Colored dissolved organic matter estimation using Sentinel-2 imagery in small-scale reservoir: Classification and...

Area of Science:

• Environmental Science
• Remote Sensing
• Machine Learning

Background:

• Imbalanced datasets pose challenges for accurately estimating Colored Dissolved Organic Matter (CDOM) in aquatic environments.
• Accurate CDOM estimation is crucial for water quality monitoring and environmental management.

Purpose of the Study:

• To integrate machine learning models with synthetic data to overcome data imbalance issues in CDOM concentration estimation.
• To evaluate the performance of various machine learning models and synthetic data generation techniques for CDOM estimation.

Main Methods:

• Utilized Sentinel-2 imagery to derive five input variables (band ratios) using the HSIC-Lasso model.
• Generated synthetic data using classification-based (SMOTE-SVM, SMOTE-Tomek) and regression-based (SmoteR) techniques to address data imbalance.
• Trained and evaluated ensemble models including Random Forest (RFR), eXtreme Gradient Boosting (XGB), Support Vector Machine (SVR), and Deep Neural Network (DNN).

Main Results:

• Deep Neural Networks (DNN) combined with SMOTE-SVM and SmoteR achieved the highest prediction accuracies (0.88 and 0.89, respectively).
• These advanced models significantly reduced errors, especially in high CDOM concentration regions.
• Classification-based oversampling techniques proved effective in addressing data imbalances, particularly at high-concentration boundaries.

Conclusions:

• Synthetic data generation and machine learning offer a powerful approach for robust water quality monitoring.
• The proposed framework provides a scalable solution for estimating CDOM concentrations, supporting environmental management and policy.
• DNN models integrated with specific oversampling techniques demonstrate superior performance in handling imbalanced aquatic datasets.