A Transformer variant for multi-step forecasting of water level and hydrometeorological sensitivity analysis based on explainable artificial intelligence technology

AI transparency matters for critical infrastructure: The study demonstrates that Explainable AI (XAI) methods can reveal which factors most influence water level predictions, making AI decisions understandable to water management professionals rather than treating the model as a "black box."

Temperature is the key meteorological driver: The research found that temperature is the most dominant meteorological factor affecting water levels, highlighting that climate-related variables deserve priority attention in flood prevention and water management strategies.

Combining multiple data types improves predictions: The model achieved better forecasting accuracy by simultaneously considering both meteorological factors (like temperature and precipitation) and hydrological factors, suggesting that siloed data approaches may miss important interactions affecting flood risk.

Advanced AI can handle complex, nonlinear relationships: The Transformer variant successfully modeled the intricate, nonlinear processes connecting weather patterns to water levels across multiple prediction timeframes, demonstrating that sophisticated AI methods can capture real-world complexity better than traditional approaches.

XAI builds trust in AI-assisted decision-making: By explaining which factors drive specific predictions, XAI technology helps water management officials evaluate whether the AI's reasoning is sound and trustworthy for critical decisions like flood warnings and resource allocation.