Optimizing Land Use Classification Using Google Earth Engine: A Comparative Analysis of Machine Learning Algorithms

Land use and land cover (LULC) mapping provides crucial information for sustainable development, urban planning, disaster risk assessment, and mitigation. Various approaches are used for LULC classification in remote sensing, but machine learning has recently gained significant popularity. This paper investigates the application of machine learning algorithms for LULC mapping in Al Ain city, UAE. The study utilizes the Gradient Tree Boosting (GTB), Random Forest (RF), Support Vector Machine (SVM), and Classification and Regression Tree (CART) classifiers within the Google Earth Engine (GEE) platform. The objective is to evaluate and compare the performance of these algorithms using Sentinel-2 imagery from 2024 while also assessing GEE’s suitability for handling both the dataset and algorithms. Various parameters influence algorithm performance. Algorithm performance is evaluated based on overall accuracy and kappa coefficient metrics along with user and producer accuracy. The results indicate that RF and GTB achieved the highest overall accuracy, with GTB's Kappa coefficient slightly lower than RF’s, followed by SVM. CART demonstrated a comparatively lower overall accuracy and Kappa coefficient than the other classifiers. These findings provide insights into the suitability of these algorithms and highlight GEE’s limitations -particularly its memory constraints- for LULC mapping in arid environments like Al Ain. This research contributes to the development of LULC mapping methodologies and their applicability in a sustainable development context.