Comparative analysis of old and new weather files: City-based versus climate zone-based approaches

Introduction

Weather files are fundamental to building performance analysis. They provide critical climatic data that influence design decisions related to energy performance, thermal comfort, daylighting, and overall building efficiency.^1–5 Accurate and representative weather data allows designers to evaluate performance of buildings under various climatic conditions and optimise building design. ⁶ Moreover, weather files are essential in assessing the thermal behaviour of buildings. ⁷

Standard weather files, such as Test Reference Years (TRY) and Typical Meteorological Years, are commonly employed to represent average weather conditions^8,9 and are typically used to estimate expected energy usage. ¹⁰ As the frequency and intensity of extreme weather events rise due to climate change, it becomes critical to incorporate extreme weather data into building performance evaluation, since extreme weather events can have a profound impact on a building’s energy consumption, structural integrity, and thermal comfort. Moreover, Jentsch et al. highlight that buildings designed using only historical average data are at risk of underperforming when faced with the extremes of heat. ¹¹ It is crucial to embed the latest trend of climate change into standard weather files to ensure that buildings designed at the current date are future-proof. Thus, the development of representative extreme weather files, specifically designed to model these tail weather conditions, is crucial for designing resilient, adaptable buildings in conjunction with an idea of the general performance as can be modelled with the typical weather file.

In the UK both TRYs and probabilistic Design Summer Years (DSYs) are used for compliance within the building regulations. Specifically, TRYs are used for Part L: conservation of fuel and power^12,13 while DSYs underpin part O: Overheating.^14–16 However, there are two key issues with the use of these files:

(1) The UK industry standard files are out of date: based on a baseline from 1984 to 2013. There has been further warming and multiple historic weather records broken across the country including the hottest temperatures.

No matter how accurate building performance simulation tools are, they will only yield reliable outcomes if the weather data used for modelling accurately reflects the conditions a building will encounter throughout its lifespan. Inaccurate or flawed weather data will lead to unreliable results, which at best may be imprecise, and at worst, could be significantly misleading. Typically, weather files are constructed from historic observation at sites, normally at airports and similar, located next to urban areas.^17,18 It is then the role of the user to determine which weather file is most appropriate for their building. In the UK for example, previous iterations of weather files have produced 16 sites (three for London) for use with building regulations, ¹⁹ but without further guidance a typical user would use the closest file which can have significantly different weather to where the building is located. ²⁰

Geographical proximity does not always entail climate similarity. To enable the creation of accurate weather files which are representative of local climate, it is critical to develop a profound understanding of geographical patterns of the climate. Climate zones can then be established to amalgamate regions with a similar climate while distinguishing those with significant variations. The conventional Köppen-Geiger climate classification system, ²¹ which divides climates into five main climate groups and 30 sub-groups based on seasonal precipitation and temperature patterns, yields very large climate zones, not able to capture local variations. While the classification system has been updated by different groups including the impacts of climate change or the incorporation of higher resolution climate data,^22,23 many countries such as the UK are split into one or two climate zones and is therefore, not suitable for applications where high resolution and granularity of climate patterns needs to be distinguished, such as weather file creation for building performance analysis. To overcome the limitations of the conventional zoning method, a hierarchical clustering method was developed and a total of 28 climate zones were created using the UK Climate Projections 2018 in our previous study. ²⁴ To the best of our knowledge, it is the first study on dividing the UK into granular climate zones to represent local climate variations and enable more accurate creation of weather files. In contrast, none of the previous concepts of zones in building performance evaluation in the UK is based on climate.

In this work we will develop up to date weather files which cover the UK with weather files suitable for benchmarking building performance which for the first time removes the ambiguity of which file should be used for a specific site location. This will be driven by climate analysis which builds on previous work that has already defined appropriate climate zones. We will refer to these as the zone-based weather files compare the developed zone-based TRY and DSY weather files with industry standard weather files, updated to the same baseline. It is vital to know the key differences between weather files based at the UK standard locations and new zone-based files from the perspective of the external climate and how it may influence building design and then whether this reflects a change in performance through building simulation so both with will be investigated. These steps will be explained in more detail through the paper.

Methodology

Data availability is fundamental to developing weather files. As per previous studies, ¹⁷ the availability of observations is limited by the network of weather stations. Whilst there are many weather stations in the UK, most do not collect appropriate hourly weather data for the modelling of buildings via building performance simulation software. Here we restrict observations to those from the land surface synoptic observations. These mostly measure dry bulb, wet bulb and dew point temperature, cloud cover, surface pressure, wind speed and wind direction on an hourly resolution. There are over 400 weather stations which have collected this data with some stations in operation since 1949 until the present day. However, some of these stations have only collected data for a short amount of time (e.g., 1 year) or have data with a resolution of less than an hour. Additionally, there are very few stations with no gaps in the observation record.

Gaps in the data will be filled using linear interpolation for weather variables such as wind direction and wind speed. Gaps in air pressure will be filled using a spline interpolation. Gaps in temperature will be filled following a two-step process. First, the daily minimum and maximum temperatures will be interpolated (both time and magnitude) using valid maxima and minima either side of the missing data, then the gaps will be filled using a spline interpolation. This follows methods used previously for the creation of industry standard weather years.^8,17

To maximise the availability of data various thresholds were employed. For hourly data, the month will be included if less than 20% of the data is missing. For bihourly data, equivalent to 49% of the data must be available. For three hourly data, at least 32% of the data must be available equivalent to less than 7 h missing over a 30-day month. In all cases, if 72 consecutive hours (or 3 days) are missing then the whole month will be considered missing for the selection of TRY months. As per previous studies contiguous months are considered more important than contiguous years for the creation of a TRY. This approach provides a balance between ensuring that enough years are available for analysis whilst ensuring that large blocks of missing data are removed.

Cloud cover is the variable most commonly unavailable with it either not being recorded or has large gaps in the data but is fundamental to aspects of the weather file such as Illuminance. Gaps larger than 72 h as stated above would not be filled due to uncertainties in data quality and this restriction alone limits the years/months available for many sites. Furthermore, linear or spline interpolation over gaps under 72 h provides an unrealistic time series. Here, gaps in cloud cover were filled from the European Centre for Medium-Range Weather Forecasts Reanalysis data V5 (ERA5) data. ²⁵ The ERA5 data is an assimilation of satellite and ground observations using climate models to recreate observations on a regular grid with a very high correlation to the observations used as boundary conditions. The underlying resolution of the dataset is 31 km × 31 km with the nearest grid box to the weather station being selected. The uncertainty of the ERA5 data is lowest for more recent years where there are more observations and better satellite coverage. Here, the average error comparing the ERA5 cloud cover to the known cloud cover from the observations from all sites is 8% demonstrating that such data is a suitable replacement whilst increasing the data availability by filling in very long periods of missing data.

Similarly, there are significant periods of time where the relative humidity is missing from the observations. Due to the complex relationship between humidity and other psychrometric variables such as dew point temperature and wet bulb temperature, larger periods of missing relative humidity will be replaced with ERA5-land hourly data with a resolution of 9 km. This data will then be used with the observed conditions to estimate the wet bulb and dew point temperature. Again, the grid box closest to the observation location is used.

In the UK, solar irradiation is not recorded at many stations while others have simply stopped recording. Satellite data is now becoming increasingly available with a high spatial and temporal resolution. The Copernicus Atmosphere Monitoring Service (CAMS) provides historic global, direct and diffuse solar irradiation from February 2004 onwards. ²⁶ The CAMs approach combines atmospheric models of aerosols, water vapour and ozone with the Meteosat Second Generation (MSG) satellite observations of cloud cover to provide an estimate of solar irradiation at the point of interest. Validation over the UK suggested the Root Mean Square Error of solar radiation over 3 years of observations in Lerwick (the limit of the field of vision for the MSG) is smaller than the Root Mean Square Error for Aldergrove using the Cloud Radiation Model as previously used in the UK. ²⁷ In addition, this data has been shown to be highly suitable for the recreation of Illuminance data within the UK. ²⁸ However, data is only available after January 2004 and therefore CAMs data can only be used to fill in the global, diffuse and direct normal irradiation for all years after January 2004. For all other years, global solar and direct solar radiation is available from the ERA5 data on single levels and will be consistent with the replaced cloud cover. The direct normal and diffuse irradiation is then calculated from the solar altitude.

The purpose of this study is to demonstrate the differences between the previous method used to generate industry standard weather and the zone-based weather files. The approach described above to generate the zone-based TRYs and DSYs will be employed here but using the locations nearest to the historically chosen urban locations.^17,29 As with previous updates, weather stations have been decommissioned while others have then come online and as per previous updates new weather stations will be sought where needed. In updating these city-based weather files most locations remain but the observation period moved to the most recent 30 years i.e., 1994–2023 (Table 1). In the case of Leeds and Manchester rather than using a third station to cover the most recent observations, a nearby site with a complete 30-year period is used. In the case of Leeds, this maximised the number of complete years available.

OPEN IN VIEWER

Table 1.

Historic and updated observation locations used for the city-based weather files.

Location	2015 city-based stations	New city-based stations
Belfast	Aldergrove (1984–2013)	Aldergrove (1994–2023)
Birmingham	Elmdon (1984–1997)	Elmdon (1994–1997)
	Coleshill (1998–2013)	Coleshill (1998–2023)
Cardiff	Roose (1984–1997)	Roose (1994–1997)
	St Athan (1998–2013)	St Athan (1998–2023)
Edinburgh	Turnhouse (1984–1998)	Turnhouse (1994–1998)
	Gogarbank (1999–2013)	Gogarbank (1999–2023)
Glasgow	Abbotsinch (1984–Apr 1999)	Abbotsinch (1994–Apr 1999)
	Bishopton (May 1999–2013)	Bishopton (May 1999–2023)
Leeds	Church Fenton (1984–1988)	Bingley No 2 (1994–2023)
	Leeds WS (1989–2013)
London	Heathrow (1984–2013)	Heathrow (1994–2023)
Manchester	Ringway (1984–2003)	Hawarden Airport (1994–2023)
	Woodford (2004–2013)
Newcastle	Newcastle (1) (1984–1990)	Newcastle (2) (1994–Feb 2003)
	Newcastle (2) (1991–Feb 2003)	Albemarle (March 2003–2023)
	Albemarle (Mar 2003–2013)
Norwich	Marham (1984–2013)	Marham (1994–2023)
Nottingham	Watnall (1984–2013)	Watnall (1994–2023)
Plymouth	Mountbatten (1984–2013)	Mountbatten (1994–2023)
Southampton	Hurn (1984–2013)	Hurn (1994–2023)
Swindon	Brize Norton (1984–2013)	Brize Norton (1994–2023)

The base observation period is used to determine the months for the TRYs and the baseline observations for determining the DSYs. In the case of the DSYs, further years from the historic observations will also be used to increase the number of extreme events available. In each case this amounts to using all previous years (back to 1961) if available of the first named station. For example, in Belfast this would mean taking all observations from 1961 to 2023 for Aldergrove and for Birmingham this would mean all observations from 1961 to 1997 for Elmdon along with those from Coleshill from 1998 to 2023. In the case of London, TM49 established DSYs from rural and urban DSYs using the same years as determined from the Heathrow location. ²⁹ TM49 used Gatwick for the rural DSYs and London Weather Station for the urban DSYs. In this case Heathrow was defined as a peri-urban site. In the update, rural DSY locations are taken from Gatwick for selected years before 1999 and Charlwood for years after. Urban DSYs come from London Weather Centre for years before 2011 and from Kew Gardens after. The peri-urban DSYs remain from Heathrow. The previous and updated locations used for baseline weather data and for determining Test Reference Years and the statistics for Probabilistic Design Summer Year are outlined in Table 1.

TRYs should consist of weather patterns which are representative of the long-term trend as observed at a particular location. Typically, Finkelstein-Schaefer (FS) statistics are used to select the appropriate months. ⁸ The FS statistic takes the sum of the absolute difference between an individual month’s (e.g., January) daily cumulative distribution function against the cumulative distribution using all previous months (e.g., against all previous January’s) for a given weather variable. The month with the smallest FS statistic is then taken as the month with the closest distribution to the long-term climate. This process is then followed for each month with the selected months then combined into a single TRY. The difference between various average weather files used around the world is the weather variables used to determine the average weather and depends on what is considered most important to determine building performance.

Here TRYs are created following the ISO method. ³⁰ This method was used previously and has been described in depth elsewhere. ¹⁷ The ISO method uses dry-bulb temperature, relative humidity and solar radiation with an equal weighting to determine the 3 months with the lowest ranking (lowest weighted sum of the FS statistics). The FS statistic for wind speed is then calculated to determine the representative month. Similar to the previous update, due to the lack of directly observed solar radiation, cloud cover is used as a proxy for solar radiation in the selection of the most representative months.

The previous method for creating DSYs involves determining extreme events, modelling these via extreme value theory to determine return period then selecting years at appropriate return periods. ³¹ In this work a similar approach will be used but to simplify the selection, only a single metric will be used – the Static Weighted Cooling Degree hours.

Firstly, at each location the threshold for overheating will be determined. Armstrong determined that deaths due to heat could begin to be determined at the 93rd percentile of the year round 2 day mean of the daily maximum temperature across England and Wales. ³² Here, thresholds were determined at each location across the UK assuming that the threshold stands under a warming climate. These thresholds are shown in Table 2 for the 14 locations. For all years the Static Weighted Cooling Degree Hours is then calculated. This is simply the sum of the difference between the dry bulb temperature and calculated threshold for all hours between April and September inclusive. In the previous approach the Generalised Extreme Value distribution was used to model the distribution of the events, importantly this kept the consistency of the approach used to model the events for TM49. The Generalised Extreme Value distribution is more typically used to model the maximal or minimal event, for example the most extreme temperature in a year. The yearly metric can be described by the exceedance over a threshold, with year being an impendent event, so in the case the yearly exceedance will be modelled using the Generalised Pareto Distribution. Return periods can then be estimated from the inverse of the exceedance probability. The first DSY (DSY1) is then given by the year with the closest to a one-in-seven return period. In the previous work each location had to be inspected manually to determine which years could be chosen as DSY2 and DSY3 from the years which had a higher return period than DSY1. Here DSY2 is defined as the most intense year. The intensity is calculated by the sum of the yearly exceedance squared divided by the number of hours where the temperature is greater than the threshold. DSY3 is then defined as the longest duration year. The DSY3 is determined from the remaining years with a higher return period than DSY1, and intensities greater than DSY1. Of these years, DSY3 is the year with the longest duration. In the instance where there is no year left with an intensity greater than DSY1, the year with the intensity closest to the DSY1 and longer in duration than DSY1 is used.

OPEN IN VIEWER

Table 2.

New SWCDH thresholds for city-based weather files by location.

Location	SWCDH thresholds/°C
Belfast	19.8
Birmingham	22.8
Cardiff	21.1
Edinburgh	20.0
Glasgow	20.0
Leeds	21.8
London (all)	24.8
Manchester	22.0
Newcastle	20.2
Norwich	23.5
Nottingham	22.4
Plymouth	21.0
Southampton	23.0
Swindon	23.2

The generation of climate zones follows the method of Xie et al where a more comprehensive description can be found. ²⁴ A two-tiered ensemble clustering method was used to create granular climate zones in the UK, using the regional climate projections with 12 km resolution from UKCP18. The first tier identifies primary spatial climate patterns in the UK and a second-tier further segments microclimate from each primary climate zone.

A dataset of 12 model projections spanning a 100-year period (1981–2080) under the RCP8.5 emission scenario from regional projections served as the raw input for the analysis. In the first tier, 41 features were extracted from each model projection, comprising 39 climatic variables and two geographical location variables, and were used as input for the ensemble clustering model. ²⁴ This model employed 12 base K-means models, where each projection in the dataset was assigned to a single K-means model using the bagging method. This process generated 12 variants of climate zoning results, which were then consolidated into primary climate zones through an agglomerative clustering process.

In this first step, by including geographical proximity in the clustering process, the method balances the trade-off between climate similarity and spatial continuity, resulting in compact and practical zone delineations. In total 14 climate zones were created preserving the broad spatial climate patterns experienced across the UK such as the colder north, warmer south; wetter west, and drier east, while improving local relevance.

In the second tier each climate zone was then examined in detail to identify localized microclimates. A similar clustering methodology was applied with a focus on temperature-related variables due to their significance in building performance assessments and their relevance to microclimate phenomena such as the urban heat island intensity. Specifically, three temperature variables—maximum temperature, mean temperature, and minimum temperature—were utilised for clustering.

These variables were used to further segment the microclimates within each primary climate zone. The outcome from this clustering process identified 14 climate zones which are then further segmented into 28 zones. This second clustering stage successfully captures key microclimatic areas—such as the urban heat island of London and mountainous regions—which are often overlooked in standard weather file classifications. These granular distinctions provide a more accurate reflection of local climate conditions critical for building energy modelling and a basis for further development.

However, note, the only consideration of the urban heat island intensity in the classification system is only that which is contained within the signature of the 12 km data. In many cases, urban areas would only consist of one or two grid squares at this resolution and are more anomalous than reflecting a wider climatic region within the main climate zone.

The next step is the identification of weather stations which are representative of the zone. Although the weather within the identified zones is already considered as somewhat similar by virtue of the clustering algorithm used there will be variation across the zone. A single grid cell which shows the highest similarity to the overall condition of the zone is identified to create weather files. Wasserstein distance of the climate variables between a single cell and the collection of all cells within the climate zone is employed as the selection criteria. As such, weather files created based on the selected grid cell can represent the climate condition of the whole zone.

When comparing this identified grid cell with observations, ideally, a single location exists which is coherent with the identified cell. However, due to limitations in the availability of observational data, the nearest weather stations with sufficient data are used as proxies for the identified representative locations. Observation files are then generated based on these stations. Here, the representative location will be selected as the weather station with at least 20 years of suitable data for generating the base statistics which is nearest to the selected representative location (Table 3). In most cases, the selected files have the full 30 years’ worth of data from the period 1994–2023. In the case of zone 12 for example, an equivalent of 26.5 years is available but with various summer months temperature data missing, only 17 complete summers could be used to determine the baseline statistics for the DSYs.

In the case of zone 26, Inverness-shire, Aonach-Mor, would be chosen as the representative location. However, far too little observation data is available, and with no other viable weather stations within the relatively small climate zone, Sky Lusa has been used for both zone 25 and zone 26. Zone 26 was separated from zone 25 in the second tier of clustering so is somewhat representative of this zone too.

Figure 1 illustrates the climate zones (a) and the locations of the approximate regions for the city-based files (b) based on the distance away from the location. The zone-based files are fully defined with zone 1 starting around Greater London and then zone numbers generally increase going from east to west, then moving north. Hence zone 5 and 6 covers Cornwall and parts of Wales and zone 7 centres around Norfolk. Local variations are captured such as urban areas around Bristol and greater London, as well as national parks such as Dartmoor and Exmoor in the southwest, to Bannau Brycheiniog in Wales as well as the highland areas in Scotland. The largest distinction can be observed in Scotland where 12 zones are used compared to two cities. Note, to enable the city zones to be easily observed on the scale used, Greater London is assumed to use the urban London DSYs (London Weather Centre), with a surrounding ring used for the suburban file (Heathrow), with the rest of the area using the rural London file (Gatwick).OPEN IN VIEWER

OPEN IN VIEWER

Table 3.

Representative locations for the zone-based weather files.

Zone number	Observation location	Zone number	Observation location
Zone 1	Greater London, Northolt	Zone 2	Kent, Manston
Zone 3	Hampshire, Middle Wallop	Zone 4	Wiltshire, Larkhill
Zone 5	Devon, Chivenor	Zone 6	Dyfed, Aberporth
Zone 7	Norfolk, Weybourne	Zone 8	Norfolk, Marham
Zone 9	Nottinghamshire, Watnall	Zone 10	Clwyd, Hawarden Airport
Zone 11	Lincolnshire, Donna Nook	Zone 12	Dyfed, Trawsgoed
Zone 13	Powys-South, Sennybridge	Zone 14	North-Yorkshire, Leeming
Zone 15	Berwickshire, Charterhall	Zone 16	Northumberland, Redesdale Camp
Zone 17	Dumfriesshire, Eskdalemuir	Zone 18	Cumbria, Shap
Zone 19	Antrim, Aldergrove	Zone 20	Antrim, Ballypatrick Forest
Zone 21	Aberdeenshire, Dyce	Zone 22	Aberdeenshire, Aboyne
Zone 23	Inverness-shire, Aviemore	Zone 24	Inverness-shire, Tulloch-Bridge
Zone 25	Western-Isles, Sky Lusa	Zone 26	Western-Isles, Sky Lusa
Zone 27	Ross-and-Cromarty, Aultbea	Zone 28	Shetland, Lerwick

The weather files will be compared in terms of the climate expected at each location across the UK. Here we will assume that a designer would choose the closest city-based weather file to their location as often better information is not available. The climate zones have clear boundaries telling the user which file they should use. For most figures the difference between the city-based and zone-based files will be the focus. In most cases we will not adjust the weather files for altitude since the main application will be compliance modelling and therefore the results will reflect the differences between the selected files as presented. This is because this is how we expect designers to use the files in practice. For example, the National Calculation Methodology makes no reference to the need to adjust the weather file given the location but simply to select the most appropriate file. ³³ In addition, standard software such would use elevation to estimate solar gains and atmospheric pressure, but not to adjust the temperature and without further automation no other adjustment would be made in practice.

The purpose of TRY files is to estimate the annual energy consumption of a building. In the UK this would be heating energy so annual heating degree days to a base of 15.5°C will be used. Part of the motivation of climate zoning is to distinguish the impacts of altitude as a result of geographical features, as evidenced by the segmented region of Exmoor, Dartmoor, Highlands, etc. Instead of correcting temperature empirically, we are using a more comprehensive method to address this issue of altitude, and other factors as well, such as proximity to sea. To distinguish the difference between the zone-based files and city files and the impact of altitude we will also calculate the heat degree days with a base of 15.5°C with all files adjusted to sea level.

On the contrary, DSYs are used to inform overheating risks and thermal discomfort during warm events. To compare the files, the annual cooling degree days with a base temperature of 15.5°C will be used. The ability of a building to remain cool during warm weather will depend on its ability to pass the likes of part O of the building regulations.^14–16 Simply, we will consider the percentage of hours where the external air temperature is above the comfort temperature for category I occupants. This means that all temperatures greater than 2.5°C above the comfort temperature will be counted. This considers the acceptable range of 2°C and the effect of rounding the difference as suggested by TM52. ³⁴

The purpose of these files will primarily be for building performance analysis so as an example we will also compare the different files from simulating a single building using EnergyPlus. The building has been used as part of a previous study and for further details readers are advised to look at Xie et al. ³⁵ In summary the building is a two-bedroom flat with a lounge area with a total footprint of 94 m² and a glazed area of 30.24 m². Both the ceiling and floors are internal surfaces while the walls and windows have U-values of 0.254 W/m²/K and 1.058 W/m²/K respectively. Overall, the building is designed to be compliant with Part L of the building regulations in London. For the TRYs the average energy use through heating requirements will be output. In this set up all internal gains such as equipment, lighting, etc is removed and the energy reported will be that to keep bedrooms at 18°C and the living area at 21°C following the UK annex to BS EN 12831:2003. ³⁰ Again, to show the impact of the zone-based approach we will also report the heating energy with all weather files adjusted for sea level.

For the use of DSYs, the ability to pass TM59 ³⁶ are seen as crucial. Here, the percentage of occupied hours where the internal operative temperature exceeds the comfort temperature will be calculated. A failure is considered when more than 3% of the occupied hours are greater than the comfort temperature. Failure is also deemed when bedroom temperatures are too high, so we will also calculate the number of hours between 22:00 and 7:00 where bedroom temperatures are above 26°C and compare these results for the two weather file location choices. Note, that TM59 considers 33 or more hours greater than 26°C as a failure. In this set up all parameter settings following CIBSE TM59. ³⁶ When reporting the results, we only consider the worst-case exceedance across the three zones of the two bedrooms and living area.

Results

Discussion and conclusion

The analysis highlights significant differences between the city-based and zone-based files with implications for building design in terms of both heating and thermal comfort.

In terms of heating degree days, much of the differences are minimal (less than 10% difference) but in general the zone-based weather files record higher heating degree days than city-based files, particularly in highland areas such as Wales, Scotland, and Northern England, where the city-based files lack resolution and extreme variations exceeding 400° days are observed. This suggests that the city-based weather files are not representative of the wider areas that they may be used for particularly for rural locations. The differences between zones and between the city and zone-based files cannot be explained by the altitude of the files. Adjusting the files to normalise all to sea level showed that correcting the temperature empirically was not a suitable replacement for appropriate consideration of the zones as more factors such as proximity to the sea also had an influence on the zone selection.

The general pattern of the cooling degree days is as expected with a peak in the cooling degree days in central London which then decreases towards the west and north. There are some anomalies such as the cooling degree days being higher for zone 3 (378) than zone 2 (313) for DSY1. This is due to the complexity of selecting years from the different zones and then comparing adjacent zones. For example, 1976 is selected as DSY1 for zone 2 and 2022 is selected for zone 3. These results prove that the spatial pattern of the UK climate is complicated, and the current city-based files are unable to capture such diverse patterns. Within the area of Scotland alone, 12 different climate zones are involved, but the city-based weather files are only provided for two locations, i.e., Edinburgh and Glasgow. Increasing the number of sites for the city-based files would be an improvement but there would still need to be better guidance of which file to use where and the closest to the site location would not be sufficient.

The results from differences between the zone-based and city-based cooling degree days are less pronounced than for the heating degree days but this is due to the UK’s building energy demand being dominated by heating and the differences between the two files having no consistent pattern. However, the highlands and moorlands generally have fewer cooling degree days than the surrounding areas; features which again show the city-based files are not representative of the surrounding areas. Of further significance here though is the use of urban, sub-urban and rural DSYs for the London area. Heathrow is significantly warmer than any other file and is highlighted by the ring around central London. It is questionable how representative Heathrow is for any other London location and therefore its appropriateness. Here the approach of selecting a zone-based file is not to choose the hottest or coolest location but the one which is more representative of the whole zone. The point of these files is to indicate risks of overheating and to benchmark performance not to precisely predict the hours of overheating where an actual weather year derived from the climate outside the building (or nearest station) for a particular period would be more appropriate.

The thermal comfort analysis for the outside temperatures reveals notable geographic trends for both the zone-based and city-based files. In each case the highest exceedance is found in London and the exceedance reduces towards the north and west. The zone-based files again show the advantages of the higher resolution by having a reduced exceedance over the highlands and moorlands. In terms of overheating and passing TM59, the weather data indicates that if the internal temperature of the building matches the external temperature for all locations, then it would be possible to pass the overheating criteria at all locations and all weather files regardless of whether the city-based or zone-based files are used.

Extending the thermal comfort analysis to within the example building model, simulations showed that it is possible to pass TM59’s criteria for hours exceeding the comfort temperature for DSY1 for all locations and all DSY2 locations for the zone-based files. The largest exceedances are found in the Southeast with limited thermal discomfort across the rest of the country as might be expected in the UK. Similar results are found for bedroom temperatures where all zone-based files would pass the TM59 criteria and the only failure for the city-based files is for the Urban and Sub-urban files. The simulations in this example show that the amount of thermal discomfort is low but follows an expected trend – there is more thermal discomfort in the south and east and less in the north and to the west.

There is no known justification for the historically selected weather stations in the UK or then how they should be used. ⁸ Most have long term weather data, but this is not consistent across all locations ¹⁷ and further closures of weather stations complicate the selection of appropriate sites as found within this work, but it remains unclear how representative locations are for a specific location. Because the zone-based files have specific boundaries a clear differentiation is seen between lowland and highland regions in all results. For example, the moorlands have lower exceedance than surrounding areas reflecting the cooler climates. Similar is found in terms of estimating heating energy where larger demands are estimated in the north and highlands which has clear advantages over the city-based files where it is less certain which files should be used for which region. Here the simplest interpretation has been taken whereby the nearest weather file has been selected. This in principle follows the National Calculation Methodology modelling guide which advises the selection of the weather file corresponding to the location closest to the building site. In cases where a microclimate is present, one of the other available weather files may be used if deemed more appropriate but justification is required. ³³ While the guidance emphasizes the importance of using a weather file that accurately represents the local climate, it does not provide specific instructions on how to determine the most representative file. A bigger issue arises for more remote locations where there may not be a more appropriate file. The use of climate zones removes any potential ambiguity that the use of the city-based files encounters. An approach may be to adjust the files for altitude to make the city-based files more representative, but it is clear from these results that adjusting for altitude alone does not account for the difference between zones.

In either case there are some distinct boundaries between different locations because the approach used does not provide spatial coherence since different years can be selected. Using zone-based files with observations does not solve this issue but the zones have already been shown to have long term weather data which are similar. ²⁴ The use of observations is the limitation here. Some hot weather events such as those experienced in 2022 manifested as short, intense heat wave events in some locations (such as zones 3 and 4) but were experienced as long less intense events in neighbouring zones (zones 1 and 2) meaning different years are used across these boundaries.

The case study building model used in this work is designed to be approximately compliant with TM59 for a 2020s weather file in London, so the low levels of exceedance are to be expected. The purpose here is to show the differences between the two approaches for providing weather files and how appropriate they are for benchmarking performance. It is clear though that the zone based-files better reflect the UK climate and specific regions than the city-based files supporting their wider adoption for use in benchmarking building performance.

In conclusion, this work has shown that zone-based weather files can better support building performance analysis and compliance modelling than the historic city-based files. Having specific zones which represent microclimates with specific boundaries will support their wider use and remove uncertainty to which file is most representative of a given location. There are clear differences between both the zone-based and city-based files, but comparisons are complicated due to the lack of specific information surrounding which city-based file should be used for which location. It is also clear that elevation alone is not sufficient to explain the difference between climate zones supporting the use of the zone-based files to better reflect the building performance as well as suggesting that representative files are more suitable than empirical corrections.

The lack of consideration of what is representative about the city-based files is a clear. Heathrow for example, is a much warmer location than the surrounding areas, and our analysis shows that it is not representative of the wider zone. This work is not about making site specific weather files but demonstrates how to create weather files that can help support the development of building designs and enable benchmarking of building performance. For comparing actual performance with the local weather, an actual weather file would be needed.

Overall, the zone-based files are cooler than the city-based files. This is largely because more cooler climates are now included and warmer sites such as Heathrow, which are not representative of the wider area are not selected. In the example of the southeastern most area as represented by zone 2, a specific climate file has been established which is much cooler than the London based file that would have been used previously. This is likely to mean that less thermal discomfort will be experienced and thus less mitigation measures will be required to pass current building regulations, and in the case where mechanical cooling is employed it will result in smaller plant sizes being specified. Subsequently, resulting in lower embodied carbon or unnecessary mitigations measures being implemented in cooler regions. The imitations of this work surround the use of observations and the limited distribution of weather stations as well as the inherent issues with data quality, missing data etc. A new approach would be required to achieve spatially coherent weather files.