CWAP‐Net: A Causal Weather‐Aware Spatiotemporal Graph Framework for Short‐Term Airport Stand Capacity Prediction

Accurate short-term forecasting of airport stand capacity is essential for resilient ground operations, adaptive gate coordination, and delay mitigation under unstable weather conditions. CWAP-Net is introduced as a domain-informed spatiotemporal framework that fuses a stand-topology spatial graph, a causally structured weather graph derived from METAR/TAF bulletins, and a multi-scale ActionFormer temporal encoder to jointly capture spatial constraints, weather-driven causal effects, and heterogeneous temporal dependencies. Model training is guided by a decision-focused value function that incorporates lead-time gain, severity weighting, soft-IoU coverage, and penalties for false alarms and missed detections, ensuring alignment with operational priorities. The model is assessed through a case study at Qingdao Jiaodong International Airport from January to March 2024 using multi-source operational data. Results show that CWAP-Net surpasses sequence and graph baselines in RMSE, MAE, and R². Robustness tests extend coverage to 2023–2024 across eight quarters and confirm the generality of each component. The framework enables interpretable forecasting under uncertainty and offers a deployable path toward intelligent airport surface management.