an analysis of future building energy use in subtropical hong kong

lable at ScienceDirect

Energy 35 (2010) 1482–1490

Contents lists avai

Energy

journal homepage: www.elsevier .com/locate/energy

An analysis of future building energy use in subtropical Hong Kong

Joseph C. Lam*, Kevin K.W. Wan, Tony N.T. Lam, S.L. WongBuilding Energy Research Group, Department of Building and Construction, City University of Hong Kong, Tat Chee Avenue, Kowloon, Hong Kong SAR, China

a r t i c l e i n f o

Article history:Received 27 May 2009Received in revised form1 December 2009Accepted 5 December 2009Available online 29 December 2009

Keywords:Principal component analysisRegressionEnergy consumptionOffice buildingsGeneral circulation modelsClimate change

* Corresponding author. Tel.: þ852 2788 7606; faxE-mail address: [email protected] (J.C. Lam).

0360-5442/$ – see front matter � 2009 Elsevier Ltd.doi:10.1016/j.energy.2009.12.005

a b s t r a c t

Principal component analysis of prevailing weather conditions in subtropical Hong Kong was con-ducted, and a new climatic index Z (as a function of the dry-bulb temperature, wet-bulb temperatureand global solar radiation) determined for past (1979–2008, measurements made at local meteoro-logical station) and future (2009–2100, predictions from general circulation models) years. Multi-year(1979–2008) building energy simulations were carried out for a generic office building. It was foundthat Z exhibited monthly and seasonal variations similar to the simulated cooling/heating loads andbuilding energy use. Regression models were developed to correlate the simulated monthly buildingcooling loads and total energy use with the corresponding Z. Error analysis indicated that annualbuilding energy use from the regression models were very close to the simulated values; the differencewas about 1%. Difference in individual monthly cooling load and energy use, however, could be up to 4%.It was also found that both the DOE-simulated results during 1979–2008 and the regression-predicteddata during 2009–2100 indicated an increasing trend in annual cooling load and energy use anda gradual reduction in the already insignificant heating requirement in cooling-dominated officebuildings in subtropical climates.

� 2009 Elsevier Ltd. All rights reserved.

1. Introduction

There is a growing concern about energy use and its implica-tions for the environment. Buildings, energy and the environmentare issues facing the building professions and energy policy makers.Recent reports by the Inter-governmental Panel on Climate Change(IPCC) have raised public awareness of energy use and the envi-ronmental implications, and generated a lot of interests in havinga better understanding of the energy use characteristics in build-ings, especially their correlations with the prevailing weatherconditions [1,2]. It was estimated that in 2002 buildings accountedfor about 33% of the global greenhouse gas emissions [3]. Insubtropical Hong Kong, the commercial sector is the largest elec-tricity end-user, and cooling-dominated office buildings accountedfor most of the sector-wide electricity consumption [4,5]. Buildingstypically have a long life span, lasting for 50 years or more. It is,therefore, important to be able to analyse how buildings willresponse to climate change in the future, and assess the likelychanges in energy use. There have been some works on the impactof climate change on the built environment based on energysimulation, which uses sophisticated building energy simulation

: þ852 2788 7612.

All rights reserved.

tool to perform hour-by-hour computation of heating/cooling loadsand energy use. In general, a weather file containing 8760 hourlyrecords of major climatic variables such as dry-bulb temperature,dew point or wet-bulb temperature, solar radiation and wind speedis required for building energy simulation. For instance, Sheppardet al. [6,7] studied the impact of climate change on commercialbuilding energy consumption in the Sydney region (Australia) usingpredicted 3-hourly data from the Australian Bureau of Meteoro-logical Research Centre’s general circulation model (GCM). Theyfound that energy use could be increased by 10–17% due to CO2

doubling in the global atmosphere. More recent works involvedstochastically generated test reference year (TRY) hourly data setsfor Portuguese buildings [8] and moderate climate [9], a sampleoffice building in 6 cities ranging from low to high latitudes(10.6�N-51.2�N) [10], and ‘morphing’ technique to stretch and shiftexisting TRY and design summer year for a number of case studiesin UK [11,12].

Archived predictions from GCMs, however, contain mostlymonthly and/or daily data (e.g. the WCRP CMIP3 multi-modeldataset [13]). Attempts were made to generate future hourly databased on the archived daily values from these climate models[14,15]. An alternative (and certainly simpler) approach would be tocorrelate building energy use directly with daily/monthly weatherdata. Although empirical or regression-based models using meandaily/monthly outdoor dry-bulb temperature and degree-days data

mailto:[email protected]

www.sciencedirect.com/science/journal/03605442

http://www.elsevier.com/locate/energy

Table 2Summary of principal component analysis (SRES B1, low forcing).

Principalcomponent

Eigenvalue Cumulativeexplainedvariance (%)

Coefficient

DBT WBT GSR

1st 2.457 81.9 0.974 0.956 0.7722nd 0.536 99.8 �0.220 �0.290 0.6363rd 0.006 100.0 �0.057 0.054 0.005

J.C. Lam et al. / Energy 35 (2010) 1482–1490 1483

tend to show good correlations between energy use and the pre-vailing weather conditions, most of them either consider only oneweather variable (e.g. dry-bulb temperature), or do not adequatelyremove the bias in the weather variables during the multiple linearregression analysis [16]. Our earlier work on existing air-condi-tioned office buildings and sector-wide electricity consumption insubtropical climates had shown that regressions models based onprincipal component analysis (PCA) of key monthly climatic vari-ables could give a good indication of the corresponding monthlyand annual electricity use [17,18], and a close correlation betweendaily PCA, day types and the corresponding daily chiller load hadbeen observed [19]. More recently, it was also found that annualbuilding energy use from PCA and regression models was very closeto that from detailed multi-year hour-by-hour simulation; thedifference was within 2% [20]. The objective of the present workwas, therefore, to investigate the likely changes in energy use infully air-conditioned office buildings in subtropical Hong Kongunder different emissions scenarios in the 21st century usingprincipal component analysis and multi-year dynamic buildingenergy simulations. The primary aim was to estimate the effects ofclimate change on future trends of energy use in office buildingsbased on the regression models developed.

2. Methodology

This study involved 5 steps/aspects:

i) Principal component analysis of historical (1979–2008)weather data measured at the local meteorological stationand future (2009–2100) predictions from GCMs to generatea new composite climatic variable, which could account forthe long-term (1979–2100) variations of major meteorolog-ical variables.

ii) Multi-year (1979–2008) hour-by-hour building energysimulation using a generic office building with designfeatures commonly found in local building stock andcomplying with the local building energy code.

iii) Correlation between 27-year (1979–2005) simulated energyconsumption with the corresponding composite climaticvariable using regression technique.

iv) Regression model evaluation using 3-year (2006–2008)simulation results and regression-predicted energy use data.

v) Estimate the likely changes in building energy use in futureyears using the regression model developed.

3. Principal components analysis (PCA) of majormeteorological variables

In the analysis of long-term meteorological variables, it is oftenadvantageous to group key weather variables directly affectingbuilding energy performance. PCA is a multivariate statistical

Table 1Summary of error analysis of predicted dry-bulb temperature (DBT), wet-bulb temperatu

DBT WBT

MBE RMSE MBE R

�C % Rank �C % Rank �C % Rank �C

BCCR-BCM2.0 �1.3 �5.7 4 2.1 9.2 2 �0.4 �1.7 2 1GISS-AOM �0.6 �2.6 2 2.6 11.3 3 �0.2 �0.9 1 2INM-CM3.0 �2.9 �12.7 5 3.6 15.7 5 �2.5 �12.0 5 3MIROC3.2-H 0.1 0.3 1 1.8 7.6 1 0.7 3.4 3 1NCAR-CCSM3.0 �1.0 �4.3 3 2.6 11.4 4 �0.8 �4.1 4 2

technique for analysis of the dependencies existing among a set ofinter-correlated variables [21,22]. Because of its ability to categorisethe complex and highly inter-correlated set of meteorologicalvariables as one or more cohesive indices, PCA tends to give a betterunderstanding of the cause/effect relationship. PCA is conducted oncentred data or anomalies, and is used to identify patterns ofsimultaneous variations. Its purpose is to reduce a dataset con-taining a large number of inter-correlated variables to a datasetcontaining fewer hypothetical and uncorrelated components,which nevertheless represent a large fraction of the variabilitycontained in the original data. These components are simply linearcombinations of the original variables with coefficients given by theeigenvector. A property of the components is that each contributesto the total explained variance of the original variables.

Initially five climatic variables were considered, namely dry-bulb temperature (DBT, in �C), wet-bulb temperature (WBT, in �C),global solar radiation (GSR, in MJ/m2), clearness index and windspeed [23]. DBT affects the thermal response of a building and theamount of heat gain/loss through the building envelope and henceenergy use for the corresponding sensible cooling/heatingrequirements, whereas WBT dictates the amount of humidifica-tion required during dry winter days and latent cooling underhumid summer conditions. Information on solar radiation iscrucial to cooling load determination and the correspondingdesign and analysis of air-conditioning systems, especially intropical and subtropical climates where solar heat gain throughfenestrations is often the largest component of the buildingenvelope cooling load [24]. Clearness index indicates the pre-vailing sky conditions while wind speed affects natural ventilationand the external surface resistance and hence U-values of thebuilding envelope. Contributions to the principal componentsfrom the clearness index and wind speed, however, were found tobe small (at least one order of magnitude smaller) compared withDBT, WBT and GSR. These 2 climatic variables were, therefore, notconsidered.

Future weather conditions were obtained from the WorldClimate Research Programme’s (WCRP) Coupled Model Intercom-parison Project Phase 3 (CMIP3) multi-model dataset [13]. Alto-gether, there were five GCMs that had archived monthly meanDBT, moisture content, and GSR. Predictions from these five GCMswere downloaded and analysed. These GCMs included the BCCR-BCM2.0 (Norway), GISS-AOM (USA), INM-CM3.0 (Russia),MIROC3.2-H (Japan), and NCAR-CCSM3.0 (USA). They covered

re (WBT) and global solar radiation (GSR) (1979–1999).

GSR Overall averagerank

MSE MBE RMSE

% Rank MJ/m2 % Rank MJ/m2 % Rank

.8 9.0 2 2.5 12.3 1 4.5 22.0 1 2.0

.0 9.7 3 6.2 30.2 5 7.2 35.3 5 3.2

.1 15.2 5 5.5 26.6 4 7.0 34.2 4 4.7

.7 8.5 1 5.1 24.8 3 6.4 31.1 3 2.0

.6 12.8 4 3.6 17.3 2 4.6 22.4 2 3.2

Table 3Summary of principal component analysis (SRES A1B, medium forcing).

Principalcomponent

Eigenvalue Cumulativeexplainedvariance (%)

Coefficient

DBT WBT GSR

1st 2.447 81.6 0.973 0.957 0.7652nd 0.547 99.8 �0.224 �0.287 0.6443rd 0.005 100.0 �0.053 0.051 0.004

J.C. Lam et al. / Energy 35 (2010) 1482–14901484

predictions for the past 10 decades (1900–1999) based on knownemissions, and future years (2000–2099 for NCAR-CCSM3.0 andBCCR-BCM2.0; and 2001–2100 for GISS-AOM, INM-CM3.0, andMIROC3.2-H) based on different emissions scenarios [25]. To get anidea about how well these GCMs could predict the temperature,humidity and solar radiation, predictions for the 21-year period(1979–1999) from these 5 GCMs were gathered and analysed. Tocompare like with like predicted moisture content was convertedto WBT. The predicted DBT, WBT and GSR were compared with thecorresponding measured monthly mean data. A summary of theerror analysis is shown in Table 1. In general, all 5 GCMs tended tohave better predictions in temperature and humidity than in solarradiation. As for DBT, MIROC3.2-H had the smallest MBE (0.3%) andRMSE (7.6%). As for WBT, GISS-AOM had the smallest MBE (�0.9%)while MIROC3.2-H had the smallest RMSE (8.5%). BCCR-BCM2.0

J F M A M J J A S O N D20

30

40

50

60

70

80

Month

Mon

thly

pri

ncip

al c

ompo

nmen

t Z

Max Mean Min

Past (1979-2008)


30

40

50

60

70

80

Mon

thly

pri

ncip

al c

ompo

nmen

t Z

Max Mean Min

Future (2009-2100) low forcing

Month

Fig. 1. Monthly profiles of principal component Z during 1979–2008 and 2009–2100for scenario SRES B1 (low forcing).

had the smallest error in GSR (MBE 12.3% and RMSE 22%).Performance (in terms of the percentage error) of the 5 GCMs wasranked and a summary is also shown in Table 1. Apparently,MIROC3.2-H tended to perform well in temperature and humiditybut only average in solar radiation among the 5 models. Its overallaverage ranking was 2, same as BCCR-BCM2.0. In this study,MIROC3.2-H was selected for 2 reasons. Firstly, temperature andhumidity greatly affect air-conditioning load, particularly latentcooling in subtropical climates. Secondly, our recent work onhuman bioclimate had found that MIROC3.2-H tended to show thebest agreement between measured data and model predictions[26].

Predictions from the MIROC3.2-H general circulation modelwere used in the PCA for future years from 2009 to 2100 for 2scenarios [13,25] – SRES B1 (low forcing, rapid change towarda service and information economy, peak global population in mid-21st century and decline thereafter, introduction of clean andresource-efficient technologies, and emphasis on global solutionsto economic social and environmental sustainability), and SRESA1B (medium forcing, very rapid economic growth, same pop-ulation trends as B1, convergence among regions with increasedcultural and social interactions, and technological emphasis ona balanced mix of fossil and non-fossil energy resources). Thepredicted monthly mean DBT, WBT and GSR were calibratedaccording to the mean bias error shown in Table 1.

1979 1999 2019 2039 2059 207940

45

50

55

60

65

70

Future:slope = 0.053 per yearaverage Z = 55.4

Ann

ual a

vera

ge Z

Year

Past:slope = 0.06 per yearaverage Z = 52.1

SRES B1 (low forcing)

2100

1979 1999 2019 2039 2059 207940

45

50

55

60

65

70

Future:slope = 0.076 per yearaverage Z = 55.8

Ann

ual a

vera

ge Z

Year

Past:slope = 0.06 per yearaverage Z = 52.1

SRES A1B (medium forcing)

2100

Fig. 2. Long-term trends of annual average principal component Z during 1979–2100.


A dataset consisting of 30-year (1979–2008) measured data and92-year (2009–2100) predictions were established for each emis-sions scenario. Altogether 122 � 12 � 3 monthly data wereconsidered in each PCA. Tables 2 and 3 show the coefficients of thethree principal components and the relevant statistics from the

1979 1999 2019 2039 2059 207915

20

25

30

35 SRES B1 low forcing SRES A1B medium forcing

Future

DB

T (

o C)

Year

Past

2100

1979 1999 2019 2039 2059 207915

20

25

30


Future

WB

T (

o C)

Year

Past

2100

1979 1999 2019 2039 2059 20795

10

15

20


Future

GSR

(M

J/m

2 )

Year

Past

2100

Fig. 3. Long-term trends of dry-bulb temperature (DBT), wet-bulb temperature (WBT)and global solar radiation (GSR) during 1979–2100.

PCA. The eigenvalue is a measure of the variance accounted for bythe corresponding principal component. The first and largesteigenvalue accounts for most of the variance, and the second thesecond largest amounts of variance, and so on. A common approachis to select only those with eigenvalues equal to or greater than one(eigenvalues greater than one implies that the new principalcomponents contain at least as much information as any one of theoriginal climatic variables [27]) or with at least 80% cumulativeexplained variance [28]. These criteria were adopted for this study.From Tables 2 and 3, the first principal component had an eigen-value greater than one with a cumulative explained varianceexceeding 81% (i.e. a one-component solution accounted for morethan 81% of the variance in the original climatic variables). The firstprincipal component was, therefore, retained, and a new set ofmonthly variable, Z, calculated as a linear combination of theoriginal three climatic variables as follows:

For SRES B1ðlow forcingÞ : Z

¼ 0:974� DBTþ 0:956�WBT

þ 0:772� GSR (1)

For SRES A1B ðmedium forcingÞ : Z

¼ 0:973� DBTþ 0:957�WBTþ 0:765� GSR (2)

Measured data for the three climatic variables were analysedand the monthly values of Z determined for the 30-year periodfrom 1979 to 2008 using Equation (1). Fig. 1 shows the maximum,long-term mean and minimum monthly values of Z during the 30-year period. The principal component profiles show distinctseasonal variations. Z tended to be at its lowest during the wintermonths (December, January and February) and peaked in thesummer (June–August). To get an idea about the climatic perfor-mance of other weather data sets, test reference year (TRY) forHong Kong was developed [29] and the corresponding monthlyprincipal component Z determined. Principal components from theTRY were then compared with those from the long-term (1979–2008) mean. The mean bias error and root mean square error were�0.28 and 3.14, representing 0.55% and 6% of the mean monthly Z,respectively.

Likewise, predictions from the GCM were used to determine themonthly Z for the low forcing scenario during 2009–2100, and themonthly profiles are also shown in Fig. 1. In general, the seasonalvariations are similar to those shown in the past 3 decades. To havea better understanding of the underlying trend, annual average Zwere determined, and the 30-year (1979–2008) and 92-year(2009–2100) long-term trends are shown in Fig. 2. Both the pastand future years show a clear (though slightly) increasing trend.Not surprisingly, medium forcing in future years has a larger slope(0.076 per year) than the 0.053 per year for low forcing, whichassumed the introduction of clean and resource-efficient technol-ogies [25]. Average Z in future years would be 55.4 and 55.8 for lowand medium forcing, respectively, representing an increase of 6.3%and 7.1% over the average Z of 52.1 for the past 30 years (1979–2008). To have a better picture of the underlying changes in theweather conditions, the annual averages of the 3 climatic variables

Table 4Summary of annual averages of dry-bulb temperature (DBT), wet-bulb temperature(WBT) and global solar radiation (GSR) during 1979–2008 and 2009–2100.

DBT (�C) WBT (�C) GSR (MJ/m2)

1979–2008 23.2 20.6 12.82009–2100 (low forcing) 25.0 22.2 12.72009–2100 (medium forcing) 25.4 22.5 12.5

Table 5Summary of key design data.

Building envelope Indoor designcondition

Internal load density HVAC

U-value (W/m2 K) Shadingcoefficient

Summer(�C)

Winter(�C)

Occupancy(m2/person)

Lighting(W/m2)

Equipment(W/m2)

AHU Cooling Heating

Wall Window Roof Window

2.01 5.6 0.54 0.4 24 21 13 25 10 VAV Centrifugal chiller(Water-cooled,COP ¼ 4.7)

Electric

Notes: HVAC ¼ Heating, Ventilating and Air-Conditioning; AHU ¼ Air-handling unit; VAV ¼ Variable-air volume.


(i.e. DBT, WBT and GSR) were determined and a summary of theirlong-term trends are shown in Fig. 3. Clear increasing trends in DBTand WBT can be observed, but not GSR. Table 4 shows a summary ofthe annual averages during the past and futures years. As for DBT,the temperature rise would be 1.8 �C (low forcing) and 2.2 �C(medium forcing) over the 1979–2008 period, and for WBT 1.6 �C(low forcing) and 1.9 �C (medium forcing). Solar radiation, however,would be slightly reduced from 12.8 MJ/m2 to 12.7 MJ/m2 (lowforcing) and 12.5 MJ/m2 (medium forcing). This seems to beconsistent with findings from investigation work on cloud cover,solar radiation and climate changes in Hong Kong [30] and else-where [10,15,31].


300

400

500

600

700

800

Month

Bui

ldin

g en

ergy

use

(M

Wh)

Max 30-year mean Min

Fig. 4. Monthly profiles of building energy use (1979–2008).

4. Multi-year building energy simulation

Hour-by-hour energy simulations were conducted for each ofthe 30 years (1979–2008) using the simulation tool VisualDOE4.1[32]. Hourly solar radiation data were not measured prior toDecember 1978. Two major inputs were considered for the simu-lation: i) 8760 hourly records of weather data (DBT, WBT, GSR, windspeed and wind direction), and ii) a generic office building and itsbuilding services system. Details of the generic building forsubtropical Hong Kong can be found in [20,24]. Briefly, a base-caseoffice building was developed to serve as a baseline reference forcomparative energy studies – a 35 m � 35 m, 40-storey buildingwith curtain walling design, 3.4 m floor-to-floor height and 40%window-to-wall ratio. The total gross floor area (GFA) is 49 000 m2

(41 160 m2 air-conditioned and 7840 m2 non-air-conditioned). Theair-conditioned space had five zones - four at the perimeter and oneinterior. The building envelope, internal loads, indoor designconditions and basic heating, ventilation and air-conditioning(HVAC) system designs were developed based on the prevailingarchitectural and engineering practices that met the requirementsstipulated in the local energy codes [33,34]. In subtropical HongKong, office buildings are cooling-dominated, where solar heat gainis by far the largest component of the building envelope coolingload. Thermal insulation to the external walls is less important andwindows tend to have small shading coefficients. The building andits lighting system operated on a 10-h day (08:00–18:00) and a 5½-day week. Infiltration rate (when the HVAC system was off) was setat 0.45 air change per hour. A summary of the key data is shown inTable 5.

The computed results were analysed and compared. Threeaspects were considered – cooling load, heating load, and totalbuilding energy use (i.e. electricity use for HVAC, lighting andequipment). To get some idea about the seasonal and yearly vari-ations in energy use due to varying climatic conditions, buildingenergy use were determined and a summary of the monthly energyconsumption for the 30 years is shown in Fig. 4. Distinct seasonalvariations can be observed, which peaked during the hot summermonths. Annual building energy use varied from 6423 MWh in

1986–6643 MWh in 2006 with a 30-year long-term mean of6505 MWh (i.e. 133 kWh/m2 gross floor area).

5. Correlation between simulated results and principalcomponent

To investigate the strength of correlation between building load/energy use and principal component, regression analysis wasconducted for the monthly simulated results (which were nor-malised to account for the difference in the number of days permonth) and the corresponding principal component. Only 27 years(1979–2005) data were used, the remaining 3 years (2006–2008)were reserved for regression model evaluation. Fig. 5 showsa summary of the correlation for the SRES B1 (low forcing) scenario.It can be seen that both the building loads and energy use corre-lated quite well with the corresponding Z. A quadratic regression(i.e. Y ¼ a þ bZ þ cZ2, where Y is the monthly load/energy use) wasobtained for the cooling load and building energy use and a 3rdorder polynomial (i.e. Y ¼ a þ bZ þ cZ2 þ dZ3) for the heating load.Similar characteristics were observed for the SRES A1B (mediumforcing) scenario. A summary of the regression statistics is shown inTable 6. It can be seen that the regressions have a rather highcoefficient of determination (R2¼ 0.93–0.99), indicating reasonablystrong correlation between the simulated building load/energy useand the corresponding principal component (i.e. 93–99% of thechanges in the simulated results can be explained by variations inthe corresponding principal component). Not surprisingly, buildingcooling load had the largest R2 because of the dominance of coolingrequirements in office buildings in subtropical climates. Coeffi-cients of the regression models for the low and medium forcingscenarios are very close to each other. Given the fact that the only

20 30 40 50 60 70 80200

400

600

800

1000

1200

Coo

ling

Loa

d (M

Wh)

Principal component Z

30 40 50 60 70-50

0

50

100

150

200

Hea

ting

Loa

d (M

Wh)


20 30 40 50 60 70 80300

400

500

600

700

800

Ene

rgy

use

(MW

h)


Fig. 5. Correlation between monthly building load/energy use and the correspondingprincipal component Z (1979–2005).

Table 6Summary of regression analysis of building loads and energy use (1979–2005).

Scenario R2 a b c d

SRES B1(low forcing)

Cooling 0.99 45.9 3.38 0.16 –Heating 0.98 1001.9 �45.85 0.69 �0.003Energy use 0.93 213.5 5.05 0.02 –

SRES A1B(medium forcing)

Cooling 0.99 46.8 3.36 0.16 –Heating 0.98 1002.7 �45.98 0.70 �0.003Energy use 0.93 213.5 5.07 0.02 –

Table 7Summary of regression model evaluation (SRES B1, low forcing).

2006 2007

MBE NMBE RMSE CVRMSE MBE NM

(MWh) (%) (MWh) (%) (MWh) (%)

Cooling load �10.6 �1.5 25.0 3.6 1.9 0.3Heating load �0.2 �0.9 3.9 20.4 1.9 12.9Energy use �9.3 �1.7 24.4 4.4 5.2 0.9


difference in the two sets of regression analysis is the slightlydifferent monthly Z value used for the 2 scenarios (i.e. Equations (1and 2)), this is not surprising.

6. Model evaluation

Performances of the regression models were evaluated. An erroranalysis was conducted by comparing the simulated results of2006–2008 with those determined from the regression equationsusing the corresponding principal component. To quantify thedifferences, mean bias error (MBE) and root mean square error(RMSE) were determined as follows:

MBE ¼P12

i¼1ðxi � yiÞ12

(3)

RMSE ¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiP12i¼1ðxi � yiÞ2

12

s(4)

where x ¼ regression-predicted monthly data (MWh); y ¼ DOE-simulated monthly data (MWh)

MBE provides information on the long-term performance of themodelled regression equation. A positive MBE indicates that thepredicted annual electricity consumption is higher than the actualconsumption and vice versa. It is worth noting that overestimationin an individual observation can be offset by underestimation ina separate observation. The RMSE is a measure of how close theregression-predicted monthly data are to the DOE-simulatedvalues. Normalised mean bias error (NMBE) and coefficient ofvariation of the root mean square error (CVRMSE) were alsodetermined by dividing MBE and RMSE by the mean simulatedmonthly values, and a summary is shown in Tables 7 and 8 for lowand medium forcing, respectively. It can be seen that, in the lowforcing scenario, NMBE for the cooling load was very small, rangingfrom 1.5% underestimation in 2006–0.3% overestimation in 2007.Cooling load CVRMSE did not vary much among the 3 years, about4%. This suggests that while regression-predicted cooling load, onan annual basis, could be very good (about 1%), individual monthlyvalues could differ from the DOE-simulated data by up to 4%.Similar characteristics can be observed for total building energyconsumption. Heating load tended to have larger percentage errors,especially in 2007. The regression models tended to underestimatethe cooling load and hence energy use except in 2007. There was,however, no clear pattern indicating whether the regression

2008

BE RMSE CVRMSE MBE NMBE RMSE CVRMSE

(MWh) (%) (MWh) (%) (MWh) (%)

28.7 4.0 �5.6 �0.8 26.0 3.74.5 30.9 0.8 3.1 2.9 11.5

25.4 4.6 �1.7 �0.3 20.6 3.8

Table 8Summary of regression model evaluation (SRES A1B, medium forcing).

2006 2007 2008

MBE NMBE RMSE CVRMSE MBE NMBE RMSE CVRMSE MBE NMBE RMSE CVRMSE

(MWh) (%) (MWh) (%) (MWh) (%) (MWh) (%) (MWh) (%) (MWh) (%)

Cooling load �10.4 �1.5 24.9 3.6 1.9 0.3 28.7 4.0 �5.6 �0.8 26.2 3.8Heating load 0.5 2.5 4.0 20.6 2.6 17.9 4.8 32.6 1.5 5.9 3.1 12.1Energy use �9.2 �1.7 24.3 4.4 5.1 0.9 25.4 4.6 �1.7 �0.3 20.6 3.8

1979 1999 2019 2039 2059 20796000

7000

8000

9000

10000

11000


Future

Ann

ual c

oolin

g lo

ad (

MW

h)

Year

Past

2100

1979 1999 2019 2039 2059 20790

100

200

300

400

500

600

700


Future

Ann

ual h

eatin

g lo

ad (

MW

h)

Year

Past

2100

1979 1999 2019 2039 2059 20795000

6000

7000

8000

9000


Future

Ann

ual

ener

gy u

se (

MW

h)

Year

Past

2100

Fig. 6. Long-term trends of annual building load/energy use during 1979–2100.


models would tend to overestimate or underestimate the buildingload/energy use. Comparing Tables 7 and 8, it can be seen that theerrors were very similar for both scenarios.

7. Future trends of building loads and energy use due toclimate change

Earlier works on long-term (1961–2000) ambient temperaturein Hong Kong [35] revealed a slight increase in the annual coolingdegree-days, and suggested that cooling requirements, henceenergy use for air-conditioning could be significantly affected ifsuch trend persisted. In this study, the regression models devel-oped were used to estimate cooling/heating loads and energy use infuture years for the 2 scenarios (i.e. low and medium forcing). Fig. 6shows the long-term trends of the building loads and energy use.As expected, cooling load and building energy use both exhibit anincreasing trend while the already small heating load becomes lesssignificant in future years. To get a better idea about the impact ofclimate change, average building loads and energy use during the1979–2008 and 2009–2100 periods were determined, anda summary is shown in Fig. 7. It can be seen that the average annualcooling load in 2009–2100 would be 9.1% and 10.7% more than thatin 1979–2008 for low and medium forcing, respectively; thepercentage increase for energy use would be 4.3% and 4.9%.Percentage reduction in heating would be rather large (about 45%for both scenarios) due mainly to the small base value (i.e. insig-nificant heating requirement in subtropical climates). The actualreduction would be small, about 150 MWh. To see whether climatechange would affect seasonal variations, standard deviations ofmonthly building loads and energy use were determined for each ofthe 123 years and a summary of the average values during the1979–2008 and 2009–2100 periods is shown in Fig. 8. It is inter-esting to see that the standard deviations would be smaller infuture years, cooling load from 226 MWh to about 200 MWh,heating load from 35 MWh to about 21 MWh and energy from

Cooling load Heating load Energy use0

2000

4000

6000

8000

10000

Ave

rage

ann

ual b

uild

ing

load

and

ene

rgy

use

(MW

h)

1979-2008 2009-2100 (low forcing) 2009-2100 (medium forcing)

Fig. 7. Comparisons of annual average building load/energy use between past (1979–2008) and the future (2009–2100).

Cooling load Heating load Energy use0

50

100

150

200

250

300

Sta

ndar

d de

viat

ion

(mon

thly

var

iati

on)

(MW

h)

1979-2008 2009-2100 (low forcing) 2009-2100 (medium forcing)

Fig. 8. Comparisons of average standard deviation (monthly variation) between past(1979–2008) and the future (2009–2100).


87 MWh to about 73 MWh for both scenarios. This suggests thatthere would be less seasonal variations in building loads andenergy use due to climate change in future years.

8. Conclusions

Principal component analysis of prevailing weather conditionsduring the past 30 years (1979–2008) and in future years (2009–2100) in subtropical Hong Kong was conducted. Three majorclimatic variables – dry-bulb temperature (DBT), wet-bulbtemperature (WBT) and global solar radiation (GSR) – wereconsidered, and a new climatic index (principal component Z)determined for 2 emissions scenarios (SRES B1 low forcing andSRES A1B medium forcing). Multi-year (1979–2008) buildingenergy simulations were carried out for a generic air-conditionedoffice building. It was found that Z exhibited monthly and seasonalvariations similar to the simulated building loads and energy use.Regression models were developed to correlate the simulatedmonthly cooling load and building energy use with the corre-sponding monthly Z. The coefficient of determination (R2) was0.93–0.99, indicating strong correlation between building load/energy use and Z. Error analysis indicated that annual cooling loadand building energy use from the regression models were veryclose to the simulated values; the difference was about 1%. Differ-ence in individual monthly cooling load and energy use, however,could be up to 4%.

It was found that both the DOE-simulated results during 1979–2008 and the regression-predicted data during 2009–2100 indi-cated an increasing trend in annual cooling load and energy useand a gradual reduction in the already insignificant heatingrequirement in cooling-dominated office buildings in subtropicalclimates. The average annual cooling load in 2009–2100 would be9.1% and 10.7% more than that in 1979–2008 for low and mediumforcing, respectively; the percentage increase for energy use wouldbe 4.3% and 4.9%. It was also found that the average standarddeviation of the monthly building load and energy profiles wouldbe smaller in future years, cooling load from 226 MWh to about200 MWh, heating load from 35 MWh to about 21 MWh andenergy from 87 MWh to about 73 MWh for both scenarios. Thissuggests that there would be less seasonal variations in buildingloads and energy use due to climate change in future years Webelieve the regression models developed can be used to estimatethe effects of climate change on future trends of energy use inoffice buildings in Hong Kong based on the monthly predictions(i.e. DBT, WBT and GSR) from general circulation or regional

climate models. This would give the building professions andenergy/environmental policy makers a good idea about the likelyorder of magnitude changes in energy consumption in thebuilding sector so that appropriate mitigation measures (betterbuilding energy codes and more energy-efficient building servicesequipment) could be considered. Although the work was con-ducted for subtropical climates, it is envisaged that the approachcould be applied to other locations with similar or differentclimates. Given the growing concerns about climate change and itslikely impact on energy use in the built environment, this mighthave important energy and environmental implications. Theregression models did not consider advances in building technol-ogies, which could affect future trends of energy use in buildings.More work is required.

Acknowledgements

T.N.T. Lam and K.K.W. Wan were supported by City University ofHong Kong Studentships. Measured weather data were obtainedfrom the Hong Kong Observatory of the Hong Kong SAR. Weacknowledge the modelling groups, the Program for Climate ModelDiagnosis and Intercomparison (PCMDI) and the WCRP’s WorkingGroup on Coupled Modelling (WGCM) for their roles in makingavailable the WCRP CMIP3 multi-model dataset. Support of thisdataset is provided by the Office of Science, U.S. Department ofEnergy.

References

[1] IPCC. Climate change. In: Solomon S, Qin D, Manning M, Chen Z, Marquis M,Averyt KB, Tignor M, Miller HL, editors. The physical science basis. Contribu-tion of the Working Group I to the fourth assessment report of the inter-governmental panel on climate change. Cambridge: Cambridge UniversityPress; 2007.

[2] Levine M, Urge-Vorsatz D, Blok K, Geng L, Harvey D, Lang S, et al. Resi-dential and commercial buildings. In: Metz B, Davidson OR, Bosch PR,Dave R, Meyer LA, editors. Climate change 2007: mitigation. Contributionof Working Group III to the fourth assessment report of the intergovern-mental panel on climate change. Cambridge: Cambridge University Press;2007. p. 387–446.

[3] Levermore GJ. A review of the IPCC assessment report four, part 1: the IPCCprocess and greenhouse gas emission trends from buildings worldwide.Building Services Engineering Research and Technology 2008;29(4):349–61.

[4] Lam JC, Ng AKW. Energy consumption in Hong Kong. Energy 1994;19(11):1157–64.

[5] Lam JC. Building envelope loads and commercial sector electricity use in HongKong. Energy 1995;20(3):189–94.

[6] Sheppard R, de Dear RJ, Rowe D, McAvaney B. The impact of climate change onenergy consumption in buildings: research in progress. AIRAH Journal 1996;50(9):24–7.

[7] Sheppard R, de Dear RJ, Rowe D, McAvaney B. The impact of climate change oncommercial building energy consumption: the Sydney region. AIRAH Journal1997;51(12):20–5.

[8] Aguiar R, Oliveira M, Goncalves H. Climate change impacts on the thermalperformance of Portuguese buildings: results of the SIAM study. BuildingServices Engineering Research and Technology 2002;23(4):223–31.

[9] van Paassen AHC, Luo QX. Weather data generator to study climate change onbuildings. Building Services Engineering Research and Technology 2002;23(4):251–8.

[10] Degelman LO. Which came first – building cooling loads or global warming?– a cause and effect examination. Building Services Engineering Research andTechnology 2002;23(4):259–67.

[11] Belcher SE, Hacker JN, Powell DS. Constructing design weather data for futureclimates. Building Services Engineering Research and Technology 2005;26(1):49–61.

[12] CIBSE TM36. Climate change and the indoor environment: impacts andadaptation. London: The Charted Institution of Building Services Engineers;2005.

[13] Meehl GA, Covey C, Delworth T, Latif M, McAvaney B, Mitchell JFB, et al. TheWCRP CMIP3 multi-model dataset: a new era in climate change research.Bulletin of the American Meteorological Society 2007;88(9):1383–94.

[14] Chow DHC, Levermore G. New algorithm for generating hourly temperaturevalues using daily maximum, minimum and average values from climatemodels. Building Services Engineering Research and Technology 2007;28(3):237–48.


[15] Guan L. Preparation of future weather data to study the impact of climatechange on buildings. Building and Environment 2009;44(4):793–800.

[16] Hadley DL. Daily variations in HVAC system electrical energy consumption inresponse to different weather conditions. Energy and Buildings 1993;19(3):235–47.

[17] Lam JC, Wan KKW, Cheung KL, Yang L. Principal component analysis of elec-tricity use in office buildings. Energy and Buildings 2008;40(5):828–36.

[18] Lam JC, Tang HL, Li DHW. Seasonal variations in residential and commercialsector electricity consumption in Hong Kong. Energy 2008;33(3):513–23.

[19] Lam JC, Wan KKW, Cheung KL. An analysis of climatic influences on chillerplant electricity consumption. Applied Energy 2009;86(6):933–40.

[20] Lam JC, Wan KKW, Wong SL, Lam TNT. Principal component analysis and long-term building energy simulation correlation. Energy Conversion andManagement 2010;51(1):135–9.

[21] Wilks DS. Statistical method in the atmospheric sciences: an introduction. SanDiego: Academic Press; 1995.

[22] Storch HV, Zwiers FW. Statistical analysis in climate research. Cambridge:Cambridge University Press; 1999.

[23] Wan KKW, Cheung KL, Yang L, Lam JC. A new variable for climate change studyand implications for the built environment. Renewable Energy 2009;34(3):916–9.

[24] Lam JC. Energy analysis of commercial buildings in subtropical climates.Building and Environment 2000;35(1):19–26.

[25] IPCC Working Group III. In: Nakicenovic N, Swart S, editors. Special report onemissions scenarios. Cambridge: Cambridge University Press; 2000.

[26] Lam JC, Wan KKW, Wong SL, Lam TNT. Long-term trends of heat stress andenergy use implications in subtropical climates. Applied Energy 2010;87(2):608–12.

[27] Kalkstein LS, Tan G, Skindlov JA. An evaluation of three clustering proceduresfor use in synoptic climatological classification. Journal of Climate and AppliedMeteorology 1987;26(6):717–30.

[28] Ladd JW, Driscoll DM. A comparison of objective and subjective means ofweather typing: an example from West Texas. Journal of Applied Meteorology1980;19(6):691–704.

[29] Hui SCM, Lam JC. Test reference year (TRY) for comparative energy study.Hong Kong Engineer 1992;20(2):13–6.

[30] Lam CY. On climate changes brought about by urban living. Hong KongMeteorological Society Bulletin 2006;16(1/2):15–27.

[31] Croke MS. Regional cloud cover change associated with global climate change:case studies for three regions of the United States. Journal of Climate1999;12(7):2128–34.

[32] DOE-2 supplement, version 2.1E, LBL-34947. Lawrence Berkeley NationalLaboratory, University of California; 1993.

[33] Amendments to the code of practice for overall thermal transfer value inbuildings. Hong Kong SAR: Buildings Department; 2000.

[34] Performance-based building energy code. Hong Kong SAR: Electrical andMechanical Services Department; 2007.

[35] Lam JC, Tsang CL, Li DHW. Long-term ambient temperature analysis andenergy use implications in Hong Kong. Energy Conversion and Management2004;45(3):315–27.

an analysis of future building energy use in subtropical hong kong

Documents