A random forest and superpixels approach to sharpen thermal infrared satellite imagery

Abstract

Thermal infrared (TIR) imagery is normally acquired at coarser pixel resolution than that of shortwave sensors on the same satellite platform. Often, TIR resolution is not suitable for monitoring crop conditions of individual fields or the impacts of land cover changes that are at significantly finer spatial scales. Consequently, thermal sharpening techniques have been developed to sharpen TIR imagery to shortwave band pixel resolutions. One of the most classic thermal sharpening technique is TsHARP. It uses a relationship between land surface temperature and normalized vegetation index (NDVI). However, there are several studies that prove that a single relationship between TIR and NDVI may only exist for a limited class of landscape. Our work hypothesis stated that it is possible to improve the spatial resolution of TIR imagery considering a relationship between vegetation and several soil spectral indexes and TIR as well the spatial context information. In this work, the potential of Superpixels (SP) combined with Regression Random Forest (RRF) is used to augmenting the spatial resolution of the Landsat 8 TIR (Band 10 and 11) imagery to their visible (VIS) spatial resolution. The SP allows to consider the contextual information over the land cover, and RF allows to integrate in a unique model the relationship between five spectral indices and TIR data. The results obtained by SP RRF approach shows the potential of this methodology, compared with classical TsHARP method.