Interpretable Approaches for Land Use and Land Cover Classification

Context: The Land Use Land Cover (LULC) undergoes various changes over time. Monitoring these changes is important for environmental disaster management, and agricultural land monitoring, among other applications. Focusing on the effects caused by environmental disasters, the Quadrilátero Ferrífero Region in Minas Gerais, Brazil, was chosen as the study area. This region has experienced dam breaks over the years, leading to an increased number of studies investigating the use of machine learning in this context. Problem: In the context of environmental disasters, making quick and effective decisions is essential, and the use of automated LULC classification systems plays a crucial role in this process. Proposed Solution: This study developed a map for the mentioned study area and presented a detailed methodology for creating a LULC classifier using machine learning techniques and discussing model interpretability in this context. SI Theory: This work is associated with Computational Learning Theory, focusing on defining and exploring automated classification methods. Method: Remote Sensing data were used to segment the area of interest, calculating indices from information obtained from different satellite bands and the digital elevation map. The models were trained with labeled samples after feature selection. Results: The tested models (Random Forest, Support Vector Machine, and K-Nearest Neighbours) demonstrated performance with no statistical difference, achieving F1-scores ranging from 0.89290 to 0.92994. Contribution: Considering the increasing research in this area, this work provides a detailed methodology that can contribute to and promote promising machine learning solutions for this field.