Ensemble machine learning paradigms in hydrology: A review

Recently, there has been a notable tendency towards employing ensemble learning methodologies in assorted areas of engineering, such as hydrology, for simulation and prediction purposes. The diversity of ensemble techniques available for implementation in hydrological sciences has led to the development and utilization of different strategies in the implementation. This review paper explores and refers to the advancement of ensemble methods, including the resampling ensemble methods (e.g., bagging, boosting, and dagging), model averaging, and stacking viz. generalized stacked, in different application fields of hydrology. The main hydrological topics in this review study cover subjects such as surface hydrology, river water quality, rainfall-runoff, debris flow, river icing, sediment transport, groundwater, flooding, and drought modeling and forecasting. The general findings of this survey demonstrate the absolute superiority of using ensemble strategies over the regular (individual) model learning in hydrology. In addition, the boosting techniques (e.g., boosting, AdaBoost, and extreme gradient boosting) have been more frequent and successfully implemented in hydrological problems than the bagging, stacking, and dagging approaches.

Automated summary

There is a growing trend in employing ensemble learning methodologies in various engineering fields, including hydrology, for simulation and prediction purposes. The diversity of ensemble techniques available in hydrological sciences has led to the development and utilization of different strategies. The general findings suggest the superiority of using ensemble strategies over traditional model learning in hydrology, with boosting techniques being more commonly and successfully implemented compared to other methods.