Abstract

SOLORZANO, Jonathan V.; MAS, Jean François; GAO, Yan  and  GALLARDO-CRUZ, J. Alberto. Evaluation of SAR and multispectral images with deep learning algorithms to monitor deforestation and forest degradation in tropical forests. Rev. cartogr. [online]. 2025, n.111, pp.79-101.  Epub Nov 11, 2025. ISSN 2663-3981.  https://doi.org/10.35424/rcarto.i111.5891.

Deforestation and forest degradation are two global change drivers that contribute to greenhouse gases emissions, biodiversity loss and reduction in the quality of ecosystemic services. Due to their ability to study large extents and count with a historical record, remote sensing has demonstrated being an essential tool to monitor these processes. Besides, the development of new methods of analysis, such as deep learning and images with higher spatial, spectral and temporal resolution open the possibility to develop methods that enhance the current monitoring capacities. In this context, the current study evaluated the performance of deep learning algorithms with multispectral and synthetic aperture radar images to classify different land use land covers, identify deforestation and forest degradation. The results show that, in general, deep learning algorithms obtain more accurate results than its machine learning counterparts, due to the incorporation of spatial and temporal dimensions. Nonetheless, this performance is conditioned by sample size and the force of the relation between the remote signal and the class or attribute being evaluated. On the other hand, the combination of multispectral and synthetic aperture radar was, in general, beneficial; although in certain cases, it did not add new useful information, in comparison with the one already contributed by the multispectral ones. The results indicate that these techniques are capable of obtaining more accurate evaluations to identify deforestation and forest degradation; thus, they represent a very attractive alternative to develop future tools to monitor these processes.

Keywords : Land use land cover classification; U-Net; Sentinel-1; Sentinel-2; biomass.

        · abstract in Spanish     · text in Spanish     · Spanish ( pdf )