High spatial-resolution satellite mapping of suspended particulate matter in global coastal waters using particle composition-adaptive algorithms

Abstract

Delivery of suspended particles, referred hereafter also to as suspended sediment, to coastal zones plays a first order control on the development and maintenance of muddy geomorphic features like river deltas, mudflats, and tidal wetlands. While sediment delivery from rivers is relatively straightforward to monitor and has been well studied, suspended sediment derived from erosion of coastal bluffs and resuspension of shallow subtidal sediments remains poorly constrained. Estimates of the concentration of suspended particulate matter (SPM) provide one of the best remotely sensed metrics for suspended sediment supply to the coast. Spaceborne ocean color sensors with coarse spatial resolution (∼1 km pixel size at nadir) are generally inadequate to resolve smaller-scale sediment dynamics in coastal waters and additionally there is a limitation associated with adjacency effect of 1-km land pixels on near-shore water pixels. In contrast, satellites dedicated primarily to land observations with a smaller pixel size (∼30 m) provide more adequate spatial resolution for observations of coastal waters. This paper presents a particle composition adaptive algorithm for retrieving SPM from ocean remote-sensing reflectance, R_rs(λ), in coastal waters which is applicable to most land observation satellites. For the algorithm development, we compiled more than 800 paired in situ spectral reflectance and SPM measurements from 12 marine sites worldwide, representing a wide range of suspended particle concentration and composition. We first classify the satellite image data into three water types: organic-rich, mineral-rich, or extremely mineral-rich based on the POC/SPM ratio that is derived from R_rs(λ). The ratio of particulate organic carbon (POC) to SPM serves as a particle composition metric. Then, SPM is estimated from R_rs(λ) using a particle composition-specific algorithm which employs the reflectance at red band for organic-rich waters and near-infrared (NIR) for mineral-rich waters. We compared the performance of this algorithm with eight previously published SPM algorithms, including empirical, semi-analytical, and machine learning approaches. Results show that our algorithm produces reliable SPM estimation with coefficient of determination (R²), root mean square error (RMSE in log space), and median absolute percent error (MAPE) of 0.91, 0.20, and 30.5 %, respectively. To examine the capability of our algorithm to study the long-term variability in coastal SPM at high spatial resolution, we implemented the algorithm to the 40-year Landsat data archive in Google Earth Engine (GEE). The Landsat mapping results of SPM were validated using both the satellite-in situ matchups of SPM data as well as in situ water turbidity measurements. Finally, we demonstrate a few scenarios of fine-scale SPM patterns as well as seasonal and long-term variability across different marine coastal environments using the satellite high spatial resolution SPM mapping. These results collectively demonstrate the promise of this new SPM retrieval algorithm for mapping and monitoring global coastal suspended sediment dynamics.