To improve localization accuracy and reduce communication energy for mobile nodes of underwater wireless sensor networks (UWSNs), the localization prediction method with mobility pattern for nodes was proposed. Considering that tidal motion is the main factor to generate movement of sea water in seashore monitoring network, Gauss radial basis function was exploited as the spatial basis function to construct the node mobility pattern. Then K-medoids method was utilized to cluster and optimize the center of Gauss radial basis function and the extended Kalman filtering algorithm was adopted to predict model coefficients. Moreover, weight coefficients related to distance were designed based on the spatial correlation between the anchor node and ordinary node. The ordinary node estimated the model coefficients based on anchor node coefficients and weight coefficients, and achieved localization with estimated speed and last location. The simulation results of UWSNs in the region from 117.25°E to 132.2°E and from 24°N to 43.45°N showed that predictable localization method with mobility pattern for nodes of UWSNs (PLM_MP) had better performances. The localization coverage rate and the localization accuracy were higher than that in SLMP scheme and MP-PSO scheme, and the average communication cost was lower. It also showed that PLM_MP scheme is suitable for localization in large mobile underwater wireless sensor networks.