Abstract
This study aims to develop an advanced deep learning model, Hydro-Informer, for accurate water level and flood predictions, emphasizing extreme event forecasting. Utilizing a comprehensive dataset from the Slovak Hydrometeorological Institute SHMI (2008-2020), which includes precipitation, water level, and discharge data, the model was trained using a ladder technique with a custom loss function to enhance focus on extreme values. The architecture integrates Recurrent and Convolutional Neural Networks (RNN, CNN), and Multi-Head Attention layers. Hydro-Informer achieved significant performance, with a Coefficient of Determination (R²) of 0.88, effectively predicting extreme water levels 12 hours in advance in a river environment free from human regulation and structures. These results demonstrate the model's robustness in identifying extreme events with minimal underestimation, essential for flood management and disaster preparedness. The study underscores the model's potential to enhance early warning systems and support timely evacuation and infrastructure planning, thereby mitigating flood impacts. Future research should explore integrating additional data sources and further refining the model to improve prediction accuracy and reliability. This work highlights the significant role of advanced deep-learning techniques in hydrological forecasting and practical applications in flood management.