Energy and Buildings 113 (2016) 74–86

Contents lists available at ScienceDirect

Energy and Buildings

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

An alternative method to predict future weather data for building

energy demand simulation under global climate change
Mingya Zhu a , Yiqun Pan a,∗ , Zhizhong Huang b , Peng Xu a
a
School of Mechanical Engineering, Tongji University, Shanghai 200092, China
b
Sino-German College of Applied Sciences, Tongji University, Shanghai 200092, China

a r t i c l e i n f o a b s t r a c t

Article history: Climate change would affect the building energy demand in the future. Building simulation is a feasible
Received 15 January 2015 way to quantitatively evaluate this impact. Since the detailed weather is dispensable for the building
Received in revised form 20 October 2015 simulation, it is important to predict the weather data for the future. Given that the uncertainties and
Accepted 13 December 2015
limitations of GCMs on regional-scale weather prediction, an alternative method of future weather data
Available online 19 December 2015
generation for future building energy demand simulation is proposed in this paper. Based on the long-
/short-term climate periodicity analysis, a Dual-Periodic Time Series Model is established to predict
Keywords:
the future monthly temperatures in Shanghai. From the fitting results and the preliminary assessment
Climate change
Climate periodicity
analysis, it is observed that the alternative forecasting method and the corresponding Dual-Periodic TSM
Time Series Model has better capability of characterizing and predicting performance for both recent and future temperature
Weather data prediction trends in Shanghai than GCM under RCP4.5. In this case, this method can be used as an alternative and
Building simulation supplementary way to the widely used GCM.
Building energy demand With consideration of three composite uncertainty scenarios, we convert the predicted monthly tem-
peratures into hourly TMYs by using Morphing method. Using the future TMYs as the weather input of
prototypical building models of Shanghai, we can see that the simulated building energy demand presents
fluctuant trends in the future periods, different from the continuous uptrends of that using IPCC RCP4.5.
© 2015 Elsevier B.V. All rights reserved.

Abbreviations: GCM, General Circulation Model/Global Climate Model; RCM, Regional Climate Models; IPCC, Intergovernmental Panel on Climate Change; RCP, Repre-
sentative Concentration Pathways, including four grades: RCP8.5 RCP6 RCP4.5 and RCP3-PD; AR5, IPCC Fifth Assessment Report; TMY, Typical Meteorological Year; HVAC,
Heating Ventilation Air-Conditioning; HDD, heating degree days; CDD, cooling degree days; DBT/dbt, dry bulb temperature; WBT/wbt, wet bulb temperature; RH, relative
humidity; WS, wind speed; SR, solar radiation; LBNL, Lawrence Berkeley National Laboratory; GHG, greenhouse gas; NCDC, National Climatic Data Center; CMIP5, IPCC’s
Coupled Model Intercomparison, Phase 5; LIA, Little Ice Age; AMO, Atlantic Multidecadal Oscillation; ENSO, El Nino-Southern Oscillation; TSM, Time Series Model/Method; TC
series, the remaining series after excluding a stochastic term (Xt) from the observation series in Time Series Method; Ht/Ht , trend term in Time Series Model; Pt/Pt , periodic
term in Time Series Model; Xt/Xt , stochastic term in Time Series Model; Yt/Yt /Yt , sum of Ht, Pt, and Xt in Time Series Model; T1/T2, two periods in the proposed alternative
forecasting model, respectively identified as the trend and periodic features of the observed temperature series; CSWD, Chinese Standard Weather Data; CTYW, Chinese
Typical Year Weather; IWEC, International Weather for Energy Calculation; SWEAR, Solar Wind Energy Resource Assessment; FS, Finkelstein–Schafer method; TMPm ,
predicted change of monthly mean dbt of m month in Morphing method; adbtm , monthly stretching factor of dbt in Morphing method; TMAXm , predicted changes of
monthly maximum temperature in Morphing method; TMINm , predicted changes of monthly minimum temperature in Morphing method; (dbt0max )m , monthly mean
daily maximum temperature of m month in existing TMY in Morphing method; (dbt0min )m , monthly mean daily minimum temperature of m month in existing TMY in
Morphing method; R2 , fitting optimization index; MAPE, Mean Absolute Percent Error; RMSE, Root Mean Square Error; S1m/S2m/S3m, uncertainty scenarios of monthly
mean temperature; S1/S2/S3, three composite scenarios, which combine the daily fluctuation scenarios with corresponding fluctuation level of the uncertainty scenarios of
monthly mean temperature.
∗ Corresponding author at: 4800 Cao’an Gonglu, Shanghai 201804, China.
E-mail address: yiqunpan@tongji.edu.cn (Y. Pan).

http://dx.doi.org/10.1016/j.enbuild.2015.12.020
0378-7788/© 2015 Elsevier B.V. All rights reserved.
 M. Zhu et al. / Energy and Buildings 113 (2016) 74–86
1. Introduction
The global warming in the 20th century is a significant con-
tribution to the climate change issue, which is a global concern.
As one of the top greenhouse gas emitters, China is under multi-
ple pressures of climate diplomacy, economy development needs,
resource shortage and environmental deterioration. Therefore, a
correct estimate of future energy demand under climate change is
significant for relieving the huge international pressure for China.
The building industry plays an important role in the energy conser-
vation plan. Nowadays, the building sector consumes one-fifth of
energy resources and the corresponding Green House Gas emis-
sions space in China [1]. With the acceleration of urbanization,
the proportion of building energy consumption in total energy use
will increase. Meanwhile, the climate change has great impacts on
the building energy performance. The weather contributes directly
and significantly to the thermal loads and the energy use of Heat-
ing Ventilation Air-Conditioning (HVAC) systems. In China, The
HVAC systems in commercial buildings take more than 50% of the
total building energy consumption. As an important part of climate
change studies, the prediction of building energy demand under cli-
mate change is greatly significant to building energy saving, urban
energy planning, urban strategy development and energy policy
formulation.
At present, the prediction researches of building energy demand
under climate change can be divided into two groups: (1) direct
prediction through statistical analysis, and (2) indirect prediction
using building simulation tool.
For direct prediction, a statistical analysis method is usually
applied to achieve the relationship between the local meteoro-
logical parameters (degree day, dry-bulb temperature/wet-bulb
temperature) and the specific building energy demand based on the
sufficient historical observed data. After a reasonable validation, the
relationship can be used as a predictive model. Christenson et al. [2]
developed a procedure to estimate the heating degree days (HDD)
and the cooling degree days (CDD) from the monthly temperature
data of four representative Swiss locations to obtain the past trends
of HDD and CDD for the period 1901–2003. The future trends of
HDD and CDD in the 21st century, which represent the change
of heating and cooling energy demand separately, were assessed
based on Intergovernmental Panel on Climate Change (IPCC) emis-
sions scenarios. The degree day method is a simple method, without
the consideration on response process of buildings to outdoor envi-
ronment. There is a basic assumption in many similar researches
as the premise of using degree day method. That is: The build-
ing energy demand has a significant linear correlation with degree
days. In other words, only for the buildings with a relatively con-
stant internal temperature, thermal gains and building properties,
this method is reliable for quantifying the building heating/cooling
energy demand. By simulating of the different building types with
different energy saving levels in Tianjin, China, Li et al. [3] found
that CDD is unable to entirely reflect the cooling loads changes of
office buildings and retail buildings, since CDD can only explain 64%
and 55% of their cooling loads respectively. Comparatively, the rela-
tionship between temperatures and building energy consumption
used in the direct prediction is more complicated than the simple
linear correlation. Lam et al. [4] developed several regression mod-
els to correlate monthly building cooling loads and the total energy
use of office building in Hong Kong with a new climatic index Z
(a function of local dry-bulb temperature, wet-bulb temperature
and global solar radiation) in 1979–2008. Future trends of build-
ing heating/cooling loads and building energy demand in the 21st
century are predicted by the regression models based on IPCC’s
emissions scenarios in 2001. The average annual cooling load in
2009–2100 would be 9.1% and 10.7% more than that in 1979–2008
respectively under low- and medium-level emissions scenario; the
75
percentage increase for energy use would be 4.3% and 4.9% accord-
ingly.
The rapid development of building simulation technology has
provided robust support for building energy demand prediction.
The indirect prediction usually needs two main parts as the
research foundation:
(1) Generation of hourly future weather data (normally in the form
of TMY) under climate change. The existed TMYs are gener-
ated from historical weather observations and not respective
of future climate conditions.
(2) Establishment of Prototypical Building Models according to
related design standards and survey data of local existing build-
ings. After calibrated by investigated actual energy use data,
the models can mainly reflect the general situation of differ-
ent building types in a specific region. As the indispensable
inputs of building simulation, hourly future weather data is a
prerequisite and a key point for building energy demand pre-
diction under climate change by taking advantage of building
simulation tool.
During 2005–2008, two separate research projects at LBNL [5,6]
on the potential impact of climate change on building energy use
in California and the US were completed using indirect predic-
tion method. Predicted climate changes for various locations were
obtained from General Circulation Models (GCMs). The Outputs of
a selected GCM were monthly and daily predicted changes of mete-
orological parameters (temperature, humidity, and solar radiation)
under standard IPCC emissions scenarios. By statistical downscal-
ing method, these changes were then combined with existed TMY
files of different locations to get new future TMYs. These new TMYs
were then used as weather inputs of prototypical residential and
commercial building models to calculate the impacts of predicted
climate changes on the heating and cooling energy use. The sim-
ulation results showed a 7% decrease in 2020s and a 18% increase
in 2080s nationally on a basis of in-site energy use in 1980–2009.
By the same method, Chan [7] developed future TMY files for
Hong Kong and used them in local prototypical buildings. The
simulated results indicate substantial increase in Air-Conditioning
energy consumption in 21st century, ranging from 2.6% to 14.3%
for office buildings and from 3.7% to 24% for residential flats,
respectively.
Currently a majority of similar prediction researches use IPCC’s
GCM results as their basic data of climate change. However, this
important basis of hourly weather data prediction may have great
uncertainties. In IPCC’s assessment of global climate change, Emis-
sion Scenarios (alternative development pathways, covering a wide
range of demographic, economic and technological driving forces
and resulting GHG emissions) are supposed to drive GCMs (numer-
ical representation of the climate system based on its physical,
chemical and biological properties) for understanding and attri-
bution of past climate variations, as well for future predictions [8].
There are some key uncertainties in the researches: (1) The attri-
bution and the prediction of regional climate change are limited by
the complicated climate variability on regional scale. (2) The predic-
tions of climate change are highly scenario- and model-dependent.
(3) Considerable differences exist in the results of different mod-
els under different feedbacks in the climate system. The focuses of
these uncertainties and related exploration from diverse research
fields, such as meteorology and geophysics, produces some con-
troversy and query on GCMs’ simulation capability for system
mechanism of climate change, especially on regional scale [8,9].
Hence there are some certain limitations to apply GCMs’ outputs
to regional-scale weather data prediction.
Given the uncertainties and limitations of the widely used cli-
mate predictions (by GCMs) for generating regional-scale hourly
76
future weather data used in building simulation, this paper puts
forward an alternative method by combining the time series fore-
casting with the climate periodicity analysis in East China. With
observed monthly average temperature from 1961 to 2010 in
Shanghai, the time series prediction model is developed by the
periodicity analysis of long-term climate change and the various
uncertainty scenarios of future monthly temperature, and then
used to predict the monthly mean temperature for Shanghai in the
21st century. The predicted weather data in this method has a simi-
lar temporal scale with the widely used results of GCMs. In order to
apply the monthly predictions to building simulation, a statistical
downscaling method, morphing, is applied to generate new future
TMYs on the basis of local existing TMY. The new TMYs are then
taken as weather input parameters of prototypical building mod-
els for regional building energy demand prediction under global
climate change in the 21st century.
2. Methodology
2.1. Basic weather data
The observed monthly mean temperature series from 1961 to
2010 in Shanghai are obtained from U.S. National Climatic Data
Center. As shown in Fig. 1, the observed average temperature in
the past 50 years is 16.3 ◦ C. All the average temperatures in 1960s,
1970s and 1980s are below average. The average temperature in
1980s is 0.3 ◦ C higher than the average temperature in 1960s and
1970s. While in 1990s and 2000s, there was a faster increase of
0.7 ◦ C/10a.
The IPCC’s Coupled Model Intercomparison, Phase 5 (CMIP5)
has published experimental outputs of many GCMs under histori-
cal climate and Emission Scenarios. These output datasets include
various weather parameters of all longitudes and latitudes around
the world by December 2099. Among a large number of CMIP5
experimental datasets, the outputs of HadGEM2-CC model under
the historical climate at the nearest node to Shanghai are chosen to
compare with above observations.
From the comparison from 1960 to 2010 (Fig. 1), the pace of
climate warming in Shanghai has obvious differences with global
warming. Based on the results of GCM, the 50-year average temper-
ature is 15.6 ◦ C, which is about 0.7 ◦ C lower than the observations.
In 1980s, 1990s and 2000s, the calculated temperatures of GCM are
all lower than the corresponding observations, and the gaps are up
to about 1.0 ◦ C. The significant warming has occurred since 1980s
in the observations, earlier than the warming based on the results
of GCM (since 1990s). Fig. 2 shows the 50-year average of tempera-
ture differences between the GCM’s calculations and observations
monthly. On average, in Shanghai, the GCM overestimates the tem-
peratures in winter by about 1.3 ◦ C, while underestimates that in
summer by about 2.3 ◦ C over the past 50 years. It is not difficult to
find that GCM’s simulation capability for regional climate change is
insufficient and its prediction results might have great uncertain-
ties.
18.0
Temperature (oC)
17.4
Annual Mean
17.0
16.0 16.7
15.7 15.7 16.3
16.0
15.5
15.0 15.6
15.4 15.3
14.0
1960s 1970s 1980s 1990s 2000s
Fig. 1. Interdecadal change comparison of annual mean temperatures of Shanghai
in 1960–2010/◦ C.
M. Zhu et al. / Energy and Buildings 113 (2016) 74–86
3.0 2.6 2.4
2.0
Temperature Difference (oC)
1.5
1.1
1.0 0.5
0.0
-1.0
-0.1
-0.5
-2.0
-2.0
-3.0 -2.3
-3.3 -3.3 -3.4
-4.0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Fig. 2. 50-year averages of temperature differences between GCM’s calculations and
observations/◦ C.
2.2. An alternative method for monthly mean temperature
prediction
An accurate prediction of climate change plays a key role in con-
vincing a building energy demand prediction. The improvement of
spatial and temporal accuracy of GCM and geophysics in climate
change prediction needs further ground-breaking researches in a
long run. Otherwise, the obvious seasonal and regional differences
of climate change in China over the latest century present a certain
inconformity with the global continued warming trend [10–15].
In this case, another perspective might work. In this paper, from
the perspective of both regionally long-term climate changes and
short-term meteorological observations, an alternative method –
time series forecasting based on climate periodicity analysis, is put
forward for monthly mean temperature prediction.
2.2.1. Regularity research of long-term historical climate change
in East China
In history, the rapid temperature changes often happen in the
alternating periods of warmth and cold. The 20th century warm-
ing is a signal of warming recovery since the Little Ice Age and
is not a sole warming in history. Over the past 2000 years, there
were four transitions with the same temperature variation range
as the warming in the 20th century in East China [16]. Based on
the temperature series in East China over the past 2000 years,
the historical analog of the 20th century warming have been ana-
lyzed from the aspects of temperature change amplitude and rate
[16–20]. The related researches indicate that the long-term histori-
cal climate change in East China has a certain periodicity. A possible
centennial-scale period with a cycle length of 200–600a and a fluc-
tuation range of about 1 ◦ C would greatly affect climate change in
East China in the 21st century.
The recent observation and simulation researches show that
the climate change in East China is significantly affected by nat-
ural variability. The impact comes from two aspects: external
climate variability and internal oscillation. Currently two most
common external climate variability factors are solar activity and
volcanic activity. The related internal oscillations are Atlantic Mul-
tidecadal Oscillation (AMO), El Nino-Southern Oscillation (ENSO),
etc. According to the researches on periodical features of related
natural variability [21–26], a possible multidecadal period with
a cycle length of 40–80a and fluctuation range of about 1.5–2 ◦ C
Observations would exist in East China in the 21st century.
GCM outputs under
historical weather 2.2.2. Time series method
The meteorological observations are typical time series. The
time series analysis is a popular forecasting method in meteorolog-
ical research. Using this method, a statistical indicator’s variation
over time is analyzed by statistical method and then applied to the
prediction model development, depends on which future trend of
the indicator is quantitatively estimated.
 M. Zhu et al. / Energy and Buildings 113 (2016) 74–86 77

Fig. 3. Time series forecasting method based on climate periodicity analysis.

The application of this method in short-term prediction has a
high reliability. When used in long-term prediction, this method
may cause a large deviation and a considerable inaccuracy. The
basic principle of time series method is that the natural conti-
nuity of thing’s development determines the regularity (found
in observed data) would be extended to the future. The inher-
ent physical properties of climate system (such as the thermal
and moisture storage capacity of soil and water, the heat preser-
vation effect of greenhouse gases) determine that some weather
parameters have temporal continuity, such as the correlation of
annual/monthly/daily temperatures at different points in time.
However, the earliest temperature observation record occurred in
1880. The observations over less than 150 years can be directly used
in weather data predictions only for the next 10 years. Hence, the
limitation of weather observation duration restricts the application
of conventional time series method in long-term weather forecast.
The key point of this problem is that the observations over less
than 150 years cannot provide an accurate regularity of long-term
climate change. Nevertheless, the regularity research of historical
climate change in East China over the past 2000 years can be help-
ful to solve this problem. Under this circumstance, this paper tries
to reasonably integrate the long-term climate regularity with time
78 M. Zhu et al. / Energy and Buildings 113 (2016) 74–86
series method, putting forward an alternative forecasting method.
The following section introduces this method in detail.
2.2.3. Alternative method: time series forecasting based on
climate periodicity analysis
The time series models mostly comprise three terms: a trend
term (Ht), a periodic term (Pt), and a stochastic term (Xt), as shown
in Eq. (1). Ht and Pt represent the deterministic change rule of
observed data, and Xt represents the uncertainty of climate change.
Yt = Ht + Pt + Xt
As a trend term, Ht is an upward or downward trend in the
conventional time series analysis. The trend shows a monotonous
change in some limited time series. While, if the time series extends
to a larger time scale, above trend (Ht) may be an ascent or
descent stage of a longer cycle. This perspective is an important
breakthrough for the development of the alternative method –
time series forecasting based on climate periodicity analysis. The
method introduces the periodicity of historical climate over the
past 2000 years as the trend term (Ht) of the conventional time
series models. In this case, the new forecasting model would have
the capability of mid- or long-term prediction. Based on the possi-
ble periods of climate change mentioned in Section 2.2.1, the time
series method is used to explore some regularities and uncertain-
ties existed in the monthly average temperature observations of
Shanghai from 1961 to 2010. Then a prediction model for future
monthly average temperatures is established based on the above
climate features (both regularity and uncertainty).
Fig. 3 shows the main implement steps of the alternative method
– time series forecasting based on climate periodicity analysis.
From the possible periods of climate change mentioned in Sec-
tion 2.2.1, a centennial-scale period (T1) and a multidecadal period
(T2) are separately selected for becoming a trend term (Ht ) and a
periodic term (Pt ) in the form of cosine functions. Then a prediction
model (Yt ) is built, as shown in Eq. (2).
Yt = Ht + Pt
Obviously, different cycle lengths of Ht (200a, 300a, 400a, 500a
or 600a) and Pt (40a, 50a, 60a, 70a or 80a) term constitute different
Yt .
After excluding a stochastic term (Xt) from the temperature
observation series, the remaining series are called TC series, includ-
ing the trend and periodic features of observation series.
By fitting different Yt to TC series, the optimal T1 and T2 are
respectively identified as the trend and periodic features of the
observed temperature series. Thus, the optimal prediction model
Yt could represents the long-term regularity of regional climate
change.
As the uncertainty factor of the observed series, the stochas-
tic term Xt has been generally considered useless and common
prediction model does not contain an Xt term in the conventional
time series forecasting. And in that way, the predictions just rep-
resent future development tendency since every predicted result
Table 2
Comparison of different versions of existing TMYs for China typical cities.
Version CSWD CTYW
Developer Tsinghua University & China ZHANG Qingyuan of
Meteorological Bureau Tsukuba University & Joe
Huang of LNBLa
Historical weather 270 stations from China 57 stations in China from
data source Meteorological Bureau NCDCb
Historical time span 1971–2003 1982–1997
a
LNBL – Lawrence Berkeley National Laboratory.
b
NCDC – U.S. National Climatic Data Center.
c
NREL – National Renewable Energy Laboratory.
Table 1
Uncertainty scenarios of monthly mean temperature.
Scenarios Descriptions Coefficient
S1m Same as Xt term of observed monthly mean 1
temperature series
S2m Weaker fluctuation than Xt term of observed 0.8
monthly mean temperature series
S3m Stronger fluctuation than Xt term of observed 1.2
monthly mean temperature series
(1)
just play a role of an average value, around which an actual value
fluctuates. However the alternative forecasting model includes a
stochastic term (Xt ) as the uncertainty feature of future climate
change. Using the Xt term of observed monthly average tempera-
ture series as a baseline, this method proposes three scenarios of
the monthly uncertainty (as listed in Table 1) to develop the dif-
ferent prediction Xt term, which is added to the above prediction
model Yt for establishment of the final prediction model (Yt ), as
shown in Eq. (3).
Yt = Ht + Pt + Xt (3)
The prediction model (Yt )
will be used to predict the monthly
mean temperatures of Shanghai in the 21st century.
2.3. Downscaling method for hourly temperature generation:
Morphing
The weather inputs of building energy simulation are the hourly
TMYs. In order to apply the above future monthly mean temper-
atures predicted by the proposed method in Section 2.2.3 to the
building energy demand predictions of Shanghai, a downscaling
method, Morphing, is utilized to generate the future TMYs on the
basis of a local existing TMY.
The Morphing method was constructed by Belcher [27] and
has been widely used by researchers from USA, UK, Australia, and
HK [7,27–31]. By means of shifting and stretching, the predicted
(2)
changes of the future monthly mean temperatures are downscaled
and combined with the hourly temperatures of the existing TMY.
In this way, the new TMYs both synthesis the climate change pre-
diction and keep the physical characteristics of practical weather
data.
Existing TMYs for China typical cities have four versions – CSWD
(Chinese Standard Weather Data), CTYW (Chinese Typical Year
Weather), IWEC (International Weather for Energy Calculation) and
SWEAR (Solar Wind Energy Resource Assessment), published on
EnergyPlus website [32], with different data sources and time spans
of historical weather, as listed in Table 2, but consistently using
Finkelstein–Schafer (FS) method proposed by Hall in 1978.
Previous research has shown a maximum 8% difference of build-
ing energy consumption among the simulation results obtained
using different TMY versions [33,34]. Therefore, any one of
above four TMY versions is reasonable for this research. In this
paper, IWEC-version existing TMY of Shanghai is selected as the
IWEC SWEAR
ASHRAE Research Project US NRELc
1015
DATSAV3 archived at 45 stations from NRELc
NCDCb
1982–1999 1973–2002
 M. Zhu et al. / Energy and Buildings 113 (2016) 74–86
baseline of the future TMYs’ generation. The IWEC-version TMY file
is downloaded from EnergyPlus website [32], and its file name is
“CHN Shanghai.Shanghai.583670 IWEC.epw”. Since the time span
of the IWEC-version TMY is from 1982 to 1999, see Table 2, the next
100 years are divided into five periods: 2000–2017, 2018–2035,
2036–2053, 2054–2071 and 2072–2089, with the same time span
to the baseline weather data, for convenient application of the Mor-
phing method.
In the Morphing method, hourly predicted dry bulb tempera-
ture (dbt) of the new TMYs is calculated on the basis of hourly
temperature in the existing TMY, as shown in Eq. (4) [7].
dbt = dbt0 + TEMPm + ␣dbtm × [dbt0 − (dbt0 )m]
In Eq. (4), (dbt0 )m is defined as monthly mean temperature of m
month in the existing TMY. dbt0 is shifted by the predicted change
of monthly mean dbt of m month (TEMPm) and stretched by the
last item. The monthly stretching factor (␣dbtm ) is calculated by
Eq. (5) [7].
TMAXm − TMINm
˛dbtm =
(dbt0 max )m − (dbt0 min )m
In Eq. (5), (dbt0 max )m and (dbt0 min )m are respectively defined as
the monthly mean daily maximum and minimum temperature of m
month in the existing TMY. TMAXm and TMINm are respectively
defined as the predicted changes of monthly mean daily maximum
and minimum temperature.
The two indicators used for calculation of dbtm , TMAXm and
TMINm , reflect the daily fluctuation situation of m month since the
difference between them is the monthly mean daily range. With
the daily fluctuation situation of temperature observations as the
baseline, three scenarios of daily fluctuation (as listed in Table 3)
are proposed.
Future TMYs in the different time periods are then generated
under three composite scenarios (S1, S2, and S3), which combine
the daily fluctuation scenarios with corresponding fluctuation level
of the uncertainty scenarios of monthly mean temperature.
The above is the proposed alternative forecasting method only
for the single climate variable – Dry Bulb Temperature. Except for
the temperature, other climate variables having influence on build-
ing energy demand, such as relative humidity, wind speed and
orientation, solar radiation, are included in a common TMY file. As
the same as the temperature, the temporal scale of other climate
variables is hourly.
When considering the effects of other climate variables, most
researches on the indirect prediction (that is combination of build-
ing models and future TMYs) of building energy demand under
climate change basically select corresponding climate data sets
(including future temperature, humidity, wind, and solar radiation
predictions) generated by appropriate GCMs or Regional Climate
Models (RCMs) under a specified future scenario [7,31,36,37].
Moreover, researchers determine whether it is necessary to use
a downscaling method (e.g. Morphing), usually according to the
temporal scale of the above selected climate data set. Here exist
two kinds of common situations. One is to use morphing method
to downscale the future MONTHLY climate data set [7,31,36]. In the
morphing method, the calculation equations for hourly predicted
Table 3
Daily fluctuation scenarios.
Scenarios Descriptions
S1d Same as daily fluctuation situation of observed
period
S2d Weaker fluctuation than daily fluctuation
situation of observed period
S3d Stronger fluctuation than daily fluctuation
situation of observed period
79
relative humidity (RH), wind speed (WS), and solar radiation (SR)
of new TMYs are introduced in [7,27,36]. The other is to directly
use the future HOURLY climate data set [37].
For both direct (that is statistical analysis method) and indi-
rect prediction of building energy demand under climate change,
the temperature variable (DBT) is the major independent vari-
able in most researches. In this application field, few researches
on the long-/short-term observation, analysis and prediction of
other variables are mature and recognized, which may be the
theoretical prerequisites of suitable statistical method (e.g. TSM)
application for future other variables generation. Therefore, this
paper emphasizes the alternative forecasting method for only tem-
(4)
perature variable and does NOT involve the prediction of other
variables in TMYs.
3. Results
3.1. Dual-Periodic Time Series Model
(5)
3.1.1. Establishment of Dual-Periodic Time Series Model
The Shanghai observed temperature dataset sourced by U.S.
National Climatic Data Center includes local Dry Bulb Temperature
records in every 3 h from 1961 to 2010. By simple pre-processing,
we got a total of 600 monthly temperatures series of 50 years, which
was used for the following Dual-Periodic Time Series Model estab-
lishment after excluding its stochastic term (Xt) as mentioned in
Fig. 3. To eliminate the interference of the seasonal trend in a year
on the modeling and predicting process, the remained TC series of
600 monthly temperature observations dataset was decomposed to
12 subsets for 12 months in a year. For month m in a year (e.g. Feb.),
its subset includes a total of 50 monthly temperatures TC series in
every year from 1961 to 2010. Therefore, the proposed forecasting
method application and the following predictive modeling were
respectively conducted for each month in a year from Jan. to Dec.
During times of fitting process with the monthly temperatures
TC series of Shanghai from 1961 to 2010, as shown in Fig. 3, the
optimal T1 and T2 that respectively represents the trend and peri-
odic features of local climate change in are 300a and 60a. The
fitting results mean that a prediction model (Yt ) consists of a trend
term (Ht ) and a periodic term (Pt ) in the form of cosine functions
respectively with 300a- and 60a-cycle could reflects the long-term
regularity of climate change in Shanghai. It is important to note that
we implemented three considerations in the selection of the opti-
mal T1 and T2 from various fitting results. The first consideration is
the fitting optimization index (R2 ) between each prediction model
(Yt ) and each monthly temperature TC series. The second principle
is that the 12 prediction models (Yt ) for 12 TC series have the same
cycle length of T1 and T2, and the same amplitude (about 1 ◦ C) of
the longer period (T1), but different amplitude of the short period
(T2) with consideration of seasonal factors. The amplitude of the
short period (T2) for winter months is about 2 ◦ C, those for sum-
mer months and transition months are respectively about 1.5 ◦ C
and 1.5–1.8 ◦ C. The third concern is the reflection degree of sev-
eral key warm and cold phases in the recent millennium by each
prediction model (Yt ). With the above considerations, the fitting
optimization index (R2 ) of the selected optimal Yt models for each
month ranges from 0.7 to 0.84, which is acceptable for the model
prediction performance.
Coefficient To compare the performance of IPCC’s GCM outputs and the
optimal fitting Yt on characterizing the temperature trend of the
1
recent period, we added the fitting Yt to Fig. 1, and increased the
0.8 time accuracy from 10-year average to 5-year average for more
accurate comparison. As shown in Fig. 4, the optimal fitting Yt
1.2 is obviously much closer to the observed temperature trend from
1961 to 2010 than IPCC’s GCM outputs. It was calculated that the
80
Fig. 4. 5-year average temperature trend comparison of GCM and optimal fitting
Yt /◦ C.
50-year Mean Absolute Percent Error (MAPE) between GCM out-
puts and the observed annual temperatures is 5.6%, while that of
the optimal fitting Yt is only 1.9% and better than the former. In the
same way, we also compared the 50-year MAPE index of GCM out-
puts and the optimal fitting Yt for Jan., Apr., Jul. and Oct. As shown
in Fig. 5, the MAPE between Yt and the observation for each month
is much less than that between GCM outputs and the observation.
The MAPE of both Yt and GCM for Jan. are much larger than other
months and annual average because of its lower monthly temper-
ature (about only 3–6 ◦ C). In order to eliminate the interference
of different monthly temperature level, we compared an absolute
deviation index-Root Mean Square Error (RMSE). As shown in Fig. 6,
the RMSE between Yt and the observation for Jan., Apr., Jul., Oct.
and Annual ranges from 0.4 ◦ C to 1.1 ◦ C, while that between GCM
outputs and the observation ranges from 1.1 ◦ C to 3.9 ◦ C. The above
comparative analysis can obviously illustrate that the optimal fit-
ting model Yt can reflect the recent monthly temperature trend
much more accurately than GCM for Shanghai region.
As mentioned in the alternative method introduction, the dual-
periodic fitting model Yt can represent the long-term regularity
of regional climate change. However, for practical and comprehen-
sive prediction, the uncertainty feature of future climate change is
60%
50-year of monthly DBT MAPE (%)
48.8%
50%
40%
30%
22.7%
20% 16.3%
12.5%
10% 8.0%
5.6%
3.0% 3.6%
0%
Jan. Apr. Jul. Oct.
Fig. 5. MAPE comparison of monthly DBT fitting by Optimal fitting Yt and GCM/%.
Fig. 6. RMSE comparison of monthly DBT fitting by Optimal fitting Yt and GCM/◦ C.
M. Zhu et al. / Energy and Buildings 113 (2016) 74–86
Table 4
Description of RCPs.
Name Radiative forcing Path shape
RCP8.5 >8.5 W/m2 in 2100 Rising
RCP6.0 ∼6.0 W/m2 at stabilization Stabilization without
after 2100 overshoot
RCP4.5 ∼4.5 W/m2 at stabilization Stabilization without
after 2100 overshoot
RCP3-PD Peak at ∼3.0 W/m2 before Peak and decline
2100 and then declines
a key factor and should not be overlooked. Here the Dual-Periodic
Time Series Model Yt totally includes the optimal fitting model
Yt , which is established and assessed above, and a stochastic term
(Xt ), which is proposed by using the Xt term of observed monthly
temperature series as a prototype and applied to reflect the uncer-
tainty feature of future monthly temperature change. Using the
Dual-Periodic Time Series Model Yt of each month in a year (from
Jan. to Dec.), we predicted the monthly temperatures of each month
for future 100 years.
3.1.2. Preliminary verification and evaluate of Dual-Periodic Time
Series Model
For preparation of Fifth Assessment Report (AR5), the IPCC has
developed new climate path series – Representative Concentra-
tion Pathways (RCPs), including four grades: RCP8.5, RCP6, RCP4.5
and RCP3-PD [35]. The number after the “RCP” means the intensity
of radiation forcing due to the human activity by 2100, as listed
in Table 4. The monthly temperature predictions of HadGEM2-CC
model under the stable path, RCP4.5 (regarded as the most pos-
sible scenario by researchers), at the nearest node to Shanghai are
identified as the IPCC RCP4.5 predictive dataset and to be compared
with the monthly temperatures predicted by the new Dual-Periodic
Time Series Model (New TSM) in this paper.
For preliminary verification and evaluation of the new Dual-
Periodic Time Series Model’s predictive performance, we collected
hourly observed Dry Bulb Temperature data of two recent years
(2013 and 2014) from local weather station. These temperature
50-year MAPE between Yt' fitting
observations were used as the baseline for the comparative anal-
and Observation
50-year MAPE between GCM ysis of the New TSM and IPCC RCP4.5 predictions. Figs. 7 and 8
outputs and Observation
respectively shows the comparison of New TSM and IPCC RCP4.5
monthly temperature predictions with observed results in 2013
and 2014. From the point of a whole year, the monthly tempera-
tures MAPE between New TSM and the observations in 2013 and
2014 are respectively 11.3% and 12.5%. Both of them are lower than
those of IPCC RCP4.5 (respectively 18.0% and 15.2%), as shown in
5.6%
1.9% Fig. 9. While from the angle of the absolute error, we calculated
that the monthly temperatures RMSE between New TSM and the
Annual
Average observations in 2013 and 2014 are respectively 2.1 ◦ C and 1.8 ◦ C,
Fig. 7. Monthly DBT comparison of New TSM and IPCC RCP4.5 prediction with the
observations in 2013/◦ C.
 M. Zhu et al. / Energy and Buildings 113 (2016) 74–86 81

4.5 4.3
New TSM
4.0 IPCC RCP4.5

RMSE of monthly DBT Prediction (oC)

3.5

3.0
2.5
2.5
2.1
2.0 1.8

1.5

1.0

0.5
Fig. 8. Monthly DBT comparison of New TSM and IPCC RCP4.5 prediction with the
observations in 2014/◦ C. 0.0
2013 2014
20.0% New TSM
18.0% Fig. 10. RMSE comparison of monthly DBT prediction by New TSM and IPCC
18.0% IPCC RCP4.5 RCP4.5/◦ C.
MAPE of monthly DBT Prediction (%)

16.0% 15.2%

14.0%
12.5%
12.0% 11.3%
10.0%
8.0%
6.0%
4.0%
2.0%
0.0%
2013
Fig. 9. MAPE comparison of monthly DBT prediction by New TSM and IPCC RCP4.5/%.
while those of IPCC RCP4.5 are respectively 4.3 ◦ C and 2.5 ◦ C, as
shown in Fig. 10. It is clear that the monthly temperatures resulted
by New TSM are much closer to the actual observations than IPCC
RCP4.5 in the recent two years, especially in 2013. From the per-
spective of every monthly prediction, the prediction performance
of the two models has significant seasonal differences. Fig. 11 shows
the MAPE comparison of DBT predictions resulted by New TSM
and IPCC RCP4.5 for different seasons in 2013 and 2014. In both
2013 and 2014, the forecast error of New TSM appeared in three
winter months: Feb., Nov. and Dec. The average MAPE of the three
winter months reached 25.0%. However, the absolute errors of the
three months range from 1.8 ◦ C to 2.6 ◦ C, not very large. The lower
winter monthly temperatures may be the reason why these MAPE
are much higher than other months. Besides, its average MAPEs
of the other months (including transition and summer months) in
2013 and 2014 are only 6.5% and 7.8%, which is accurate enough for
the following building energy demand prediction. While for IPCC
RCP4.5, its forecast error not only appeared in winter months (Jan.,
Feb. and Dec.) but also in summer and transition months. The aver-
age MAPEs of the three winter months resulted by IPCC RCP4.5
in 2013 and 2014 are 22.3% and 29.6%, lager than those of New
TSM. Besides, the average MAPEs of two transition months (Apr.
and May.) resulted by IPCC RCP4.5 in 2013 and 2014 are 21.8% and
2014 23.8%. The prediction accuracy of IPCC RCP4.5 for summer months
is not stable. The average MAPE of three summer months (Jun., Jul.
and Aug.) resulted by IPCC RCP4.5 in 2013 is 24.3%, while that in
2014 is 7.5%. Generally speaking, the above preliminary analysis
of the two models the prediction demonstrates that the predictive
performance of the New TSM is better than the IPCC RCP4.5 both
throughout a whole year and on a season scale, especially in spring
and summer.
3.2. Regional climate change prediction in the 21st century
After the comparison and evaluation of the new Dual-Periodic
Time Series Model’s performance on characterizing and predicting
the temperature trend of the recent period, the new Dual-Periodic
Time Series Model is applied to predict monthly temperatures
under three uncertainty scenarios for future five time spans
(2000–2017, 2018–2035, 2036–2053, 2054–2071 and 2072–2089).

35.0% New TSM

29.6%
MAPE of seasonal DBT Prediction (%)

IPCC RCP4.5
30.0%

23.8% 24.3%
25.0% 22.3% 21.9%
21.8% 21.6%
20.0%

15.0%
10.3% 9.8%
10.0% 8.9%
7.5%
6.1% 6.4% 6.3%
5.3%
4.3%
5.0%

0.0%
Spring in Spring in Summer Summer Autumn inAutumn in Winter in Winter in
2013 2014 in 2013 in 2014 2013 2014 2013 2014

Fig. 11. MAPE comparison of seasonal DBT prediction by New TSM and IPCC RCP4.5/%.
82
4.0
Annual Temperature Changes ( oC)
3.0
2.0
1.0
0.0
-1.0
2000-2017 2018-2035 2036-2053 2054-2071 2072-2089
Fig. 12. Annual temperature changes relative to baseline period in Shanghai/◦ C.
The predicted temperatures under S1m, S2m, and S3m fluctuate
within certain ranges in the five time spans.
As shown in Fig. 12, the annual temperatures of the five time
spans basically have the same changing trend, fluctuating within
a limited range of 1.5 ◦ C around the baseline level. Relative to
the baseline weather condition in the time span of 1982–1999,
the annual temperatures of both 2000–2017 and 2054–2071 are
higher, and the annual temperatures of other three time spans are
lower. The effect of uncertainty scenarios is more obvious on the
time span of 2054–2071 than other periods.
As shown in Fig. 13, the seasonal temperatures of summer basi-
cally have the same changing trends with annual temperatures,
fluctuating within a limited range of 1.0 ◦ C around the baseline
level. However, there have a couple of minor differences. The low-
est annual temperature occurs in the time span of 2018–2035,
while the lowest summer temperature occurs in the time span
of 2036–2053. Although both of them have fluctuant downtrend
in the five periods, the annual temperatures fluctuate around the
baseline level, while the summer temperatures are higher than the
baseline level.
As shown in Fig. 14, the seasonal temperatures of winter rel-
atively have more complicated trend, fluctuating within a limited
range of 2.0 ◦ C around the baseline level. The changes of winter
temperatures under different uncertainty scenarios are obviously
different. Under S1m and S2m, the winter temperatures have fluc-
tuant downtrend in the five spans. While the second peaks under
S1m and S2m separately occurs in the time span of 2036–2053 and
2054–2072. The occurrence time spans of the winter temperature
4.0
Summer Temperature Changes ( oC)
3.0
2.0
1.0
0.0
-1.0
2000-2017 2018-2035 2036-2053 2054-2071 2072-2089
Fig. 13. Summer temperature changes relative to baseline period in Shanghai/◦ C.
M. Zhu et al. / Energy and Buildings 113 (2016) 74–86
4.0
RCP4.5 RCP4.5
S1m S1m
S2m 3.0
S2m
Winter Temperature Changes ( oC)
S3m S3m
2.0
1.0
0.0
-1.0
-2.0
2000-2017 2018-2035 2036-2053 2054-2071 2072-2089
Fig. 14. Winter temperature changes relative to baseline period in Shanghai/◦ C.
peaks and valleys under S2m and S3m are the same. However,
under S3m, the second peak is higher than the first one. That is to
say, the winter temperatures under S3m have a fluctuant uptrend.
Figs. 12–14 respectively show that either the annual or the sea-
sonal temperatures have steady increase trend under RCP4.5, with
obviously higher temperatures than S1m, S2m, and S3m.
3.3. Regional building energy demand prediction in the 21st
century
Based on the local existing IWEC-version TMY as the baseline
condition, the monthly temperature predictions in Section 3.2 are
downscaled to generate the future TMYs in the five time spans
(2000–2017, 2018–2035, 2036–2053, 2054–2071 and 2072–2089)
by Morphing method under the three composite scenarios (S1, S2,
and S3). These future TMYs are used as weather inputs of three
prototypical building models (high-rise office building, hotel and
shopping mall) of Shanghai for regional building energy demand
prediction. Meanwhile, the monthly outputs of the selected GCM
(HadGEM2-CC) under RCP4.5 are downscaled by Morphing method
with consideration of the daily fluctuation scenarios, and the cor-
responding future TMYs in the five time spans are generated on
the basis of the same existing baseline TMY. Then, a comparative
analysis is conducted between energy simulation result of the pro-
totypical models using the future TMYs under S1, S2, and S3 and
that under RCP4.5.
3.3.1. Regional prototypical building models
RCP4.5
The approach for establishing prototypical building models is
S1m basically mature and widely used. According to related design stan-
S2m dards and sufficient survey data of existing buildings in Shanghai,
S3m we developed three prototypical building models by EnergyPlus
in this paper. After calibrated by investigated building energy use
data, these models can generally reflect the local energy use situa-
tion of different building types. This paper focuses on three building
types – office building, hotel and shopping mall, as shown in Table 5.
There exists certain limitation when we use the existed pro-
totypical models as the constant building energy model for the
future building energy demand prediction, without consideration
of the expected evolution of building stock. Both advanced building
designs (e.g. optimized building envelope) and building energy effi-
ciency improvement (e.g. efficient building services) are promising
approaches to reducing/utilizing energy consumption impacts of
climate change. Some studies discussed the additional effects of
excepted building evolutions on future building energy demand
 M. Zhu et al. / Energy and Buildings 113 (2016) 74–86 83

Table 5
Building information of prototypical models.

Type Floors Floor-floor Area (M2 ) WWRa Heat transfer coefficient (W/M2 K) Shape HVAC system
height (M)
Exterior wall Roof Window

Office 12 4 19,200 0.40 0.888 0.638 2.67 Chiller & boiler FCU + OA

Hotel 12 3.5–5 23,316 0.42 0.791 0.552 2.959 Chiller & boiler FCU + OA

Shopping mall 7 above, 1 under 3.5 33,800 0.11 1.304 0.653 3.667 Chiller & boiler AHU

a
WWR – window–wall ratio.

under climate change [39,40] by means of existed prototypical studies use econometric modeling of building energy demand
models [5–7,38–40]. In this case, some actual national/regional based on databases of existing buildings.
energy programs (e.g. Department of Energy’s energy efficiency Here this paper emphasizes the future weather data predicted
programs) are considered as the driver for various adaptive strate- by the proposed alternative forecasting method and the corre-
gies and technologies to climate change. sponding building energy demand prediction under the expected
Except for the evolution of building stock, two socio-economic climate change. Given this, we use the existing prototypical build-
factors also influence the prediction of future building energy ing models as the constant building model for the future energy
demand, either in the national or in the regional scale. One is the demand prediction, without consideration of either building stock
demographic shift and growing. Another is the consumer reac- evolution or socio-economic factors. Further research work is being
tion to climate change. Most of relevant studies have two research continued on investigation into the multifactorial impact of climate
routine [5]. Roughly half of the studies use building energy simu- change, advanced building design, and efficient building services
lation tools and explicitly account for the future population trends on future building energy demand prediction on the scale of both
and adaption strategies by means of scenario planning. The other individual building and urban district.

60 2000-2017
2018-2035
50 46.1
2036-2053
41.3
2054-2071
Change Percentages of Office Building Energy Demand (%)

40
2072-2089
30.1
30

19.2
20 17.4
13.2 13.9
11.8
10.5
10 5.3 5.7
3.4 3.0 2.4 2.6 2.6
2.3
0.9 1.3 1.2
0
-1.0 -1.1 -0.6
-2.6

-10 -6.5

-20
-19.2

-30 -28.3

-40 -37.7

-44.3
-50
-50.7

-60
Cooling Heating Whole Cooling Heating Whole
Building Building
RCP4.5 S1

Fig. 15. Change percentages of office building energy demand under RCP4.5 and S1 relative to baseline period in Shanghai/%.
84 M. Zhu et al. / Energy and Buildings 113 (2016) 74–86

3.3.2. Building energy simulation using future TMYs building energy use is larger than that of heating energy consump-
In this paper, the energy simulation result of the prototypical tion in Shanghai.
models using the existing TMY (1982–1999) is regarded as the Under S1, with the temperatures fluctuating, both the cool-
uniform energy demand baseline. In that way, the effects of cli- ing and heating energy demand have corresponding fluctuant
mate change on building energy demand are mainly reflected in the trends. Relative to the baseline period, the temperature increases
changes of building heating and cooling energy demand. The build- of both summer (1.0 ◦ C) and winter (0.5 ◦ C) in 2000–2017 cause
ing energy simulation results using the future TMYs under S1, S2, that the reduction of heating demand offsets the most of cooling
S3 and RCP4.5 are compared and analyzed from the following three demand increase. In this case, the change percentage of whole-
aspects: change percentage of whole-building energy demand rel- building energy demand is not obvious in this time span. By
ative to the baseline period, change percentage of heating demand 2036–2053, a decrease of 0.3 ◦ C in summer and an increase of
relative to the baseline period, and change percentage of cooling 0.3 ◦ C in winter cause the simultaneous decline of both heating
demand relative to the baseline period. and cooling demand, which make whole-building energy demand
Both RCP4.5 and S1 are the steady-development scenarios of three building types smaller than that in the baseline period and
among their own scenario series. Fig. 15 shows the comparison reach their valleys among the five periods. While in 2072–2089,
of office building energy simulation results under RCP4.5 and S1. the situation is exactly contrary. An increase of 0.4 ◦ C in sum-
Under RCP4.5 (that is, steady warming trend), the cooling energy mer and a decrease of 1.2 ◦ C in winter cause the simultaneous
demand of office building, hotel and shopping mall have contin- rise of both heating and cooling demand. Relative to the baseline
uous uptrends and the change percentages are 46.1%, 79.7% and period (1982–1999), the largest increase percentages of the whole-
22.3% in 2072–2089 respectively, relative to in 1982–1999. The building energy demand of office building, hotel and shopping mall
heating energy demand of office building, hotel and shopping under S1 are respectively 2.6%, 3.1% and 1.4%. All the change per-
mall have continuous downtrends and the change percentages centages are much lower than that under RCP4.5.
are 50.7%, 19.5% and 99.2% in 2072–2089 respectively, relative to The effects of the three composite scenarios (S1, S2, and S3) on
in 1982–1999. After the offset of cooling demand increase and building energy demand are similar in the three building types.
heating demand decrease, the whole-building energy demand of Fig. 16 shows the comparison of office building energy simulation
office building, hotel and shopping mall still have continuous results under S1, S2, and S3. The fluctuation ranges of both the sea-
uptrends and the change percentages are 5.7%, 10.4% and 5.7% sonal energy demand and the whole-building energy demand have
in 2072–2089 respectively, relative to in 1982–1999. The reason certain increases with the enhancement of the monthly uncertainty
is that the proportion of cooling energy consumption in whole and daily fluctuation.

25 2000-2017
2018-2035
Change Percentages of Office Building Energy Demand (%)

2036-2053
20 19.2
2054-2071
17.5 17.4
16.2 16.3 2072-2089
15.2
15 13.9
13.2 13.3

11.3
10.5

10 8.9

7.4 7.7
6.4

5 3.5
3.0 3.0 3.0
2.3 2.4 2.62.6 2.3 2.4
1.9 2.0
1.2 0.9 1.01.2 1.2

0
-0.6 -0.4
-1.0 -1.1
-1.8
-2.6

-5 -3.8
-5.3
-5.9
-6.5
-7.6

-10
-10.3
-11.3

-15
Cooling Heating Whole Cooling Heating Whole Cooling Heating Whole
Building Building Building
S1 S2 S3

Fig. 16. Change percentages of office building energy demand under S1, S2 and S3 relative to baseline period in Shanghai/%.
 M. Zhu et al. / Energy and Buildings 113 (2016) 74–86
4. Conclusion
In IPCC’s detection and attribution analysis, the long-lived
greenhouse gas (mainly means CO2 , N2 O and CH4 ) came from
anthropogenic activity is the leading radiative forcing component
causing the global climate change. In this case, the greenhouse
gas change trends under different emission scenarios become the
significant driving force of GCM, which is the key tool of IPCC’s
future global climate change projections. As a result, all of the
monthly, the seasonal and the annual temperature of Shanghai
have steady warming trend with large temperature increases under
IPCC’s RCP4.5, the most possible scenario by researchers.
IPCC’s research work has attracted widespread attention as the
raised concern of climate change around the world in recent years.
Due to this global issue involving various researches across multiple
disciplines, some controversy and debate has come out. The main
argument revolves around the attribution of climate change, either
nature or anthropogenic activity. Correlational study around this
argument is extremely complicated [8,9] and not yet been deter-
mined. However, the weather data prediction research in this paper
does NOT focus on the cause of climate change, but the regional
long- and short-term temperature observations and their regular-
ity, which can be seen as the expression of various complex reasons
of climate change.
This study proposed an alternative forecasting method, time
series forecasting based on climate periodicity analysis, and applied
this method to the future monthly temperatures prediction of
Shanghai. This alternative method combines the centennial-scale
periodicity (T1) over the past 2000 years and the multidecadal peri-
odicity (T2) over the past several centuries with the features of
trend term (Ht) and period term (Pt) in time series model. In this
way, the improved time series method is suitable for long-term
prediction.
By using this alternative method, we established a Dual-Periodic
Time Series Model that well fitting with the observed monthly DBT
series from 1961 to 2010 of Shanghai. The fitting results indicate
that the New TSM has better capability of characterizing the tem-
perature trend of the recent period in Shanghai than the selected
IPCC GCM under RCP4.5. Besides, the New TSM performs more
accurate predictions in the following recent two years than IPCC
GCM under RCP4.5 in the preliminary verification and assessment.
Therefore, we believe that, the alternative forecasting method and
the corresponding Dual-Periodic TSM has greater prediction per-
formance for future temperature trends in Shanghai than IPCC GCM
under RCP4.5.
The future monthly temperature predictions of the New
TSM present the dual-cycle fluctuant trends with certain ranges
(1.0–2.0 ◦ C) in the 21st century. Using the future TMYs (gener-
ated with the monthly temperatures predicted by the alternative
method) as the weather input of different prototypical buildings,
the simulated building energy demand also presents fluctuant
trends in the future periods. Relative to the baseline time span
of 1982–1999, the largest increase percentages of whole-building
energy demand of office building, hotel and shopping mall are
respectively 2.6%, 3.1% and 1.4% in the 21st century. The change
percentages are lower than that under RCPs scenario.
In this paper, the alternative method is proposed from the per-
spective of the long-term historical climate change regularity of
East China. Given the limitations of the GCMs on regional-scale
weather data prediction, we can use this method, time series fore-
casting based on climate periodicity analysis, as an alternative and
supplementary way for more accurate weather data prediction of
East China. For the other regions, the local climate change period-
icity analysis is an essential premise to develop locally appropriate
Dual-Periodic TSM in the similar way, as well as sufficient temper-
ature observations.
85
References
[1] Building Energy Conservation Research Center, Tsinghua University, Annual
Report on China Building Energy Efficiency, China Building Industry Press,
Beijing, 2009.
[2] M. Christenson, H. Manz, D. Gyalistras, Climate warming impact on
degreedays and building energy demand in Switzerland, Energy Convers.
Manage. 47 (2006) 671–686.
[3] M. Li, J. Guo, J. Shi, et al., Applicability evaluation of cooling/heating
degree-days in analyzing building energy consumption changes, Adv. Clim.
Change Res. 9 (2013) 43–48.
[4] J.C. Lam, K.K.W. Wan, T.N.T. Lam, et al., An analysis of future building energy
use in subtropical Hong Kong, Energy 35 (2010) 1482–1490.
[5] T.J. Wilbanks, V. Bhatt, D.E. Bilello, et al., Effects of climate change on energy
production and use in the United States. U.S. Climate Change Science
Program, Synthesis and Assessment Product 4.5, 2008.
[6] P. Xu, Y.J. Huang, N. Miller, Effects of global climate changes on building
energy consumption and its implications on building energy codes and policy
in California, Pier final project report, 2009, CEC-500-2009-006.
[7] A.L.S. Chan, Developing future hourly weather files for studying the impact of
climate change on building energy performance in Hong Kong, Energy Build.
43 (2011) 2860–2868.
[8] IPCC, Climate Change 2007: The Physical Science Basis, Contribution of
Working Group I to the Fourth Assessment Report of the Intergovernmental
Panel on Climate Change, Cambridge University Press, Cambridge, UK and
New York, NY, USA, 2007.
[9] S. Fred Singer (Ed.), Nature, Not Human Activity, Rules the Climate: Summary
for Policymakers of the Report of the Nongovernmental International Panel
on Climate Change, The Heartland Institute, Chicago, IL, 2008.
[10] W. Shaowu, Global climate warming and its trend in the future, Quat. Sci. 3
(1991) 269–276.
[11] W. Shaowu, Y. Jinlin, G. Daoyi, et al., Construction of mean annual
temperature series for the last one hundred years in China, Q. J. Appl.
Meteorol. 9 (1998) 392–401.
[12] W. Shaowu, Z. Zongci, A study of the trend of climatic change during the
period of next 50 years, Q. J. Appl. Meteorol. 6 (1995) 333–342.
[13] W. Shaowu, Z. Zongci, T. Guoli, Focal topic: climate change and China’s
national security, Int. Polit. Q. 30 (2009) 1–11.
[14] W. Shaowu, Y. Jinlin, An analysis of global warming during the last one
hundred years, Sci. Atmos. Sin. 19 (5) (1995) 545–553.
[15] Y. Rucong, Z. Tianjun, L. Jian, et al., Progress in the studies of
three-dimensional structure of interdecadal climate change over eastern
China, Chin. J. Atmos. Sci. 32 (2008) 893–905.
[16] J. Zheng, Q. Ge, X. Fang, Seeing the 20th century warming from temperature
changes of winter-half-year in eastern China for the last 2000 years, Acta
Geogr. Sin. 57 (2002) 631–638.
[17] T.J. Crowley, T.S. Lowery, How warm was the medieval warm period? AMBIO
29 (2000) 51–54.
[18] D. Zhang, A study on the Medievel Warm Period in China, Quat. Res. 1 (1993)
7–13.
[19] Q. Ge, X. Fang, J. Zheng, Warm-cold change in millenarian cycle derived from
natural proxy data in China during the past 3000 years, Adv. Earth Sci. 17
(2002) 96–103.
[20] Q. Ge, J. Zheng, J. Liu, Amplitude and rhythm of winter half-year temperature
change in eastern China for the past 2000 years, Adv. Clim. Change Res. 2
(2006) 108–112.
[21] E. Friis-Christensen, K. Lassen, Length of the
solar cycle: an indicator of solar activity closely associated with climate, Sci.
New Ser. 254 (1991)
698–700.
[22] M. Tang, Y. Liu, W. Guo, Climatic condition and Chinese history (I): SCL and
Chinese climate, Plateau Meteorol. 20 (2001) 369–373.
[23] X. Gao, Y. Liu, W. Dong, et al., New evidences for the relationship between
climate change and geomagnetic field, Plateau Meteorol. 22 (2002)
395–401.
[24] S. Li, Y. Wang, Y. Gao, A review of the researches on the Atlantic Multidecadal
Oscillation (AMO) and its climate influence, Trans. Atmos. Sci. 32 (2009)
459–465.
[25] Z. Lin, S. Wang, EMD analysis of northern hemisphere temperature variability
during last 4 centuries, J. Trop. Meteorol. 20 (2004) 90–96.
[26] Y. Wang, Z. Gong, L. Xu, et al., Analysis of long trend of temperature and
precipitation in China and its influencing factors, J. Appl. Meteorol. Sci. 16
(2005) 85–91.
[27] S.E. Belcher, J.N. Hacker, D.S. Powell, Constructing design weather data for
future climates, Buil. Serv. Eng. Res. Technol. 26 (2005) 49–61.
[28] D.B. Crawley, Creating weather files for climate change and urbanization
impacts analysis, Proc. Build. Simul. (2007) 1075–1082.
[29] M.F. Jentsch, A.S. Bahaj, P.A.B. James, Climate change future proofing of
buildings – generation and assessment of building simulation weather files,
Energy Build. 40 (2008) 2148–2168.
[30] L. Guan, Preparation of future weather data to study the impact of climate
change on buildings, Build. Environ. 44 (2009) 793–800.
[31] A.L.S. Chan, Developing a modified typical meteorological year weather file
for Hong Kong taking into account the urban heat island effect, Build. Environ.
46 (2011) 2434–2441.
86 M. Zhu et al. / Energy and Buildings 113 (2016) 74–86
[32] http://apps1.eere.energy.gov/buildings/energyplus/cfm/weather data3.cfm/
region=2 asia wmo region 2/country=CHN/cname=China.
[33] D. Seo, J. Huang, M. Krarti, Impact of typical weather year selection
approaches on energy analysis of buildings, ASHRAE Trans. 116 (2010) 1.
[34] Z. Shen, H. Tan, S. Lu, E. Kazuo, Research on the typical meteorological year
data of Shanghai for the building energy consumption calculation, HV&AC 40
(2010) 89–94.
[35] R.H. Moss, J.A. Edmonds, et al., The next generation of scenarios for climate
change research and assessment, Nature 463 (2010) 747–775.
[36] H. Wang, Q. Chen, Impact of climate change heating and cooling energy use in
buildings in the United States, Energy Build. 82 (2014) 428–436.
[37] T. Berger, C. Amann, H. Formayer, A. Korjenic, B. Pospichal, C. Neururer, R.
Smutny, Impacts of urban location and climate change upon energy demand
of office buildings in Vienna, Austria, Build. Environ. 81 (2014) 258–269.
[38] Y.J. Huang, The Impact of Climate Change on the Energy Use of the U.S.
Residential and Commercial Building Sectors, LBNL-60754, Lawrence
Berkeley National Laboratory, Berkeley, CA, 2006.
[39] M.J. Scott, J.A. Dirks, K.A. Cort, The adaptive value of energy efficiency
programs in a warmer world, in: International Energy Program Evaluation
Conference, Madison, WI, United States (US), 2005, Print.
[40] M.J. Scott, D.L. Hadley, L.E. Wrench, Effects of climate change on commercial
building energy demand, Energy Sources 16 (3) (1994) 339–354.