River ice breakup timing prediction through stacking multi-type model trees.

Area of Science:

• Hydrology and Environmental Engineering
• Climate Science and Meteorology

Background:

• River ice breakup events have significant ecological and economic impacts.
• Accurate forecasting of breakup timing is essential for managing river-ice related flooding and supporting emergency responses.

Purpose of the Study:

• To propose and evaluate a novel stacking ensemble tree model (SETM) framework for river ice breakup timing forecasting.
• To investigate the influence of temperature and water flow conditions on breakup timing.
• To compare the performance of SETM with individual Classification and Regression Tree (CART) and M5 models.

Main Methods:

• Developed a two-level stacking ensemble tree model (SETM) framework.
• Utilized historical data from the Athabasca River at Fort McMurray over 36 years.
• Employed the leave-one-out cross-validation method for model performance assessment.

Main Results:

• Key indicators influencing breakup timing include temperature and water flow conditions during pre-breakup, freeze-up, and mid-winter periods.
• The M5 model simplifies tree structure compared to CART, with similar optimal performance.
• The SETM framework, particularly using a simple average ensemble, improved prediction accuracy over base CART and M5 models by over 13%.

Conclusions:

• The SETM framework offers improved accuracy for river ice breakup forecasting.
• Base models within SETM provide interpretable relationships between environmental indicators and breakup timing.
• Ensemble methods, like simple averaging, enhance predictive performance in river ice forecasting.