Comparison of reservoir routines for large-scale hydrological models

A correct representation of reservoirs in a large-scale hydrological model is crucial for simulating reliable streamflow, particularly in strongly managed catchments. The challenge lies in developing a simple, universally applicable routine for reservoirs worldwide, regardless of their size, purpose or climate. However, a thorough analysis of reservoir routines is hindered by the limited availability of in situ observations.

To address this limitation, we created a harmonised dataset of reservoir operations —including inflow, storage, and release— using observations in the US (Steyaert et al., 2022), Mexico, Brazil and Spain. The dataset also includes meteorological time series and static attributes —such as reservoir and dam characteristics, water use, and climate indices— that can be used to train data-driven models or regionalise model parameters.

Using that dataset, we compared five reservoir routines from the literature: a simple linear reservoir, the routines implemented in the hydrological models LISFLOOD-OS (Burek et al., 2013), CaMa-Flood (Hanazaki et al., 2022), and mHM (Shrestha et al., 2024), as well as the Starfit routine (Turner et al., 2021). Our comparison analysed the ability of these routines to model both reservoir storage and release, as well as their potential for implementation in continental or global operational systems.

Our results indicate that the Hanazaki routine strikes the best balance between storage-release performance and complexity, as it minimises the number of parameters to be calibrated and has limited data requirements. Consequently, we have implemented the Hanazaki reservoir routine in the LISFLOOD-OS v5 model, which will be used in future versions of both the European Flood Awareness System (EFAS) and Global Flood Awareness System (GloFAS).