Mapping Continental Water Bodies in the Peruvian Andes Using Machine Learning and Sentinel-2 Imagery

To map continental water bodies in the southern Andean region of Arequipa, Peru, using satellite images and machine learning algorithms to generate accurate information to facilitate their monitoring, conservation, and sustainable management. The study employed the CRISP-DM methodology, encompassing six phases: business understanding, data understanding, data preparation, modeling, evaluation, and deployment. Sentinel-2 multispectral satellite imagery was processed using ArcGIS Pro. Three supervised machine learning algorithms (Random Forest [RF], Support Vector Machine [SVM], and K-Nearest Neighbor [KNN]) were trained and tested on raster data to classify features into distinct categories and map inland water bodies. Accuracy was assessed using confusion matrices and random sampling validation points. K-Nearest Neighbor algorithm outperformed RF and SVM, achieving over 74% precision and accuracy in detecting inland water bodies, compared to nearly 70% for the other algorithms. The results demonstrated the feasibility of using high-accuracy machine learning techniques to classify and map inland water bodies, specifically lakes and lagoons, by identifying key features contributing to classification accuracy. This research demonstrates the effectiveness of integrating satellite imagery and machine learning in mapping continental water bodies. Generating detailed spatial data supports informed decision-making for preserving Arequipa's continental water bodies, contributing to sustainable environmental management practices.