Spatio-Temporal Modeling of Near-Surface Air Temperature Based on Satellite-Derived LST in the Mediterranean Region, Türkiye

Accurate estimation of near-surface air temperature (Ta-est) is of great importance, particularly in regions characterized by complex land cover and sparse meteorological station networks. Although meteorological station measurements provide high temporal accuracy, interpolation approaches limited by station density and distribution cannot adequately represent microclimatic variability under heterogeneous land conditions. To overcome these limitations, this study estimates air temperature at the pixel scale by modeling land surface temperature (LST) derived from high-resolution Landsat 8 satellite imagery together with air temperature measurements (Ta) obtained from meteorological stations. The analysis is based on monthly Landsat 8 data for the 2016–2019 period and air temperature observations from 11 meteorological stations. To model the LST–Ta relationship, linear, logarithmic, exponential, and second-order polynomial regression models were developed, and their performances were evaluated using RMSE, MBE, and R² metrics. The results demonstrate that the linear regression model exhibits a more balanced and generalizable performance than the other models, characterized by consistently low error levels, negligible bias, and stable explanatory power throughout the year. While all models show high explanatory capability during winter months, performance declines are observed in the nonlinear models during summer and transitional seasons due to increasing heterogeneity in surface–atmosphere energy exchanges. The Ta-est maps generated using the linear regression model show a strong correlation with observed Ta values, with absolute differences predominantly remaining below 1 °C. Overall, the methodology and findings of this study demonstrate that microclimatic variability can be successfully modeled and provide a practical framework for producing spatially detailed air temperature datasets.