Mental health and natural land cover: a global analysis based on random forest with geographical consideration

Abstract

Abstract Natural features in living environments can help to reduce stress and improve mental health. Different land types have disproportionate impacts on mental health. However, the relationships between mental health and land cover are inconclusive. In this study, we aim to accurately fit the relationships, estimate the impacts of land cover change on mental health, and demonstrate the global spatial variability of impacts. In the analysis, we show the complex relationships between mental health and eight land types based on the random forest method and Shapley additive explanations. The accuracy of our model is 67.59%, while the accuracy of the models used in previous studies is usually no more than 20%. According to the analysis results, we estimate the average effects of eight land types. Due to their scarcity in living environments, shrubland, wetland, and bare land have larger impacts on mental health. Cropland, forest, and water could improve mental health in high-population-density areas. The impacts of urban land and grassland are mainly negative. The current land cover composition influences people’s attitudes toward a specific land type. Our research is the first study that analyzes data with geographical information by random forest and explains the results geographically. This paper provides a novel machine learning explanation method and insights to formulate better land-use policies to improve mental health.