Machine learning-based prediction for land degradation mapping using multi-source geospatial data in the Batanghari watershed, Sumatra, Indonesia

Abstract

Abstract The study area is a tropical environment located in the Batanghari watershed, Sumatra, Indonesia. The existence of the environmental problems and damages in the study area can be identified based on land degradation. It can be interpreted as a complex process and is influenced by human activities, climate change, and natural events. This study proposes the latest Geospatial Artificial Intelligence (Geo-AI) model using multi-sources geospatial data that is specifically used to address challenges and phenomena related to the identification of land degradation in the study area. The novelty of this study is that it is the first time to integrate the 6 (six) main variables of multi-source geospatial data - Topographical, Biophysical, Bioclimatic, Geo-environmental, Global human modification, and Accessibility - in predicting potential land degradation in the tropical environment, such as Indonesia. Machine learning-based prediction Support Vector Machine (SVM), Minimum Distance (MD), Classification and Regression Trees (CART), Gradient Tree Boost (GTB), Naïve Bayes (NB), Random Forest (RF) algorithms were used to predict and to map land degradation in the study area. The overall accuracy of the results of comparison and evaluation of machine learning-based predictions on the RF, CART, GTB, SVM, NB, and MD in the study area are 86.2%, 85.8%, 81.2%, 52.8%, 36.3%, and 34.5%, respectively. Therefore, the study concluded that the RF, CART, and GTB algorithms are proposed to be applied to produce land degradation map in the study area.