Projected Mean Temperatures over India under RCP Scenarios: A Comparison of Statistical and Dynamical Downscaling

We evaluate and compare two downscaling methods for reproducing observed temperature statistics across weather stations in India: (i) a dynamical downscaling approach using the Weather Research and Forecasting (WRF) model, and (ii) an Empirical–Statistical Downscaling (ESD) method. Both methods are driven by the same global climate model, CESM1-CAM4, allowing a consistent evaluation. Owing to its computational efficiency, the ESD method is also applied to the full CMIP5 multi-model ensemble to explore a broader range of climate futures.Our analysis includes multi-scale evaluation at global, regional, and local levels, with a focus on seasonal and annual mean air temperature. Once validated, both methods are used to assess projected temperature changes at national and station scales. Despite a modest cold bias in WRF simulations (less than 0.5°C), both downscaling approaches project annual warming of approximately 1.5±0.5°C by 2041–2060 and 3.0±1.0°C by 2081–2100 under RCP8.5. Seasonal warming may reach up to 10°C in parts of India, indicating significant potential impacts on health, agriculture, and infrastructure. We find that increasing WRF resolution from 45km to 15km reduces cold biases, though projected temperature changes are more sensitive to the choice of downscaling method and input data than to model resolution.