A hybrid intelligence approach to predicting and explaining land use and land cover change using Bayesian classifiers

The future health of land systems depends on our ability to quantify and understand the causes of land cover change. This requires the integration of data of different types and spatio-temporal resolutions and the creation of accurate, explainable prediction models. To achieve this, a framework is presented where all variables are converted to binomial form at the same spatio-temporal resolution, co-occurrences between classes of interest, C , and a set of predictive features, X , are computed and converted into a Bayesian classifier P ( C | X ) , where each feature X i j is quantified by its effect size, coverage, statistical significance and potential causality. As a usecase, a prediction model for deforestation at the municipal level in Mexico in 2020 is created using a set of 98 Climatic, Social, Economic and Governmental factors as putative causes, showing that besides the impact of alternative land uses, the models infer an important role for the Sembrando Vida government program.