Projected temperature changes in the Eastern Mediterranean for the 21st century from downscaled GCMs

The Middle East is a "hot spot" for climate change. Global climate models (GCM) have low spatial resolution and cannot capture the spatial variability of the region's climate. We developed a statistical downscaling (SD) method to assess expected temperature changes in diverse climate zones of the Eastern Mediterranean. The SD algorithm is based on finding past synoptic conditions ("analogues"), also known as the K-nearest neighbors (KNN) method. This method links coarse spatial resolution GCMs with past local measurements of temperature to provide the local future estimates.

The SD was trained on ERA5.1 reanalysis to characterize past synoptic conditions and 30 homogenized past local observations of surface temperature for the period 1979-2014. The algorithm optimization and validation were achieved through cross-validation against the historical observations. The process showed that gridded surface temperatures from global models are enough for optimal accuracy. The algorithm has been applied on 10 CMIP6 models for the historical period, the SSP245 [greenhouse gases (GHG) emissions rate decreases till the end of the 21^st century] and the SSP585 (GHG emissions significantly increase till the end of the 21st century) scenarios.

Cross-validation of ERA5.1 and CMIP6-models downscaled results for the historical period show that the minimum and maximum temperature distributions generated by the SD algorithm closely fit those at the 30 measurement stations. Moreover, they are significantly more accurate than those derived from the coarse-resolution CMIP6 models, which show cold-biased thinner distributions.

After validation of the algorithm, we apply it to downscale CMIP6 models for the 2080-2100 period. The downscaled SSP245 scenario shows that minimum and maximum temperatures increase by up to 5^o C and 4^o C with respect to the reference historical period, respectively. For the downscaled SSP585 scenario maximum changes are as large 7^o C and 5^o C for minimum and maximum temperatures, respectively.  More significant warming is observed during the cold season, in agreement with previously reported studies for the Northern hemisphere extra-tropics. The downscaled temperature estimations show their usefulness in projecting future temperatures at fine spatial resolution (well below the GCMs spatial resolution) that capture different climate characteristics (e.g., urban versus rural locations, low elevation versus mountain terrain), not possible to appropriately estimate from coarse GCMs.

On-going work focuses on additional future time periods and scenarios, further improvement of the algorithm in dealing with out-of-sample data, geographic transferability and use of the projected downscaled temperatures in human health studies.