Water Stress Weighted Metric for Sustainable Computing

This document is a research article that presents SCARF, a framework designed to evaluate the water impact of computing by incorporating spatial and temporal variations in water stress. The article outlines the increasing importance of water consumption in computing sustainability, particularly with the rise of AI workloads. It discusses the limitations of current water impact assessments, which often neglect the context of water stress. SCARF introduces an Adjusted Water Impact (AWI) metric that accounts for both the volume of water consumed and the local water stress over time. The framework follows a four-step process: modeling raw water consumption, mapping facilities to hydrological basins, aggregating temporal water stress, and calculating the AWI. The article includes case studies on large language model serving, datacenters, and semiconductor fabrication plants, demonstrating how location and timing can significantly influence water impact. The findings aim to promote water-sustainable computing practices by providing a comprehensive evaluation method for water consumption in various computing contexts.