Land suitability modeling integrating geospatial data and artificial intelligence

Context: Sustainable agricultural practices are critical in a world grappling with climate change and pressure on natural resources. Unplanned agricultural expansion often harms ecosystems and the services they provide. Balancing food production with environmental protection demands sophisticated tools like spatial analysis and artificial intelligence to inform land-use decisions. Objective: This study introduces an AI-driven process to assess land suitability for agrosilvopastoral systems, going beyond traditional methods by incorporating a broader spectrum of landscape characteristics. Our approach integrates climate, water resources, soil properties, morphological features, and accessibility to enhance the accuracy of suitability mapping. Methods: We constructed a data cube comprising 100 geospatial layers representing diverse landscape attributes. Field observations from two watersheds in Minas Gerais, Brazil, were used to train and validate a Random Forest classification model. We evaluated the model's accuracy and quantified the influence of each attribute group on suitability determination. Results: Integrating climate, water, edaphic, and morphological attributes significantly improved the model's accuracy and provided a more nuanced understanding of agrosilvopastoral suitability compared to using only soil class, lithology, and slope. Climate and edaphic variables emerged as key drivers of suitability. This approach identified a more constrained, yet potentially more sustainable, distribution of suitable land. Significance: Our findings highlight the need to transition from conventional land suitability assessments towards more holistic, data-driven approaches that consider the complex interplay of environmental factors. This model offers a valuable tool for guiding sustainable land-use planning, potentially mitigating environmental impacts while optimizing agrosilvopastoral production.