Global loss in global hydropower supply under droughts using a hybrid model

Hydropower is considered as an important source of renewable energy due to its flexibility and storage capabilities. However, hydropower faces significant challenges with climate change and especially the increasing risks of extreme weather events such as droughts.

In this study, we analysed the impact of historical droughts on hydropower at a global scale by developing a hybrid model that combines a physically based hydropower model with a machine learning model. This integrated approach enables us to capture important features affecting hydropower generation beyond water availability, considering the details of local specific conditions at hydropower plant sites while it can be applied across the globe. A new open-source global dataset is developed that contains key information of the hydropower plant characteristics and their reservoir attributes by merging various plant sources with a global reservoir database. The hybrid model is trained against observed monthly hydropower generation data at the power plant level. By employing this approach, we aim not only to enhance the realism of simulating hydropower output compared to the simplistic physically based equation but also to leverage the flexibility of machine learning. Additionally, this method enables us to circumvent detailed power system modelling which requires significant computing power and extensive data.

We found that the performance of our hybrid hydropower model surpasses the simple physics-based hydropower equation at most hydropower plant sites. Key findings highlight the significant losses of hydropower generation during major historical drought events across the globe.