Analysing Natural Gas Consumption with Heating Degree Days (HDD)

When the natural gas bill rises from one month to the next, the first question is always the same: did the building run less efficiently, or was the weather simply colder? A total cubic metre or kWh figure cannot answer that, because heating demand moves directly with outdoor temperature. Degree-day analysis strips the weather effect out of consumption so that months, years and buildings can be compared on the same footing. This guide walks energy managers and engineering teams through the HDD calculation, base temperature selection, the regression model and the data problems that show up in practice.

What Is a Heating Degree Day (HDD)?

A heating degree day is a simple indicator of how much heating demand a day creates. When the daily mean outdoor temperature falls below the chosen base temperature, the difference is that day's HDD value; if the mean temperature is equal to or above the base, HDD is zero:

HDD = max(0, T_base − T_mean)

Monthly or seasonal HDD is the sum of the daily values. A cold month produces a high HDD total and a mild month a low one. Cooling works the same way in reverse: cooling degree days (CDD) count how far the mean temperature rises above a cooling base. The daily mean is usually taken either as the average of hourly readings or as the average of the daily minimum and maximum; whichever method you choose, use it consistently across every period you compare.

How to Choose the Base Temperature

The base temperature represents the outdoor temperature at which the building starts to need heat. The common convention is 18 °C, but that is a starting point rather than a law of nature. In a well-insulated building with high internal gains from lighting, occupants and equipment, heating starts at a lower outdoor temperature; in an older, uninsulated structure it starts earlier. This threshold is known as the building's balance point.

The practical way to find a suitable base is to generate HDD series for several candidate base temperatures, regress each one against consumption and keep the one with the highest explanatory power (R²). Once chosen, the base temperature should stay fixed; changing it mid-year makes historical comparisons meaningless.

Consumption = a + b·HDD: The Regression Model

At its core, degree-day analysis is a simple linear regression. Daily or monthly consumption is modelled against HDD for the same period using ordinary least squares (OLS):

Consumption = a + b · HDD

Both coefficients carry concrete operational meaning. The intercept (a) is the weather-independent load: domestic hot water, kitchens, laundry or process steam that runs all year. The slope (b) is the extra consumption added by each additional degree day, and it describes the building's heating sensitivity. Because HDD approaches zero in summer, summer consumption anchors the intercept while the winter months define the slope.

A model built on daily data contains more points and is therefore more stable; a model built on monthly bills rests on few points, and a single faulty month can noticeably skew the slope. Days that look abnormal because of breakdowns, maintenance or data gaps should be removed before fitting.

Comparing Years with Weather-Normalized Consumption

Once the model is in place, its real value appears in comparison. If this winter was colder than last winter, a rise in raw consumption does not by itself mean a loss of performance. Normalization brings both periods to the same weather conditions: either each period is expressed as consumption per degree day (for example kWh/HDD), or the model coefficients are used to answer the question “what would consumption have been if this year's weather had occurred last year?”

To compare buildings of different sizes, floor area enters the equation as well. Dividing total heating-season consumption by the heated area and by the season's HDD total produces an intensity indicator cleaned of both climate and size effects. It shows a portfolio manager which building stands out on heating; buildings with missing weather or consumption days, however, should be left out of the ranking.

Another practical use is decomposing cost changes. The bill difference between two periods can be split into three parts: the unit price change, the weather effect (slope × HDD difference) and the remaining usage effect. Management then gets the answer to “why did the bill go up?” broken into its components instead of a single figure.

Verifying Results After an Efficiency Measure

Whether the measure is a boiler replacement, burner tuning, hydronic balancing, weather-compensated control or facade insulation, the degree-day model acts as a baseline for judging whether it actually worked. The model is fitted on the pre-measure period; after the measure, the observed HDD is fed into that model to calculate the consumption that would have been expected without the change. The difference between expected and actual consumption is the weather-adjusted result.

This approach is consistent with the energy baseline and energy performance indicator concepts in ISO 50001 and with the logic of the International Performance Measurement and Verification Protocol (IPMVP). For verification to be credible, the pre-measure period should cover at least one full heating season, and operating conditions in both periods (working hours, occupancy, setpoints) should be documented.

Data Problems and Common Mistakes

• Billing period drift: The distributor's meter reading dates rarely coincide with calendar months. If HDD is summed by calendar month while consumption follows the reading period, the model breaks; align HDD with the days between reading dates.
• Mixing Sm³, m³ and kWh: Measured volume (m³) and standard volume corrected for pressure and temperature (Sm³) are not the same thing, and the calorific value used for kWh conversion can change between periods. Keep the series in one unit with a consistent conversion.
• Gas day boundaries: Distributor data may be aggregated by a gas day that closes at a fixed morning hour rather than at midnight. Account for this offset when matching daily consumption with weather data.
• Estimated readings: When the meter cannot be read, the bill may carry an estimated value that is corrected at the next actual reading, distorting two months at once.
• Mixed end uses: If heating, hot water and kitchens share one meter, the intercept grows and the slope weakens. Tracking the summer base load separately is worthwhile in that case.
• Weather data source: A distant weather station may not represent real conditions, especially in coastal or elevated locations. Use location-based weather data and flag missing days before filling them.
• Operational changes: A new wing, changing occupancy or longer operating hours sit outside the model; comparisons made without noting them will mislead.

Degree-Day Analysis in Argus EMS

Argus EMS does not read a physical Modbus gas meter. In projects with access to the İGDAŞ AVE customer portal, the distributor's main meter measurements are pulled hourly and daily and written to the project's virtual gas meter; where that access is not available, gas is tracked through invoice entry. On the weather side, daily temperatures for each project location come from the Open-Meteo archive, and HDD is calculated with a default heating base of 18 °C, while CDD uses a 22 °C cooling base.

The energy-weather sensitivity model solves the last year of daily consumption as base + slope·HDD + slope·CDD using least squares, stores the R² value, excludes outlier days and clips physically meaningless negative slopes to zero. The portfolio gas analysis provides a monthly consumption-per-HDD slope, a cost bridge that splits changes into price, weather and usage components, the summer base load and a heating-season ranking normalized per unit area and per HDD. In the month-end gas projection the weather term is applied only once the sensitivity quality gates are passed, and on the budget side no speculative price escalation is added to future months: the last known unit price is carried forward flat.