Land Cover Mapping Using Automated Machine Learning with Landsat Data

This study explores the efficacy of AutoML in developing machine learning models for pixel-wise classification of land cover in Landsat images, focusing on Boracay Island, Philippines.Using the MLJar AutoML tool, high-performing algorithms including Neural Network, XGBoost, CatBoost, Extra Trees and LightGBM were integrated into an ensemble classifier through iterative selection and tuning processes.Evaluation on the training image demonstrated superior performance with strong precision and recall metrics across various land cover classes.However, variability in classifier performance was evident when applied to images from different dates or sensors, particularly affecting built-up areas and less prevalent classes.Despite this variability, significant land cover trends from 2008 to 2024 were discerned in Boracay Island, showing a substantial increase in built-up areas (23% to 38% of total area) and a decline in vegetation cover (59% to 45%).These findings underscore the dynamic changes occurring on the island and highlight the practical applications of this study, such as urban planning, environmental monitoring, and policymaking to balance development with environmental preservation, ensuring the island's long-term sustainability.