It is vital to model the topography of the river bed for the geographical characteristics of the river and thus to monitor water quality systematically. This paper provided a solution involving point cloud filtering, interpolation and modelling of bathymetry data for most of the waters of Jinji Lake in Suzhou. Empirical filtering and statistical filtering are performed to address the problems that its undulating topography and many islands pose to the data. Through a literature review, we selected the inverse distance weighting method and the ordinary kriging method for comparison. Through cross-validation, the more accurate data was used to construct a 3D terrain model through the Poisson modelling method. The experimental results show that the ordinary kriging method possesses higher accuracy. In the end, a riverbed model containing 78949 triangular surfaces was completed in 1.1s. The final mesh has no anomalous triangular pieces visible to the naked eye and the normal vectors are generous.