What is a weather-normalised baseline - and why does a credible saving need one?

A defensible baseline estimates what the building would likely have consumed under the weather and operating conditions that actually occurred - so the season does not take credit for your team’s work.

ENERGY PERFORMANCE · LAST REVIEWED 26 AUGUST 2026 · 6 MIN READ

Last year is not a baseline when the conditions changed.

A raw year-on-year comparison mixes operational performance with weather, occupancy, opening hours and building changes. It can reward a mild winter, punish a hot summer or hide a genuine improvement.

If heating energy falls by 8% while the winter is 10% milder, the building may not have improved at all. The reverse is equally damaging: a better-controlled building can use more energy during a severe winter and look worse in a simple budget comparison.

A baseline solves a narrower, more useful question: given the conditions in the reporting period, what would the building probably have consumed before the improvement?

Turn weather into a variable the model can use.

Heating and cooling degree days summarise how far outdoor temperature sat from an agreed base. They let the model compare periods by weather demand rather than by calendar label alone.

A heating degree-day calculation needs a stated base temperature, daily weather data and the building’s actual location. Cooling degree days apply the same logic above a cooling base. The source, base and missing-weather treatment must remain visible because each choice changes the result.

Degree days are not automatically the right driver for every load. Servers, lighting and hot water may behave as a base load; hotels may need occupied room-nights; factories may need production volume; offices may need operating hours or occupancy. Use only variables supported by reliable data and a defensible relationship.

Define the periods and assumptions before fitting the model.

A baseline is a documented model built from the building’s own metered history. Its purpose, meters, period, variables, exclusions and acceptance tests should be agreed before anyone sees the claimed saving.

  • Baseline period - the historical interval used to fit expected consumption.
  • Reporting period - the later interval in which performance is evaluated.
  • Relevant variables - weather, occupancy, production or operating conditions used by the model.
  • Static factors - material building changes that require a documented non-routine adjustment.
  • Data boundary - the exact meters, utilities and exclusions covered by the result.

A credible baseline is measured, tested and frozen before the claim.

A model becomes evidence only when its inputs trace to source readings, its fit is good enough for the stated use and its approved version cannot move after the improvement is known.

  • Measured - consumption comes from identified meters with visible data-quality states.
  • Tested - the quality of the fit is shown, not implied.
  • Frozen - the approved model, variables and period are dated before the reporting result is calculated.
  • Reviewable - exclusions, substitutions and non-routine adjustments keep their reason and approver.

Worked example: an 8,800 kWh drop is not an 8,800 kWh saving.

Consider a hypothetical office heating model with a 12,000 kWh monthly base load plus 95 kWh per heating degree day. The numbers below illustrate the method, not a client result.

  • Baseline January: 400 heating degree days; measured use 50,000 kWh.
  • Reporting January: 340 heating degree days; measured use 41,200 kWh.
  • Expected use at 340 degree days: 12,000 + (95 × 340) = 44,300 kWh.
  • Weather explains 5,700 kWh of the raw reduction; the modelled operational improvement is 3,100 kWh.
  • The defensible saving is about 7% of expected use - not the raw 17.6% year-on-year drop.

A real report must also show tariff treatment, model uncertainty, data exclusions and any change in occupancy, schedule or building use. The arithmetic is simple; the governance around it is what makes the result credible.

Reject the model when the evidence is not good enough.

A baseline should be allowed to fail. Publishing a confident saving from short history, weak fit or an unrecorded building change is worse than admitting that the current data cannot support the claim.

  • The history covers only one season or too few operating states.
  • The weather source is too distant or the base temperature is unexplained.
  • A large base load is incorrectly treated as fully weather-driven.
  • Occupancy, production, floor area or schedule changed without a documented adjustment.
  • One whole-building model is used to attribute several simultaneous measures to individual actions.

Keep the baseline beside the readings and actions it evaluates.

Volts can use the same authorised meter history for performance analysis, operational investigation and reporting, reducing the risk that separate spreadsheets produce separate versions of the saving.

Volts keeps the baseline, the readings behind it and the approval that fixed it on one record.

Software can maintain evidence; it cannot certify an energy-management system or guarantee an auditor’s acceptance. Certification remains the responsibility of the organisation and its auditor.

See energy-management workflows

Weather-normalised baselines without the black box.

How much metered history do I need?

Enough to represent the seasons and operating states the model must predict. Twelve months is a common starting point for whole-building weather models, but the acceptable minimum depends on interval, data quality and building behaviour.

Does this work for cooling as well as heating?

Yes. Cooling degree days or another temperature relationship can model cooling-sensitive use, provided the data shows a stable and testable connection.

Where do the degree days come from?

They are calculated from local weather data for the site.

Is a baseline the same as a benchmark?

No. A baseline compares a building with its own modelled past. A benchmark compares it with peers, a standard or another reference group.

What model-fit statistics should we inspect?

No single statistic proves validity; residual pattern, data coverage and intended use also matter.

What if the building changes after the baseline is frozen?

Record a non-routine adjustment for material changes such as floor area, tenant mix, production or operating schedule. Preserve the original model, reason, evidence, approver and effective date.

Can one baseline prove which retrofit created the saving?

Not automatically. A whole-building result can show total performance change, but attributing simultaneous measures requires separate action records, submeter evidence or a more specific M&V design.

Does a Volts baseline guarantee ISO 50001 acceptance?

No. Volts can provide the baseline and the evidence behind it. Certification depends on the organisation’s complete system, documented method and auditor.

Related reading

← Back to Learn

Build the baseline before you publish the saving.

Bring one site’s meter history, weather source and known operating changes. We will test whether the available evidence can support a credible performance period - and show the gaps when it cannot.