Intra-annual characterization of soil mean temperature at 5 and 10 cm depths based on remote sensing data, at country scale

Soil temperature (ST) is an important physical property that influences all soil processes; it is a relevant component of the climate system and impacts terrestrial ecological, hydrological and biogeochemical processes. Its variability presents challenges, limiting studies the spatiotemporal distribution and prediction of ST. This study describes the calibration of a model that uses land surface temperature (LST), the Normalized Difference Vegetation Index (NDVI), and daily solar declination (Ds), along with data from meteorological stations, to predict the covered soil temperature at depth of 5 cm (SMTc5cm) and 10 cm (SMTc10cm). Iterations were performed using combinations of 18 LST and 10 NDVI treatments derived from MODIS images, with the aim of selecting and applying validated models for the intra-annual characterization of SMTc5cm and SMTc10cm across Uruguay. The Results showed that the models for SMTc5cm and SMTc10cm had R2 values of 0.84 and 0.87-0.89, respectively, and RMSE values of 2.3 °C for SMTc5cm and 2.1-1.8 °C for SMTc10cm. Comparisons with observed SMTc5cm and SMTc10cm, and SMTc5cm with observed ST at 20 cm depth and uncovered soil, indicated that the models accurately predicted soil temperature and maintained phase shifts, with minor variations in the timing of profile intersections. Models that used averages of daytime and nighttime LST observations, along with filtered NDVI series, achieved better fits than those using the original data. Preliminary observations highlight the importance of further investigating the effects of forests, soil composition, and subsurface characteristics on soil temperature.