Skip to main content

You are viewing a project that is currently in draft state for the Standards Working Group in the Green Software Foundation. This project should not be considered finished or officially supported in any way by the Green Software Foundation or its members.

Stress Characterisation — Practical Application

Stress Characterisation — Practical Application

The specification states that stress adjustment applies at the location of electricity generation, never at the computing facility. This page covers how to put that into practice.

Apply at generation, never at the compute facility​

This is worth stating plainly because it is easy to get backwards: the characterization factor for indirect operational water is drawn from where the power plant is, not from where the software's servers are. If a workload runs in one region but draws power that was generated elsewhere — common in interconnected grids — the stress factor follows the generation location. The operating location of the software is never a valid substitute, even when it would be more convenient to use.

This generalises across all three water components: direct operational water is characterised at the facility's own watershed (since that's where the water use actually happens), embodied water at the location of the relevant lifecycle process, and indirect operational water at the generation location as above.

Plant level, then aggregate — never the other way round​

Characterization factors are applied per power plant, using watershed-boundary coordinates to map each plant to the correct stress region. Aggregation up to a grid-region figure happens after stress adjustment, not before. Aggregating first would average away exactly the plant-to-plant variation — different watersheds, different hydro classifications (see Hydropower Classification) — that stress characterisation exists to capture.

Granularity: annual vs. monthly​

Power plant water consumption is not constant through the year — it varies with season and ambient temperature. WRI Aqueduct provides monthly stress data specifically to let implementers capture that variation rather than smoothing it away with an annual average.

Pending backtesting

Whether the difference between an annual and a monthly approach is large enough to matter in practice — and therefore whether monthly granularity should be a default recommendation rather than a nice-to-have — is the subject of a backtest committed to within the SSWG but not yet completed. Until that backtest concludes and a recommendation is documented here, treat monthly granularity as available and preferable where feasible, not as a settled requirement. See Open Methodological Questions for status.

Whatever temporal resolution is actually used, disclose it — the specification requires this explicitly, and it matters more than usual here given the pending question above.


Please submit any comments you have here.