Abstract
Water quality monitoring in shallow and sheltered sub-tropical coastal water bodies like the Mississippi Sound is crucial for understanding ecosystem dynamics and supporting management decisions, especially when considering major river diversion projects. Application of machine learning (ML) techniques offers promising cost-effective new approaches utilizing archived remote sensing data for analyzing complex environmental data and predicting water quality parameters accurately and efficiently. The aim of this research was to leverage Landsat satellite imagery and ocean optics data from Aqua MODIS in conjunction with ML techniques to enhance the accuracy and efficiency of chlorophyll-a (Chla) estimation in the Mississippi Sound with a focus on variability driven by seasonal patterns, riverine inputs, and ocean biogeochemical parameters. Using a robust ML model based on an ensemble model, Extra Trees (ET), we estimated Chla concentrations across twelve months and evaluated the model’s performance against other ML regression-based models. The ET model consistently provided accurate and reliable predictions, achieving an R² of 0.999 and a root mean square error of 0.187 mg/m³. By capturing complex interactions influencing Chla variability, the ET model demonstrated superior performance compared to traditional empirical and regression-based methods. Model outputs showing lower Chla concentrations observed during winter months align with established seasonal trends in temperate coastal ecosystems. Conversely, the higher Chla concentrations observed along the coast are attributed to increased nutrient inputs from rivers such as the Pearl, Pascagoula, and Mobile Rivers, as well as coastal runoff and freshwater diversions from the Mississippi River. The influx of freshwater increased levels of nutrients, total suspended solids, phytoplankton, and total organic carbon, which resulted in higher light extinction and diminished light penetration to the seabed. This research improves our comprehension of Chla fluctuations in the Mississippi Sound and showcases the promise of cutting-edge machine learning methods for monitoring and forecasting coastal ecosystems.
