Abstract: Objectives To address the limitations of the IMO second-generation intact stability assessment method—particularly its failure to account for the spatiotemporal variability of real ocean environments and its limited ability to accurately evaluate ship instability risk—this study investigates a sensitivity index forecasting approach based on real-sea environmental conditions, aiming to enhance the navigational safety of container ships. Methods Combining the Second Generation Intact Stability Criteria with deep learning, we develop an accurate regional sensitivity index prediction model, and taking typical container vessels with recorded stability accidents as study cases, the predictive accuracy of the sensitivity indices was systematically validated. Results The proposed model significantly improves temporal and spatial resolution compared to the IMO method, enabling accurate identification of the specific location and timing of instability risks. The predicted results show high consistency with actual risk distributions and do not exhibit risk overestimation. Compared to the IMO method, the model better reflects the variability of stability risks under complex sea conditions. Conclusions The developed sensitivity index forecasting model fully accounts for the spatiotemporal variability of real ocean environments and enables rapid prediction of second-generation ship stability under realistic sea conditions. It improves both the accuracy and practical utility of risk forecasting, offering valuable decision-support information for container ship safety assurance and route planning. This research holds significant implications for safeguarding ships and cargo at sea.