Direct Operational Water
Direct Operational Water
In short: direct operational water (W_O^dir) is water used directly
by the facility in which the software executes — cooling towers,
humidification, and similar facility-level operations. Unlike indirect
operational water, it doesn't depend on where the electricity came from;
it depends on how the facility itself is cooled. It scales with how much
electricity your workload used, multiplied by how thirsty that particular
data centre is per unit of electricity — the same shape as "miles driven ×
gallons per mile = gallons used," just water and electricity instead of
fuel and distance.
Shape: energy used × water-per-unit-of-energy (the facility's WUE) = direct operational water.
Worked: 300 kWh × 0.5 L/kWh = 150 L. (illustrative placeholder, not a real measurement)
Measurement and allocation
The specification's preference order is straightforward: use software- or workload-level measurements where they exist; where they don't, allocate facility-level water to the software system using a documented, disclosed allocation method. There is no single mandated allocation method — what matters is that whatever method is used is consistent and disclosed.
Estimating with WUE
Water Usage Effectiveness (WUE) — a facility's water consumption per unit of energy, in L/kWh — is one option for estimating direct operational water consumption, not a required input and not the only valid method. This is the precise version of the formula already introduced above:
W_O,c^dir = E × WUE
where E is the energy the software system consumed, in kWh. If a
facility reports WUE, this is usually the easiest available estimate. If
it doesn't, or if better workload-level data exists, use that instead.
The withdrawal data gap
Here is the limitation to keep in view: WUE characterizes consumption, not withdrawal. There's currently no agreed way to do the same sum for withdrawal — that's a real gap, not something this page is glossing over. No widely adopted facility-level metric equivalent to WUE exists yet for allocating direct operational withdrawal to a software workload. If a facility measures and discloses its own withdrawal, that measured data can be used, or an organization-specific allocation method can be applied where one is available and documented — but absent one of those, direct operational withdrawal simply cannot be quantified with the same rigor as consumption today.
This is a known, structural gap in current methodology, not something this
guidance can resolve by picking a workaround. Where an assessment cannot
quantify direct operational withdrawal, the specification requires that
omission — and what it means for the reported withdrawal results — to be
disclosed rather than silently left out. Reported SWI_w results should be
read with this gap in mind: an unquantified direct operational withdrawal
component is not the same thing as a zero one.
Please submit any comments you have here.