Integrating Multi-Variable Driving Factors to Improve Land Use & Land Cover Classification Accuracy using Machine Learning Approaches: A Case Study from Lombok Island

Abstract

Listen

Translate article

Accurate classification of land cover is essential for effective land management and environmental monitoring. This study aimed to enhance land cover classification for Lombok Island using advanced machine learning algorithms. The models employed include Random Forest, Gradient Boosting, Decision Tree, and Naive Bayes, integrating a wide range of variables, such as Landsat satellite imagery, spectral indices, physiographic, climatic, and socio-economic data. Among these, Random Forest demonstrated the highest model accuracy at 82%, followed by Gradient Boosting at 80%, Decision Tree at 73%, and Naïve Bayes at 61%. In field validation assessments, comparing the predictions of these machine learning models with ground truth data, Random Forest was the most reliable, achieving an overall accuracy of 88%. This superior performance is largely due to the multi-variable approach, which allows the model to mitigate issues like cloud cover in satellite images. The key variables that significantly influenced the land cover classification on Lombok Island include proximity to settlements, temperature, and distance to roads. These results provide essential insights for land management strategies, enabling policymakers and stakeholders to make informed decisions on sustainable development, urban planning, and environmental conservation in rapidly changing landscapes.