Evaluating the Accuracy of Land-Use Change Models for Predicting Vegetation Loss Across Brazilian Biomes

Land-use change models are used to predict future land-use scenarios. Various methods for predicting changes can be found in the literature, which can be divided into two groups: baseline models and machine-learning-based models. Baseline models use clear change logics, such as proximity or distance from spatial objects. While machine-learning-based models use computational methods and spatial variables to identify patterns that explain the occurrence of changes. Considering these two groups of models, machine-learning-based models are much more widely used, even though their formulation is considerably more complex. However, the lack of studies comparing the performance of models from these two groups makes it impossible to determine the superiority of one over the other. Therefore, this article aims to evaluate and compare the accuracy of baseline and machine-learning-based models for study areas in three Brazilian biomes. Four baseline models (Euclidean distance from anthropic uses, Euclidean distance from vegetation suppressions, null change model, and random change model) and four machine-learning-based models (TerrSet artificial neural network, TerrSet SimWeigth, Weights of Evidence–Dinamica Ego. and Random Forest model) were trained considering the environmental context of the period from 1995 to 2000. The objective was to predict natural vegetation suppression from 2000 to the years 2005, 2010, 2015, and 2020. The predicted maps were evaluated by comparing them with reference land-use maps using rigorous accuracy methods. The results show that, regardless of the underlying method, the models presented similar performance in all situations. The results and discussions provide a contribution to understanding the strengths and weaknesses of various change models in different environmental contexts.