Energy certificates overestimate gas use in Turin by up to 25-fold, metered data reveal

Energy performance certificates in Turin dramatically overstate how much natural gas buildings actually burn, according to a new city-scale comparison between certificate estimates and metered consumption records. In the least efficient class G buildings not connected to the district heating network, certificates put average gas use at 226.54 kilowatt-hours per square metre per year, while the fiscal registry of actual metered deliveries recorded just 10.00 kilowatt-hours per square metre per year for the same class. Across the entire non-district-heated stock, the gap between the theoretical and the measured figure was even starker: an overall certificate-based mean of 151.22 kilowatt-hours per square metre per year against a metered mean of only 10.41. The study, published in Energy Reports, is among the first to translate this discrepancy into empirical correction factors designed specifically for urban-scale energy modelling.

The research team, led by M.V. Airoma with S. Anselmo and M. Ferrara at the Polytechnic of Turin, set out to answer three questions: how large the discrepancy between certificate-derived and measured gas consumption is at city scale, how it varies with energy class, building use and district heating connection, and whether empirically derived correction factors can improve certificate-based estimates for urban building energy models. Their answer to the first question is unambiguous. Certificates systematically overestimate gas demand across every energy class examined, and the overestimation grows steadily worse as efficiency ratings decline. In class A buildings without district heating, the correction factor stood at 0.27, meaning metered consumption was roughly a quarter of the certified estimate. By class G, that factor had collapsed to 0.04, a twenty-five-fold overestimate.

To build the comparison, the researchers stitched together three datasets that are rarely joined at this scale. Three-dimensional building geometry came from the Municipal Technical Map of Turin, which catalogued 128,225 volumetric units, each a three-dimensional entity defined by a uniform footprint and height. Energy estimates came from the Piedmont Region’s certificate database, extracted from regional files covering energy data, heating plants and energy consumptions. Actual consumption came from SIATEL, a national fiscal registry holding electricity and gas records for active meters, which the Municipality of Turin released under a collaboration framework aimed at developing an Urban Digital Twin. The team matched certificates and meters through street addresses, then linked both to the volumetric units so that metered consumption could be paired with an energy class and a district heating status.

The matching demanded considerable data engineering. Certificate records reported gas in mixed physical units, so the team converted standard and normal cubic metres into kilowatt-hours using the higher heating value of 38.10 megajoules per standard cubic metre, yielding roughly 10.6 kilowatt-hours per cubic metre. Consumption was normalised to the climate of 2023, the year of the metered data, by applying the ratio between the actual heating degree days recorded in Turin that year, 2,648, and the standard value of 2,617 for the city’s climate zone E. Specific consumption was then calculated by dividing delivered gas by heated floor area. Records falling below the 5th or above the 95th percentile of specific consumption were excluded within each subset to blunt the effect of data-entry errors, a filter that proved essential: before filtering, the mean for class G buildings without district heating reached an implausible 846.43 kilowatt-hours per square metre per year, driven by severe recording mistakes, and settled at 226.54 afterwards.

The filtering cascade left 163,517 certificate records in the total dataset, alongside 13,290 records for certificates issued after renovation and 2,095 for certificates covering entire buildings. On the metered side, 73,465 volumetric units with SIATEL data were refined to 29,400 after the same statistical treatment, and 20,546 of those could be matched to at least one certificate. Because metered records carry no floor area, the team estimated heated floor area by multiplying each building’s footprint by its number of above-ground floors and applying a reduction factor of 0.75, the average ratio between heated useful floor area and gross floor area derived from the city’s certificates. Because metered records also carry no energy class, the researchers conservatively assigned each volumetric unit the worst class among its available certificates.

That conservative choice was tested. In a sensitivity analysis, the team instead assigned each unit the average class of its certificates, which redistributed the sample away from its heavy concentration in class G toward the intermediate categories, with most non-district-heated units falling into classes E and F. Yet the metered consumption profile barely moved. Mean specific gas consumption for units without district heating stayed between 9.96 and 11.12 kilowatt-hours per square metre per year regardless of class assignment, and medians for the district-heated subset ranged only from 2.59 to 4.36. The authors conclude that the striking absence of correlation between certificate ratings and actual gas use is not an artefact of their class-assignment method.

The shape of the metered data is as notable as its magnitude. Certificate estimates rise steadily and physically sensibly as efficiency worsens, from 38.95 kilowatt-hours per square metre per year in class A4 to 226.54 in class G among non-district-heated units in the total dataset. Metered consumption, by contrast, is almost flat: class A units without district heating averaged 15.58 kilowatt-hours per square metre per year, class G units 10.00, with the intermediate classes clustered between roughly 10 and 12. The researchers suggest several partial explanations. Certificates rely on standardised and default parameters that do not reflect the real physical, technological and morphological characteristics of individual buildings, while metered data capture actual operating conditions and occupant behaviour. They also raise the possibility of energy or heat rationing, documented in the literature as household consumption reductions under financial constraint, though they caution that their dataset contains no variables on income, deprivation, energy poverty or occupancy density, so this interpretation must be treated with care.

District heating complicates the picture in an instructive way. Turin’s network, which began operating in the late 1970s and is estimated to meet about 55 percent of the city’s building heat demand, supplies the primary thermal load for space heating, leaving connected buildings to draw gas mainly for domestic hot water and cooking. In the SIATEL data, 13,144 of the 29,400 filtered volumetric units, roughly 45 percent, were classified as district-heated, and their metered gas consumption was correspondingly tiny, with an overall mean of 4.64 kilowatt-hours per square metre per year. Certificate estimates for district-heated buildings were also far lower than for the rest of the stock, ranging only from about 29.51 to 37.94 across classes, so the gap between estimated and metered values is narrower there. The authors warn, however, that this narrower gap cannot be read as evidence that certificates are more accurate for these buildings, since gas represents only a residual carrier in them.

From these comparisons the team computed correction factors, defined as the ratio of metered to certificate-derived specific consumption, separately for mean-based and median-based metrics, for each energy class, building use and district heating status. For units without district heating in the total dataset, mean-based factors fell from 0.27 in class A to 0.044 in class G, with an overall factor of 0.119 across classes. Renovation certificates fared somewhat better, with factors from 0.301 in class A to 0.057 in class G, suggesting that certificates issued after energy efficiency interventions better reflect current building conditions. Median-based factors were generally lower and more stable, and for district-heated buildings the median approach proved considerably more homogeneous across datasets than the mean, which the authors attribute to the mean’s sensitivity to high-consumption outliers.

The factors were then stress-tested through a holdout validation: the total dataset was randomly split, with 80 percent used to recalculate the factors and 20 percent held back as an independent test set. Applied to the uncalibrated certificate estimates of the test set, the calibrated predictions tracked the metered values closely for the classes that dominate the building stock. For non-district-heated buildings in classes D through G, prediction errors stayed below 1 kilowatt-hour per square metre per year in absolute terms. Larger deviations appeared in thinly populated classes, including an underestimation of 10.886 kilowatt-hours per square metre per year for class A without district heating and an overestimation of 5.497 for district-heated class A, which the authors attribute to small sample sizes and the random composition of the test sample. The team stresses that the coefficients are empirical, Turin-specific calibration values rather than universally applicable parameters, and notes key limitations: metered consumption is aggregated at meter and address level while floor areas are estimated independently, district heating connections are inferred rather than directly observed, only the single year 2023 is covered, and certificates exist for just 22,551 of the 128,225 volumetric units, a sample that cannot be considered statistically representative. Even so, the work demonstrates a practical route for turning the well-documented energy performance gap into usable calibration coefficients for urban-scale energy modelling.

Subject of Research: Comparing EPC-estimated and metered natural gas consumption to derive correction factors for urban energy modelling in Turin

Article Title: Assessing EPC-derived gas consumption estimates using metered data: A case study in Turin

Article References: Airoma, M. V., Anselmo, S., & Ferrara, M. (2026). Assessing EPC-derived gas consumption estimates using metered data: A case study in Turin. Energy Reports, 16, Article 109810. https://doi.org/10.1016/j.egyr.2026.109810

Image Credits: AI Generated

DOI: 10.1016/j.egyr.2026.109810

Keywords: energy performance certificates, energy performance gap, natural gas consumption, urban building energy modelling, Turin, SIATEL metered data, district heating, correction factors, building energy efficiency, Piedmont EPC database, holdout validation, decarbonisation