Secchi disk depth (SDD) reflects the turbidity and transparency of seawater and serves as an intuitive indicator of water quality. As a key parameter in marine ecological environment monitoring, it is closely related to the physicochemical properties of seawater, fishery production, and issues such as water eutrophication. It also plays a significant role in the study of the optical characteristics of water bodies. While traditional in-situ measurement methods can indeed provide accurate information on transparency, these methods are resource-intensive and labor-consuming, which makes it difficult to meet the demands of dynamic spatiotemporal monitoring of transparency. With the advancement of satellite remote sensing technology, satellite imagery enables the acquisition of water transparency data on a much larger spatial and temporal scale. An increasing number of ocean color sensors with varying radiation accuracies are also being used to estimate transparency. Transparency is influenced by environmental conditions and shows complex spatiotemporal variations. Currently, inversion models for transparency are highly regional and local in nature. Traditional remote sensing inversion methods have mostly been based on spectral data. The spectra themselves are influenced by many factors, such as satellite sensor band settings, bandwidth, and signal-to-noise ratio. The consistency of transparency inversion methods between different satellites requires further validation. Currently, there is no universal transparency inversion model applicable to different multispectral satellites, which makes achieving consistency and comparability in transparency estimates from different satellites challenging. Water chromatic parameters are closely related to water body scattering, the aquatic environment, and water color components. They contain rich environmental information that can also be used for transparency inversion. Unlike spectral-based transparency remote sensing inversion methods, the advantage of using chromatic parameters for transparency inversion is that they are less affected by spectral influences, which makes them more suitable for water transparency inversion. With the development of remote sensing technology, the acquisition of ocean color satellite data has continuously increased, and data processing algorithms have advanced. Machine learning techniques have been widely applied in the inversion of water quality parameters. In this study, we construct the coefficient matrix for Aqua-MODIS, S3A-OLCI, S3B-OLCI, and NOAA-20-VIIRS ocean color satellites based on in-situ data from the Bohai Sea. Using the CatBoost machine learning algorithm, we develop a transparency inversion model for the Bohai Sea, aiming to obtain consistent and comparable transparency inversion results from multiple satellite sources.

Based on in-situ spectral and transparency cruise data from the Bohai Sea, we use chromatic parameters as key variables and apply the CatBoost machine learning algorithm to develop the transparency inversion model for the Bohai Sea. The model’s accuracy is validated using the leave-one-out cross-validation method. Using hyperspectral remote sensing reflectance (R_rs) collected during the cruise, we simulate spectral bands R_rs for Aqua-MODIS, S3A-OLCI, S3B-OLCI, and NOAA-20-VIIRS ocean color satellites. A coefficient matrix between the central wavelengths of the four satellites and the XYZ tristimulus values is constructed to calculate the chromatic parameters for each satellite. We analyze the precision and consistency of the chromatic parameter inversion results and transparency inversion results for different satellites. Using S3A data, we study the spatiotemporal variation characteristics of transparency in the Bohai Sea for the four seasons, along with the annual average temporal variation of transparency in the Bohai Sea from 2019 to 2024.

We use Aqua-MODIS, S3A-OLCI, S3B-OLCI, and NOAA-20-VIIRS R_rs to calculate the chromatic parameter information. The constructed transparency inversion model for the Bohai Sea is then applied to obtain the transparency estimation results for the four satellites. The accuracy and consistency of the chromatic parameter inversion results and transparency inversion results for different satellites are analyzed. The hue angle and saturation inversion results from the four satellites are as follows: R²>0.97, MAPE<3%, compared to in-situ measurements (Figs. 4 and 5). The results indicate that the chromatic parameter inversion method used in this study demonstrates excellent consistency for multispectral satellites with different central wavelengths and bandwidths. The accuracy of the transparency inversion model developed in this study is as follows: for model training, the R² is 0.97, the Pearson correlation coefficient is 0.98, the RMSE is 0.24 m, and the MAPE is 14.3% (Fig. 6); for model validation, the R² is 0.87, the Pearson correlation coefficient is 0.93, the RMSE is 0.48 m, and the MAPE is 25.6%, which indicates high validation accuracy (Fig. 6). The model demonstrates high accuracy for transparency inversion in the Bohai Sea. For the transparency inversion models of the Bohai Sea developed by other researchers, the same in-situ data used in this study are applied. After calibrating the model parameters, the leave-one-out cross-validation method is performed for inversion accuracy verification. The comparison between the accuracy of other inversion models and the accuracy of the inversion model in this study shows that the Bohai Sea transparency inversion model in this study demonstrates higher accuracy (Fig. 7, Table 3). The consistency analysis of transparency inversion results from different satellite sensors is as follows. In terms of inversion accuracy for the four satellites, the average RMSE values are 0.47, 0.30, 0.30, and 0.51 m, the average MAPE values are 17.73%, 14.23%, 14.25%, and 19.63%, and the average Pearson correlation coefficients are 0.94, 0.98, 0.98, and 0.93, respectively. Regarding the transparency inversion results among different satellites, the max RMSE is 0.31 m, the max MAPE is 7.3%, and the Pearson correlation coefficients are all above 0.97 (Fig. 8). The chromatic parameter inversion model and transparency inversion model for the Bohai Sea developed in this study effectively solve the limitations of single-satellite sensor inversion models in satellite spectral bands. From 2019 to 2024, in terms of time, the highest average transparency of the Bohai Sea occurred in summer, approximately 3 m, while the lowest occurred in winter, around 1.5 m. Spatially, the transparency near Qinhuangdao is higher than that in other regions. Near the coast, the transparency is lower, and further offshore, the transparency is higher (Figs. 9 and 10).

We utilize chromatic parameters and the CatBoost machine learning algorithm to establish the transparency inversion model in the Bohai Sea. The spectral band simulation method and the chromatic parameter coefficient matrix for the four multispectral satellite sensors—Aqua-MODIS, S3A-OLCI, S3B-OLCI, and NOAA-20-VIIRS—are used to obtain long-term transparency information of the Bohai Sea. The inversion results for hue angle and saturation from the four satellites, compared with in-situ measurements, show that R² values are greater than 0.97 and MAPE values are lower than 3%. This demonstrates that the chromatic parameter inversion method used in this study exhibits excellent consistency for multispectral satellites with different central wavelengths and bandwidths. The transparency inversion model in this study shows the following validation results: model training shows R² of 0.97, RMSE of 0.24 m, MAPE of 14.3%, and the Pearson correlation coefficient of 0.98; model validation shows R² of 0.87, RMSE of 0.48 m, MAPE of 25.6%, and the Pearson correlation coefficient of 0.93, indicating high accuracy (Fig. 6). The transparency inversion results for the Bohai Sea from the four satellites, compared with in-situ measurements, show that the average RMSE for all satellites is below 0.6 m, the average MAPE is below 20%, and the average Pearson correlation coefficient is above 0.9. This indicates that the model developed in this study provides high accuracy and strong consistency for transparency inversion across different multispectral satellites. The multi-source satellite transparency inversion model for the Bohai Sea based on chromatic parameters holds significant value and necessity. As parameters reflect the color characteristics of water bodies, chromatic parameters are not constrained by the satellite’s central wavelength. Additionally, high-resolution imaging equipment can also capture chromatic parameters, which highlights the practical application value of the model developed in this study. Future research will explore inversion algorithms for other water quality parameters, aiming to obtain the water quality parameters using water chromatic parameters and apply them in water quality monitoring.