Abstract
The objective of this study is to investigate the vulnerability of different land‐cover types to climate change. To this end, land‐cover specific temperature change factors are quantified for the southern Saskatchewan region using a novel statistical downscaling model: physical scaling ( SP ). SP model considers large‐scale climate and regional physical characteristics like land‐cover, elevation in its formulation and hence can be used to predict future temperature for different land‐cover types under changing large‐scale climatic and land‐cover conditions. The model is validated by assessing its ability to downscale North American Regional Reanalysis ( NARR ) derived surface (skin) temperature from an initial resolution of 32 km to 500 m. The downscaled NARR data are evaluated using a cross‐validation approach over the period 2006–2013 with reference to MODerate ‐resolution Imaging Spectroradiometer ( MODIS ) derived surface temperature estimates and satisfactory model performance is obtained (average RMSE = 0.03 K). The validated model is used to predict future surface temperature across the study region. Future land‐cover projections are derived by downscaling land‐use projections for Representative Concentration Pathways ( RCPs ) 2.6 and 8.5 made by integrated assessment models: IMAGE and MESSAGE , respectively. An analysis of land‐cover specific temperature changes between historical (2006–2013) and future (2081–2100) timelines indicate variations of up to 2 K between different land‐cover classes. Vulnerability pattern of different land‐cover classes differ significantly between day‐ and night‐time. Further, variations of upto 1 K in projected changes are observed among different forest cover types. Closed shrubland is obtained as the most vulnerable forest‐cover class whereas evergreen broadleaf forest is found to be the least vulnerable.