See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/245383326

New algorithm for generating hourly temperature values using daily

maximum, minimum and average values from climate models

Article  in  Building Service Engineering · August 2007

DOI: 10.1177/0143624407078642

CITATIONS READS

35 2,946

2 authors:

D.H.C. Chow G. J. Levermore

University of Liverpool The University of Manchester
37 PUBLICATIONS   318 CITATIONS    93 PUBLICATIONS   1,489 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

The development and assessment of the residential buildings energy code in the Kingdom of Saudi Arabia View project

All content following this page was uploaded by D.H.C. Chow on 03 March 2017.

The user has requested enhancement of the downloaded file.

 Building Services Engineering
Research and Technology http://bse.sagepub.com/

New algorithm for generating hourly temperature values using daily maximum, minimum and
average values from climate models
D.H.C. Chow and Geoff J. Levermore
BUILDING SERV ENG RES TECHNOL 2007 28: 237
DOI: 10.1177/0143624407078642

The online version of this article can be found at:

http://bse.sagepub.com/content/28/3/237

Published by:

http://www.sagepublications.com

On behalf of:

The Chartered Institution of Building Services Engineers

Additional services and information for Building Services Engineering Research and Technology can be found at:

Email Alerts: http://bse.sagepub.com/cgi/alerts

Subscriptions: http://bse.sagepub.com/subscriptions

Reprints: http://www.sagepub.com/journalsReprints.nav

Permissions: http://www.sagepub.com/journalsPermissions.nav

Citations: http://bse.sagepub.com/content/28/3/237.refs.html

Downloaded from bse.sagepub.com at University of Nottingham on July 27, 2011

Building Serv. Eng. Res. Technol. 28,3 (2007) pp. 237–248

New algorithm for generating hourly temperature

values using daily maximum, minimum and
average values from climate models
DHC Chow BSc BEng PhD MCIBSE MASHRAE and Geoff J Levermore BSc ARCS PhD DIC CEng DMS
FCIBSE MASHRAE
a
Division of Built Environment, Sheffield Hallam University, UK
b
School of Mechanical Aerospace and Civil Engineering, University of Manchester, UK

The use of building simulation programmes for predicting building performance is increasing
all the time especially with the advent of cheap, fast computers. Hourly weather data,
in particular outdoor dry bulb temperature (DBT) and solar radiation values, are required for
simulation programmes for building performance. When hourly values are not available
there are algorithms for generating hourly temperature values from daily values. These use
the daily maximum temperature TMAX, and daily minimum temperature, TMIN. However,
climate prediction models, such as HadCM3 and HadRM3 also provide the daily
average dry-bulb temperature TAVE as well as the daily maximum and minimum.
The average temperature is important for selecting weather years and also because the
average temperature is often different from the average of the maximum and minimum,
assumed in the simpler algorithms. Buildings being designed now will need to perform
under future weather conditions with climate change, so the downscaling of daily values
from climate prediction models to hourly values is required. This paper describes a new,
more accurate algorithm for generating hourly temperature values in the UK that uses all
three temperature parameters from climate change models, and demonstrates the
improvement of the quality of the generated values against traditional algorithms that use
just the daily maxima and minima.
Practical application: The proposed algorithm for generating hourly DBT values from
daily maximum, minimum and average values is intended primarily for deriving hourly data
for running building simulation programmes, as some weather stations and future climate
prediction models only provide daily values of weather parameters.
Climate change is affecting all aspects of human life, and as well as being affected by
climate change, buildings can also affect the degree of climate change, since well-designed
buildings will require less energy to run, thus minimising the amount of carbon dioxide
emitted. As the climate is predicted to change significantly in the next 100 years, if buildings
are designed to last, it is important for designers to know how buildings will respond and
perform then. Building simulation programmes are useful for this, but they require hourly
weather data which are not provided by most climate prediction models. By having a

Address for correspondence: DHC Chow, Built Faculty of

Development and Society, Environment Division, School of
Environment and Development, Sheffield Hallam University,
Owen Building 457, City Campus, Howard Street, Sheffield.
S1 1WB, UK
E-mail: D.Chow@shu.ac.uk
Figures 2–16 appear in colour online: http://bse.sagepub.com

ß The Chartered Institution of Building Services Engineers 2007 10.1177/0143624407078642

Downloaded from bse.sagepub.com at University of Nottingham on July 27, 2011

238 Algorithm for generating hourly temperature values

quality-assured algorithm for down-scaling the raw daily values to hourly values, data from
climate prediction models, such as HadCM3 and HadRM3 can be used for building
simulations for any location in the world.
The hourly DBTs derived from using the algorithm suggested in this paper may also be used
in conjunction with other hourly weather data, such as wet-bulb temperature, solar
irradiance, cloud cover and wind speed, derived from the same dataset.

1 Introduction temperature values to hourly values. It also

describes a new algorithm that uses values of
Building designers often now predict the the average daily temperature, TAVE, which is
performance of buildings using building sim- a weather variable available from climate
ulation programmes that require hourly models, together with daily maximum tem-
weather data. A number of UK weather perature, TMAX, and daily minimum temper-
stations provide hourly values of weather ature, TMIN. When the hourly temperatures
variables, such as dry-bulb temperature are symmetrical about the average value then
(DBT), global solar irradiance (GSR), and the average can be calculated from the mean
wind speed and direction. These data can be of the minimum and maximum. However, in
used for selecting test reference years (TRYs) the UK the weather is not often like this.
and design summer years (DSYs)1 for analy- Figure 1 shows the significant error in
ses of energy use for thermal comfort, plant using TMAX þ TMIN/2 to estimate TAVE for
sizing, and the performance of buildings in DBT data for Heathrow between 1976 and
hot summers. However, not all weather 1995. The error can be as large as
2.58C,
stations provide hourly data. Many provide and the average error per day is
0.58C
daily data, usually in the form of averages for regardless of the month of the year.
variables, such as wind speed and solar
radiation; and daily maxima and minima for
temperatures. To obtain hourly weather data, 2 Existing algorithms for generating
algorithms are required. hourly temperatures
This is particularly important as predicted
future weather data becomes available from Although daily minimum and maximum
climate prediction models, such as the UK values of temperature are given, their time
Met Office-developed Hadley models of occurrence is not. Many of the existing
HadCM3 and HadRM32 which provide algorithms to downscale daily temperature
daily data. Simulation with such data is data to hourly values place the maximum and
useful as buildings designed now will last minimum values at designated times during
between 50 and 100 years, and will have to the day, and then link them together by
cope with the future climate. HadCM3 and sinusoidal fits, as shown in Figure 2.
HadRM3 provide daily values for various
weather parameters. This is due to computa- 2.1 CIBSE method (1982)
tional constraint, especially as most models The CIBSE method3 algorithm (similar to
are global models. In order to make full use of the ASHRAE method4) for generating hourly
these predicted future weather data in build- DBT uses simply the daily maximum temper-
ing simulation programmes, algorithms for ature, TMAX, and daily minimum tempera-
downscaling daily values to hourly values are ture, TMIN. This method relies on the
required. This paper looks at current algo- strategically allocating times (tmax and tmin)
rithms for downscaling daily outdoor for which TMAX and TMIN occur in the day

Downloaded from bse.sagepub.com at University of Nottingham on July 27, 2011

 DHC Chow and GJ Levermore 239

3.00

2.00

1.00
Max pos. error
Error (°C)

Max neg. error

0.00
Average error band
Average error band
−1.00

−2.00

−3.00
1 2 3 4 5 6 7 8 9 10 11 12
Month

Figure 1 Monthly errors of predicting average daily temperature using the average of TMAX and TMIN for Heathrow 1976–1995

Sinusoidal fit for known Tmax and Tmin

22.00

20.00
Temperature (°C)

18.00 Known Tmax value

16.00

14.00

12.00
Known Tmin value

10.00
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Hour

Figure 2 Downscaling of temperature values using sinusoidal fits

(See Table 1) and uses two sinusoidal curves where: For t5tmin:
to fit the data.
Hourly temperatures, T(t) are calculated cosð
ðtmin  tÞ=24 þ tmin  tmax Þ þ 1
f1 ¼ : ð2Þ
using Equation (1). 2
For tmin5t5tmax:
TðtÞ ¼ f1  TMIN þ f2  TMAX ð1Þ
cosð
ðt  tmin Þ=tmax  tmin Þ þ 1
f1 ¼ : ð3Þ
where f1 and f2 are factors given in a table in 2
the guide, and TMIN and TMAX are the daily For tmin5t:
minimum and maximum temperatures respec-
tively. These are related to sinusoidal inter- cosð
ð24 þ tmin  tÞ=24 þ tmin  tmax Þ þ 1
f1 ¼
polations, and can be expressed more 2
mathematically in Equations (2)–(5). ð4Þ

Downloaded from bse.sagepub.com at University of Nottingham on July 27, 2011

240 Algorithm for generating hourly temperature values
Table 1 TMAX and TMIN times suggested by CIBSE Guide rather than treating them separately. This
A2 (1982)3
would produce a smoother transition between
Month TMAX time (tmax) TMIN time (tmin) days. The equation for calculating this is
January 14 6 given by Equation (6).
February 14 6

March 14 5 TempðnextÞ þ TempðprevÞ
April 15 5 TðtÞ ¼
May 15 4 2
June 16 4 "

July 15 4
August 15 5
TempðnextÞ  TempðprevÞ

September 15 5 2
October 14 6
November
December
14
14
6
7
#

ðt  tðprevÞ Þ
 cos ð6Þ
ðtðnextÞ  tðprevÞ Þ
and where Temp(next) is the next known
f1 þ f2 ¼ 1: ð5Þ temperature value (TMAX or TMIN);
Temp(prev) is the previous known temperature
value (TMAX or TMIN); t(next) is the time for
2.2 Sin (14R-1) method the next known temperature value; t(prev) is
The CIBSE method was adapted with the time for the previous known temperature
different designated times for TMAX and value; and t is the time.
TMIN. As solar radiation is the most signifi-
cant source of heating to Earth, the lowest
temperature often occurs just prior to sunrise. 3 Q-Sin method
This method is called the Sin(14R-1) method.
Using standard formulae for calculating Building simulations use TRYs for estimating
sunrise and sunset times5 for different loca- heating and cooling loads, and DSYs, for
tions according to the latitude of the site, determining the extent of any summer over-
sunrise hours are calculated, and the hour heating with natural ventilation. As the TRY
before sunrise occurs was designated to be the is the most average year and the DSY is
times when TMIN occurs. TMAX was assumed selected on its average summer temperature,
to occur at hour 14, regardless of the time of TAVE is very important for the selection
the year, as it takes typically 2 hours for air processes. For this a quarter-sine (Q-Sin)
temperature to react to the maximum solar method was developed to utilize values of
irradiation at noon. TMAX, TMIN and TAVE. To ascertain the
correct average value a point is selected
2.3 Linking days together between the TMAX and TMIN (at a
The above two algorithms just produce variable time) and quarter-sine waves fitted
data for one day only. If the days are simply between the three points, as shown in more
put together, as would be necessary in detail later. Hence this is referred to as the
constructing a year of data, there will often Q-Sin method.
be steps between the ends of days and the The values of daily maximum temperature,
starts of the next days. However, the TMAX of TMAX, and daily minimum temperature,
any day can also be linked, using a sinusoidal TMIN, are first allocated in hour 14 and the
curve, with the TMIN of the following day, hour before sunrise respectively. Then as with

Downloaded from bse.sagepub.com at University of Nottingham on July 27, 2011

 DHC Chow and GJ Levermore 241

the Sin (14R-1) method, hourly temperatures

are first generated using Equation (6). TMAX

There is a slight possibility, especially in

winter that with the linking of days some of
the generated hourly temperature will be
above the daily maximum temperature
TMIN
(TMAX), or below the daily minimum
value (TMIN). This can happen when TMIN
of the following day is higher than the
TMAX. For this reason, capping is intro- tmin tmax
duced, whereby if the generated value
is higher than TMAX, then the value is Figure 3 Half-sine wave between tmin and tmax
changed to TMAX; and if the value is lower
than TMIN, then it is changed to TMIN. The shaded area in Figure 3, Shade 1, is given
This will not affect the final generated value in Equation (8).
of TAVE, as the correction procedures to fit Z tmax

the given TAVE value occur later on. TMAX  TMIN
Shade 1 ¼
Originally, there will be a difference t 2
between the given value of TAVE and the min
 
ðt  tmax Þ

daily average of generated hourly data, so a  cos þ1
way of correcting the values in the gener- tmax  tmin


ated data is required. This can be done by TMAX  TMIN
¼  ðtmax  tmin Þ:
the application of the quarter-sine wave. 2
ð8Þ
3.1 Application of the quarter-sine wave If the time when the mid-point between TMAX
The quarter-sine wave is used between tmin and TMIN, (TMAX þ TMIN/2), changes
and tmax, since this is the only part of the day from midpoint between tmin and tmax,
where there is a half-sine wave. The temper- (tmax þ tmin/2), to a time, , as shown in
ature between the times of tmin and tmax is Figures 4 and 5, then two quarter-sine curves
generated as in Equation (7): would be produced as a consequence, hence

 the name ‘‘Quarter Sine’’, and the shaded area
TMAX  TMIN can be changed to fit the given value of TAVE.
TðtÞ ¼
2 The temperature between tmin and  is

  generated as shown in Equation (9):
ðt  tmax Þ


 cos þ 1 þ TMIN : TMAX þ TMIN
tmax  tmin TðtÞ ¼  TMIN
2
ð7Þ 
 
ðt  f þ ð  tmin Þg

Applying Equation (7) with only TMIN and  cos þ 1 þ TMIN
2ð  tmin Þ
TMAX, a half-sine wave is generated, as in

Figure 3. TMAX  TMIN
¼
The temperature values between tmin and 2
tmax can be manipulated so that the daily 
 
ðt  2 þ tmin Þ

average temperature would fit the TAVE given  cos þ 1 þ TMIN :
2ð  tmin Þ
in HadCM3 or HadRM3. The shaded
area in Figure 3 would alter as a result. ð9Þ

Downloaded from bse.sagepub.com at University of Nottingham on July 27, 2011

242 Algorithm for generating hourly temperature values

The shaded area in Figure 4, Shade 2 is given The temperature between  and tmax is
in Equation (10). generated as shown in Equation (11):

Z 


TMAX  TMIN TMAX þ TMIN
Shade 2 ¼ TðtÞ ¼  TMIN
2 2
tmin 
 
ðt  tmax Þ


   cos þ 1 TMIN
ðt  2 þ tmin Þ
2ðtmax  Þ
 cos þ1

2ð  tmin Þ TMAX  TMIN
¼

 2
TMAX  TMIN 
 
¼ ðt  tmax Þ

2  cos þ 1 þ TMIN : ð11Þ
2ðtmax  Þ


2 The shaded area in Figure 5, Shade 3, is given
 ð  tmin Þ  1  : ð10Þ in Equation (12).


Z tmax 

The situation from  to tmax is shown in TMAX  TMIN
Figure 5. Shade 3 ¼
2

 
ðt  tmax Þ

 cos þ1
2ðtmax  Þ


TMAX  TMIN
T MAX ¼  ðtmax  Þ
2
T MAX + T MIN


2
2  1þ : ð12Þ


T MIN
Figure 6 shows the difference between the
half-sine wave and the quarter-sine wave
methods. Using the quarter-sine wave and
tmin l tmax
varying the position of  allow the correct
Figure 4 Quarter-sine wave between tmin and  TAVE to be achieved with the given TMAX and
TMIN.

T MAX
T MAX
Quarter-sines
T MAX + T MIN
T MAX + T MIN
2
Half-sine
2
T MIN
T MIN

tmin l tmax
tmin l tmax
Figure 6 Difference between half-sine wave and quarter-sine
Figure 5 Quarter-sine wave between  and tmax waves

Downloaded from bse.sagepub.com at University of Nottingham on July 27, 2011

 DHC Chow and GJ Levermore 243

The main purpose of this is to find the 4 Comparison of algorithms

position of  to compensate for the differ-
ences between the value of TAVE given by the Having studied various conventional down-
climate model and the generated daily tem- scaling algorithms, together with the new
perature from the original downscaling using proposed algorithm, it is important to mea-
sinusoidal fits. sure the significance of improvement of the
If the difference between these two values is new algorithm, if there are any, to determine
D, the total amount needed to be amended is whether the extra procedures in the new
24  D since there are 24 hours in a day. The algorithm are justified.
shaded area in Figure 6 is the difference
between the 2-Quarter sine waves (using ) 4.1 Methods of comparing algorithms
and the single half-sine wave. The performance of the algorithms for
generating hourly temperature values can be
[ Total difference ¼ Equation ð12Þ analysed in a number of ways. In this study,
daily averages, maximums and minimums are
þ Equation ð10Þ extracted from 20 real observed years from
Heathrow, from 1976 to 1995. For each year,
 Equation ð8Þ hourly temperature values are generated from

 each of the algorithms discussed previously.
TMAX  TMIN The resulting sets of hourly values are then
[ 24  D ¼  ð  tmin Þ
2 statistically analysed in three ways:


 Comparing the root mean square of errors
2 TMAX  TMIN (RMSE), comparing the difference from the
 1 þ

2 real observed cumulative and comparing their

 average hourly values. The analyses are
2 conducted for each individual month of the
 ðtmax  Þ  1 þ

year, as some algorithms may be better for

 some months than others.
TMAX  TMIN
  ðtmax  tmin Þ 4.1.1 Comparing root mean square of
2
errors (RMSE)

 Each hourly value generated from the
ðtmax  tmin Þ 12D2

[¼  : algorithms, ŷ is compared to the correspond-
2 ðTMAX  TMIN Þ ing value from the real observed dataset, y.
ð13Þ For N entries, the RMSE is given in
Equation (14).
Having calculated where  is to be situated to vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
correct for the differences, hourly tempera- u N
u1 X 2
tures can be generated keeping the TMAX, RMSE ¼ t yi  y^i^ : ð14Þ
TMIN and also agreeing with the value of N i¼1
TAVE from the model.
There are occasions when  actually lies
outside the times between tmin and tmax. This 4.1.2 Comparing cumulative curves
can cause problems, and TMAX, TMIN or In order to compare the overall distribu-
TAVE would not be kept. However, these are tion of temperatures for each algorithm,
very rare and considered not to affect results cumulative plots are required for each
greatly. algorithm, and also that for the real

Downloaded from bse.sagepub.com at University of Nottingham on July 27, 2011

244 Algorithm for generating hourly temperature values

observed data. Absolute errors at each whole and Q-Sin algorithms for each individual year
degree for the cumulative distribution of the between 1976 and 1995. The performance of
generated data from real observed data are the new proposed Q-Sin algorithm is signifi-
summed and compared for each month. cantly better than the other algorithms for all
20 years, across all three analyses. It should
4.1.3 Comparing averaged hourly data also be noted that the algorithms where the
The third analysis of the algorithms days are linked together rather than treated
uses temperature at hour 1, hour 2, separately also performed better, having
hour 3, . . . hour 24 for each month (January, smaller errors.
February, March, . . . December). These are
compared with the real monthly averaged
values, and the sum of the errors from each 4.3 Effect of fitting average temperature
hour compared. Figures 10–12 show the performances of
the algorithms for each month, by averaging
4.2 Effect of linking days the errors for all 20 years for each month. The
Figures 7–9 show the average errors for the analysis of comparing cumulative plots
three tests for linked days and individual days (Figure 11) still shows the new proposed

1.80
CIBSE individual
1.70 days
Yearly average error (°C)

1.60
14R-1 individual
1.50 days
1.40 CIBSE linking days
1.30
1.20 14R-1 linking days
1.10
1.00 Q-Sin
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995

Year

Figure 7 Yearly RMSE analysis of hourly temperature algorithms for Heathrow 1976–1995

35%
Yearly cumulatve percentage error

CIBSE individual
30% days

25% 14R-1 individual

days
20%
CIBSE linking days
15%

10% 14R-1 linking days

5%
Q-Sin
0%
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995

Year

Figure 8 Yearly cumulative error analysis of hourly temperature algorithms for Heathrow 1976–1995

Downloaded from bse.sagepub.com at University of Nottingham on July 27, 2011

 DHC Chow and GJ Levermore 245

Yearly average hourly error (°C) 0.85

CIBSE individual
0.80 days
0.75 14R-1 individual
days
0.70
CIBSE linking days
0.65

0.60 14R-1 linking days

0.55
Q-Sin
0.50
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
Year

Figure 9 Yearly average hourly error analysis of hourly temperature algorithms for Heathrow 1976–1995

1.00
0.95 CIBSE individual
days
Monthly average error (°C)

0.90
0.85 14R-1 individual
days
0.80
0.75 CIBSE linking days
0.70
0.65 14R-1 linking days
0.60
0.55 Q-Sin
0.50
1 2 3 4 5 6 7 8 9 10 11 12
Month

Figure 10 Monthly RMSE analysis of hourly temperature algorithms for Heathrow 1976–1995

20.00%
Monthly cumulative percentage error

18.00% CIBSE individual

days
16.00%
14.00% 14R-1 individual
days
12.00%
10.00% CIBSE linking days
8.00%
6.00% 14R-1 linking days
4.00%
2.00% Q-Sin
0.00%
1 2 3 4 5 6 7 8 9 10 11 12
Month

Figure 11 Monthly cumulative error analysis of hourly temperature algorithms for Heathrow 1976–1995

Downloaded from bse.sagepub.com at University of Nottingham on July 27, 2011

246 Algorithm for generating hourly temperature values

Monthly average hourly error (°C) 0.60

CIBSE individual
0.50 days

14R-1 individual
0.40 days

0.30 CIBSE linking days

0.20 14R-1 linking days

0.10
Q-Sin

0.00
1 2 3 4 5 6 7 8 9 10 11 12
Month

Figure 12 Monthly average hourly error analysis of hourly temperature algorithms for Heathrow 1976–1995

2.10
CIBSE individual days
1.90
Monthly average error (°C)

1.70 14R-1 Individual days

1.50
CIBSE linking days
1.30
14R-1 linking days
1.10

0.90 Q-Sin (14R-1)

0.70
Q-Sin (CIBSE TmaxR-1)
0.50
1 2 3 4 5 6 7 8 9 10 11 12
Month

Figure 13 Revised monthly RMSE analysis of hourly temperature algorithms for Heathrow 1976–1995

algorithm having significantly fewer errors. changed to the CIBSE values for the Q-Sin
However, Figures 10 and 12 suggest that for algorithm, then results may improve.
two summer months (June and July), the For May, June, July and August, the times
CIBSE linked and individual days algorithm for TMAX were changed from hour 14 to 15,
performed better. This could be due to the 15, 16 and 15 respectively. The results are
difference in the designated times for the daily shown in Figures 13–15. This change appears
maximum temperature, TMAX. to give an over-riding success to the new
Figures 10 and 12 show that the 14R-1 proposed Q-Sin algorithm for downscaling
linking days algorithm generally perform daily temperature values to hourly values.
better than the CIBSE linking days algorithm, Figures 16 shows the maximum, minimum
with the exception of May, June, July and and average error range of hourly tempera-
August. The reason could be due to the ture for each month using the Q-Sin algo-
designation of TMAX times. If these were rithm. On average, the generated

Downloaded from bse.sagepub.com at University of Nottingham on July 27, 2011

 DHC Chow and GJ Levermore 247

35.00%
CIBSE individual days
Monthly cumulative percentage error

30.00%
14R-1 individual days
25.00%

20.00% CIBSE linking days

15.00% 14R-1 linking days

10.00%
Q-Sin (14R-1)

5.00%
Q-Sin (CIBSE TmaxR-1)
0.00%
1 2 3 4 5 6 7 8 9 10 11 12
Month

Figure 14 Revised monthly cumulative error analysis of hourly temperature algorithms for Heathrow 1976–1995

1.20
CIBSE individual days
Monthly average hourly error (°C)

1.00
14R-1 individual days
0.80
CIBSE linking days
0.60
14R-1 linking days
0.40
Q-Sin (14R-1)
0.20
Q-Sin (CIBSE TmaxR-1)
0.00
1 2 3 4 5 6 7 8 9 10 11 12
Month

Figure 15 Revised monthly average hourly error analysis of hourly temperature algorithms for Heathrow 1976–1995

temperatures fall within
0.58C throughout minimum and average temperatures,

the year, and absolute errors are within re-produced the hourly temperature better

2.08C, with the winter months being more than algorithms using simply the maximum
erroneous. This was expected, as the daily and minimum values.
profiles of winter temperature are more The specified times for when the daily
chaotic. maximum and minimum temperatures occur
are vital to the accuracy of the remodelled
hourly values. It was evident from this
5 Discussions and conclusion study that the best times for the daily
minimum temperatures is the hour before
From the three analyses, it can be concluded sunrise of the day, and the times for the
that the Q-Sin method, using daily maximum, daily maximum temperatures appear to be

Downloaded from bse.sagepub.com at University of Nottingham on July 27, 2011

248 Algorithm for generating hourly temperature values
Monthly range of differences
2.0

1.5

1.0
Difference in (°C)

0.5 Max
Min
0.0 Average
Average abs err +
−0.5 Average abs err −

−1.0

−1.5

−2.0
1 2 3 4 5 6 7 8 9 10 11 12
Month

Figure 16 Maximum, minimum and average error of generated hourly temperatures using Q-Sin algorithm for Heathrow 1976–1995

those described by the CIBSE Guide in require methods of downscaling from daily
Table 1. values to hourly values. Current and pro-
These revised times are not specific to the posed algorithms for these are to be covered
Q-Sin algorithm, and can also be applied to in a future paper as this paper concentrates on
improve tradition algorithms where only daily the more complex algorithm of downscaling
maximum and minimum temperature values daily DBT values to hourly values.
are available.
The use of this new algorithm is useful
particularly for generating TRYs and DSYs
for future time-slices, as most climate predic- References
tion models do not provide hourly values for
temperature, but the daily maximum, mini- 1 CIBSE Guide J Weather Solar and Illuminance
mum and average. Since the selection of Data, CIBSE, London, March 2002.
2 Hulme M et al. Climate change scenarios for
TRYs and DSYs use daily average tempera-
the United Kingdom: The UKCIP02 Scientific
tures, it would be useful to keep this variable Report, Tyndall Centre for Climate Change
when generating the resulting hourly values. Research, School of Environmental Sciences,
Since the improvement in the results is University of East Anglia, Norwich, 2002.
significant, it would provide a more accurate 3 CIBSE Guide A2 Weather & Solar Data, 1982.
simulation of the effect of climate change on 4 ASHRAE Handbook of Fundamentals,
buildings. Chapter 26, page 6, ASHRAE, Atlanta, 1993.
Building simulation programmes also 5 Sharmer K, Greif J. The European Solar
require hourly values of other weather para- Radiation Atlas Vol 2: Database and
meters such as wet-bulb temperature, solar Exploitation Software 2000; Paris, Les Presses
irradiance and wind speed. These will also de l’Ecole des Mines.

Downloaded from bse.sagepub.com at University of Nottingham on July 27, 2011

View publication stats