Satellite-based and mesoscale regression modeling of monthly air and soil temperatures over complex terrain in Turkey

Simple regression algorithms were developed to quantify spatio-temporal dynamics of minimum and maximum air temperatures (Tmin and Tmax, respectively) and soil temperature for a depth of 0–5 cm (Tsoil-5cm) across complex terrain in Turkey using Moderate Resolution Imaging Spectroradiometer (MODIS) data at a 500-m resolution. A total of 762 16-day MODIS composites (127 images × 6 bands) between 2000 and 2005 were averaged over a monthly basis to temporally match monthly Tmin, Tmax, and Tsoil-5cm from 83 meteorological stations. A total of 60 (28 temporally averaged plus 32 time series-based) linear regression models of Tmin, Tmax, and Tsoil-5cm were developed using best subsets procedure as a function of a combination of 12 explanatory variables: six MODIS bands of blue, red, near infrared (NIR), middle infrared (MIR), normalized difference vegetation index (NDVI), and enhanced vegetation index (EVI); four geographical variables of latitude, longitude, altitude, and distance to sea (DtS); and two temporal variables of month, and year. The best multiple linear regression models elucidated 65% (RMSE = 5.9 °C), 65% (RMSE = 5.1 °C), and 57% (RMSE = 6.9 °C) of variations in Tmin, Tmax, and Tsoil-5cm, respectively, under a wide range of Tmin (−34 to 25 °C), Tmax (0.2–47 °C) and Tsoil-5cm (−9 to 40 °C) observed at the 83 stations.

► A practical utility of devising a simple model with readily available inputs from satellite images was explored.
► A total of 762 16-day MODIS composites were averaged to match monthly minimum and maximum air and soil temperatures.
► A total of 60 linear regression models were developed as a function of a combination of 12 explanatory variables.
► Best regression models elucidated variations by 65% in Tmin and Tmax, and 57% in Tsoil-5cm.