Our research program is organized around two complementary pillars: understanding how AI and digital infrastructure consume water, and applying AI to help utilities and watersheds use water more efficiently.
Quantifying the direct and indirect water footprint of AI compute and digital infrastructure growth — and linking it to basin-scale hydrological capacity.
Quantifying the direct and indirect water footprint of data center growth in Virginia's "Data Center Alley" and linking it to basin-level hydrological capacity.
Virginia's status as the global hub for data centers is straining local water, energy, and community resources, intensified by aging infrastructure and competing municipal demands. Currently, there is a lack of standardized water-use disclosure covering both direct cooling and indirect electricity generation — preventing regulators from assessing cumulative impacts or planning for sustainability.
This research develops a science-based, basin-scale framework to quantify industrial water demand and align it with hydrological capacity, supporting sustainable resource management in high-growth regions.
Water intensity varies by workload and technology, yet current planning overlooks indirect water use from power generation — which often exceeds direct cooling. Corporate "water positive" goals frequently fail to align with local hydrological realities. Because much of data center cooling water is lost to evaporation, facilities have a disproportionately high impact on consumptive water use relative to their share of total withdrawals.
Build a spatial inventory using satellite imagery and utility filings to identify cooling systems and electrical loads.
Model building footprints and energy-water intensity to estimate direct cooling and indirect power-generation withdrawals.
Use USGS and state records to evaluate municipal water system exposure to industrial demand.
Analyze local exposure to noise, water withdrawals, emissions, and infrastructure stress under current regulations.
Develop a GIS-based tool to visualize facility-level water footprints and watershed stress indicators.
Establish a metric for the gap between facility-level water demand and local basin sustainability thresholds.
Our approach utilizes a data-first platform containing verified records of nearly one billion gallons of water use. By employing established benchmarks like Power Usage Effectiveness (PUE) and Water Usage Effectiveness (WUE), alongside AI and cloud workload modeling, we translate facility scale into actual resource demand — linking siting decisions to hydrological vulnerability.
Prototype for visualizing basin-level water footprints and grid constraints.
Statewide frameworks for quantification and community impact assessment.
Findings shared through webinars and technical briefings with stakeholders.
FOIA-acquired utility records integrated into a basin-scale water-use database.
The decision-support platform from this research is available now at Data Center Water Leaks.
Applying machine learning and digital platforms to reduce losses and strengthen verification — from non-revenue water reduction in distribution networks to basin-scale water accounting for crediting and ESG reporting.
A research and deployment program applying machine learning to detect leaks, quantify losses, and strengthen operational decision-making in municipal water distribution systems.
Non-revenue water (NRW) — water lost to leaks, pipe breaks, metering inaccuracies, and unauthorized consumption — is one of the most persistent inefficiencies in municipal water systems. In water-stressed regions, every cubic meter lost to NRW is a cubic meter that cannot reach a household or business in a system already operating at a deficit, making NRW reduction a climate adaptation intervention as much as an operational one.
This research program develops a custom, end-to-end AI-NRW platform grounded in a utility's own sensor data, hydraulic characteristics, and stakeholder requirements — combining leak detection, District Metered Area (DMA) water balance accounting, and an operator-facing dashboard with a structured capacity-building and scale-up program.
The global market for AI in water is projected to grow rapidly across every region, yet field-scale deployment lags far behind investment. Lack of in-house AI expertise and uncertainty about financial payback remain the most-cited barriers to adoption. This research treats successful AI adoption as fundamentally a people problem as much as a technical one — building operator trust through visible, incremental wins rather than opaque automation.
Isolation Forest and LSTM autoencoder models trained on SCADA, AMI/AMR, acoustic, and pressure data to learn a network's normal signature and flag anomalies proactively.
Continuous sensor-based monitoring to detect quality anomalies that can mask or compound NRW signatures in the distribution system.
Predictive analytics on flow, pressure, and pump performance to optimize PRV settings and pump stations — typically the highest-ROI NRW interventions.
AI-assisted analysis of availability, usage patterns, and infrastructure performance to support allocation decisions under scarcity.
Generative models (TimeGAN, VAEs) create realistic training data where historical sensor coverage is incomplete — critical for smaller networks.
Natural-language interfaces allow operators to query sensor data, maintenance history, and model outputs conversationally — lowering the AI skills barrier.
NRW reduction is fundamentally an Operational Performance Management (OPM) challenge — optimizing the system as a whole rather than individual assets. The methodology follows a three-step cycle, continuously repeated as conditions change: