Forest cooling is not natural-vegetation cooling: pair-specific and environmentally conditioned land-cover thermal contrasts across India

Land cover is often treated as a fixed warming or cooling agent, with natural vegetation assumed to be cooler than human-modified surfaces. Here, we challenge this assumption by quantifying land-surface temperature contrasts (ΔLST) among seven major land-cover classes across India using two decades (2003–2022) of satellite observations and a spatially matched moving-window framework complemented by environmental-gradient, statistical and machine-learning analyses. Of 21 possible pairwise contrasts, 17 were sufficiently co-located for analysis. During daytime, forests were the coolest surface, cooler than cropland in 99.6% of matched pairs (mean ΔLST = -2.32°C) and urban land in 97.6% (-2.57°C). However, this cooling was forest-specific: savanna, shrubland and grassland were collectively cooler than cropland in only 33.0% of comparisons and averaged 0.73°C warmer. At night, 11 of 17 contrasts reversed sign, forests became warmer than cropland (+ 0.40°C), and urban land was warmer than all vegetated surfaces. Thermal contrasts varied across biogeographic zones and responded nonlinearly to environmental gradients, with breakpoints at background temperatures of 29–44°C and evapotranspiration-to-potential-evapotranspiration ratios of 0.09–0.40. Forest cooling relative to cropland peaked at intermediate evaporative conditions, whereas water-stressed non-forest vegetation was substantially warmer than neighbouring cropland or urban land. Biophysical attribution identified albedo as the dominant daytime predictor for 12 of 16 contrasts, whereas vegetation structure, productivity and latent heat exchange governed others. These findings establish that land-cover thermal effects are pair-specific, temporally variable and environmentally conditioned, challenging generalized assumptions about natural-vegetation cooling and informing land-surface modelling and nature-based cooling strategies.