Validation of a desertification monitoring model in a semiarid region with the support of machine learning techniques

Desertification is a global and concerning phenomenon resulting from the interplay of climatic factors, inadequate human activities, and unsustainable use of natural resources. Its historical roots are linked to intensive agriculture, deforestation, and land-use changes, making the development of mathematical models crucial to sustainably identify and manage these areas. These models incorporate variables such as climatic patterns, land use, and degradation indicators, enabling an accurate assessment of the risk and extent of desertification in specific regions. The objective of this study was to evaluate the effectiveness of the RisDes_Index model in identifying areas affected by desertification and assessing the severity of environmental degradation. The model was developed based on orbital information and in situ data collected from Caatinga environments, wetlands, and areas undergoing desertification. The study was conducted in the Sertão Central region of Brazil, covering the municipalities of Floresta, Cabrobó, Belém do São Francisco, Carnaubeira da Penha, Itacuruba, and Orocó—an area known to be affected by desertification. The model demonstrated high efficacy in identifying desertified environments. One key feature that allows the RisDes_Index model to be applied to various global regions is its low computational power requirement, unlike machine learning and random forests, which, despite their high identification capacity, demand significant computational resources. However, the RisDes_Index model requires a higher operational capacity from researchers, which may render certain studies unfeasible due to a lack of necessary data. No correlations were found between the RisDes_Index model and vegetation indices (NDVI, SAVI, LAI, albedo, TGSI, and TSoil).