Water clarity monitoring in complex coastal environments: Leveraging seagrass light requirement toward more functional satellite ocean color algorithms

Abstract

Florida's Big Bend region is home to the second largest contiguous seagrass bed in the continental United States, but the extent and integrity of this seagrass bed is declining as a result of degraded water clarity. Seagrass beds provide important ecological services and support high levels of biodiversity, thus monitoring the spatial and temporal variability of water clarity in the Big Bend region, particularly during the seagrass growth season (April to October), is of vital importance. Satellite remote sensing of coastal water clarity is often challenging due to the optical complexity and heterogeneity of coastal regions as well as frequent clouds and sun glint in the satellite imagery. As such, we developed machine learning (ML) approaches to derive the diffuse attenuation coefficient at 490 nm [Kd(490)] from satellite data with partial atmospheric correction. Traditionally, such algorithms would be developed to capture the full range and variation of the parameter of interest. Here, we demonstrate deficiencies in this method, and instead advocate for an alternate approach which leverages seagrass light requirements to improve the algorithm design. Specifically, we developed and implemented a 2-stage ML approach, which first classifies the coastal water into two types, then retrieves Kd(490) for the water type within which measurable changes in water clarity may impact seagrass health. We developed and compared the traditional ML approach and the proposed 2-stage ML approach based on data from MODIS/Aqua, OLCI/Sentinel-3A, and VIIRS/SNPP. Given the outperformance of the 2-stage ML approach, we applied it to the MODIS/Aqua data archive (2003−2020) to generate time-series Kd(490) maps and analyzed how often different areas across the Northern Big Bend region are suitable for seagrass growth each year. The derived information is important toward understanding how the spatial and temporal variation in light availability impacts seagrass growth and distribution in the study region, while the 2-stage ML approach appears applicable to other coastal regions with similar optical complexity.