Machine learning methods: Modeling net growth in the Atlantic Forest of Brazil

Abstract

Tree growth models are an important and essential part of modeling forest dynamics and valuable tools for management planning and biodiversity conservation strategies. We applied three different machine learning models, namely Artificial Neural Networks (ANN), Support Vector Machine (SVM) and Random Forest (RF) models to predict tree growth at the plot-level in the Atlantic Forest of Brazil. Forest attributes, land use history, landscape, soil and climatic characteristics were used in the modeling. Recursive Feature Elimination (RF) was used to select the best subset of predictor variables. We found that edaphic, forest attributes and climatic variables are important in shaping growth in the Brazilian Atlantic Forest. Soil acidity was the most important characteristic. The machine learning methods were efficient. The Random Forest method showed superiority over the others for modeling growth in the Atlantic Forest. The Nemenyi test pointed out that the difference between the RF model and the other techniques was greater than the calculated Critical Difference (CD). Machine learning can be an important tool for modeling growth in forest fragments in the Brazilian Atlantic Forest. They can help in understanding the biome and in developing management strategies aimed at recovering biodiversity and reducing the deleterious effects of fragmentation.