Energy Lens

Energy management made easy

Degree days are a simplified form of historical weather data. They are commonly used in energy monitoring and targeting to model the relationship between energy consumption and outside air temperature.

**Weather normalization of energy consumption** is one of the most common such uses of degree days. In theory, weather normalization (or "*weather correction*") enables a like-for-like comparison of energy consumption from different periods, or from different places with different weather conditions.

At the risk of stating the obvious, weather normalization is only appropriate for energy consumption that is affected by the weather, which mainly means building heating and cooling, and refrigeration. Since a lot of energy usage (like that of most electrical equipment) is *not* weather dependent, there is more to energy-data analysis than the weather-related techniques described in this article. However, since heating/cooling energy consumption is substantial in most buildings, its analysis is undoubtedly important.

Weather-normalization techniques are often based around *regression analysis of past energy-consumption data*, a method that is frequently used with degree days to:

**Calculate or prove energy savings**after the implementation of specific measures like installing new insulation.**Monitor ongoing energy usage**for signs of waste (excess consumption or "*overspend*") and to track ongoing progress at reducing it. This ongoing*monitoring and targeting*(or "*M&T*") involves comparisons of recent consumption with a past-performance-based estimate of*expected*consumption (usually based on a regression model). It is sometimes done with the help of CUSUM analysis – a charting technique that makes it easier to spot deviations from typical behaviour.

Although much of the theory behind these methods is sound, bad implementation and poor understanding often lead to results that are less accurate than many people realize. Degree-day analysis is central to many energy management programmes, and, consequently, important decisions are frequently based on figures that can be misleading to those who put too much faith in them.

This article looks at the problems with the ways in which degree days are commonly used, and suggests a few ways these problems can be avoided or mitigated. Its aim is not to denigrate the degree-day-based methods that are widely used in energy management, but to highlight the major sources of inaccuracy, suggest improvements, and encourage the reader to consider the validity of their results before using them as a foundation for decision making.

The underlying issue is that heating/cooling energy consumption of buildings is complicated. Degree days are invaluable for providing a relatively simple way to deal with this complexity, they are much better than temperature data in most cases, and much more accessible than full building simulation, but still they need to be handled with care. This article explains why, and how. We hope you find it useful!

If you are looking for degree-day data (rather than an article about it), you might want to head over to another of our websites: **Degree Days.net**.

First, this article looks at the basics of degree-day theory (you might want to skip this if you already know the basics):

- Weather, energy consumption, and weather-normalization basics
- Introducing degree days – historical weather data made easy
- Common methods for using degree days in energy monitoring and targeting

Next, this article highlights problems with the ways in which degree days are commonly used, and then suggests ways in which those problems can be avoided or mitigated:

- Problems with common degree-day-based methods
- Suggestions for improvement: using degree days wisely
- Prefer regression-based methods to simple ratio-based methods
- Use the best energy data you can
- Use the most appropriate degree-day data you can
- Prefer a yearly timescale for comparisons of weather-normalized data
- Be particularly sceptical of normalized figures for periods with an ideal outside temperature
- Remember the level of accuracy
- Look at proportional differences before looking at absolute numbers

- Conclusions

In heated or cooled buildings, energy consumption tends to depend on the outside air temperature:

- The colder the outside air temperature, the more energy it takes to heat a building to a comfortable temperature.
- The warmer the outside air temperature, the more energy it takes to cool an air-conditioned building to a comfortable temperature.

"Weather normalization", or "weather correction", allows you to *adjust* your energy-consumption figures to factor out the variations in outside air temperature. In theory, you can then compare the normalized figures fairly.

Weather normalization is commonly used when analyzing changes in a building's energy consumption, and, when combined with other normalization techniques (such as normalizing for occupancy and building size), when comparing the energy consumption of different buildings.

Let's say you have several years' worth of energy-consumption data for a building, and you want to compare the energy consumption in 2019 with that in 2020, to see if there was an improvement in energy efficiency.

Year | Total energy consumption (kWh) |

2019 | 175,441 |

2020 | 164,312 |

The raw figures show that the building used less energy in 2020 than it did in 2019. However, let's say you know that a large proportion of the building's energy consumption went on heating, and you also know that 2020 was a warmer year than 2019. You would therefore *expect* less energy to have been used in 2020 than in 2019, as the warmer outside temperatures in 2020 meant that less energy was needed to heat the building.

So, less energy was used in 2020 than in 2019, but did energy efficiency improve, or was it just because 2020 had warmer weather?

This is where weather normalization can help. Provided you have the appropriate historical weather data (most probably *heating degree days*), you can calculate the *weather-normalized* energy consumption in 2020, and the *weather-normalized* energy consumption in 2019. These two figures can then, in theory, be compared on a like-for-like basis, enabling you to see whether or not there was an improvement in the building's energy efficiency.

Degree days are essentially a simplification of historical weather data – outside-air-temperature data to be specific. Degree-day data is easy to get hold of, and easy to work with. This makes degree days popular amongst energy consultants and energy managers, certainly when compared with other forms of past weather data such as hourly temperature readings.

Degree days can come in all sorts of breakdowns, but daily, weekly, and monthly figures are common. You can sum daily figures together to make figures that match whatever periods of energy consumption you are dealing with. For example, weekly and monthly figures are simply the sum of the daily figures for the days within them, and you can sum daily or monthly figures to make quarterly or yearly figures (if you are working with quarterly or yearly energy consumption).

There are two main types of degree days: heating degree days (HDD) and cooling degree days (CDD). Both types can be Celsius based or Fahrenheit based.

Heating degree days (HDD) are used for calculations that relate to the heating of buildings. For example, HDD can be used to normalize the energy consumption of buildings with central heating.

Heating-degree-day figures come with a "*base temperature*", and provide a measure of how much (in degrees), and for how long (in days), the outside temperature was below that base temperature. In the UK, heating degree days have traditionally come with a base temperature of 15.5°C; in the US, it's 65°F. Though nowadays it is easy to get HDD in any base temperature.

An example calculation: if the outside temperature was 2 degrees below the base temperature for 2 days, there would be a total of 4 heating degree days over that period (2 degrees * 2 days = 4 degree days). In reality, the process of calculating accurate degree days is complicated^{*} by the fact that outside temperatures vary throughout each day. You are unlikely to need to calculate degree days yourself, but do try to get them from a source that properly takes these within-day temperature variations into account.

^{*}*We understand these complexities well from our work on Degree Days.net, which calculates degree days from detailed temperature readings taken throughout each day. This gives the most accurate degree days but it does require a lot of data processing. So we can appreciate why many other sources are still providing approximate degree days estimated from simple daily max/min temperatures, even though we would love for them to modernize their systems!*

Cooling degree days (CDD) are used for calculations relating to the cooling of buildings. For example, CDD can be used to normalize the energy consumption of buildings with air conditioning.

Cooling-degree-day figures also come with a base temperature, and provide a measure of how much, and for how long, the outside temperature was *above* that base temperature.

Although this article talks mainly about heating degree days, much of the information is also applicable to calculations involving cooling degree days.

Celsius-based degree days are calculated using a base temperature measured in Celsius, and outside temperatures measured in Celsius.

In contrast, Fahrenheit-based degree days are calculated using a base temperature measured in Fahrenheit, and outside temperatures measured in Fahrenheit. Fahrenheit-based degree days are common in the US, where they typically come with a base temperature of 65°F (equivalent to 18.3°C). Since a temperature difference of 1°C is equivalent to a temperature difference of 1.8°F, Fahrenheit-based degree days are 1.8 times bigger than their equivalent Celsius-based degree days.

The theory and arguments presented in this article are equally applicable to both Celsius-based and Fahrenheit-based figures and calculations.

Degree days are commonly used in monitoring and targeting of energy consumption. As an understanding of the popular methods is necessary for an understanding of the bulk of this article, they are briefly explained below:

Heating degree days are often used to *normalize* the energy consumption of a heated building so that, in theory, the normalized figures can be compared on a like-for-like basis. So, for the example given above, heating degree days would enable you to calculate *normalized* energy-consumption figures for 2019 and 2020 that, in theory, could be compared fairly.

The simplest way to normalize energy-consumption figures is to calculate the *kWh per degree day* for each kWh energy-consumption figure in question. Simply divide each kWh figure by the number of degree days in the period over which that energy was used. In theory, dividing by the degree days factors out the effect of outside air temperature, so you can compare the resulting *kWh-per-degree-day* figures fairly.

**Warning**: although this ratio-based method is simple, and commonly used, there are several issues that prevent it from working well in many circumstances. Please do read this full article before using it yourself.

The following figures continue the example explained above, using real heating degree days for Gatwick airport in the UK, with a base temperature of 15.5°C:

Year | Total energy consumption (kWh) | Total heating degree days | kWh per degree day | Normalized kWh |

2019 | 175,441 | 1,889 | 92.8751 | 176,184 |

2020 | 164,312 | 1,742 | 94.3238 | 178,932 |

From the table above you can see that the *kWh-per-degree-day* values increased from 2019 to 2020, and you can calculate this increase as a percentage: +1.56%. This suggests that, after the weather has been accounted for, 2020 used 1.56% more energy than 2019. In other words, energy efficiency was actually slightly *worse* in 2020 than in 2019. That's the theory anyway!

The rightmost column of the table has *normalized-kWh* figures that represent the weather-normalized energy usage for each year. For many people these *normalized-kWh* figures make more intuitive sense than the *kWh-per-degree-day* figures, though they show the same key thing: that weather-corrected energy usage increased by 1.56% from 2019 to 2020.

To calculate the *normalized-kWh* figures, we multiply the *kWh-per-degree-day* figures by a constant number of degree days. As we're dealing with yearly figures in this example, we use the *average-year* heating degree days. These represent the HDD in a typical year. We use the same weather station (Gatwick airport), and the same base temperature (15.5°C), and we use a 10-year average covering 2011 to 2020 i.e. the average of the yearly HDD values for those 10 years. This translates to a value of 1,897, so in this case we multiply the *kWh-per-degree-day* figures by 1,897 to get the *normalized-kWh* figures.

NB Some people use 5-year-average degree days as the multiplier, some use 10- or 20-year-average degree days, and some might use "*standard degree days*" (to normalize figures in such a way that they can be compared with other similarly-normalized figures for different buildings in different locations). You could also use the 2019 degree days as the multiplier, to leave the 2019 energy usage unchanged and to normalize the 2020 energy usage to 2019 weather conditions. Provided you use just one multiplier (e.g. do not use rolling averages that would be different for each period) it should not matter much what multiplier you use, as your figures will at least be proportionally comparable.

Regression analysis is used widely in energy monitoring and targeting. Central to this is the assumption that energy consumption is caused by a "*driving factor*" (or "*driver*") – this could be the widgets produced by a production line, or, in the more common case of heating or cooling, the degree days. So, for a heated building, it is assumed that the energy consumption required to heat that building for any particular period is proportional to (or *driven by*) the number of heating degree days over that period.

Typically you would select a "*baseline*" set of energy-consumption data: this would usually be weekly or monthly data from the past year or two. For each figure of energy consumption, you need a corresponding figure for the degree days (or whatever driving factor you are using) – you would typically then plot these two sets of figures on a scatter chart.

For example, the scatter plot below shows a years' worth of monthly kWh (y-axis) plotted against monthly degree days (x-axis). Specifically, the chart uses 65°F-base-temperature heating-degree-day figures for San Francisco airport in 2020, and shows a very good correlation (an R^{2} close to 1):

The "*regression line*" is the line of best fit through the points in the scatter chart. It is often known as the "*trend line*" or the "*performance characteristic line*". Degree days are expected to show a linear relationship with energy consumption, so we want a straight line, not a curve.

The "*regression equation*" is the equation that describes the regression line. Our article on regression analysis of energy consumption and degree days in Excel explains this in much more detail, though we suggest you read the rest of this article first.

With the regression equation you can calculate the *baseline*, or *expected*, energy consumption from the degree days. So, each time you obtain a new figure for the degree days (typically each week or month), you can plug it into the regression equation to get the expected energy consumption. You can compare this figure with the actual energy consumption for the period, to determine whether more or less energy was used than expected.

**Warning**: regression analysis is an important technique, but it needs to be applied carefully to get good results. Please do read this full article before implementing it yourself.

There are a number of other techniques that revolve around degree-day regression analysis like that described here. The proximity of the points to the regression line is often used as an indication of the accuracy of the heating control (the greater the scatter, the worse the control), and it is common for people to plot a CUSUM chart of the difference between actual and expected consumption. Estimating "*baseload*" energy consumption is another typical application:

It is very common for a single energy meter to measure both weather-dependent and non-weather-dependent energy consumption together. For example, a building with electric heating might have a single electricity meter measuring all its electricity consumption (heating, lighting, office equipment etc.).

In degree-day analysis, energy consumption that does *not* depend on the weather is often referred to as "*baseload*" energy consumption. It generally comes from energy uses that are *not* directly involved with heating or cooling the building; examples include electric lights, computer equipment, and industrial processes. For the purposes of degree-day analysis, it is usually assumed that a building's baseload energy consumption is constant throughout the year.

It is worth nothing that the term "baseload" is often used to describe the total kW power of all the equipment that is on *constantly*, including when the building is closed. However, the baseload energy that we are talking about here is a total of *all* the non-weather-dependent energy consumption (including energy consumption from equipment that is only used during occupied hours), and is usually expressed as an average daily, weekly, or monthly kWh value.

Anyway, baseload energy consumption complicates the simple ratio-based approach to weather normalization that we described further above. You can only apply that method to energy consumption that is 100% degree-day dependent, so, if the raw energy-consumption figures contain baseload energy consumption as well as degree-day-dependent energy consumption, you need to subtract the baseload kWh from the raw figures before applying the ratio-based method. (You would typically add the baseload kWh back on to your normalized figures afterwards.)

There are two methods that are commonly used to calculate the baseload energy as a monthly kWh value:

- Regression analysis: when you plot monthly degree days (x-axis) against monthly energy consumption (y-axis), you can estimate the monthly baseload energy consumption from the point at which the regression line crosses the y-axis. For example, the scatter plot above shows a baseload (y-axis intercept) of around 1,870 kWh per month.
- If the building's heating is switched off over the summer, you can estimate the monthly baseload by taking an average of the monthly energy consumption over that period.

The same methods are often used on a weekly basis too, if weekly data is available. And in fact this is preferable for reasons we will cover later.

We've given a brief overview of the motivation and methods that are commonly associated with degree days in energy monitoring and targeting, and we shall now move on to the substance supporting the main theme of this article:

When applied to real-world buildings, common degree-day-based methods suffer from a number of problems that, if care is not taken to avoid or mitigate them, can easily lead to inaccurate, misleading results.

To explain how significant inaccuracies occur, following is an explanation of several major problems with the ways in which degree days are commonly used:

In degree-day theory, the *base temperature*, or "*balance point*" of a building is the outside temperature above which the building does not require heating. Different buildings have different base temperatures.

In the UK, for example, a lot of energy professionals primarily use degree days with a base temperature of 15.5°C. This is partly because 15.5°C-base-temperature degree days are the historical norm in the UK, and partly because, unlike degree days with other base temperatures, 15.5°C figures have always been readily and freely available.

Use of a 15.5°C base temperature is often justified with arguments along the lines of:

- Buildings are typically heated to a temperature of around 19°C.
- The heating system does not need to supply
*all*the heat necessary to ensure that the building is heated to 19°C: some heat comes from other sources such as the people and equipment in the building. These sources contribute to an "average internal heat gain" that is typically worth around 3.5°C. - If you subtract the typical average internal heat gain from the typical building temperature (19°C − 3.5°C) you get a base temperature of 15.5°C. This is effectively the temperature that the heating system needs to heat the building to, as the average internal heat gain supplies the difference. 15.5°C is therefore an appropriate base temperature for degree-day analysis of the energy consumption of the heating system.

The method of calculating an appropriate base temperature by subtracting the average internal heat gain from the building temperature is a sensible approximation. However, the figures used in the above arguments are where the problems lie:

**Different buildings are heated to different temperatures**. Although it's*often recommended*that office buildings be heated to 19°C, in reality they are often several degrees warmer. Industrial buildings are often several degrees cooler.**Average internal heat gain varies greatly from building to building**. Clearly a crowded office packed with people and equipment will have a much higher average internal heat gain than a sparsely-filled office with a high ceiling. Clearly the internal heat gain from industrial processes depends greatly on the processes in question.

In reality, 15.5°C is rarely the most appropriate base temperature to use for degree-day analysis. This is important, as the results of degree-day analysis can be greatly affected by the base temperature of the degree days used.

The "default" base temperature actually varies from country to country. For example, whilst in the UK it is 15.5°C, in the US it is 65°F which is equivalent to 18.3°C – nearly 3 Celsius higher. This alone is a pretty strong indication that the one-base-temperature-fits-all approach to degree-day analysis is inappropriate!

The following chart is based on 2020 heating-degree-day figures for San Francisco airport. Figures for three different base temperatures (55°F, 60°F, and 65°F) are displayed as percentages of the January values, January being the coldest month in 2020 for that location – the month with the most heating degree days in all three base temperatures. This makes it easier to see how the proportional month-to-month differences change with base temperature. In most degree-day analysis, proportional differences are more important than absolute values.

The chart shows how the base temperature of degree days can have a big effect on the proportional difference between the degree days of one month and the degree days of the next. The effect will be smaller in climates that are less seasonal than San Francisco's, but it will always be present. It is important to be aware of this if you are weather-normalizing monthly energy-consumption figures (or weekly or daily), as an inappropriate choice of base temperature can easily give you misleading results.

To complicate things further, the base temperature of most buildings actually varies a little throughout the year. It is affected by the sun (solar heat gain), the wind, and the patterns of occupancy, all of which typically vary to some extent throughout the year. Even the internal temperature of the building will typically vary a little unless the building's heating control system is working perfectly. In most buildings the effect of these factors will be relatively small, so our point is not that you should give up, or do a full building simulation that takes *everything* into account. Perfection is the enemy of good enough. Just bear in mind that the simplified model of degree days (and base temperature) is not a perfect representation of reality.

As this section has shown, it's important to pick an appropriate base temperature for degree-day analysis, and the most appropriate base temperature is unlikely to be the one that happens to be the traditional "default" in your country. As a building's base temperature typically varies a little throughout the year, even the most appropriate base temperature is usually only an approximation.

The concept of baseload energy is important, and very useful. But it can be difficult to calculate the baseload energy accurately.

Regression analysis (as described above) is one way to calculate the baseload energy. In a chart of degree days on the x-axis against energy consumption on the y-axis, the baseload energy can be taken from the point at which the regression line crosses the y-axis. However, the accuracy of this method is highly dependent on the degree days having an appropriate base temperature, which introduces the base-temperature problems described above.

To illustrate the effect of base temperature on the calculated baseload energy, the chart below extends the example we used above by adding plots of the same energy data against heating degree days with base temperatures of 60°F and 55°F. This is, in fact, made-up energy data (kWh and degree days don't correlate perfectly in the real world like they do here with the 60°F base temperature degree days!), but the degree days are real, and show the striking effect of base temperature on the baseload energy calculated by this method.

You can see from this chart that HDD in all three base temperatures correlate pretty well with the energy consumption – the R^{2} values are all high, and, if you saw any one plot without the others, you could easily assume it was fine (with the possible exception of the 55°F data that shows a bit of a curve, which is a sign that the base temperature is wrong). The 60°F data clearly fits best, but the 65°F data has an excellent fit too, yet it gives a baseload energy that is around 2.5 times greater than that given by the "perfect" 60°F data! Choosing the right base temperature clearly makes quite a difference to the baseload energy!

In fact, the whole concept of baseload energy is usually an approximation, as typically at least some of the energy consumption that contributes towards it is variable. For example: lighting energy consumption typically depends on the level of daylight, which varies seasonally and from day to day.

Baseload energy is certainly not suited to being wrapped up as a monthly kWh figure, as months are very different in calendar terms (this is explained further in our article on energy performance tracking). The simple fact that, in common 365-day years, March is over 10% longer than February makes it pretty clear that baseload energy consumption is unlikely to be constant from month to month.

Many buildings are only heated to full temperature intermittently, usually to fit around occupancy hours (e.g. 09:00 to 17:00, Monday to Friday). However, degree days cover a continuous time period: 24 hours a day, 7 days a week. This means that, for intermittently-heated buildings, degree days are not a perfect representation of the outside air temperatures that are most relevant to heating energy consumption.

Consider, for example, a building that is unheated overnight. Colder nighttime temperatures will have a *partial* effect on daytime energy consumption, as a colder night will typically mean more energy is required to bring the building back up to temperature in the morning. Thanks to the complicated ways in which buildings retain heat/coolth, to account for this well requires sophisticated building simulation. But the point here is that, whilst colder nighttime temperatures have only a *partial* effect on energy consumption, they are *fully* represented by degree days.

To some extent intermittent heating can be accounted for by lowering the base temperature of the heating degree days. This approach does work reasonably well in practice, but it doesn't solve the problem perfectly.

This problem is not only limited to day/night intermittent heating, as many buildings are also unheated through weekends, public holidays, and shutdown periods. When a particular weekend is uncommonly warm or cold, the degree-day total for that week or month will be affected accordingly, even though, for a building that is unheated on weekends, the outside temperature on that weekend will only have a partial effect on energy consumption (through its impact on the energy required to bring the building to temperature on Monday).

With degree-day analysis of monthly figures, such intermittent heating also introduces a calendar mismatch: although monthly degree days cover the entire month, the proportion of days for which a building is heated typically depends on the calendar of the month in question. Consider the following example monthly figures for a building that is heated on weekdays only:

Month | Total no. days | No. unheated days (weekends) | No. heated days (weekdays) | Proportion of days that are heated |

Feb 2019 | 28 | 8 | 20 | 71.43% |

Mar 2019 | 31 | 10 | 21 | 67.74% |

Apr 2019 | 30 | 8 | 22 | 73.33% |

May 2019 | 31 | 8 | 23 | 74.19% |

These example figures show that, for a building that is heated on weekdays only, the proportion of days that are heated varies quite considerably from month to month. This is a simple result of the way that the calendar works. Monthly totals of degree days do not take this calendar mismatch into account, giving another reason why they should not be expected to correlate perfectly with monthly heating energy consumption (kWh). When public holidays and shutdown periods are considered, the effect of this calendar mismatch becomes more marked, and weekly degree-day analysis is also affected.

As mentioned previously, degree days commonly come as weekly or monthly values. In order to compare or correlate energy consumption with weekly or monthly degree days, you need meter readings that are taken at the start of each week or month. If you are taking these meter readings manually, you should take them at midnight, and, for monthly data, often on weekends.

Of course, it is rarely convenient to take manual meter readings at midnight or on weekends, so it is common for such readings to be taken up to several days early or late. This can introduce a significant inaccuracy into degree-day analysis.

For example, the month of July 2020 ended at midnight on Friday 31st July. If, for convenience, the meter reading was taken at 09:00 on Monday 3rd August, July's energy consumption would cover a period that was around 8% longer than it should be, and August's energy consumption would cover a period that was around 8% shorter than it should be. With traditional monthly degree days, the degree-day figures would match the calendar months exactly, but the energy-consumption data would not. It's not difficult to see that this would introduce a significant inaccuracy.

*(This is as much a symptom of the base-temperature problem and the intermittent-heating problem as it is a problem in its own right.)*

When the outside temperature is close to the base temperature of the building, the building will often require little or no heating. Degree-day analysis is typically less accurate under such circumstances:

- Results are particularly sensitive to the base temperature of the degree days used, and, as explained above, the base temperature is difficult to pin down precisely. It's important to estimate it well, but there's rarely a perfectly "correct" base temperature for any given building.
- The problem is exacerbated if there is intermittent heating. As explained above, intermittent heating is usually accounted for by lowering the heating base temperature, but the flaws in this approximation become more apparent on days when it is touch and go as to whether heating is needed at all.

Matters are complicated further if the building has both heating and air conditioning. Under such conditions (an outside temperature close to the base temperature), a building will often require both heating and cooling to maintain a constant inside temperature over the course of a single day.

An energy-efficient building would usually sacrifice the maintenance of a constant temperature by ensuring that the temperature above which the air conditioning came on was a few degrees higher than the temperature below which the heating came on (a "comfort zone"). However, it is rare for real-world buildings to have perfect HVAC control. In fact, poor HVAC control can often result in a building being heated at the same time as it is being cooled – not very energy efficient at all!

Anyway, the upshot of these complexities is that, when the outside temperature is close to the base temperature of the building, degree-day analysis is typically less reliable. The inaccuracies introduced by the base-temperature problem and the intermittent-heating problem are exaggerated.

Fortunately, this ideal-temperature problem does not tend to affect year-to-year comparisons so much, because:

- most climates have temperatures that are considerably higher or lower than the ideal temperature for much of the year; and
- when temperatures are around the ideal temperature, heating/cooling energy consumption is typically much lower than usual, so inaccuracies introduced by these ideal-temperature periods typically only have a small effect on yearly totals.

It does, however, become more of a issue if you are doing ongoing energy-performance tracking on a daily, weekly, or monthly timescale. It makes it more difficult to place confidence in figures for the days/weeks/months that had an ideal temperature for much of the time, especially if there is intermittent heating.

This article has highlighted several problems with popular degree-day-based methods. The greatest danger with these methods is that they are used without an awareness of their shortcomings, as inaccurate figures that are thought of as accurate can easily lead to bad decision making.

Following are a few suggestions for how you can avoid or mitigate the problems highlighted above, and improve the accuracy of your degree-day analysis:

Regression is key to most good analysis of heating/cooling energy consumption. It is more complicated than the simple ratio-based normalization described further above, but, when done well, it is much more powerful and reliable.

A baseline regression model defines both the base temperature (or base temperatures plural if it covers both heating and cooling) and the baseload energy consumption, both of which are critically-important variables in degree-day analysis. It also provides goodness-of-fit statistics like R^{2}, which simpler ratio-based methods cannot. Getting a good regression model is imperative for most degree-day-analysis processes, like the one we recommend to calculate energy savings.

Several years after publishing the first iteration of this article, we developed the Degree Days.net regression tool to make it easier to get a good baseline regression model. We recommend you make good use of it!

If you have a choice, weekly energy data is usually a good one. Buildings typically operate on a weekly schedule, so factors like occupancy and baseload energy usage tend to be fairly constant from week to week. Also, weeks are all 7 days long, and this uniformity makes it easier to accurately calculate baseload energy consumption through regression analysis.

Monthly energy data is inferior to weekly data. You can use it, but be aware of the calendar mismatch that it introduces, and the fact that baseload energy consumption can't accurately be wrapped up as a monthly total. The Degree Days.net regression tool does regression in an improved way that helps with this.

Daily energy data can work well for buildings that operate similarly on all 7 days of the week, but in other cases it complicates things, as to use it well you have to group days by operating pattern and analyze each group separately.

If you are taking manual meter readings, try your best to take them at the set interval you have chosen, and as close to midnight as possible.

**Interval metering** can solve the meter-reading problem by automatically taking readings at the start of each day, week, and month (in fact, they typically take readings every 15 or 30 minutes). Our Energy Lens software is useful for splitting interval data into daily, weekly, or monthly totals for degree-day analysis. If you have interval metering already, do make use of the detailed interval data it records; if you don't, you might consider getting it installed.

NB Many people instinctively assume that, if they have detailed interval data, they should make full use of its detail when analyzing their heating/cooling energy consumption (e.g. using hourly temperature data instead of degree days). However, there are complicated time lags between changes in outside air temperature and their effect on heating/cooling energy consumption, and, unless you are doing a full building simulation, daily or weekly energy-usage totals and degree days are actually better for analysis of heating/cooling energy consumption. That said, analysis of detailed interval data can, in seconds, reveal patterns of energy wastage that could never be revealed by weekly or monthly regression analysis, so it is worth doing as well.

Interval metering helps further if multiple interval meters are fitted. **Interval submetering** can significantly reduce or even eliminate the baseload-energy problem. If the heating energy consumption is metered separately, any baseload energy will be minimal, so there will be minimal potential for inaccuracies to be introduced by a fluctuating or poorly-estimated baseload. You should therefore be free to perform degree-day analysis on the heating-energy-consumption data *without* having to worry about inaccurate estimates of baseload energy.

Submetering can also give you greater confidence in your analysis of the non-heating energy consumption of the building. If the non-heating energy consumption is metered separately, it will be unlikely to be affected by variations in outside temperature, and you should be able to make more accurate comparisons of week-on-week and month-on-month energy performance, without need for degree-day normalization techniques.

You should aim to use degree days that are:

- From a weather station near the building you are analyzing.
- Calculated accurately from high-quality temperature readings.
- Aligned with the dates of your energy data.
**In the most appropriate base temperature for your building**. The base temperature will almost certainly be different for heating than it is for cooling (cooling would usually have a higher base temperature). Internal heat gains will push both the heating and cooling base temperatures downwards. Intermittent heating will effectively push the heating base temperature down, and intermittent cooling will effectively push the cooling base temperature up. With experience, many energy professionals can estimate a building's base temperature(s) pretty well (and we have an article on choosing base temperatures for guidance), but it's often a good idea to try a multi-base-temperature regression analysis to see what fits best. The Degree Days.net regression tool does this automatically.

When we originally wrote this article, it was hard to get data that satisfied the above criteria. It was hard to find *any* degree days in a range of base temperatures, let alone *accurately-calculated* degree days in a range of base temperatures. We consequently included Hitchin's Formula, which provided a way to get a rough estimate of degree days in any base temperature, from monthly mean-temperature values. This was far from ideal, but it was better than nothing. However, since then we have set up the Degree Days.net website, which offers free heating and cooling degree days, in any base temperature, in a daily, weekly, monthly, or custom breakdown, for locations worldwide. Given that this makes it easy for anyone to get good data, we decided there wasn't much point in including Hitchin's Formula any more.

When comparing weather-normalized figures from one week or month to the next, inaccuracies caused by the base-temperature problem, the baseload-energy problem, and the intermittent-heating problem are likely to be exaggerated by seasonal variations. Such comparisons should be treated with caution.

However, provided the size and general operational patterns of the building do not change from one year to the next (energy-efficiency improvements aside), a comparison of yearly weather-normalized figures is less likely to suffer to the same extent, as the seasonal changes in any one year will usually be approximately repeated every year. Comparisons of year-on-year weather-normalized energy consumption are not infallible, but they should be less prone to the problems highlighted in this article than weekly or monthly comparisons.

NB It doesn't have to be calendar years – any periods covering around 365 days (or a multiple thereof) will have the same advantages. Also, this advice only applies to weather-normalized comparisons, which would typically be calculated using a baseline regression model. As explained above, weekly data or similar is best for actually getting a good baseline regression model (though ideally it should also cover a year or a full multiple of years, so as to represent all seasons equally).

The ideal-temperature problem (explained above) occurs when the outside temperature is such that the building requires minimal heating or cooling. Since the standard approaches to degree-day analysis are particularly inaccurate under such circumstances, it makes sense to be particularly sceptical of weather-normalized figures for these periods.

Remember that the figures calculated using degree-day-based methods are usually only approximate.

Plugging numbers into a formula will almost always give a result, but the accuracy of that result can only be trusted if the formula is sound and appropriate for the situation, and the numbers that went into it are accurate themselves. This is essentially the main problem with many common uses of degree days in monitoring and targeting: the calculations have no difficulty producing results, but the combined effect of the problems highlighted in this article means that, unless care is taken to avoid or mitigate them, the accuracy of those results can be very low (despite the fact that they may appear with several figures after the decimal point).

After reading about the inaccuracies introduced by the problems highlighted in this article, it should not be difficult to see how justifiable changes to a weather-normalization process or its input parameters (like the base temperature) could easily turn a supposed 3.74% month-on-month improvement in energy efficiency into a supposed 1.12% overspend *(figures after the decimal point are shown here to highlight the misleading nature of numbers that appear more accurate than they really are)*.

Even when you *do* follow best practices, and take all reasonable steps to avoid or mitigate the problems highlighted in this article, figures calculated using degree-day-based methods will usually *still* be approximate. This is just the nature of this sort of analysis.

The underlying issue is that heating/cooling energy consumption is complicated. In most buildings it's plenty large enough to deserve attention, but still, it's complicated. Degree-day analysis is the best option for dealing with the complexity (short of full building simulation which is far more involved), so we are not suggesting that you should give up on it, just that it is important to do the analysis carefully, and with an awareness of its approximate nature.

The degree-day-based techniques described in this article are just as applicable to buildings with consumption figures in the millions of kWh as they are to buildings with consumption figures in the hundreds of kWh. However, the magnitude of the absolute numbers has no bearing on their accuracy – it's the proportional differences that matter.

For example: if you use regression analysis to calculate that, over the last month, a building consumed 2% more energy than expected (a relatively low proportional difference), that might be more likely to be a symptom of inaccuracies in the calculation than an indication of a real problem. And, if you have little confidence that a 2% difference means anything, it's irrelevant whether that 2% difference equates to 200 kWh or 2,000,000 kWh.

In contrast, a 20% proportional difference may well be high enough for you to consider it meaningful. If you have confidence in the proportional difference, you can also have confidence in the corresponding absolute numbers, and can confidently use them for deciding priorities and so on. The critical point is that only the proportional difference can enable you to judge whether the result of a calculation is likely to be meaningful, and, until you have judged that, the absolute numbers can only be a distraction.

By looking at the general accuracy of your calculation process (e.g. how many of the problems highlighted in this article apply), and by looking at the quality of your baseline regression, you should be able to get a feel for the level of proportional difference that makes a figure worthy of your attention. It's only worth looking at the absolute numbers (e.g. excess kWh or cost) once you've determined from the proportional differences (e.g. percentage above expected) that they are likely to be meaningful (and not just a symptom of calculation inaccuracies).

Heating/cooling energy consumption is considerable in most buildings, and its analysis is an important part of most energy-management programmes. Degree days are central to this analysis, and rightly so, but to use them well is more complicated than it appears on the surface, simply because heating/cooling energy consumption is complicated. Consequently, degree days are commonly used in ways that lead to inaccurate, misleading results. People can easily find themselves chasing excess consumption that doesn't really exist, and highlighting improvements that haven't really been made.

This article has hopefully helped you understand the main sources of inaccuracy in degree-day analysis, how to avoid or mitigate them, and gain a sense for the reliability of the figures on which you base decisions. This improved understanding should help you separate the real indicators from the blind alleys, focus your time as effectively as possible, and achieve significantly greater energy savings as a result.

First, congratulations for reaching the end of this article – it's a long one! If you found it useful, might you consider telling your colleagues or mentioning it on your website?

Our Degree Days.net website provides free degree-day data for locations worldwide, and has a number of other articles on degree days and how best to use them. We actually built Degree Days.net a few years after publishing the first iteration of this article, after realizing there was a lot more we could do to help people save energy by using degree days more effectively.

If you have interval energy data such as half-hourly data or 15-minute data you might also be interested in the articles on energy management with interval data that we have here on our Energy Lens website, and the Energy Lens software for interval-data analysis. Fine-grained analysis of detailed interval energy data is different to the sort of longer-timescale degree-day analysis of heating/cooling energy consumption described in this article, but it is just as useful. To maximize energy savings it makes sense to do both.

© 2006–2022 BizEE Software – Contact Us | About Us | Privacy Policy