Analysis of disaggregation techiques applied to satellite images for the estimation of surface termal parameters at different scales
During the last years, both the technological development and the greater availability of geospatial information have led to the emergence of new application areas for remote sensing techniques. This is also relevant in the case of thermal remote sensing. Applications such as crop tracking require a greater availability of thermal information with spatial resolutions appropriate for a more local level scope. However, and despite the increasing availability of remote sensing products that have appeared and are expected to appear in the coming years, thermal infrared data continue to be available at lower spatial resolutions than the visible and nearinfrared data. Numerous authors have developed or tested methods to extract information at the sub-pixel level by using complementary remote sensing products with suitable results for using in applications at higher scales. Most of these methods are based on correlations between some vegetation indexes, such as NDVI, and radiative temperatures for a given cover. They are based on traditional mathematical models, such as linear or quadratic regression. Despite newer analysis tools like Support Vector Machines (SVM) or Neural Networks (NN) have become relevant in the last decade, their application on thermal remote sensing is in an relatively early stage of research and the use of traditional methods remains nowadays. The objective of this study is carrying out a comparison of these methods. A downscaling process from a MODIS temperature product scene has been developed using different methodologies. The results have been evaluated using “in situ” (ground-truth) temperature measurements showing an estimate of the accuracy and the potential of two different techniques.