Global and regional models for identification of cooling technology in thermal power generation for water demand estimations in water-energy nexus studies

Water-energy nexus studies aim to connect the process of power generation with the corresponding water de-mand. The lack of information on the currently installed cooling systems at individual power plants is a challenge for the assessment of the water use on the power plant-level, which complicates decision-making in water management, especially in water-stressed regions. In this study, we investigate the spatial and temporal trends in cooling technology installations globally. Based on that, we propose a machine learning model for cooling technology identification on a regional and global level, which uses a combination of feature selection and classification algorithms. The global model demonstrates an average test set accuracy of 85.42%, which corre-sponds to only a minor underestimation of the actual global water footprint of 1.78% when the cooling tech-nology and water footprint of individual power plant units is unknown. Apart from that, a special emphasis was placed on regions characterized by high and extremely high water stress, where mistakes in water policy planning and water management may lead to an unsustainable water use or even to an overexploitation of water resources. In these regions, the calculated test set accuracy was 80.83%, which is significantly larger than the average accuracy of a majority class model. The results and the method proposed in this study enable cooling system identification in individual power units using information available from other sources, such as the water stress score or seasonal freshwater availability in the region.

Automated summary

The water-energy nexus is an important area of study that aims to understand the connection between power generation and water demand. This study focuses on the lack of information on cooling systems in power plants, which hinders the assessment of water use and decision-making in water management. The researchers propose a machine learning model that can identify cooling technologies globally, with an average accuracy of 85.42%. The model also performs well in water-stressed regions, where mistakes in water policy planning can have significant consequences. This study provides a valuable method for identifying cooling systems in power plants and highlights the importance of considering water stress in water management.