High-Precision Surfacing Position Prediction for Underwater Gliders via Coordinate Transformation

The accurate prediction of the surfacing position of underwater gliders (UGs) is critical for mission success and cost-effective retrieval. However, current state-of-the-art (SOTA) methods often rely on complex multi-model integrations or large volumes of ocean current data, thereby increasing operational costs and system complexity. In this study, we systematically introduce—for the first time—a coordinate-transformation-based prediction framework, originally applied in other navigation contexts, into the UG surfacing-position-prediction task. By projecting both the glider’s entry and surfacing positions into a Universal Transverse Mercator (UTM) planar coordinate system and treating the resulting displacement as the prediction target, we avoid dependence on heavily parameterized current models, simplify the training process, and maintain robust predictive accuracy. Our approach combines common machine learning predictors (e.g., AdaBoost, LGBM, gradient boosting, random forest, decision trees) instead of advanced deep learning architectures, thus reducing computational overhead. Experiments on two real-world sea trial datasets (containing 2159 and 1456 profiles, respectively) show that, compared with direct regression approaches, this method improves positioning accuracy by up to 50% within a 500-meter range, yet requires minimal multi-source data. Overall, this study integrates the concept of coordinate transformation into the task of predicting the surfacing position of underwater gliders, effectively streamlining the method without sacrificing accuracy. The result is a highly flexible and cost-effective approach, providing theoretical support for future optimizations of underwater glider navigation systems.