Validation of standard ocean-color chlorophyll-a products in turbid coastal waters: A case study on statistical evaluation and quality control tests in the Persian Gulf

Abstract

Selecting the appropriate in-situ and satellite matchups is a critical task for evaluating and application of ocean color products in optically complex coastal waters. This article investigates the performance of five single-sensor and two merged multi-sensors Chl-a products in the Persian Gulf. In this study, bio-optical and Chl-a measurements collected from 531 stations in the northern Persian Gulf from 2008 to 2019 were used. In-situ samples were initially controlled to filter out inappropriate datasets. Results indicated that surface measurements may not be representative of satellite-derived Chl-a, and satellite estimations were mostly consistent with Chl-a concentrations in the first optical depth (Chlopt). Statistical analysis showed that all satellite-derived Chl-a products overestimated the Chlopt by 48%–170%, in which Chl-a from Neural-Network and OC5 algorithms yielded the best agreement. Afterwards, the QC tests were designed based on the remote sensing reflectance (Rrs) at 555 nm as a proxy of SPM, Rrs(412)/Rrs(443) as a proxy of CDOM, and Rrs(560)/Rrs(490) as a proxy of Chl-a, to select the matchups with the least interference of Colored Dissolved Organic Matter (CDOM) and Suspended Particulate Matter (SPM) on Chl-a. The correlation between Chlopt and satellite-derived Chl-a improved significantly after applying the QC tests (R2 = 0.80–0.89). The valid pixels were determined using matching the results of QC tests and standard level-2 quality flags. Finally, statistical calibration was used to create the final quality controlled calibrated Chl-a maps, using regression coefficients obtained from training/validation exercises and bootstrapping-like techniques applied to the quality-controlled datasets. The findings showed that in turbid coastal waters, standard ocean color Chl-a products cannot be validated only by performing statistical methods, and quality control experiments are necessary for their applications.