An object-based custom ensemble framework for automated water body extraction using Sentinel-2 imagery.

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Environmental Monitoring

Background:

• Pixel-based classification methods for surface water monitoring in India face limitations due to complex landscapes.
• Existing methods often struggle with accuracy and generalization across diverse geographical regions.

Purpose of the Study:

• To develop and validate an automated object-based framework for enhanced surface water extraction using Sentinel-2 imagery.
• To improve the accuracy and regional generalizability of water monitoring in India.

Main Methods:

• Integration of DBSCAN clustering with supervised machine learning classification (XGBoost) for object-based image analysis.
• Training and validation on a multi-site dataset covering 20 diverse geographical locations across India.
• Benchmarking against traditional NDWI thresholding and pixel-based XGBoost classification.

Main Results:

• The proposed DBSCAN-XGBoost framework achieved a superior overall accuracy of 97% and an F1-score of 0.96.
• XGBoost outperformed other machine learning models (Random Forest, SVM) in complex spectral zones.
• The object-oriented approach significantly reduced "salt-and-pepper" noise and topographic shadow errors.

Conclusions:

• The object-based DBSCAN-XGBoost framework offers a significant advancement for automated surface water extraction in India.
• This approach demonstrates improved accuracy and robustness compared to traditional methods.
• Further multi-site validation could enable high-fidelity water auditing across diverse Indian landscapes.