Development of Empirical CDOM Algorithm for Sentinel-2 Using the Gloria Dataset

Water quality is crucial for the long-term health of undersea biological ecosystems, including elements like Colored Dissolved Organic Matter (CDOM). Gathering field data to characterize CDOM is expensive and time-consuming. To address this, the optical aquatic research community has compiled the GLORIA dataset, which includes measurements of water quality indicators such as chlorophyll a, total suspended solids, absorption by dissolved substances, and Secchi depth. This dataset aids in routine monitoring of high-priority sites, algorithm development, and data validation. In this study, we employed the CDOM data from the GLORIA dataset to develop an empirical CDOM algorithm using Sentinel-2 imagery. The GLORIA dataset encompasses 7,572 stations globally, but for this study, only 92 stations were utilized to construct a tropical water CDOM algorithm. This algorithm was then calibrated with CDOM measurements from the Derawan Archipelago. The developed empirical algorithm is based on a random forest regression model. The algorithm, derived from the GLORIA dataset, demonstrated promising training data accuracy (RMSE = 0.42, R-Square = 0.37). However, the validation accuracy was lower (RMSE = 0.41, R-Square = 0.23), and the tests on the Derawan CDOM dataset indicated even poorer accuracy. These results highlight the challenges in developing a global CDOM algorithm based on multispectral imagery.