Summary
- A peer-reviewed Newcastle University paper models data centre locations across 295 English local authority districts.
- Its combined population and water-resilience scenarios generally shift selected locations away from London and the South East.
- The model excludes actual site cooling demand, grid connections, land availability, and measured communications latency.
A new peer-reviewed study from Newcastle University has modelled the competing demands of population coverage and water resilience when selecting potential data centre locations across England. Published on 13 October in AQUA – Water Infrastructure, Ecosystems and Society, the research covers 295 local authority districts and identifies a marked shift towards northern areas when water stress is given greater weight.
Researchers Jacob Allen and Manuel Herrera developed a spatial optimisation model rather than assessing a single proposed facility. Their method combines demographic information with a district-level Water Resilience Index built from rainfall, groundwater productivity, and river-flow reliability. It then compares potential location sets to determine how population coverage changes as constraints relating to water are introduced.
The model identifies a trade-off between serving populous districts and favouring districts with more resilient water conditions. London and parts of the South East score well for population coverage but poorly under the water-resilience measures, creating a trade-off when the model is asked to optimise both objectives.
What the researchers actually modelled
The study uses integer linear programming to select hypothetical districts from a national set. The authors examined configurations with six to 15 major facilities, using ten as a reference case. These are analytical scenarios for future infrastructure siting, not an inventory of projects with planning consent or an assumption that ten specific data centres will be built.
The Water Resilience Index combines separate hydrological indicators after putting them on a common scale. The approach is designed to distinguish locations along a spectrum rather than treating a whole water company region as either suitable or unsuitable. It also allows a decision-maker to see how the preferred mix of locations changes when population reach or water stress is emphasised.
Across the baseline and future scenarios, the balanced configurations tend to shift away from the South East towards northern regions. The result is robust to alternative weightings of the underlying water indicators, according to the authors’ sensitivity testing. Such consistency provides useful evidence about how the model behaves, but it cannot validate every input or prove that the selected districts would be better sites under all real-world conditions.
The study also examines different future climate-pressure assumptions between 2050 and 2100. These are scenario-based stress adjustments rather than a detailed forecast of the water available to a named proposed facility in any particular year.
A screening model, not planning permission
Several material siting variables are absent. The model does not include actual grid connection capacity, substation constraints, site acquisition costs, the availability of appropriate land, or measured network latency. It also does not assign facility-specific water abstraction or cooling demand to individual data centre designs. Those are decisive practical considerations for real development decisions.
A liquid-cooled facility may use different heat rejection arrangements from an air-cooled installation, and water consumption can vary substantially by operating conditions and chosen design. A district-level water resilience score does not therefore establish that an individual developer would receive a water connection, or that a facility could operate without additional environmental constraints.
The population coverage measure is likewise an approximation. Data centre customers can be located across national and international markets, and low-latency applications may require specific fibre routes and interconnection availability, not simply proximity to large residential populations. The study acknowledges these simplifications explicitly.
Regional comparison using water-resilience indicators can inform early site screening, provided grid, fibre, permitting and cooling demand are investigated separately. An investor could combine a water-stress indicator with separate evidence on electricity, fibre, permitting, and projected cooling performance rather than treating the optimisation output as a final project ranking.
The three reference scenarios show how changing the weighting between population coverage and water resilience alters the selected district set. In the authors’ model, the maximum-coverage option reaches a greater share of England’s baseline population but accepts higher water stress, while the water-conservative option limits stress at the cost of lower population coverage. These percentages reflect the model’s chosen definition of coverage and hypothetical location set, not the proportion of real data centre traffic that could be served from those districts.
The authors call for more facility-level water consumption reporting and models that incorporate grid capacity, land, latency, and individual cooling requirements. Without those additional variables, their findings are evidence of a spatial trade-off rather than a recommendation to move an existing London data centre or cancel any particular planning application.
