Modeling and mapping of Inselberg habitats for environmental conservation in the Atlantic Forest and Caatinga domains, Brazil

Abstract

Inselbergs are rocky outcrops generally composed of granitic or gneiss matrix, with monolithic proportions and expressive levels of biodiversity and endemism in different climatic domains. Brazil has one of the largest concentrations of Inselbergs on the planet, which require research to highlight the representative sites for conservation in the urban and rural environment. In this context, the present study evaluated the performance of machine learning algorithms in the mapping of Inselbergs in different environments. Specific objectives correspond to predicting and spatializing Inselbergs using algorithms, as well as selecting important covariates for their spatial distribution in the Atlantic climatic domain of Brazil. Thus, image classification techniques, machine learning algorithms and geospatial modeling were used to map the distribution of Inselbergs and select relevant sites for ecological and environmental conservation, considering anthropogenic and climate change threats. According to the results, the employed models presented kappa and accuracy values ranging from 0.76 to 0.91 and 0.89 to 0.96, respectively. Gradient Boosting Machine (GBM), Support Vector Machines with Radial Basis Function Kernel (SVM), C5.0 and Random Forest (RF) algorithms presented the best results in the mapping of the typical Inselbergs in the study area, since they were able to model spectral signatures of different land use and land cover classes. Statistically, they do not depend on the normal distribution of data and, at the same time, algorithms are able to work with a high number of covariates. The employed methodology has great potential to help in the selection of typical Inselberg landscapes in the Atlantic Forest and Caatinga biomes, which can support the protection and efficient management of biodiversity and geodiversity. The proposed methodology can be adapted to different areas and biomes of the world.