A novel chlorophyll-a retrieval model based on suspended particulate matter classification and different machine...

Area of Science:

• Remote sensing of inland aquatic ecosystems
• Water quality assessment using satellite data
• Machine learning applications in environmental science

Background:

• Chlorophyll-a (Chla) is a key optical parameter for assessing inland water health and enabling early algal bloom warnings.
• MOD09 satellite products offer high temporal and spatial resolution, crucial for water color remote sensing.
• Accurate Chla concentration retrieval is vital for effective aquatic ecosystem management.

Purpose of the Study:

• To develop a high-accuracy machine learning model for estimating Chla concentration in inland waters using MOD09 products.
• To investigate if classifying water bodies by suspended particulate matter (SPM) concentration improves Chla retrieval accuracy.
• To compare the performance of ten common machine learning models for Chla estimation.

Main Methods:

• Developed a machine learning model for Chla estimation, incorporating a novel approach of classifying water bodies based on SPM concentration.
• Evaluated ten machine learning models, including Random Forest Regressor (RFR), Deep Neural Networks (DNN), Extreme Gradient Boosting (XGBoost), and Convolutional Neural Network (CNN).
• Filtered 41 basic bands and 820 band ratios based on their correlation with Ln(Chla) and selected bands for model input. Identified specific bands (B3, B20, B32) for SPM classification with 0.9 accuracy.

Main Results:

• Classifying water bodies by SPM concentration significantly improved the correlation between spectral bands/ratios and Ln(Chla), boosting model verification R² from 0.41 to 0.83.
• The Random Forest Regressor (RFR) model demonstrated superior performance, indicated by higher R², lower RMSE, and lower MAPE.
• Band B3 showed the highest contribution to Chla estimation across different SPM-classified groups.

Conclusions:

• SPM classification is an effective strategy for enhancing the accuracy of satellite-based Chla concentration retrieval in inland waters.
• The RFR model, utilizing selected spectral bands, provides a robust and accurate method for Chla estimation.
• The developed model shows significant potential for application in other inland water bodies and serves as a valuable reference for future research.