Evaluating Tropical Chlorophyll-a Algorithms from Sentinel-2 Satellite Imagery: A Comparative Study with In Situ Observation Data in Freshwater Bodies of Thailand

Abstract

In Thailand, eutrophication remains a critical issue due to high nutrient loading that promotes rapid phytoplankton proliferation. Assessing eutrophication using satellite-derived data is essential, offering a faster, more cost-effective, and less labor-intensive alternative to conventional field-based methods. In this context, the present study evaluates chlorophyll-a estimation algorithms from Sentinel-2 imagery through two approaches: (1) refinement of four previously published algorithms originally developed for temperate or subtropical regions, and (2) development of new region-specific algorithms using multiple linear regression (MLR) and polynomial regression (PO). The results from the first approach indicated that the adjusted chlorophyll-a (Chla) algorithm, validated through 10-fold cross-validation, yielded the best performance among the four, achieving an average R² and adjusted R² = 0.64, RMSE = 46.25 µg/L, mean bias = -0.58 µg/L, and Index of Agreement (IoA) = 0.88. In the second approach, the PO algorithm exhibited superior predictive accuracy, with R² = 0.73, adjusted R² = 0.72, RMSE = 39.19 µg/L, mean bias = -4.28 µg/L, and IoA = 0.91. The findings reveal that previously published algorithms developed for temperate or subtropical regions cannot be directly transferred to tropical freshwater lake systems in Thailand, due to differences in ecological dynamics, climate variability and phytoplankton species. The limited performance of unadjusted existing algorithms emphasizes the necessity of locally calibrated algorithms. Therefore, the proposed PO-based empirical algorithm, developed with regional datasets, demonstrates substantial potential for accurate, scalable, and cost-effective chlorophyll-a monitoring.