Kalender

Ocean colour for Norwegian coastal water.

by Edson Silva, NERSC

Abstract:

Ocean color remote sensing has supported the monitoring of ocean water quality since the 1980s. However, bio-optical algorithms designed for the open ocean exhibit poor accuracy in coastal waters due to the inflow of dissolved and particulate matter from land, necessitating the calibration of regional bio-optical models. In this study, we calibrated bio-optical models for chlorophyll-a (Chl-a) concentration and Secchi depth (Zsd) in Norwegian coastal waters using Sentinel-3 OLCI. Support Vector Machines (SVM) and Multi-Layer Perceptron (MLP) models were employed for calibration, utilizing in situ observations of Chl-a and Zsd along with Sentinel-3 OLCI optical bands processed through three atmospheric corrections: Basic Atmospheric Correction (BAC), Polymer, and Acolite. The calibrated regional models outperformed standard models across all atmospheric corrections. For Chl-a estimation, the Pearson correlation coefficient improved from 0.19 (BAC-OC4ME) to 0.6 (Polymer–SVM), while Zsd estimations improved from 0.27 (BAC-OC4ME derived) to 0.65 (Polymer–SVM). Error rates were also reduced by more than half with the regional models. The application of the regional models led to two key improvements: (1) correction of Chl-a overestimations in inner coastal areas, likely caused by colored dissolved organic matter, and (2) improved representation of spring and summer bloom phenology along the coast. For Zsd, the main improvement was the reduction of overestimations in Northern Norway. In conclusion, the regional models, particularly Polymer–SVM, provide more accurate estimates and better represent the dynamics of Chl-a and Zsd along the Norwegian coastline, which can support water quality monitoring in the future up to 300 m resolution and daily frequency."