Abstract: Based on the water quality sampling data and hyperspectral data of suspended sediment in the Pearl River Estuary, this paper constructs artificial neural network (ANN), support vector regression machine (SVR) and random forest (RF) suspended sediment inversion models. The results show that the prediction accuracy of the models is RF,ANN,SVR, and the random forest model has the best fitting effect. The random forest model is used to invert the suspended sediment concentration in the Pearl River estuary. The results show that the suspended sediment concentration at the Pearl River port is high in the west and low in the east, and it gradually decreases from the nearshore to the offshore. It is mainly due to the trumpet-shaped topography of the Pearl River estuary. Under the combined action of monsoon and tide, the top area of the Pearl River estuary is strongly mixed by tide and wind direction. In the area far away from the top of the Pearl River estuary, the terrain is relatively open, and the suspended sediment of each tributary runoff is concentrated and accumulated in the southwest region of the Pearl River estuary, resulting in a high suspended sediment concentration in the southwest region of the Pearl River estuary.

Key words: Pearl River Estuary, suspended sediment, machine learning, remote sensing inversion, regularity analysis