An Expanded Three Band Model to Monitor Inland Optically Complex Water Using Geostationary Ocean Color Imager (GOCI)

Abstract

Due to strict spectral band requirements, the three-band (TB) chlorophyll-a concentration (Cchla) estimation algorithm cannot be applied to GOCI image, which has great potential in frequently monitoring inland complex waters. In this study, the TB algorithm was expanded and applied to GOCI data. The GOCI TB algorithm was subsequently calibrated using an in-situ dataset which contains 281 samples collected from 17 inland lakes in China between 2013 and 2020. MERIS TB and GOCI band ratio (BR) models were selected as comparisons to assess the proposed model. The results showed that the proposed GOCI TB model has similar accuracy with MERIS TB model and overperformed GOCI BR model. The root mean square error (RMSE) of the GOCI TB, MERIS TB, and GOCI BR algorithms are 14.212 μg/L, 12.096 μg/L, and 20.504 μg/L, respectively. The mean absolute percentage error (MAPE) (when Cchla is larger than 10 μg/L) of the three models were 0.377, 0.250, and 0.453, respectively. Similar conclusion could be drawn from a match-up dataset containing 40 samples. Finally, a simulation experiment was carried out to analyze the robustness of the models under various total suspended matter concentration (CTSM) conditions. Both the in-situ validation and simulation experiment indicated that the GOCI TB factor could effectively eliminate the optical influence of CTSM. Furthermore, the broader spectral range requirement of GOCI TB model made it proper for many other multispectral sensors such as Sentinel two Multispectral Instrument (S2 MSI), Moderate Resolution Imaging Spectroradiometer (MODIS) (onboard the Terra/Aqua satellite), and Visible Infrared Imaging Radiometer Suite (VIIRS) (onboard the National Polar-orbiting Partnership satellite). Compared with the GOCI BR algorithm, the GOCI TB algorithm has stronger stability, better accuracy, and greater potential in practice.