A statistical framework for estimating air temperature using MODIS land surface temperature data

Remote sensing has shown an immense capability for large‐scale estimation of air temperature (Tair), one of the most important environmental state variables, using land surface temperature (LST) data. Following recent investigations on the Tair–LST relationship, in this article, we propose an advanced statistical approach to this realm. We tested the approach for estimation of Tair in eastern part of Iran using MODIS daytime and nighttime LST products and 11 auxiliary variables including Julian day, solar zenith angle, extraterrestrial solar radiation, latitude, altitude, reflectance at various visible and infrared bands and vegetation indices. Fourteen statistical models constructed through a stepwise regression analysis were evaluated along a 5‐year period (2000–2004) using MODIS and meteorological station data. Results of this study indicated that the statistical approach performed reasonably well, where our final proposed model could estimate average Tair with validation mean absolute error of 2.3 and 1.8 °C at daily and weekly scales, respectively. Nighttime LST, Julian day, altitude and solar zenith angle indicated to be the most effective variables capturing most variations of Tair in the study region. Variables influenced by land surface and land cover properties including reflectance at different bands and vegetation indices showed a negligible effect on the Tair‐LST relationship within the study area. It was indicated that the proposed models generally performed better for lower altitude regions.