Download Weather Files and Future Climate

Weather Files for Current and Future
Climate
1.
Weather Files for Building Simulation ..................................................................................................2
i) Introduction .............................................................................................................................................2
ii) The Example Weather Year (EWY): ...................................................................................................2
iii) The Typical Metrological Year (TMY): .............................................................................................2
iv) International Weather Years for Energy Calculations (IWECs): ......................................................3
v) Test Reference Years (TRY) and Design Summer Years (DSY): .....................................................3
vi) Considerations when using Reference Years......................................................................................6
Climate change and impacts..........................................................................................................................10
i) Introduction ...........................................................................................................................................10
ii) Projections............................................................................................................................................10
iii) Impacts ................................................................................................................................................14
iv) Other Considerations ..........................................................................................................................21
2.
Creation and Availability of Future Weather Years.........................................................................22
3.
Comparison between UKCP09 and UKCIP02 ...................................................................................23
4.
The morphing algorithms ......................................................................................................................25
5.
Creation of Future Years using a Weather Generator .....................................................................26
i) Creation of Future Weather Years .......................................................................................................26
ii) Creation of probabilistic weather data................................................................................................27
iii) Implications for Low Carbon Building Design ................................................................................28
6.
Appendices ...............................................................................................................................................29
i) Typical Metrological Year Creation ....................................................................................................29
ii) International Weather Years for Energy Calculations (IWECs) Creation .......................................30
iii) Test Reference Years (TRY) and Design Summer Years (DSY) Creation ....................................31
iv) The Morphing Algorithms .................................................................................................................31
v) Creation of Future Weather Files Using the UKCP09 Weather Generator .....................................34
Weather Files for Current and Future Climate
Page 1 of 37
1.
Weather Files for Building Simulation
i) Introduction
Building thermal modelling packages such as IES, TAS, Energy Plus etc. need representaion of
weather for the location in question. This is used to generater representaive values for energy
usage and to show compliance with various building regulations. A reference weather year
(variously known as a Test Reference Year (TRY), a Typical Meteorological Year (TMY), a
Standard Weather Year or an Example Weather Year (EWY)) is a single year of hourly data
(8760 hours), selected to represent the range of weather patterns that would typically be found
in a multi-year dataset, i.e. it is an average or typical year for a given location and time frame. It
is intended to allow more economical simulation than multi-year datasets, and to form an
equitable basis for comparing the predicted typical energy consumptions of different building
designs and, in some cases, the typical performance of solar collectors.
Definition of a Reference Year depends on its satisfying a set of statistical tests relating it to the
multi-year parent dataset from which it is drawn. Some groups have preferred to identify a
continuous, 12-month period as typical, whereas others have applied the criteria to individual
months, subsequently assembled into a composite 12-month set. Some common types of
reference year and their characteristics are listed below.
ii) The Example Weather Year (EWY):
The first basis for a UK reference Year was developed by CIBSE (Holmes and Hitchen, 1978).
This example weather year (EWY) sought to match the characteristics of an entire year in terms
of the means and standard deviations of its monthly data to the average monthly values for
many years of data (ie. the most average year out of a set of many years of data). An entire
contiguous year was chosen to avoid discontinuities in the weather sequence. However, by
selecting an entire year as the average there is only a small chance of finding a truly typical year
in ~20 years worth of data, which will accurately represent future weather patterns. However,
the files are still available for a large number of locations around the world and are still used
where there is insufficient data to create a different format of weather file.
• Total irradiation on horizontal surface
• Diffuse radiation on the horizontal
• Daily mean wind speed
• Daily maximum dry-bulb temperature
• Daily minimum dry-bulb temperature
• Daily mean dry-bulb temperature
• ‘Infiltration number’ (a function of wind speed and dry-bulb temperature)
The selection criteria are relatively simple: to reject any contiguous year with one or more
monthly mean values departing more than 2 standard deviations from the long- term monthly
mean (e.g. the mean of ~20 or so Januarys under consideration), and then to reject from any
remaining years those with the highest deviations from long-term monthly means until only one
year remained.
iii) The Typical Metrological Year (TMY):
The TMY data sets are primarily for the United States. The original TMY data sets were created
in 1978 for 248 locations using long-term weather and solar data from the 1952–1975. Later
updated in 1994 using data from the 30-year period 1961–1990. The current TMY3 data sets are
produced using input data for period 1976-2005 for 239 sites and using data from the period
1991-2005 for another 950 sites. The files are created using a statistical method to chosen the
Weather Files for Current and Future Climate
Page 2 of 37
most representative months from the 15 or 30 years of data. These months are then combined to
form a composite year of weather data. The details of how this is done is given in the appendix.
iv) International Weather Years for Energy Calculations (IWECs):
The ASHRAE International Weather Years for Energy Calculations (IWECs) are available for
227 locations outside North America and are compiled from up to 18 years worth of data
depending on location. The selection criteria for the IWEC files is similar to the TMY selection
process but instead uses nine weighted weather parameters, which are shown in the appendix.
v) Test Reference Years (TRY) and Design Summer Years (DSY):
In the UK we have the Chartered Institution of Building Services Engineers (CIBSE) Test
Reference Year (TRY) and Design Summer Year (DSY). The current versions of the TRY and
DSY are available for 14 locations around the UK. The TRY is composed of 12 separate
months of data each chosen to be the most average month from the 23 years of data, typically
1983 to 2005 but this varies depending upon data availability. The most average months were
chosen based on the cumulative distribution functions of the daily mean values of the three
parameters: dry bulb temperature (DryT), the global solar horizontal irradiation (GlRad) and
wind speed (WS). The daily mean values were determined from the hourly values of each of the
parameters for all the months in the years considered. The most average months are again
chosen using the Finkelstein-Schafer (FS) statistic to compare the cumulative distribution
functions. The FS statistic compares sums the absolute difference between the values for each
day in an individual months cumulative distribution function and the overall cumulative
distribution function for all the months considered. The months with the smallest FS statistic are
chosen as the most average. Hence the average month chosen using the FS statistic can be
considered representative of all the years. This process is followed for each month of the year
for each parameter in turn.
In contrast to the relatively complex method of assembling a TRY, the method of selecting a
DSY is very simple. The DSY is a single contiguous year rather than a composite one made up
from average months. The CIBSE procedure for selecting a DSY is to calculate the daily mean
DryT in the period April to September for each of the 23 years. The DSY is the year with the
third hottest April to September period. TRYs and DSYs were produced for 14 locations around
the UK, Belfast, Birmingham, Cardiff, Edinburgh, Glasgow, London (Heathrow), Swindon,
Leeds, Manchester, Newcastle, Norwich, Nottingham, Plymouth and Southampton.
Weather Files for Current and Future Climate
Page 3 of 37
Figure 1 Map of the UK on a 5km grid showing mean annual air temperature for the period 1961-1990.
The legend shows the temperature scale and the number of grid squares that fall into each temperature
range.
Figure 1 shows the range of mean annual temperatures across the UK. This is based upon the
Perry and Hollis data set, which is available at a 5km resolution for the whole of the UK for the
period 1961-1990. The Perry and Hollis data set uses measurements from many locations
around the UK and then spatially interpolates between them accounting for altitude, longitude
and distance from the coast. As the figure shows there is significant temperature variation cross
the UK, not only moving North to South but also variation with altitude and distance form the
sea.
Figure 2 Map of the South West Region on a 5km grid showing mean annual air temperature for the
period 1961-1990. The legend shows the temperature scale and the number of grid squares that fall into
each temperature range.
Weather Files for Current and Future Climate
Page 4 of 37
Focusing more closely on a specific region, in this case the South West (Figure 2) we can
observe significant temperature variation just across the region. There is an obvious decrease in
temperature on Dartmoor, Exmoor and Bodmin moor, which is not unexpected. The magnitude
of the temperature difference is surprising though; the difference in mean temperature for
instance between Plymouth and Dartmoor equivalent to the difference in temperature between
Plymouth and the South of France.
The weather data used for the CIBSE TRY/DSY files is not however based upon the
interpolated Perry and Hollis data set, it is based upon real observations of data and hence is
limited in the number of weather stations that record all the necessary weather variables. As
mentioned previousl TRY/DSY tpe files are avaiable for 14 loactions arround the UK. Figure 3
shows the location of each of the weather stations used to create the CIBSE files. Since for
compliance purposes thermal modelling needs to use the geographically closest CIBSE weather
file we can draw an area of influence on the map as well. As Figure 3 shows the area
represented by each weather file varies considerably. Due to the limited availability of data
some areas of the UK are better covered than others, for a country with such a large variation in
temperatures over such a small distance this presents a problem.
Figure 3 Map of the UK on a 5km grid overlaid with each of the 14 CIBSE TRY/DSY locations and their
range of influence. The legend shows how many 5km grid squares are within each range of influence.
Comparison between Figure 1 and Figure 3 shows that there is likely to be significant
discrepancy between the weather data in the CIBSE file and the weather where the building is
actually located. The magnitude of this difference is shown in Figure 4, as we can see the
majority of the UK is cooler than the mean temperatures of the 14 TRY/DSY locations.
Weather Files for Current and Future Climate
Page 5 of 37
Figure 4 Map of the UK on a 5km grid showing the difference in temperature between each 5km grid
square and the associated CIBSE TRY/DSY location. The legend shows the difference in temperature and
the number of grid squares in each temperature range (based on 1961-1990 data).
vi) Considerations when using Reference Years
As shown in Figure 4 the use of a limited number of locations for reference weather years can
result in a significant discrepancy between the weather file temperature data and the
temperatures at a given location. This is partly due to the relatively small number of files
available for the UK and also the locations of the weather stations used to create the CIBSE
TRYs and DSYs are generally located next to cities. This may seem to be a good idea since this
is where the majority of people live and where the majority of buildings are located. However,
the weather stations are typically located on hills on the edge of the city, in parks or on the tops
of buildings. This means that while the temperatures represented in the weather file are warmer
then the surrounding rural areas they are cooler than the temperatures in the centre of the city.
This change in temperatures is due to the urban heat island effect, the reduction of green space
and improved drainage coupled with an increase of surface area to absorb solar radiation results
in urban areas being warmer than their surrounding rural areas. The urban heat island will be
covered further later on.
Another consideration when using reference weather years representative of the average
weather and climate over many years is that the weather file gives no indication of the range of
possible conditions. While this is not a massive problem for determining average energy use or
typical carbon emissions there are several issues when considering, peak heating loads, human
comfort or overheating levels.
The following figures show results of a thermal simulation of a typical office design adapted
from CIBSE TM36. Data shown for the Plymouth TRY and DSY as well as the base set of
weather years from which they were chosen (1983-2004), 2005-2007 are also shown for
reference. Figure 5 shows the annual heating energy requirement for the office (bars) and the
peak boiler load (squares) for the different weather years.
Weather Files for Current and Future Climate
Page 6 of 37
The figure shows that the TRY gives a reasonable estimate of the average heating energy used
but underestimates the peak load. The use of a single weather file in this case gives no
information about the variation between different years.
Figure 5 Plot of annual heating energy usage and peak boiler load for different weather years.
Figure 6 and Figure 7 show how the mean external and internal temperature vary with different
weather years. Remember that the DSY is chosen as the 3rd warmest summer period (April to
September inclusive) from the base set of weather years, for Plymouth 1990 is used as the DSY.
It is interesting to note that the DSY/1990 is the 3rd warmest outside for the period 1983-2004
but only 5th warmest internally falling to 6th if 2006 is included.
Weather Files for Current and Future Climate
Page 7 of 37
Figure 6 Plot of mean external temperature (dry-bulb) over the summer period for different weather
years.
Figure 7 Plot of mean internal temperature for different weather years.
Figure 8 shows the variation in hours of overheating for the different weather years. There is a
large variation in the hours of overheating between the different years. It is interesting to note
that the greatest levels of overheating do not come from the warmest years. This could be due to
a number of factors, the highest temperatures do not occur during occupied hours, cloud cover
Weather Files for Current and Future Climate
Page 8 of 37
limits solar gains during high temperature events in some years or alternately there is sufficient
ventilation during high temperature events in some years to cool the building. While the cause
of why 1983 has 300 hours more overheating than the DSY (an increase of ~60%) despite being
a cooler year externally would require further investigation, the implications are clear. The use
of a single reference year to estimate internal conditions and levels of thermal comfort and
overheating has its flaws, there is no information about the range of response of the building to
changes in weather and hence way to estimate the risk of building failure and the risk to
occupants.
Figure 8 Hours of overheating (>25°C) for different weather years.
Table 1 Implications of using representative weather files for building simulation.
Description
Only one or two files used
(e.g. TRY and DSY)
Pros
Simple, allows easy
comparison with other
buildings.
Based
on
previous
observations of weather at a
location.
Data is realistic, as it has
already happened.
Files are compiled from
several years of weather.
Files like the TRY are
representative of the average
weather and climate.
Only a few locations available
Easy to choose which file to
use.
Weather Files for Current and Future Climate
Cons
Can give unrealistic values for
energy usage and overheating
etc.
Limited data available to
create representative files.
The files are out of date; files
like the DSY no longer
represent a near extreme
warm year.
Average files like the TRY
ignore natural variation in the
weather, no information about
the range of variation.
Weather at file location is
unrepresentative of where
building is located.
Page 9 of 37
Climate change and impacts
i) Introduction
There is unequivocal evidence that the climate is changing. Predictions of the world’s climate
point to an increasingly warmer world, with greater warming across land and away from the
equator (IPCC, 4th Assessment Report, 2007). Climate is defined as the long-term averages and
ranges of different weather variables. As such climate change is a difficult concept for people to
deal with as generally they think in terms of weather, short-term variations and extremes. The
heat wave of 2003 and the associated deaths raised awareness of climate change (the frequency
and intensity of heat waves is predicted to increase) and showed that the built environment is an
area where adaptation measures are needed. Urban areas have long been known to exhibit their
own microclimate dependent on the climate of surrounding rural areas and the radiative heat
balance within the urban area. With the current emphasis on low energy buildings and
sustainability there is a need to understand how these buildings will fare in the future, will these
design continue to be sustainable and use little energy? Increased levels of insulation can save
heating energy now and into the future but at the cost of increased overheating risk,
consideration needs to be given to how the weather and climate may change and the potential
implications of this. The material covered in this lecture will touch upon these issues and
provide insight into both the climate change material available and how it can be used to assess
climate based risks to buildings and their occupants.
ii) Projections
IPCC:
there is a 50:50 chance that global warming will not exceed 2 °C for stabilisation of CO2-eq at
450ppm.
The statement of the Intergovernmental Panel on Climate Change (IPCC) that a stabilisation of
atmosphereic greenhouse gases at concentration equivalent to 450ppm of CO2 is often
misquoted and misunderstood. If we stabilise at these levels then there is a 50:50 chance that
global warming will be less than 2 °C, this also means that there is a 50:50 chance that global
warming will be greater than 2 °C. Currently we are at ~430ppm CO2-eq and rising.
Figure 9 shows the observed trend in atmospheric CO2 concentration measured at Mauna Loa
Observatory Hawaii. The yearly oscillation in CO2 concentration is due to trees, since the
majority of the Earths land mass is located in the northern hemisphere the CO2 concentration
increases over the winter months while there are less leaves on trees and reduces again in the
spring.
Weather Files for Current and Future Climate
Page 10 of 37
Figure 9 Measured trends in atmospheric CO2, source: Earth Systems Research Laboratory
http://www.esrl.noaa.gov/gmd/ccgg/trends/
Globally temperatures have increased 0.8°C since the late 19th century and have risen by 0.2°C
per decade over the past 25 years with the 10 warmest years on record all being since 1995. It is
very likely that most of the warming has been caused by anthropogenic greenhouse gas
emissions and with the current climate change mitigation policies the warming trend is likely to
continue with a global temperature rise of 4°C under the A1FI (high) emissions scenario.
However recent research has suggested that the actual temperature could be much higher
(Anderson, K. and Bows, A., 2008, Philosophical Transactions A, 366, 3863-3882).
Some level of climate change is inevitable even if CO2 emissions were to cease tomorrow. This
is due to CO2 and other green house gases remaining in the atmosphere for many years and the
fact that the climate system is slow to adapt to changes in green house gas concentrations.
Current CO2 emissions and typical climate change emissions scenarios are shown in Figure 10
the emissions scenarios were created to represent different socio-economic and political trends
and hence different levels of carbon emission. We are currently emitting CO2 at a higher rate
than the A1FI emissions scenario, commonly known as the ‘high emissions scenario’ or
‘business as usual’.
Weather Files for Current and Future Climate
Page 11 of 37
Figure 10 Plot of CO2 emissions predicted under different emissions scenarios overlaid with actual CO2
emissions. CO2 emissions are currently greater than the A1FI emissions scenario. Adapted from Climate
Change Scenarios for the United Kingdom: The UKCIP02 Scientific Report.
The levels of climate change projected by the IPCC have been adapted for the UK by the Met
Office Hadley Centre and the UK Climate Impacts Programme. Their findings are presented in
the UK Climate Impacts Programme 2002 Report (UKCIP02) and in the latest 2009 UK
Climate Projections (UKCP09).
The projected levels of climatic change presented in UKCIP02 and UKCP09 use the period
1961-1990 as a basis. The maps shown in Figure 11 indicate the change in mean annual air
temperature as projected by UKCIP02 and UKCP09 under the high emissions scenario by the
2080’s. Not only does the UKCP09 data offer better resolution but it also encompasses a greater
range of possible values.
Weather Files for Current and Future Climate
Page 12 of 37
Figure 11 Comparison of changes in seasonal mean temperature, summer and winter, by the 2080s under
High emissions scenarios, from the UKCIP02 report (far left panels) and as projected in UKCP09 (10,
50 and 90% probability level). (© UK Climate Projections, 2009).
Probabilistic projections of climate change such as those available from UKCP09 are created
using many runs of climate models. A probability density function (PDF) can then be created,
giving information about the range and relative likelihood of a certain amount of climate
change. This differs from the results presented in UKCIP02, which represent a single outcome.
A brief description of the idea of probabilistic climate projections follows, taken from the
UKCP09 Briefing Report (available from http://ukclimateprojections.defra.gov.uk/).
Weather Files for Current and Future Climate
Page 13 of 37
Figure 12 Illustration of how many climate models can give information about the probability of different
amounts of climate change. (© UK Climate Projections, 2009).
By ordering the results of the different model runs for example by change in mean temperature
it is possible to create a cumulative distribution function (CDF) which can be expressed as a
percentage. The different UKCP09 percentiles shown in Figure 11 are a representation of the
data contained within the CDF curve. The climatic change represented by the 10th percentile is
no more or less likely than the 90th percentile, (UKCP09 does not attribute likelihoods to
different model runs, each is equally likely) instead the different percentiles the position in the
CDF curve. If there are 10000 runs of climate models represented by a CDF curve then the 90 th
percentile is the data point for which there are 1000 results with greater climate change and
8999 with less climate change. The UK Climate Impacts Programme defines the different CDF
percentiles as follows: unlikely to be less than (10%), unlikely to be greater than (90%), 50% is
the central estimate or median, everything between 33% and 66% is generally considered to be
equally likely.
iii) Impacts
There are many possible impacts of climatic change such as increased levels of UV radiation
from reduced cloud cover, sea level rise, subsidence, increased intensity of storm events and
changing rainfall patterns. However, the main climate change impacts relevant to this lecture
series include; changes to mean, maximum and minimum temperatures, increased solar gains,
changes in relative humidity and changes wind speed/direction. Changes to wind direction was
not included in UKCIP02 and UKCP09 gives no information about wind speed or direction,
however, wind data can be derived from other variables [M. Eames, T. Kershaw and D. Coley,
2011, Building Serv Eng Res Technol. In Press].
Changes to these variables can have an effect on building energy usage and levels of
overheating which in turn influences human health, comfort and productivity. Figure 13 shows
possible changes in temperature of the warmest day and the warmest night. It is entirely
possible that these may both occur within a short amount of time during a heat wave. The effect
of the reduced diurnal temperature swing on the internal environment of a building could be
profound. There are also implications for human health with high temperatures sustained over
prolonged periods. During 2003 ~35000 people across Europe died of heat related illness
largely caused by the failure of buildings to cool adequately.
Weather Files for Current and Future Climate
Page 14 of 37
A typical adult human doing light work produces approximately 150W of heat, as temperatures
and relative humidity increases it becomes more difficult for the human body to lose heat either
by radiation or sweating. Hence core body temperatures can increase if heat loss is inadequate
resulting in heat stress and potentially death if prolonged. The termperature increases shown in
Figure 13 are a possible result of climate change. Since both of these events are likely to occur
during a short period of time (a heat wave) the implications for humans occupying buildings are
severe, especially for vulnerable groups such as the elderly or infirm.
Figure 13 Plots of change in temperature of the warmest day (left) and warmest night (right). Data
shown for the 2080’s, 90th percentile high emissions scenario.
There are two main ways to produce future weather files for use in building simulation
software: mathematical transformation of historical weather data (morphing) and using a
weather generator. Both of these methods have advantages and disadvantages. Weather data
from the period 1961-1990 can be mathematically transformed using the monthly estimates of
climate change available from UKCIP02/UKCP09. This morphing method produces future
weather years with the same weather patterns allowing the impacts of climate change to be
assessed independently of weather. However, this method has the disadvantage that if the
underlying weather is particularly hot or cold then the morphing method can exaggerate this to
the point where the future weather seem unlikely to be realistic. The weather generator can
produce future weather years for different future time periods and different emissions scenarios
representative weather for the period 1961-1990 can also be produced. The resulting weather
files will each have different weather patterns as well as different climate; this may well be
more realistic but also more complex to interpret.
Weather Files for Current and Future Climate
Page 15 of 37
Figure 14 Radiance image of the thermal model used for these examples presented in this lecture.
The house shown in Figure 14 is based upon a passivhaus design with air tightness, heating
energy and U-values conforming to passivhaus standards. The design utilises large south facing
windows to capture winter solar gains and brise soleil to limit summertime solar gains. Figure
15 and Figure 16 show plots of external and internal air temperatures respectively for different
projections of climate change.
Figure 15 Plots of external air temperature for two weeks in August for 2080’s future weather files
created using both the morphing method and using the weather generator. Both the 50th and 90th
percentiles are shown for each method.
Weather Files for Current and Future Climate
Page 16 of 37
Figure 16 Plots of internal air temperature for two weeks in August for 2080’s future weather files
created using both the morphing method and using the weather generator. Both the 50th and 90th
percentiles are shown for each method.
What is shown in Figure 15 and Figure 16 is that the morphing method produces future weather
patterns that are easy to compare with other emissions scenarios and current climate as the
weather is always the same, if the 10th August is hot in the base data then it will also be hot in
the future data as well. This makes the comparison of different climate change scenarios easier
and the impacts easier to identify. However, as you can see the morphing procedure can have a
tendency to exaggerate extreme events. If the 10th August is very hot then the 10th August in the
future weather files will be extremely hot, this is a side effect of the morphing algorithms,
which are included later on in these notes for completeness. The weather generator on the other
hand produces future time series of weather data that are all individual, this makes comparison
between scenarios and time periods difficult and the identification of climate impacts harder to
identify. However, since each time series is generated independently they are not prone to the
exaggeration exhibited by the morphed files. The inclusion of different weather patterns also
results in a somewhat more ‘realistic’ time series of data for modelling purposes as the response
of the building is tested by more than one set of data.
Weather Files for Current and Future Climate
Page 17 of 37
Figure 17 Histograms of internal temperature data for the house (living room) for the 1960-1990 base
line data and the 2080’s under the high emissions scenario, showing both the median and upper
percentiles for weather generator data and morphed weather data.
The histograms shown in Figure 17 show the response of the building to the different future
weather files. The histograms show the temperature in °C versus the number of hours in the
year than that temperature occurs. As the climate scenarios become more extreme the width of
the distribution of temperatures experienced becomes wider. The distribution is stretched not
just shifted. In the future we will not only have to deal with hotter temperatures but also a wider
range of temperatures.
Weather Files for Current and Future Climate
Page 18 of 37
Figure 18 Plot of Heating Degree Days for the Plymouth TRY and future weather years for high and
medium emissions scenarios at 10th, 50th and 90th percentiles.
Figure 19 Plot of Cooling Degree Days for the Plymouth TRY and future weather years for high and
medium emissions scenarios at 10th, 50th and 90th percentiles.
Weather Files for Current and Future Climate
Page 19 of 37
Figure 20 Plot of heating energy usage for the Plymouth TRY and future weather years for high and
medium emissions scenarios at 10th, 50th and 90th percentiles.
Figure 21 Plot of summertime overheating for the Plymouth TRY and future weather years for high and
medium emissions scenarios at 10th, 50th and 90th percentiles.
Figure 18 show changes in heating degree days (HDD), cooling degree days (CDD) and some
of the common metrics used for assessing building performance for a standard domestic house.
Weather Files for Current and Future Climate
Page 20 of 37
Three different percentiles of projected future climate are shown for different time periods for
the medium and high emissions scenarios. The figures show the level of uncertainty in the
climate change predictions, this is not to be confused with errors. Assuming climate models are
correct the actual amount of climate change will lie some where between the different points
indicated on these graphs. From the values of the 10th, 50th and 90th percentiles shown in the
figures we can deduce the shape of the distribution, its width and hence the range of likely
climate change impacts for which we must account for in building design. In order to access this
information however, we need to run more than a single file representative of future climate.
iv) Other Considerations
It has long been recognised that urban areas have their own climate and are typically warmer
than surrounding non-urban areas. Increased temperatures can lead to heat stroke and other heat
related illnesses especially amongst vulnerable groups such as the elderly. The 2003 heatwave
caused ~2000 extra deaths across England and Wales. The effect was greatest in urban areas. In
London the number of deaths of people aged over 75 increased by 59%.
Briefly, buildings store heat gained during the day, both from solar radiation and from human
related activity such as traffic exhaust and heat loss from buildings. Also generally, there is a
lack of trees and plants in cities (green space), which both provide shade and cool the
surrounding environment through the endothermic transpiration of water. Paved areas within the
urban area also increases run off reducing evaporation of water and improved drainage limits
the amount of blue space (ponds and lakes) in the urban area. The amount of heat absorbed
depends greatly upon the orientation of the buildings, their size and the materials used in their
construction as well as the amount of anthropogenic heat released from buildings and even
metabolic heat from people. Heavy weight materials such as concrete or stone with a high
admittance are capable of absorbing heat readily for storage and re-emission later. The rate of
absorption and re-emission depends upon the emissivity of the material not only in the visible
region but also in the infra-red regions. The majority of the heat that contributes to the urban
heat island comes from absorption of solar radiation which has approximately half of its power
located in the visible regions and the other half in the near infra-red region. This is known since
the magnitude of the urban heat island is similar in summer to that in winter, so we can deduce
that anthropogenic heat release in the urban area is a relatively minor contributor to the heat
island effect.
The stored heat is radiated at night artificially warming the urban area above surrounding rural
areas. Many buildings are designed to store heat in this way to regulate their internal
temperatures limiting heating (and cooling) energy required. The geometry of most urban areas
(tall buildings and narrow streets) means that buildings provide large areas for absorption of
heat but limit the ease with which the heat can be lost via radiation and convection (the “canyon
effect”). The temperature difference between an urban area and the non-urban (rural)
surroundings is referred to as the urban heat island (UHI), and has a maximum value at night
when air temperatures are at their minimum. Non-urban areas warm more quickly than urban
areas, but also cool more rapidly at night.
The urban heat island for London was measured to be ~2°C almost 200 years ago, 4-6°C in the
1960’s and was measured to be as high as 9°C during the heatwave of 2003. This increase is
due not only to the increase in energy used by buildings within urban areas but also typically
taller buildings that are more densely packed. This increase of the urban heat island and the
associated deaths is an indication of the failure of buildings and urban planning to regulate
internal temperatures within safe limits. It is important to note that climate change predictions
do not include any information about the urban heat island and that the UKCP09 weather
generator and historical observations of weather used in weather files typically do not contain
the urban heat island due to the fact that the observing weather station is usually located outside
the urban area.
Weather Files for Current and Future Climate
Page 21 of 37
There has been some research into the use of roof top gardens to cool buildings and negate the
effects of the urban heat island. The research has showed that the greening of a roof reduced
surface temperatures by up to 60°C and the heat flux through the roof due to solar radiation by
up to 366 kJ/m2 over a day; this in turn reduced internal air temperatures. Other studies have
considered the effects of increased urbanisation in Germany, the research considered the
reduced evaporation of water as a result of increased urbanisation (replacing green fields) and
equated this to an effective heat release of 50 TWh per annum per km2. Clearly this net increase
in heat present in the urban area will have profound effects on the urban heat island and on the
buildings within the urban area.
There are however remedies, in Ho Chi Minh City (Vietnam) urban planning guidelines are in
place for urban expansion to improve air quality and to remediate the urban heat island effect
within such a large and rapidly growing urban area. Sectors of the city have been laid out on a
grid with parks surrounded by low-rise buildings, with building size increasing radially. The
aim being that all buildings have a direct line of sight to the green space. This layout is intended
to allow buildings to shed heat effectively and to be exposed to the cooling effects of the green
space. This is a similar idea to the courtyards and pools present in Moorish architecture.
However, details of the effectiveness of these measures are scarce and the effects of such a
small area on the urban heat island of a city the size of Ho Chi Minh are currently unknown.
Since the magnitude of the urban heat island and projections of climate change are comparable
then adaptation measures for one should be applicable to the other. Increased use of airconditioning will only further exacerbate the problem so it may be possible to consider the
adaptation of the surrounding urban environment to cool buildings. Such adaptation measures
that increase the amount of green and blue space around buildings and potentially within
buildings could prove to be a cost effective measure against climate change by cooling the
surrounding area and providing shade. However, current thermal simulation software packages
do not allow for the explicit modelling of evaporation or green and blue space.
2.
Creation and Availability of Future Weather Years
The two sets of climate change data currently available for the UK are UKCIP02 and UKCP09
both created by the Met Office in conjunction the UK Climate Impacts Programme (UKCIP).
The UKCIP02 scenarios consist of a set of climate change data for the 2020s, 2050s and 2080s
for four different global carbon emission scenarios; low, medium-low, medium-high and high
emissions which are related to the original IPCC scenarios B1, B2, A2 and A1F1 respectively.
The baseline period to which the UKCIP02 climate change data relates is a simulated timeframe
representing the years 1961 to 1990. CIBSE sell future weather files created using the UKCIP02
data and the 14 TRY/DSY files, since the base data carries copyright the future files do also
since the weather patterns have not changed hence there is a cost attached to obtaining these
files. However, Southampton University have a free tool that can be used to morph TRY and
DSY files, this tool is available from the UKCIP02 website or from Southampton University.
Note you still need to have copies of the base TRY and DSY data.
Currently the only UKCP09 future climate files compatible with building simulation models
available is from the University of Exeter created as the result of a funded research project
(PROMETHEUS). The UKCP09 data as discussed earlier is probabilistic and can be used to
provide far more information than the UKCIP02 data. The method used to create these files uses
the UKCP09 weather generator rather than a morphing method so the resultant files are free of
copyright and are available for download from the PROMETHEUS website
(http://centres.exeter.ac.uk/cee/prometheus/downloads.html). The methodology used to create
these files and generate the missing variables is included later for completeness. Since the
UKCP09 weather generator has a resolution of 5km over the whole of the UK it is possible to
Weather Files for Current and Future Climate
Page 22 of 37
create many more weather files than possible if using observed data. Currently future weather
files are available for 35 locations around the UK and more will be added as they are created,
Figure 22 shows the distribution of available future weather files created by Exeter University.
The methodology for the creation of these files has been peer-reviewed and has been published
(M.Eames, T.Kershaw and D.Coley, 2011, Building Serv. Eng. Res. Technol. In Press).
Figure 22 Map of the UK showing the locations of available UKCP09 future weather files.
A single run of the UKCP09 weather generator outputs 3000 weather years of data. This
comprises of 100 samples of future climate each augmented with 30 years of generated weather.
Since the climate is the same for each sample it is possible to create TRY and DSY type files
from each set of 30 years using the methodology described earlier. This results in the creation of
100 TRY and 100 DSY type files, one for each sample of climate but representing either typical
weather patterns or a near extreme summer the same as the original CIBSE TRY/ DSY files.
To create probabilistic future weather files (the probabilistic information is the main benefit of
UKCP09 over UKCIP02) the 100 TRY and DSY files are disassembled into their constituent
months and ordered according to increasing monthly mean air temperature (Dry Bulb).
Probabilistic files are then assembled by selecting different percentiles, i.e. to create the 50th
percentile file, from the ordered Januarys the 50th file is selected and combined with the 50th
February, March etc, this process is termed pointwise intervals. This is done for 10th, 33rd, 50th,
66th and 90th percentiles to map out the whole distribution. The entire process can be
summarised by the flowchart in the appendix.
3.
Comparison between UKCP09 and UKCIP02
Here we show a comparison between the resultant files created using the different methods.
Figure 23 shows the distribution of heating energy requirement for the house. As can be seen
from the figure the UKCP09 weather generator TRY files show a distribution of monthly
Weather Files for Current and Future Climate
Page 23 of 37
heating energy requirement, which is similar to that produced by the CIBSE future TRY based
on UKCIP02. It also demonstrates that the 90th percentile TRY uses the least heating energy in
every month as expected due to warmer external air temperatures.
Figure 23 Plot of the monthly heating energy requirement of the building for the different weather
generator (WG) TRY files and the CIBSE future TRY (UKCIP02). Data shown is for London in 2050
under high emissions scenario.
Weather Files for Current and Future Climate
Page 24 of 37
Figure 24 Plot of the mean internal air temperature for the different future DSY and TRY percentiles
(black circles) and the equivalent CIBSE future files (blue circles). Data shown is for Edinburgh under
high emissions scenario.
The distribution of mean internal temperatures for all the different weather generator files and
the CIBSE future files for Edinburgh is shown by Figure 24. The shape of the distribution of the
different percentiles can be seen, indicating that the shape of the CDF curve of the sorted
weather generator data is translated into the internal environment in the thermal simulation. It is
interesting to note that the CIBSE future TRY produces results towards the higher percentiles of
the weather generator as shown by Figure 23 and Figure 24. This may be an indication of the
over estimation mentioned earlier as a result of the files being morphed since the climate may
have changed between the 1961-1990 period and the period from which the TRY is chosen,
typically 1983-2004.
4.
The morphing algorithms
Based on the UKCIP02 data output, Belcher et al [Building Serv. Eng. Res. Technol. 26 49
(2005)] have developed a methodology for transforming CIBSE TRY and DSY weather files in
to climate change years. Hourly CIBSE weather data for the current climate is adjusted with the
monthly climate change prediction values of the UKCIP02 emission scenario datasets. This
methodology is termed ‘morphing’. The basic underlying process for the morphing of the
weather files consists of three different algorithms depending on the parameter to morphed.
1) a ‘shift’ of a current hourly weather data parameter by adding the UKCIP02 predicted
absolute monthly mean change.
x = x0 + Δxm
where x is the future climate parameter, x0 the original present-day parameter and Δxm the
absolute monthly change according to UKCIP02. This method is, for example used for
adjusting atmospheric pressure.
2) a ‘stretch’ of a hourly weather data parameter by scaling it using the UKCIP02 predicted
relative monthly mean change.
x = αmx0
where αm is the fractional monthly change according to the UKCIP02 scenarios. This is
used for example to morph present-day wind speed values.
3) a combination of a ‘shift’ and a ‘stretch’ for current hourly weather data. In this method a
current hourly weather data parameter is shifted by adding the UKCIP02 predicted absolute
monthly mean change and stretched by the monthly diurnal variation of this parameter.
x = x0 + Δxm + αm(x0 – 〈x0〉m) = 〈x0〉m + Δxm + (1 + αm)( x0 – 〈x0〉m)
where 〈x0〉m is the monthly mean related to the variable x0 and αm is the ratio of the monthly
variances of Δxm and x0. This method is for example used to adjust the present-day dry bulb
temperature. It uses the UKCIP02 predictions for the monthly change of the diurnal mean,
minimum and maximum Dry T in order to integrate predicted variations of the diurnal
cycle.
Weather Files for Current and Future Climate
Page 25 of 37
It needs to be noted however, that the approach of morphing present-day weather data with
monthly climate change predictions misses details of potential future changes in diurnal weather
patterns or the extent of future extreme weather events such as heat waves etc. The morphing
algorithms used for each of the different weather variables can be found in the appendix.
5.
Creation of Future Years using a Weather Generator
Based on the UKCP09 probabilistic change factors over land, a stochastic climate change
weather tool has been produced by the University of Newcastle to generate weather files with
either a daily or hourly time series. The weather signal is created by using daily precipitation as
the primary variable while other variables are created using mathematical and statistical
relationships with daily precipitation and the previous day’s values. The general procedure uses
the baseline climate (1961 – 1990) to calibrate the weather generator rainfall model, then;
change factors are applied to generate the future precipitation. Finally the other variables are
calculated conditioned by the precipitation and appropriate UKCP09 change factors. The hourly
data is then disaggregated from the daily signal using relationships derived from observations. It
must be noted that although the climate change information is given on a 25 km grid, the
addition of a 5 km grid for the weather generator allows for changes in local topology and is
based on observations which have been spatially interpolated onto the same 5 km grid but does
not give any further climate information other than the 25 km grid.
The weather generator outputs nine variables for the daily signal: daily precipitation, maximum
temperature, minimum temperature, sunshine fraction, vapour pressure, relative humidity, direct
radiation, diffuse radiation and potential evapotranspiration (PET). Where as the hourly signal
contains the variables hourly precipitation, temperature, vapour pressure, relative humidity,
sunshine fraction, diffuse radiation and direct radiation. However, to create a weather file of the
same format as the CIBSE reference years, wind speed, wind direction, air pressure and cloud
cover need to be generated in a consistent manner with the rest of the weather signal. These
variables will be discussed in the appendix.
i) Creation of Future Weather Years
The key difference between the UKCIP02 and UKCP09 sets of projections is the use of
probabilistic information within UKCP09, such probabilities represent a random sampling of a
probability distribution function and hence, the likelihood of a certain amount of climate
change. At each location (25 km grid square), decade and emissions scenario, 10,000 equiprobable realisations (samples of the probability density function) have been generated relating
changes in climatic parameters. This makes the creation of future weather files more
complicated than UKCIP02 as now many future weather years can be realised from the vast
number of change factors available. The weather generator, for each run, randomly samples
from the 10,000 change factors available and creates a stationary thirty year time series, thirty
years of naturally varying weather with the addition of a single sample of climate change on
top. The UKCP09 weather generator will output 100 randomly chosen samples of climate data
chosen from the set of 10,000. One hundred samples results in an hourly time series of 3,000
equi-probable future weather years. However, it is difficult to visualise what the complete set of
3,000 years of data looks like and there is a large computational burden of using all 3,000 years
within modelling software so an appropriate method for the selection of future reference
weather years is required.
To create future weather years from the weather generator output a similar method is employed
to create test reference years and future design summer years. The UKCP09 weather generator
produces 100 sets of 30 years on a daily time series. Then using a disaggregation procedure an
hourly time series is then produced. Each set of thirty years, although stochastically produced to
include natural variability, is stationary with regard to the climate change signal incorporated
within it. This means that at the beginning of the weather generator run the future climate signal
Weather Files for Current and Future Climate
Page 26 of 37
is selected randomly from the probability density function of possible future climates including
the inter variable relationships between different weather parameters. So each sample of 30
years includes a different realisation of a future climate.
For each sample of 30 years, the design summer year is calculated in a similar manner to the
one used to order the observations. For each year in the set, the mean temperature from April to
September inclusive is calculated. The original design summer year was chosen as the third
warmest from a set typically of 21 years (i.e the 19 th) and is the 90th percentile. Since the
weather generator produces a larger set, to maintain the same relationship the 90th percentile is
chosen, but in this case this is the fourth warmest April to September period in the thirty year
set. The CIBSE test reference year method considers the most average months based on daily
means of dry bulb temperature, global solar radiation and wind speed using the FS statistical
method. Using the weather generator the mean daily dry bulb temperature is calculated by
taking the average of the daily minimum temperature and the daily maximum temperature and
the wind speed is calculated from PET using a rearrangement of eqn. 1. For most of the original
test reference years, global radiation was not available and was calculated from solar radiation
models and the cloud cover and thus, originally cloud cover is used within the FS statistics. This
is similar in the weather generator where global solar radiation is not a direct output of the
weather generator but calculated from solar models and sunshine hours and therefore the
variable sunshine hours is used within the FS statistic for our method.
By finding the most average 12 months within each set of thirty years (in the case of the TRY),
the tails of the distribution (extremes) of natural variability are removed leaving an idea of the
climate signal in an average year while making an estimate of a near extreme year (design
summer year) gives an idea of the extremes of natural variability on top of the climate signal.
The end result is a set of 100 test reference years and 100 design summer years, one from each
of the thirty-year samples, each with a different climate signal.
ii) Creation of probabilistic weather data
When the weather generator samples from the complete probability density function, the
climate change factors (differences between the base climate from 1961 to 1990 and the
required future time period) include many inter-variable relationships. For the weather generator
this includes monthly, seasonally and annual changes to precipitation, relative humidity, mean
air temperature, maximum air temperature, minimum air temperature and total cloud cover. For
a realisation of a future climate these inter-variable relationships must be maintained. Also a
probabilistic weather file must give careful consideration to changes over the entire year. For
instance, considering just air temperature; a yearly change in mean air temperature does not
give an idea of the distribution of the change in mean monthly air temperature. An overall
warming of 2°C over the year could be the result of all months being slightly warmer or a colder
first six months combined with extreme positive change factors for the second six months could
lead to the same overall yearly change. This could raise problems when the weather files are
used, especially if only part of the year is under consideration, i.e. summer overheating.
To create probabilistic weather years, which maintain the climate signal and keep a consistent
weather signal over the year, appropriate percentiles can be chosen using pointwise intervals for
each month using a single climatic variable. If each variable is ranked separately and combined,
although the central estimate for each variable would be used, it would discount joint
probabilistic information. i.e. combining the 50th percentile of mean monthly temperature
change with the 50th percentile of relative humidity change could be incredibly unlikely and
could give an unrealistic climate change. In this paper we will consider pointwise intervals on
the mean monthly temperature only. It is just as valid to use any of the other variables as
discussed above dependent on which variable is required for the analysis of risk but for
simplicity we are considering mean temperature only. For both the sets of 100 design summer
years and 100 test reference years each month is ordered according to the mean monthly
Weather Files for Current and Future Climate
Page 27 of 37
temperature (the average of the mean daily temperatures over that month), ranked from lowest
to highest and then the required percentile (e.g. 50th, 90th etc) are selected. The process is
repeated for each month with data selected at the required percentile to produce either a
probabilistic test reference year or probabilistic near extreme weather year as required. For
every month the mean monthly temperature is ordered and the required percentile is selected.
The other variables, which correspond to this value of mean monthly temperature, are then also
taken as the complete data file for the month. The months with the corresponding hourly time
series are then joined together to form the future weather year using the method outlined by
Levermore, i.e. from the set of 100 TRY files ordered by mean monthly temperature the 50th
percentile January is joined to the 50th percentile February etc to create a composite year. Using
this method the temperature series show a clear trend where the mean monthly temperature is
always greatest for the 90th percentile and the 50th percentile is the central estimate and always
above the 10th percentile. However, no trend will be found for the other weather variables. This
method gives temperatures across the year, which are consistent (i.e median January followed
by a median February etc) but maintains the climate signal of the variables. Using this method it
is extremely unlikely that the concurrent months occur from the same generated test reference
year or design summer year (with 100 different samples, there is a 1 in 10,000 chance of a
median January followed by a concurrent median February) but does give a good indication of a
median temperature change for the future. This method ensures that the whole year is ‘median’
(50th percentile) eliminating extremes for the given percentile. In this method using a high
percentile test reference year such as the 90th gives an indication of the extent of likely future
warming (UKCP09 defined the 90th percentile as unlikely to be greater than) and likewise the
10th percentile give an idea of the likely minimum change (unlikely to be less than) with current
theories and models. Using this method a 90th percentile design summer year gives a near
extreme weather year in terms of natural variability (90th percentile of natural variability) with a
near extreme climate signal.
iii) Implications for Low Carbon Building Design
Table 2 Pros and Cons of using future weather files to model buildings and possible implications for low
carbon building design.
Description
Weather files available for
future climate.
More locations available.
Files are probabilistic.
The files available in TRY
and DSY format.
Distributed freely.
Pros
Allows design decisions to be
made to account for future
conditions. Can design
buildings to be low carbon in
the future as well as now.
Weather files more
representative of where the
building is located.
Allows the full range of
possible climate change to be
explored. Risk based analysis.
Allows easy comparison with
files used to show
compliance. Files contain
average weather allowing
probability levels to explore
climate only.
Easily available, no cost
except that of time for extra
simulations.
Weather Files for Current and Future Climate
Cons
Extra files can be confusing.
The concept of probabilistic
climate change can be
confusing.
Representative weather files
lose information about the
range of possible weather.
Heatwaves, cold spells etc.
Page 28 of 37
6.
Appendices
i) Typical Metrological Year Creation
Step 1 - For each month of the calendar year, five candidate months with cumulative
distribution functions (CDFs) for the daily indices that are closest to the long-term (15 or 30
years depending on location) average CDFs are selected. Candidate monthly CDFs are
compared to the long-term CDFs by using the following Finkelstein- Schafer (FS) statistics
(Finkelstein and Schafer 1971) for each index.
n
( )#"
FS = 1 n
i
i=1
where,
δi = absolute difference between the long-term CDF and the candidate month CDF on any given
day.
! in a month.
n = the number of daily readings
The FS method is superior to just using means to chose the most average as it chooses months
with less extreme values that have a cumulative distribution function closer to that of all the
years considered. Since some weather parameters are judged more important than others, a
weighted sum (WS) of the FS statistics is used to select the five candidate months that have the
lowest weighted sums.
WS = " w i FSi
where,
wi = weighting for index
FSi = FS statistic for each weather parameter.
!
Table 3 TMY3 FS statistic weighting factors for different weather parameters.
Global radiation
5/20
Direct radiation
5/20
Wind velocity (maximum)
1/20
Wind velocity (mean)
1/20
Dew point (maximum)
1/20
Dew point (minimum)
1/20
Dew point (mean)
2/20
Dry bulb temperature (maximum)
1/20
Dry bulb temperature (minimum)
1/20
Dry bulb temperature (mean)
2/20
Weather Files for Current and Future Climate
Page 29 of 37
Step 2 - The five candidate months are ranked with respect to closeness of the month to the
long-term mean and median.
Step 3 - The persistence of mean dry bulb temperature and daily global horizontal radiation are
evaluated by determining the frequency and length of runs of consecutive days with values
above and below fixed long-term CDF percentiles. For mean daily dry bulb temperature, runs
above the 67th percentile (consecutive warm days) and below the 33rd percentile (consecutive
cool days) were determined. For global horizontal radiation, the runs below the 33rd percentile
(consecutive low radiation days) were determined. The persistence criteria excludes the month
with the longest run, the month with the most runs, and the month with zero runs. The
persistence data are used to select from the five candidate months the month to be used in the
TMY. The highest-ranked candidate month from Step 2 that meets the persistence criteria is
used in the TMY.
Step 4 - The 12 selected months were concatenated to make a complete year and
discontinuities at the month interfaces were smoothed for 6 hours each side using curve
fitting techniques.
ii) International Weather Years for Energy Calculations (IWECs)
Creation
The weighting actors for the IWEC files are slightly different from those used for the TMY
files, the weightings for differtn weather variables are shown in Table 4
Table 4 Weather parameters and their relevant weightings used for the ASHRAE IWECs.
Maximum dry bulb temperature
5/100
Minimum dry bulb temperature
5/100
Mean dry bulb temperature
30/100
Maximum dew point temperature
2.5/100
Minimum dew point temperature
2.5/100
Mean dew point temperature
5/100
Maximum wind speed
5/100
Mean wind speed
5/100
Total horizontal solar radiation
40/100
Weather Files for Current and Future Climate
Page 30 of 37
iii) Test Reference Years (TRY) and Design Summer Years (DSY) Creation
The candidate, most average months for the TRY were assessed from the sum, FSsum, from their
minimum FS statistics, FSmin, to give a weighted index for selection combining all three
parameters (DryT, GlRad and WS).
FSsum = w1FSmin (DryT) + w 2 FSmin (GlRad) + w 3 FSmin (WS)
where w1, w2 and w3 are weighting factors for each weather parameter. The weighting factors
add up to unity and the exact values are chosen depending on each parameters relative
!
importance.
Since the candidate month with the lowest FSmin for WS might not have the lowest
FSmin for DryT the sum is taken and each candidate most average month is input into eqn. 1. The
most average month is the one with the lowest FSsum, and hence the most average for the three
parameters considered. This is done for each month of the year in turn. For the TRY files
created for the UK a value of 1/3 was chosen for each weighting factor. The deviation of these
weighting factors from those used for TMY and IWEC files is due to the prevalence of natural
ventilation in the UK. Natural ventilation is less common in North America and humidity is
more important than in the UK, the equal weightings for the CIBSE data were believed
adequate, as natural ventilation, solar gain and fabric heat transmission were all considered
equally important factors. Also it is difficult to be too prescriptive with the weightings for the
different parameters since buildings will have different sensitivities to each parameter and hence
their relative importance varies.
In order to make the TRY usable the 12 most average months need to be combined. Often there
is a mismatch in the parameters at the join between different months. The method used to
overcome this discontinuity simply takes out the middle four values for each parameter in the
weather file between 21:00 and 02:00 and replaces with a linear interpolation for 22:00, 23:00,
00:00 and 01:00.
Due to the nature of the data collection data points in the files will some times be missing from
the desired most average month. Where missing values are found in a month that had less than
15% missing values (108 hourly values in a 30 day month) in total for that variable,
interpolations were used to fill the gaps. Different interpolation techniques were used depending
on the nature of the data and the size and place of the gap. Special care has to be taken is a
parameter maxima or minima is likely to occur within the gap. If there were gaps of more than
15% of any single variable in a month, then all the values for this variable were assumed to be
invalid. Further details of the interpolation methods and other error checking and correction,
along with a more detailed description of the data selection procedures for the TRY data files
can be found in the report by GL Levermore and JB Parkinson. “Analyses and algorithms for
new Test Reference Years and Design Summer Years for the UK”, Building Serv. Eng. Res.
Technol. 27 311 (2006).
iv) The Morphing Algorithms
This section contains the different morphing processes involved for each of the weather file
parameters. The notation for variables is consistent with CIBSE Guide J for weather data
variables and with UKCIP02 for climate variables. It should also be noted that this methodology
can also be used with the probabilistic climate change values output from UKCP09 once all the
necessary parameters have been calculated.
Solar irradiance on horizontal, gsr (Wm–2h):
The UKCIP02 scenarios give an absolute increment for the monthly average solar shortwave
flux received at a surface. Note, that the CIBSE solar irradiance is the integrated irradiation over
one hour, so that the units are Watts per metre^2 hours (Wm–2h). The UKCIP02 shortwave flux
Weather Files for Current and Future Climate
Page 31 of 37
is the total increase in monthly mean irradiation. This is the variable corresponding to solar
irradiance on the horizontal plane in the weather files. However, for the morphing procedure we
need to stretch the data not shift it (use method (2)) but not uniformly otherwise the sun would
shine at night! The appropriate scaling factor can be obtained from the absolute change and the
monthly mean from the observed base climate.
"gsrm = 1+ (#DSWFm / gsr0 m )
This scaling factor is then applied to all the months m in the time series using method (2) above
to give:
!
gsr = "gsrm # gsr0
This transformation gives the correct absolute increase in monthly means for the morphed
timeseries. Note that according to this method there is increased solar radiation of sunny days
! is not increased.
but the number of sunny days
Diffuse solar irradiation on horizontal, dsr (Wm–2h):
The UKCIP02 scenarios do not give information regarding the change to diffuse irradiation,
dsr, so an indirect method must be used. It is assumed that dsr varies according to gsr.
dsr = "gsrm # dsr0
Sunshine duration: radiation site sfr (h):
Sunshine duration is not a variable given directly by UKCIP02. The related variable that is
given is total cloud cover !
in longwave, TCLW, which is related to the weather file variable
cloud cover cc. sfr is obtained from the morphed time series of cc using an empirical
relationship between cc and sfr (given in CIBSE Guide J).
sfr = a(0) + a(1) " cc + a(2) " cc 2
Sunshine duration, synoptic site sfs (h):
! as sfr above.
This variable is adjusted
Cloud cover, cc (oktas):
Cloud cover is generally observed visually using judgement and recorded on an integer scale of
0–8 in oktas (1/8 of the sky). In UKCIP02, the variable is recorded as a percentage of TCLW,
and the increment is given as an absolute amount in percentage sky covered. The first step is to
convert the UKCIP02 percentage change to oktas.
(
"cc m = int "TCLW m # 8
)
100
Where ‘int’ denotes an integer value. This increment is then added to the timeseries using
method (1).
!
cc = cc 0 + "cc m
Dry bulb temperature dbt (°C):
! daily mean temperature, TEMP, daily maximum temperature,
UKCIP02 gives changes for
TMAX and daily minimum temperature, TMIN. These three parameters are used to change two
statistical parameters of the timeseries of temperature, namely the mean and the variance. This
Weather Files for Current and Future Climate
Page 32 of 37
is achieved by shifting by the UKCIP02 value for mean temperature and stretching by the
diurnal range TMAX – TMIN. The required scaling factor required for the stretch is:
"dbt m =
#TMAX m $ #TMIN m
dbt 0 max m $ dbt 0 min m
The required shift is the UKCIP02 increment ΔTEMPm. As such the required transformation is:
! dbt = dbt 0 + "TEMPm + #dbt m $ (dbt 0 % dbt 0 )
m
It can be confirmed that this transformation preserves the UKCIP02 changes to TEMP and
TMAX – TMIN (but not TMAX and TMIN independently). Since the changes in TMAX and
TMIN are not
! large this does not unduly bias the morphed data.
Wet bulb temperature, wbt (°C):
To obtain wbt, dbt is combined with either the specific or relative humidity. Note that neither of
these latter two quantities is included explicitly in the existing weather files, while increments
for both quantities are given in UKCIP02. Here we shall use specific humidity, the first step is
to calculate a historical time series for specific humidity s0.
Use dbt0 and wbt0 from the existing timeseries to derive the historical timeseries for moisture
content g0 and from this the historic timeseries for specific humidity s (in g of water per kg of
moist air). Next use method (2) to stretch the humidity data s to derive a future timeseries for s.
The UKCIP02 changes in specific humidity SPHU are given as a percentage, therefore:
s = (1+ SPHU m, /100) " s0
Next, calculate the future timeseries for moisture content g from s. The future timeseries for wet
bulb temperature wbt can then be calculated using g and dbt (same as for dbt above). The
psychometric formulae are given in CIBSE Guide C.
!
Atmospheric pressure, atpr (mb):
This is calculated by simply shifting the present-day timeseries by the UKCIP02 projections for
increment in atmospheric pressure, MSLP:
atpr = atpr0 + MSLPm
Wind speed, ws (m/s):
!
Wind speed changes in UKCIP02, WIND, are given as a percentage. Hence the new wind speed
time series is obtained using method (2):
ws = (1+ WINDm /100) " ws0
Wind direction, wd (degrees):
!
Since it is assumed that there is no change in the underlying weather, there is no change to the
wind direction.
Rainfall amount, ra (mm):
Rainfall changes in UKCIP02 are given as a percentage, so using method (2):
ra = (1+ PRECm /100) " ra0
Rainfall duration, rd (h):
No change.
!
Weather Files for Current and Future Climate
Page 33 of 37
Present weather code, pwc:
No change.
Solar altitude, solalt (degrees):
No change.
v) Creation of Future Weather Files Using the UKCP09 Weather
Generator
The calculation of missing variables and the creation of future weather files from the UKCP09
weather generator is discussed below.
Wind speed:
Although there is no wind information as a direct output from the weather generator, the
calculation of PET requires wind speed. A variant of the Penman-Monteith PET developed by
the Food and Agricultural Organisation is given by,
eqn. 1
PET =
900
U (e # e )
T + 273.16 2 a d ,
" + $ (1+ 0.34U 2 )
0.408" ( Rn # G) + $
where Rn is the net radiation at the crop surface, G is the soil heat flux, T is the mean
temperature,
! ea is the saturation vapour pressure at the air temperature, ed is the actual vapour
pressure, Δ is the slope of the vapour pressure curve, γ is the psychrometric constant and U2 is
the wind speed at a height of 2m. Since all other variables are either provided as an output or
can be calculated, a simple rearrangement of eqn. 1 can give the daily mean wind speed
consistent with the daily weather signal.
For the purpose of thermal modelling of buildings an hourly wind speed is required. This can be
achieved by comparing the calculated wind speed with daily mean wind speed calculated from
hourly observations at the given location. The hourly-observed signal, which corresponds to the
observed mean wind speed that best matches the calculated value from PET is inserted in to the
weather file. Although instantaneous values will not necessarily be maintained, the daily,
monthly and yearly averages are consistent.
Wind direction:
The outputs of the weather generator have no indication of wind direction and without
information about the pressure systems dominating the generated weather it is impossible to
calculate the pressure from first principles. To generate a wind direction signal, observational
relationships between wind speed, season and wind direction are used to create a probabilistic
distribution at the weather station located closest to the weather generator grid square. It is
unknown how weather systems will change in the future or the position of storm-tracks will
move in relation to the United Kingdom. Current consensus suggests that weather patterns will
not change vastly in the future and natural variability is much greater than the effects of climate
change. Therefore basing the future wind direction on a probabilistic distribution from the
observations is justified. In this method the wind direction is randomly generated every six
hours from the probabilistic distribution based on the season and the hourly wind speed. The
missing data is then linearly interpolated to generate a full hourly time series. The choice of
Weather Files for Current and Future Climate
Page 34 of 37
every 6 hours is used as a compromise to maintain the probabilistic wind direction distribution
and to allow realistic hourly changes and is based on a spectral analysis of observed UK wind
data.
Figure 25 shows the yearly-observed wind direction distribution for the weather generator base
climate (1961 to 1990), Chartered Institute of Building Services Engineers (CIBSE) test
reference year wind direction distribution and a generated wind direction. The generated wind
direction is created by stripping out the wind field (wind direction and wind speed) in the test
reference year and using the methods described above to recreate an hourly wind speed and
wind direction. Although there are minor differences at each location, the generated wind
direction matches the observed weather signal well. The differences are due to the simplistic
method used where one hour is not dependent on the next and because the interpolation does not
follow the same probabilistic distribution (although this makes very little difference to a thermal
model in the UK since the probability of the wind coming from any one direction is small as
shown in Figure 25). The differences however are small and are not found to be statistically
significant. Both the test reference year and generated wind field are just snap shots of the wind
field and in the case of the generated wind field are biased by the calculated hourly wind speed.
It must also be noted that the TRYs used are based on observations but are based on a different
period (typically 1983 to 2004) compared to the base period of the weather generator. The
differences in this case are even smaller, showing that weather patterns did not change
substantially in this period and the differences result from the natural variability of the chosen
observations, which is a subset of the true distribution.
Figure 25 Polar plots of wind direction against probability for observed weather (1961 – 1990), CIBSE
test reference year and generated wind directions for four different locations from across the United
Kingdom.
Weather Files for Current and Future Climate
Page 35 of 37
Air pressure:
Similar to the wind direction, the weather generator does not report air pressure. The air
pressure has little effect on thermal modelling of buildings in terms of overheating risk (since
most equations work as a pressure difference across a building) although it does influence the
absolute humidity and condensation risk. It is possible to generate a probabilistic value for air
pressure from observations but this leaves the questions of over what time scale should the
value be generated (the periodicity of the pressure systems) and on what variables should the
pressure be based, as there is no set speed with which pressure systems move across the UK. As
stated above, it is unknown how the pressure systems and the position of storm tracks will
change with regard to climate change and therefore it is impossible to also know how pressure
will relate to the future climate, therefore, for simplicity the pressure will be generated from a
distribution relating pressure to the daily mean wind speed and time of year obtained from
observations (1961 – 1990). The periodicity of the generation however is much harder to
determine.
Examining the pressure oscillations present in historical observations for Cardiff we noted that
there are both high frequency and low frequency oscillations in atmospheric pressure. For our
simple model we are not concerned with the higher frequency signal but the time between the
peaks and troughs of the underlying signal. Smoothing the signal using an FFT (fast Fourier
transform) filter removing any oscillation of the order of 2.5 days (60 hours) and below was
found to give a good representation of the true pressure. The average time between the peaks
and troughs (average half-wavelength) of the smoothed data is found to be 125 hours. Similar
analysis performed on other locations gave an average time of 116 hours for Edinburgh, 104
hours for London and 117 hours for Manchester. This implies a value between 4.3 and 5.2 days
can be used in the simple model. For simplicity we have considered that a pressure randomly
generated every 5 days (120 hours) with the probability of a given air pressure occurring with
the daily wind speed and season calculated from observations is adequate for use in building
thermal simulation codes.
Cloud Cover:
Muneer has derived empirical relationships between the sunshine fraction and cloud cover in
the UK. Using the sunshine fraction from the hourly signal, the cloud cover during daylight
hours can be generated with 8 Oktas implying no sunshine in a given hour and 0 Oktas when
there is a full sunshine hour. All other hours are linearly interpolated.
Weather Files for Current and Future Climate
Page 36 of 37
Figure 26 Flow diagram showing the method of creating future probabilistic weather years from the
UKCP09 weather generator
Weather Files for Current and Future Climate
Page 37 of 37