Integration of hydrological models with entropy and multi-objective optimization based methods for designing specific needs streamflow monitoring networks

Integration of hydrological models with entropy and multi-objective optimization based methods for designing specific needs streamflow monitoring networks


Ursulak, J., & Coulibaly, P. (2021). Integration of hydrological models with entropy and multi-objective optimization based methods for designing specific needs streamflow monitoring networks. Journal of Hydrology, 593, 125876.

Abstract
Water resource managers depend on the collection of accurate hydrometric data for various modeling and planning projects. An essential use of hydrometric data includes hydrologic modelling and forecasting to support decision making in water resources planning and management. It is, therefore, essential to design hydrometric monitoring networks while considering the relationship between data collection and model application. A new model-based network design strategy is proposed that embeds hydrological models into a multi-objective evolutionary algorithm, facilitating direct optimization according to the model-based design objectives. This method is compared to the traditional model-based approach used to design hydrometric monitoring networks. The traditional approach is to first conduct optimization using secondary design objectives, that are not model based, to identify a set of optimal networks. Hydrological models are then applied as a post-processing mechanism to identify which of the optimal networks best satisfy the model orientated design objectives or users’ needs. In this investigation, the well-established dual-entropy multi-objective optimization (DEMO) approach was employed to conduct the initial network design based on the principles of information theory, followed by post-processing with rainfall-runoff models. Two case studies are evaluated, a monitoring network reduction in the Fraser River basin and a network augmentation in an upstream subsection of the Churchill River basin. Results show that embedding models in the optimization algorithm consistently yields better network configurations compared to those identified using the traditional method. It is shown that a smaller size optimal network that outperforms larger size networks can be identified directly by the proposed method. The models and model performance criteria used in the design process can be readily adapted, allowing for a user-directed design capable of addressing problem-specific objectives on a case by case basis.