Machine learning models for river flow forecasting in small catchments.

Area of Science:

• Environmental science
• Hydrology
• Data science

Background:

• Climate change necessitates advanced tools for hydrogeological risk mitigation.
• Small catchments are particularly vulnerable to rapid hydrological changes.
• Effective forecasting is crucial for timely risk management by local authorities.

Purpose of the Study:

• To investigate deep learning models for forecasting hydrometric height in rapid hydrographic basins.
• To assess the accuracy and lead time of these models for predicting extreme events.
• To evaluate the benefit of incorporating confidence intervals for improved prediction accuracy.

Main Methods:

• Application of deep learning models to forecast hydrometric height.
• Testing various input datasets to optimize model performance.
• Integration of machine learning models for prediction confidence intervals.
• Evaluation of model performance using metrics like Root Mean Square Error (RMSE).

Main Results:

• Deep learning models achieved very small errors (centimeters) with several hours of forecasting.
• Accurate prediction of extreme events with 4-6 hour lead times (RMSE 10-30 cm) was demonstrated.
• Incorporating confidence intervals improved prediction accuracy over longer forecasting horizons.
• Combining different models provided a more comprehensive outlook on river flow evolution.

Conclusions:

• Deep learning models are effective tools for forecasting river flows in fast-responding basins.
• These models enable objective and rapid application, facilitating the development of user-friendly software.
• The technology supports local authorities in making informed decisions for hydrogeological risk mitigation.