Satellite-Based Retrieval of Water Quality Parameters in Inland Waters Using Mixture Density Network Models

Retrieving water quality (WQ) parameters from remotely sensed imagery in inland and coastal waters is challenging as, unlike the open ocean, dissolved and mineral substances often do not covary with phytoplankton. In this study, we tested the accuracy of Mixture Density Networks (MDN) algorithms in estimating WQ parameters⸺Chlorophyll-a  (Chla), Secchi, total suspended solids, total nitrogen (TN), and total phosphorus⸺from Landsat imagery in Missouri reservoirs, USA. Landsat 5-8 data were acquired from the LimnoSat database and paired with in situ measurements taken within 1 day of sampling from up to 150 water bodies (matchups, N ranged from 929 to 966). Each MDN was trained using 50% of the total samples, while the remaining data was used to evaluate the model's performance. Across the suite of WQ parameters, the best performing MDN model (41% error, low bias) was, perhaps unexpectedly, TN. In contrast, the MDN error for estimating Chla, by far the most studied WQ parameter from space, exceeded 100%. Model accuracy was modestly improved when MDNs were trained to estimate multiple parameters at the same time, rather than MDNs trained to estimate a single parameter. That satellite-based TN estimates have much lower error than Chla estimates is noteworthy not only from an applied nutrient management perspective, but also because only a few studies to date have attempted to estimate nutrient concentrations from space. The primary processing scheme used by LimnoSat is LaSRC; however, alternative atmospheric correction processors (e.g., ACOLITE) may improve MDN estimates of WQ.

Presentation Preference: Oral

Primary Presenter: Lorena Pinheiro-Silva, University of Missouri-Columbia (lsilva@umces.edu)

Authors:

Lorena Pinheiro-Silva, University of Missouri  (lsilva@umces.edu)

Rebecca North, University of Missouri  (northr@missouri.edu)

Greg Silsbe, University of Maryland Center for Environmental Science  (gsilsbe@umces.edu)