Improvements in the estimation of daily minimum air temperature in peninsular Spain using MODIS land surface temperature

Air temperature (T-a) is a key variable in many environmental risk models and plays a very important role in climate change research. In previous studies we developed models for estimating the daily maximum (T-max), mean (T-mean), and minimum air temperature (T-min) in peninsular Spain over cloud-free land areas using Moderate Resolution Imaging Spectroradiometer (MODIS) data. Those models were obtained empirically through linear regressions between daily T-a and daytime Terra-MODIS land surface temperature (LST), and then optimized by including spatio-temporal variables. The best T-mean and T-max models were satisfactory (coefficient of determination (R-2) of 0.91-0.93; and residual standard error (RSE) of 1.88-2.25 K), but not the T-min models (R-2 = 0.80-0.81 and RSE = 2.83-3.00 K). In this article T-min models are improved using night-time Aqua LST instead of daytime Terra LST, and then refined including total precipitable water (W) retrieved from daytime Terra-MODIS data and the spatio-temporal variables curvature (c), longitude (lambda), Julian day of the year (JD) and elevation (h). The best T-min models are based on the National Aeronautics and Space Administration (NASA) standard product MYD11 LST; and on the direct broadcast version of this product, the International MODIS/AIRS Processing Package (IMAPP) LST product. Models based on Sobrino's LST1 algorithm were also tested, with worse results. The improved T-min models yield R-2 = 0.91-0.92 and RSE = 1.75 K and model validations obtain similar R-2 and RSE values, root mean square error of the differences (RMSD) of 1.87-1.88 K and bias - 0.11 K. The main advantage of the T-min models based on the IMAPP LST product is that they can be generated in nearly real-time using the MODIS direct broadcast system at the University of Oviedo.