Estimation of Air Surface Temperature From Remote Sensing Images and Pixelwise Modeling of the Estimation Uncertainty Through Support Vector Machines

The knowledge of air temperature near the Earth´s surface plays a relevant role in weather and climate studies as well as in the framework of solar energy management; e.g., for identifying the most suitable locations for a new solar installation or monitoring the performance of existing systems. Remote sensing allows air temperature to be estimated on a spatially distributed basis, thus complementing the spatially sparse observations collected by ground micro-meteorological stations. In this paper, a novel approach to periodic (e.g., daily or monthly) air temperature estimation from satellite images based on support vector machines (SVMs) is proposed. A recently developed SVM-based approach to supervised land and sea surface temperature estimation using satellite images is generalized to the case of air temperature and integrated with case-specific techniques aimed at computing periodic statistics of air temperature using the expectation-maximization algorithm. The method is fully automated and allows the statistics of the estimation error to be modeled on a pixelwise basis. This last result is accomplished by combining nonstationary multidimensional stochastic processes and Clark´s variance approximation. The method is experimentally validated with MSG-SEVIRI data acquired over Provence-Alpes-Côte d´Azur (France).