‐neighbour framework, to project daily maximum temperature (T max ), minimum temperature (T min ), and diurnal temperature range (DTR) at 32 homogenized meteorological stations. The method is trained with ERA5 and ERA5.1 reanalyses and station observations for 1979–2014 and applied to daily output from ten CMIP6 models under SSP2‐4.5 and SSP5‐8.5. The method is evaluated using distributional comparisons, yearly trend analysis, split‐period testing, and RMSE/bias diagnostics. Relative to raw GCM output, which is strongly cold‐biased and often too narrow, downscaling substantially improves T max and T min distributions at station scale. Absolute seasonal biases decrease from about 4°C–9°C to less than 1°C, and RMSE decreases from about 5°C–8°C to 3°C–5°C. Trend analysis shows that the downscaled ensemble reproduces the observed direction and seasonality of historical warming. A split‐period test confirms that the method captures transient warming between 1979–1999 and 2000–2014 rather than imposing artificial stationarity. Late twenty‐first‐century (2080–2100) projections show robust warming across all stations and seasons, with stronger changes under SSP5‐8.5. Under SSP2‐4.5, most seasonal changes are approximately + 1°C to +4°C, whereas under SSP5‐8.5 many stations exceed +6°C in summer, with the largest increases at inland, elevated, and arid stations. T min warming is especially coherent in summer, indicating reduced nocturnal cooling, while T max warming is more geographically heterogeneous. DTR generally decreases in summer under SSP2‐4.5 but is more heterogeneous under SSP5‐8.5; autumn is mainly positive, while winter and spring are scenario‐ and location‐dependent. These station‐scale projections reveal local contrasts inaccessible to coarse GCMs and provide information relevant for heat stress, energy demand, agriculture, ecosystems, and adaptation planning.