In order to achieve high-precision remote sensing inversion of suspended particulate matter concentration in the seas surrounding the Yellow River Estuary, this paper constructs seasonal models for spring, summer, and autumn, as well as a cross-seasonal model, utilizing GOCI-I image data and based on the WOA-BP algorithm. These models are compared with multiple algorithms such as Catboost, RF, KNN, BP and so on. The results reveal that within each seasonal model, the WOA-BP algorithm exhibits superior performance on both the training and testing sets, with the average relative errors for the respective seasonal testing sets being 24.18%, 25.97%, and 29.42%. When the cross-seasonal testing set is employed to evaluate the three models, and their accuracy is found to be significantly lacking, which indicates that seasonal models are not applicable across different seasons. In the cross-seasonal model, the WOA-BP algorithm again demonstrates the highest accuracy, with an overall average relative error of 26.96%. The average relative errors when testing with the three seasonal testing sets are 25.80%, 21.90%, and 37.17%, respectively. While the accuracy for summer is improved, the accuracy for the other two seasons falls below that of the corresponding seasonal models, with autumn experiencing the greatest decline in precision. Therefore, it is suggested that the cross-seasonal model be employed for spring and summer, whereas the appropriate seasonal models are recommended for autumn.