Sentinel-2 chlorophyll-a algorithms performed best after adjustment

What the study found: The adjusted chlorophyll-a algorithm from Sentinel-2 imagery performed best among four previously published algorithms after refinement, with 10-fold cross-validation showing average R² and adjusted R² of 0.64, RMSE of 46.25 µg/L, mean bias of -0.58 µg/L, and Index of Agreement of 0.88.
Why the authors say this matters: The study suggests that satellite-derived chlorophyll-a estimates can support eutrophication assessment in Thailand, offering a faster, more cost-effective, and less labor-intensive alternative to conventional field-based methods.
What the researchers tested: The researchers evaluated chlorophyll-a estimation algorithms for freshwater bodies in Thailand using Sentinel-2 satellite imagery and in situ observation data. They used two approaches: refining four previously published algorithms developed for temperate or subtropical regions, and developing new region-specific models with multiple linear regression (MLR) and polynomial regression (PO).
What worked and what didn't: Among the four refined algorithms, the adjusted chlorophyll-a algorithm performed best. The abstract also states that new region-specific algorithms were developed with MLR and polynomial regression, but the provided text ends before reporting their performance.
What to keep in mind: The abstract only provides detailed performance results for the first approach, and the description of the second approach is incomplete in the supplied text. The study is limited to freshwater bodies in Thailand and the abstract does not describe other limitations.

Key points

• The best-performing refined model was the adjusted chlorophyll-a algorithm from Sentinel-2 imagery.
• That model’s 10-fold cross-validation results were R² and adjusted R² of 0.64, RMSE of 46.25 µg/L, mean bias of -0.58 µg/L, and IoA of 0.88.
• The authors say satellite-based chlorophyll-a assessment may support eutrophication monitoring in Thailand.
• The study compared refined published algorithms with new region-specific models using MLR and polynomial regression.
• The abstract does not report the performance of the region-specific models in the supplied text.