Reinforcement Learning Applications in Water Resource Management: A Systematic Literature Review

Linus Kåge ^1* Vlatko Milic ^1,2 Maria Andersson ¹ Magnus Wallén ¹

• ¹ Linköping University, Linköping, Sweden
• ² University of Gävle, Gävle, Sweden

Climate change is increasingly affecting the water cycle. Droughts and floods pose significant challenges for agriculture, hydropower production, and urban water resource management, driven by growing variability in the factors influencing the water cycle.Reinforcement learning (RL) has demonstrated promising potential in optimization and planning tasks by training models on historical data or in simulations, allowing them to generate new data through interaction with the simulator. This systematic literature review examines the application of RL in water resource management across various domains. A total of forty papers were analyzed, revealing that RL is a viable approach for water resource management due to its ability to learn and optimize sequential decision-making processes. The results show that RL agents are predominantly trained in simulated environments rather than directly on historical data. Among the algorithms, deep Q-networks are the most commonly employed. Future research should address the challenges of bridging the gap between simulation and real-world applications and focus on improving the explainability of the decision-making process. Future studies need to address the challenges of bridging the gap between simulation and real-world application. Furthermore, future research should focus on the explainability behind the decision-making process of the agent, which is important due to the safety-critical nature of the application.