Genetic Algorithm Optimization of Ensemble Learning Approach for Accurate Land Cover and Land Use Mapping

Accurate Land-Cover and Land-Use (LCLU) classification is critical for biodiversity conservation and environmental management. An ensemble Machine Learning (ML) approach for LCLU mapping in the Talassemtane National Park (TNP) in Morocco using Remote Sensing (RS) data has been developed and optimized. Sentinel-2 satellite imagery was used and processed to extract six spectral features and six vegetation indices. Google Earth Engine (GEE) as a processing platform. Multiple ML classifiers including Random Forest (RF), Support Vector Machines (SVM), Naive Bayes (NB), Classification and Regression Tree (CART), Minimum Distance (MinD), and Gradient Tree Boost (GTB) were compared to identify the most suitable ML algorithm for LCLU classification. Accuracy was assessed using multiple metrics to provide a robust evaluation of model performance. RF Ensemble Classifier was trained and optimized using a Genetic Algorithm (GA) to tune its hyperparameters. Accordingly, optimal settings improved classification accuracy by 4.51% and achieved an overall accuracy of 97.63% and a kappa coefficient of 95.69%. The selection of covariates (input features and vegetation indices) tailored to distinguish different land cover types significantly improved classification results. This research demonstrates that optimized ensemble ML represents a promising approach for accurate LCLU mapping from RS data to inform conservation planning and management in protected areas. The investigation of GA as a tuning strategy for RF hyperparameters provided insights into effectively optimizing ensemble classifiers for RS applications. The optimized LCLU map provides valuable information to guide conservation planning and management efforts within the protected areas.