New approach to determining the surface temperature without thermal band of satellites

Abstract

Abstract Surface temperature (LST) is one of the main variables that involve the physical processes present on the surface of a planet and is one of the variables of greater relevance in the process of energy and water balance. Due to the importance of LST in the energy balance equation and in irrigated agriculture management, it is important to develop an estimation of methods for orbital sensors, such as Sentinel 2A and 2B, that do not have the capacity to capture the electromagnetic radiation emitted in the thermal infrared region. Thus, the aim of this work was to estimate the surface temperature from the MSI/Sentinel 2A sensor using the LST obtained from the TIRS (Thermal Infrared Sensor) present on the Landsat-8 platform as a reference. In this context, we used different methodologies of pattern recognition algorithms to establish the best model: SVMlinear (support vector machine linear), SVMradial (support vector machine radial), LM (linear regression), RIDGE, RF (random forest), Cubist, PLS (partial least squares), PCR (principal components regression), GBM (generalized boosted regression), and BRNN (Bayesian regularized neural network). After the LST was estimated by the different techniques, a set of statistical metrics was used to select the best model, which in this case was the RF. With the best model defined, we calculated the actual and reference evapotranspiration ratio (r) using the LST of this model and compared it with the LST that was obtained by Landsat-8, considered a standard, to verify the applicability of our methodology.