Reliability of satellite, reanalysis and observation-based gridded temperature datasets for climate change impact studies in Bhutan

The impacts of global warming are pronounced in mountainous regions, yet a scarcity of long-term climate data hinders robust documentation. Reanalysis (ERA5, ERA5-Land, MERRA2), gridded observational (CRU TS), and satellite-derived (MODIS LST) datasets serve as alternatives, but their reliability for local-scale impact studies remains uncertain without rigorous evaluation. Here, we present the first comprehensive assessment of these datasets across Bhutan's complex topography, comparing them to in-situ observations (1996–2023) using systemic statistical metrics, which is a critical prerequisite for their applications. Results reveal that pre-corrected datasets contain severe systematic cold bias increasing with elevation at 3.1–4.2 °C/km, culminating to bias up to −19 °C in the high-altitude areas. The post-correction analysis reveals that elevation-corrected reanalyses data reduces mean bias by a maximum of 31 %. However, enhancement of spatial representativeness of temperature through dynamically estimated lapse rate on in-situ temperature markedly reduces mean bias across all datasets including MODIS-derived air temperature. The altitudinal bias gradient, depending on reanalyses data, is reduced to 0.1°C–0.8 °C/km. Despite these notable improvements in accuracy, MODIS LST and reanalyses/CRU datasets continue to exhibit over- and underestimation, respectively. These findings suggest that limitations of accuracy stem not only from model assimilation or interpolation, but also from limited spatial representativeness of station observations. Our findings underscore that the use of these datasets directly in climate impact studies is impractical without prior corrections. This work provides a framework for evaluating temperature products in mountainous regions, ensuring their utility for adaptation planning in Bhutan and analogous terrains globally.