Integrating Google Earth Engine and machine learning for urban land use and land cover dynamics analysis

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Machine Learning

Background:

• Accurate land use and land cover (LULC) classification is crucial for urban planning and environmental management.
• Data-scarce urban regions present significant challenges for LULC classification due to computational and accuracy limitations.
• Traditional LULC classification methods often fall short in complex urban environments.

Purpose of the Study:

• To address the research gap in LULC classification for data-scarce urban areas.
• To introduce and evaluate robust machine learning algorithms within the Google Earth Engine platform.
• To comparatively analyze the performance of four machine learning classifiers for LULC mapping.

Main Methods:

• Utilized Google Earth Engine (GEE) for LULC classification using Sentinel satellite data (2020-2024).
• Implemented and compared four machine learning algorithms: Classification and Regression Tree (CART), Minimum Distance (MiD), Random Forest (RF), and Support Vector Machine (SVM).
• Evaluated classifier performance using accuracy assessment metrics, including overall accuracy, Kappa coefficient, Producer's Accuracy (PA), User's Accuracy (UA), Mathew Correlation Coefficient (MCC), and F1 score, with 70% training and 30% testing data splits.

Main Results:

• Random Forest (RF) and CART classifiers achieved the highest overall accuracy (95%) and Kappa coefficients.
• CART and RF demonstrated excellent performance with PA and UA exceeding 90% for all classes.
• MiD exhibited the weakest performance among the tested classifiers. McNemar test indicated no significant difference between CART, RF, and SVM, but CI confirmed CART and RF superiority.

Conclusions:

• RF and CART are highly effective and transferable machine learning algorithms for LULC classification in complex urban environments.
• Google Earth Engine provides a robust platform for implementing and comparing advanced ML algorithms for geospatial analysis.
• The study provides reliable LULC classification data for the urbanized region of Peshawar, overcoming previous limitations.