Artificial Intelligence for Water Consumption Assessment: State of the Art Review

Authors: MORAIN, ALMANDO - Florida A & M University; Delhom, Christopher; ANANDHI, AAVUDAI - Florida A & M University

Submitted to: Water Resources Management
Publication Type: Literature Review
Publication Acceptance Date: 3/12/2024
Publication Date: 4/25/2024
Citation: Morain, A., Delhom, C.D., Anandhi, A. 2024. Artificial Intelligence for Water Consumption Assessment: State of the Art Review. Water Resources Management. https://doi.org/10.1007/s11269-024-03823-x.
DOI: https://doi.org/10.1007/s11269-024-03823-x

Interpretive Summary: Increased demand for freshwater resources has increased the risk of severe water stress. Many researchers have turned to artificial intelligence (AI) as an alternative to linear methods to assess water consumption. This study contributes to the existing literature by focusing on the involvement of AI in understanding water consumption from various viewpoints, such as innovation, the application sector, sustainability, and machine learning applications. The study provides valuable insights into both standalone and hybrid AI models used for estimating water consumption, understanding the process of assessing AI model performance, highlighting the advantages, disadvantages, and challenges associated with certain AI models, and identifying research needs and knowledge gaps. This review considered 219 publications. This study highlights the various benefits of standalone and hybrid AI models, such as time-saving abilities, accuracy, convenience, flexibility, and capacity to handle complex systems and large amounts of data. However, challenges related to reproducibility, method standardization, data availability, uncertainty, and privacy remain. Another challenge is choosing an appropriate model with high performance for forecasting water consumption. There is no one-size-fits-all AI model, and this study suggests utilizing hybrid AI models. These models offer flexibility in terms of efficiency, accuracy, interpretability, adaptability, and data requirements. They can address the limitations of individual models, leverage the strengths of different approaches, and provide a better understanding of the relationships between variables.

Technical Abstract: In recent decades, an increase in the demand for freshwater resources has increased the risk of severe water stress. With the growing prevalence of artificial intelligence (AI) in various fields, many researchers have turned to it as an alternative to linear methods to assess water consumption. This study contributes to the existing literature by focusing on the involvement of AI in understanding water consumption from various perspectives, such as innovation, the application sector, sustainability, and machine learning applications. It also provides valuable insights into the standalone and combined/hybrid AI models used for estimating and forecasting water consumption, understanding the process of assessing AI model performance, highlighting the advantages, disadvantages, and challenges associated with certain AI models, and identifying research needs and knowledge gaps associated with AI applications in water consumption studies. Using the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) framework, this systematic review utilized 190 screened publications, and 29 other publications were identified through snowball sampling. This study highlights the various benefits of standalone and combined/hybrid AI models, such as their time-saving abilities, accurate estimates and forecasts, convenience and flexibility, and capacity to handle complex systems and vast amounts of data. However, challenges related to reproducibility, method standardization, data availability, uncertainty, and privacy remain unresolved. Another significant challenge is choosing an appropriate model with high performance for estimating and forecasting water consumption. As there is no one-size-fits-all AI model, this study suggests utilizing hybrid AI models as alternatives. These models offer flexibility in terms of their efficiency, accuracy, interpretability, adaptability, and data requirements. They can address the limitations of individual models, leverage the strengths of different approaches, and provide a better understanding of the relationships between variables.