Identification of priority areas for forest restoration using self-organizing maps neural network

Abstract

The aim of this work was to identifying priority areas for forest restoration and analyze variables related to such areas at two distinct spatial scales using Self-Organizing Maps neural network (SOM). Initially, a SOM analysis was conducted to detect a watershed suitable for forest restoration within the Management Unit for Hydrological Resources of the Paraiba do Sul river, located in São Paulo State, southeast of Brazil. The variables employed in this analysis were environmental connectivity and forest cover. The Jaguari watershed, located in the municipality of Igaratá, was selected as study area in the second stage of analysis. In the permanent preservation areas along riversides within this watershed, a new SOM analysis was performed to detect suitable areas for forest restoration. At this more refined scale, the regarded variables were distance to forest fragments, urban areas, paved roads, and rural constructions, as well as the NDVI (the Normalized Difference Vegetation Index) and the natural soil erosion potential. At both scales, the priority areas for forest restoration were assessed based on cluster histograms of SOM. Finally, a contributive map of samples for the best matching units was elaborated, and thatenabled an insightful approach for the analysis of the generated clusters.