Abstract: When underwater gliders perform missions in complex marine environments, the uncertainty of waypoints (surfacing points) significantly affects the effectiveness and accuracy of mission planning. To improve the prediction accuracy of waypoints, in response to the limitations of existing machine learning methods in predicting performance under complex terrains and small sample sizes, this paper proposes an underwater glider waypoint prediction model that integrates terrain and ocean current data, namely the Terrain-Conditional Generative Adversarial Network-Attention-weighted-Random Forest (Terrain-CGAN-Att-RF) model. The model introduces topographic information as environmental data to enhance its capability to analyze the motion states of underwater gliders under diverse terrain conditions. Secondly, a Conditional Generative Adversarial Network (CGAN) is purposefully employed, utilizing DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clustering labels as conditional inputs for sample augmentation. This approach alleviates data scarcity while improving the diversity and rationality of the generated samples. Furthermore, an attention-weighted neighborhood enhancement module is designed to calculate the spatial and environmental similarity between target and neighborhood samples, for extracting more relevant input features. Finally, a Random Forest-based prediction model is constructed using these enhanced features as inputs. The validation results based on sea trial data from the South China Sea demonstrate that the proposed Terrain-CGAN-Att-RF model maintains superior performance even under small-sample conditions. For dead reckoning distance prediction, the Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Coefficient of Determination (R²) are 291.34 m, 401.88 m, and 0.9356, respectively. For heading prediction, the MAE, RMSE, and R² are 24.28°, 33.55°, and 0.8504, respectively.