Available online at www.sciencedirect.

com

ScienceDirect
Procedia Engineering 118 (2015) 842 – 852

International Conference on Sustainable Design, Engineering and Construction

Energy Use Intensity Estimation Method Based on Façade Features

Chao Yanga, Joon-Ho Choib*
a
University of Southern California, Building Science, School of Architecture, Los Angeles, CA 90089, United States
b
University of Southern California, Building Science, School of Architecture, Watt Hall #318, Los Angeles, CA 90089, United States

Abstract

In the U.S., over 72% of the total generated power is consumed by commercial and residential buildings. Among a building’s
envelope, system, and control, which significantly influence the building’s energy performance, façade is a major parametric
element that accounts for 70% of its energy performance. Compared with the internal mechanical system and operation schedule,
façade features information is relatively easy to obtain from the visual aspects of a building. By adopting several key façade
attributes, a customized energy use intensity baseline model can be generated by considering building design features. Therefore,
instead of using traditional and complicated simulation methods, a mathematical model can be established to estimate EUI
baselines based on sufficient existing building practices data. In a national building performance survey, data such as CBECS and
building energy usage are collected for a large database to provide performance guidance for new or renovation building projects.
Unfortunately, averaged performance data are too aggregated and generic to identify specific conditions for each building
category in a specific climate condition. In this research, a vision-based performance prediction model was developed to
estimate building energy consumption based on simplified façade attribute information and weather conditions. Data about
building façade features, including orientation, façade area, window-to-wall ratio, volume, surface-to-volume ratio, etc., were
collected along with energy use public disclosure. A prediction model, based on this dataset, was established to estimate building
energy use intensity as a function of façade features. This prediction approach will provide a more realistic EUI estimation tool
for calculating an energy use baseline and will enable real-time energy usage monitoring and management of each target
building.
©©2015
2015TheThe Authors.
Authors. Published
Published by Elsevier
by Elsevier Ltd.
Ltd. This is an open access article under the CC BY-NC-ND license
Peer-review under responsibility of organizing committee of the International Conference on Sustainable Design, Engineering
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of organizing committee of the International Conference on Sustainable Design, Engineering and
and Construction 2015.
Construction 2015

Keywords: Energy Use Intensity; Façade Features; Regression Model;

* Joon-Ho Choi. Tel.:+1-213-740-4576.

E-mail address: [email protected]

1877-7058 © 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of organizing committee of the International Conference on Sustainable Design, Engineering and Construction 2015
doi:10.1016/j.proeng.2015.08.522
 Chao Yang and Joon-Ho Choi / Procedia Engineering 118 (2015) 842 – 852 843

1. Introduction

In 2010, the U.S. consumed 97.8 quads of energy, which accounted 19% of global energy consumption [1]. In the
United States, the buildings sector, including both residential and commercial buildings, represented about 41% of
primary energy consumption. Among different building energy uses, space cooling, space heating and lighting are
the dominant end uses, which accounted for about 52% of total energy consumed by buildings sector. Façade
features, such as exterior wall type, glazing type, shading type, window-to-wall ratio etc. have a great influence on
space heating, cooling and even lighting demand [2]. To reduce energy demand a good building façade design is
greatly significant by determining the optimum façade features according to local climate.
Energy Use Intensity (EUI) presents a building’s energy use in terms of its function, size and other
characteristics, which is calculated by dividing annual building energy consumption in one year by the total gross
floor area as kBtu/sf. EUI is a very important indicator [3] to evaluate building energy performance and energy
saving potential. Annual EUI could also be used as the baseline indicator for building owners and designers to set a
comparable energy reduction goal for the following years. In addition, demands from urban planners and building
designers require a new method to predict building energy use by using a simple way at the early design stage, which
could use easily accessible information like building façade features.
Benchmarking indicates tracking and recording each building’s energy use to establish a baseline of energy
performance. By using an equitable metric to compare each building energy use with its past performance as well as
equivalent and similar buildings, building owners and managers will be capable of knowing their building energy
use more deeply and the potential of improving efficiency and making the most cost effective decision. Over 35,000
buildings used U.S. Environmental Protection Agency’s (EPA’s) ENERGY STAR Portfolio Manager to benchmark
energy use [4].
Architecture 2030 [5] was established to promote energy reduction by changing buildings into a solution of global
energy crisis. This action adopts the Commercial Buildings Energy Consumption Survey (CBECS) 2003 data, which
provides national and regional medians as the baseline. CBECS is a national building sample survey [6] that collects
information on the stock of U.S. commercial buildings, including their energy-related building characteristics and
energy usage data. Figure 1 presents an example of national median reference EUI of selected building types. Energy
use intensity (EUI) baseline currently relies on a national or local energy usage average based on census division,
climate zone, building size or year constructed, which can hardly represent the specific physical condition of each
building and other building characteristics, since it doesn’t consider any individual building feature. Besides, the
average value from certain census division, climate zone or HDD/CDD (heating degree day/cooling degree day)
range, is too general to categorize weather condition.

Nomenclature

EUI energy use intensity

CBECS Commercial Buildings Energy Consumption Survey
HDD heating degree day
CDD cooling degree day
DC data collection
DP data processing
MD model development
WWR window-to-wall ratio
V volume
FA façade area
SA site area
MLR multiple linear regression

One of the methods is to use regression model incorporating basic visualized building façade features to estimate
building energy consumption instead of using average data from survey or running simulation by using complicated
844 Chao Yang and Joon-Ho Choi / Procedia Engineering 118 (2015) 842 – 852

software. The main objective is to develop a customized building energy use baseline estimation tool by using
mathematical method, considering specific façade features and local climate condition. Compared with software
simulation, the tool is simpler and quicker in terms of time cost. The model would be applicable to set a reasonable
EUI reduction baseline for building performance management and improvement. In addition, the tool will analyze
the impact of basic façade features on energy performance in different climate zone by sensitivity analysis, in order
to provide a guideline of how façade features could influence certain building energy use in a specific climate
condition. At the public level, the result could also draw more attention on the significance of building energy use
disclosure from ongoing benchmarking policy. Customized baseline could be more acceptable for building owners to
know building energy saving potential and adopt measures to improve energy efficiency, which in turn will benefit
energy conservation for the whole society.

Fig. 1. National median reference EUI of selected buildings types [7].

2. Methodology

Variable regression models are developed to determine the most significant façade features which are relevant to
energy aspect as well as to predict energy performance by entering a minimum number of data. Instead of using
details of building information, like construction thermal properties, mechanical system, operation schedule, etc.,
which are used as basic input for simulation and other estimation methods, multiple linear regression and stepwise
regression are adopted with easily accessible façade features, which include building height, orientation, volume,
floor area, façade area, site area, window-to-wall ratio, volume-to-façade area ratio, etc.

Fig. 2. Methodology
 Chao Yang and Joon-Ho Choi / Procedia Engineering 118 (2015) 842 – 852 845

As the figure above illustrates, there are mainly three parts of the methodology: data collection (DC), data
processing (DP), and model development (MD). The predicted outcome of this research is a new EUI estimation
tool, which in this paper refers to the annual office EUI estimation model in New York City.

x For data collection, generally two types of data are supposed to be collected. One is real energy use data, another
is corresponding façade features of the same buildings. Energy use data is presented by Energy Use Intensity
(EUI) as the target metric are from building energy benchmarking and disclosure data by local government. On
the other hand, façade features are collected by using different methods which contain manual estimation (visual
reading and physical model rebuilding), existing building model (SketchUp, etc.) and direct information
collection from design drawing or specification. Other potential factors like built year and HDD/CDD could be
easily obtained from open resources.

x Data processing section is served as a preparation for the following model development. For annual EUI model
development, this step could be skipped since annual EUI data is the basic data provided by different building
energy resources. When it comes to monthly EUI model, simulation tool could be used to estimate monthly
energy performance first. As a result, monthly data could be estimated accurately after calibrating the simulation
model by real energy bill or annual EUI data.

x Finally, multiple linear regression and stepwise regression are used to develop the EUI estimation tools based on
collected façade information and EUI data. In this section, the significance of each parameter and correlation
between predictors and response could also be analyzed with the consideration of local code requirements, design
strategies and best practices. In the end, all regression models should be validated by appropriate method, for
example, cross-validation is used to validate Partial Least Square (PLS).

2.1. EUI data collection

Building energy benchmarking is used to obtain and record building energy data as baselines to compare to other
properties performance. The building energy consumption and reduction potential could be reflected clearly by
giving owners an opportunity to get the benchmarking data for a time period. To accomplish the task of
benchmarking, it is necessary to monitor and measure utilities and the data should be submitted by using a common
format to be available to put into database. The most commonly used tool is Portfolio Manager developed by EPA
[8], which could be used to track and evaluate energy use for commercial buildings. The benefits of using
benchmarking [9] to keep track of building energy use are listed in the following figure.

Fig. 3. (a) U.S. Building benchmarking and transparency policies. Source: [10]; (b) benchmarking benefits.

Currently, there are 9 cities in U.S. [10] which committed to implement energy benchmarking and disclosure
programs for commercial buildings [11], including Seattle, San Francisco, Austin, Minneapolis, Cambridge, Boston,
846 Chao Yang and Joon-Ho Choi / Procedia Engineering 118 (2015) 842 – 852

New York City, Philadelphia, Washington, DC, etc. In New York City, benchmarking policy of Local Law 84
(LL84), part of Greener, Greater Buildings Plan (GGBP) was adopted in 2009 [12], which requires all non-
residential buildings with floor area over 50,000 square feet to submit and disclose their building energy and water
data to the city. The results show that the median source EUI for office properties in 2010 and 2011 are 213.3
kBtu/sf and 207.3 kBtu/sf and the median Energy Star score increased from 64 to 67.
In this paper, office building energy benchmarking data in New York City are used to develop an exemplary
regression model. It could be used to predict annual energy use for office buildings in New Your City. 99 office
buildings in Manhattan, New York City from the benchmarking database are firstly selected. Then 28 buildings with
existed SketchUp model are further sorted out in order to read the façade features easily and accurately. In most
selected buildings there are 2 years of reported energy data available (24 of them have both years). In total 50
datasets with full information of both real EUI and façade features are available for the further regression analysis.

2.2. Façade feature definition

All building façade features used in regression model are supposed to be easily readable without knowing
detailed information. Generally, unlike thermal performance, geometry attributes are the basic predictors. Roof or
wall R-value, window U-value and SHGC, etc. are not used since the fabric information are not accessible without
the permission from owner or designer. The original 17 assumed predictors (including height, floors, orientation,
operable window, volume, window-to-wall ratio (WWR), window area, façade area, site area, floor area, volume-to-
façade area ratio, volume-to-site area ratio, façade area-to-site area ratio, weather condition, surrounding context and
built year) are showed in the table 1 explaining the definition of each parameter.
In this research, a basic assumption is that EUI could be estimated only based on simple façade features as well as
a few other factors, like HDD/CDD which represents dynamic local weather condition. To consider other aspects of
building energy use, built year as an additional predictor is used to incorporate all the requirements by code in each
time period into consideration, which means after the first national/local building energy code established a building
had to meet the requirements of corresponding codes or standards, including fabric thermal performance, system
efficiency, ventilation rate requirements, etc. The built year is easy to obtain from urban zoning or public service
information. In addition, since in urban context, adjacent building will cast shades on target buildings which in turn
will influence heat gain through the façade especially glazing area, adjacency is used as another additional factor
which is collected for regression analysis.

Table 1. Predictor’s definition and explanation.

No. Façade feature Definition Unit Category
1
1 Height From open air pedestrian entrance to highest occupied floor Feet Basic
2 Floors Total occupied stories or levels2 - Basic
3 Orientation Positing of a building with respect to the North3 - Basic
4 Operable window Window could be open or close based ventilation need4 - Basic
5 Volume Inner space volume enclosed by external envelope CF Basic
6 WWR Window-to-wall ratio (total window area/total exterior wall area) - Basic
7 Window Area Total glazing area SF Area
8 Façade Area Total area of all parts of the structure’s façade SF Area
9 Site Area Total site area within fixed boundaries SF Area
10 Floor Area Total floor area inside the building envelope SF Area
11 V/FA Ratio of volume to façade area - Ratio
12 V/SA Ratio of volume to site area - Ratio
13 FA/SA Ratio of façade area to site area - Ratio
Degree
14 HDD Heating degree day (the demand for energy to heat a building) Weather
Days
15 CDD Cooling degree day (the demand for energy to cool a building) Degree Weather
 Chao Yang and Joon-Ho Choi / Procedia Engineering 118 (2015) 842 – 852 847

Days
16 Adjacent Building If adjacent building exists to cast shading on objective building†5 - Additional
17 Built Year Year of construction complete year Additional

2.3. Regression

Many tools could be used to develop the regression models, like SPSS Statistics [14], MATLAB [15], etc. In this
research, another statistical analysis tool, Minitab® 17 [16] is adopted for data analysis and regression model
development. By using Minitab, a large amount of data can be processed [17] for basic statistical analysis, regression
and correlation analysis, hypothesis tests, model validation, prediction, and graphs making, etc. All façade features,
additional factors and EUI data can be input as basic training samples. The correlation between each factor and EUI
could be analyzed by calculating Pearson’s correlation coefficient. Then different regression models could be
compared and used to determine the most accurate model which is sufficient to predict response values for new
observations.
Rather than only using one independent variable as predictor in regression, multiple linear regression (MLR) has
multiple independent variables. The same purpose as simple linear regression is to develop the relationship between
response and predictors and predict the new response with a new set of predictors at an acceptable confidence level.
The multiple linear regression is presented as the following form:

U a  b1 x1  b2 x2  b3 x3  ...  bk xk  H (1)

Where a is the constant while b1,∙∙∙, bk are the regression coefficients, x1,∙∙∙, xk are the significant predictors and ε
is the random error.
In addition, when there are a large number of predictors to be used in regression, stepwise regression is also used
to remove the least significant predictor at each step. The order of removed predictors also indicate the significance
of each predictor, which in turn indicates which façade feature is the most important one in a certain area. This is
also called backward elimination [18]. To analyze the results of regression models, multiple indicators could be
calculated to evaluate the characteristics of the corresponding models. The main indicators are listed in the following
table 2.

Table 2. Predictor’s definition and explanation.

No. Indicator Explanation Accepted Range
1 Pearson Correlation Whether 2 continuous variables are linearly related (-1,1)/closer to 1
2 P-value The probability of obtaining a test statistic (0,1)/closer to 0
3 VIF Multicollinearity (correlation between predictors) NA
4 R2 Pct. of response variable variation can be explained (0,100%)/closer to 100%
2 2
5 R (adj) R adjusted for the number of predictors in the model (0,100%)/closer to 100%
6 R2 (pred) Models predictive ability (0,100%)/closer to 100%
7 Durbin-Watson whether the correlation between adjacent error terms is 0 (1,3)/closer to 2
8 Error rate discrepancy between the estimated values NA/closer to 0

1
Height is measure from the level of the lowest, significant, open-air, pedestrian entrance to the finished floor level of the highest occupied
floor within the building [13].
2
Floors refer to the total levels of a building which could be used by occupants.
3
Long axis along with North-South is quantified as 1, NE-SW is 2, E-W is 3, SE-NW is 4.
4
With operable window is quantified as 1, without operable window is quantified as 0.
5
No adjacent building is quantified as 0, while adjacent building on the north side is 1, others are clockwise defined by 2 to 8.
848 Chao Yang and Joon-Ho Choi / Procedia Engineering 118 (2015) 842 – 852

3. Results and discussion

3.1. Basic data analysis

All 50 raw datasets with façade features are firstly analysed by dividing into different groups. The results
represent the correlation between reported site EUI with each predictor through interval plotting. The confidence
interval is 95% by default which indicates 95% probability from the future experiment within this interval.

Fig. 4. (a) Interval plot of site EUI and construction year; (b) site EUI and building height.

Fig. 5. (a) Interval plot of site EUI and WWR; (b) site EUI and operable window.

Fig. 6. (a) Interval plot of site EUI and V/FA; (b) site EUI and orientation.
 Chao Yang and Joon-Ho Choi / Procedia Engineering 118 (2015) 842 – 852 849

Figure 4 (a) indicates the correlation between site EUI and construction year, which divides the datasets into 2
groups (before and after 1980), since the first New York state energy code was established in 1979 [19]. Office
buildings that were built before 1980 have higher mean value of 102.06 kBtu/sf than 92.74 kBtu/sf after 1980. Even
the confidence intervals are slightly overlapped, but with more strict requirements of building performance from
improved energy code, buildings consume lower energy as expected. Tall buildings are grouped into megatall (more
than 600 ft), supertall (300 to 600 ft) and tall (165 to 300 ft) for the analysis of height [13]. Figure 4 (b) shows the
significant difference of energy use for different height tall buildings. Megatall buildings consumes the highest
energy, followed by super tall and tall buildings. National median site EUI of 67.3 kBtu/sf is only in the tall building
EUI range. The overall 40% of WWR for prescriptive fenestration requirement [20] is used to divide all datasets
into 2 groups and the results present that WWR is a significant factor to influence office building energy use in terms
of heating and cooling load by solar heat gain. The mean value of buildings with over 40% WWR is 107.88 kBtu/sf
compared to 84.81 kBtu/sf for lower WWR buildings. Buildings with operable windows consumes less energy since
the mixed mode of natural ventilation and mechanical ventilation is more energy efficient, which is proved by the
fact that the mean value 84.9 kBtu/sf for buildings with operable window is lower than 104.25 kBtu/sf for buildings
without operable window. V/FA ratio stands for the compactness which has significant impact on heating load.
Figure 6 (a) indicates that buildings with V/FA less than 40 have the lower mean EUI of 89.03 kBtu/sf. Figure 6 (b)
shows there is no significant difference of EUI between N-S orientation and NE-SW orientation while buildings with
NW-SE have the highest mean EUI value of 111.01 kBtu/sf. It is because that the main façade faces south west has
more heat gain through direct sun exposure.

3.2. MLR and stepwise regression results

EUI can be predicted by the façade features through 2 methods: MLR and Stepwise Regression. The results are
showed in table 3. Total façade area is replaced by 8 different direction façade area. In MLR, all predictors are the
R2 value indicates that all 25 predictors are included in the every model. The R 2 value indicates that all predictors
could explain 77.64% of the variance in EUI while the adjusted R 2 means only 56.18% of EUI variable variation that
is explained by its relationship with predictor variables, adjusted for the number of predictors in the model. D-W
statistic is closer to 2, which means there is no significant autocorrelation. Only orientation and floor area are
significantly related to annual EUI at an α-level of 0.05 since P-values are close to 0. VIF values for coefficients are
greater than 10 which means the regression coefficients are poorly estimated due to severe multicollinearity.
By comparison, R2 from stepwise regression means 88.15 % of the variance in EUI. The adjusted R 2 is also
improved when compared to MLR. The predicted R2 value is 77.72% which indicates the model does not appear to
be overfit and has adequate predictive ability since it’s close to R2 and adjusted R2. D-W statistic is 1.989 which is
also closer to 2. All P-values of corresponding predictors are less than 0.05 while VIFs are less than 10 except south
and west façade areas are slightly higher than 10. The results show the advantage by using stepwise regression is not
only to improve each indicators of accuracy but also to identify a useful subset of predictors. The stepwise process
systematically adds the most significant variable or removes the least significant variable during each step. As a
result, predictors including height, WWR, orientation, operable window, floor area, V/SA ratio, HDD as well as
south and west façade area are the most important factors which have greater impact on energy use for office
buildings in New York City.

Table 3. MLR and stepwise regression coefficients and indicators.

Determination Multiple Linear Regression Stepwise Regression
R2/ R2 (Adj)/ R2 (pre) 77.64% 56.18% - 88.15% 84.66% 77.72%
D-W 2.022 1.989

Predictors Coef P-value VIF Coef P-value VIF

Constant 27302 0.174 -75.3 0.047
Height 0.087 0.593 83.84 0.1553 0.000 3.85
Floors 0.06 0.979 78.14 - - -
Built year -0.339 0.586 17.67 - - -
WWR 0.542 0.507 25.16 0.719 0.000 2.03
Orientation 26 0.033 25.61 18.77 0.000 4.53
850 Chao Yang and Joon-Ho Choi / Procedia Engineering 118 (2015) 842 – 852

Operable Window -29.9 0.15 12.2 -19.65 0.000 2.11

Volume 0 0.995 605.78 - - -
Window Area 0.000149 0.55 100.77 - - -
Site Area 0.00035 0.729 54.2 - - -
Floor Area -0.00007 0.031 29.78 -0.000054 0.000 8.55
V/FA -0.84 0.809 127.38 - - -
V/SA 0.185 0.515 132.69 0.1352 0.001 4.52
FA/SA -10.29 0.11 77.31 -9.47 0.000 8.61
Adjacency -1.85 0.502 12.44 - - -
HDD 5.86 0.178 53879.79 0.0324 0.006 1.02
CDD -22.7 0.181 53885.9 - - -
N Façade Area -0.01101 0.201 6298.99 - - -
S Façade Area 0.125 0.23 1023528.62 0.001340 0.000 11.46
W Façade Area -0.00249 0.2 598.28 -0.000634 0.009 13.83
E Façade Area -0.0889 0.243 862326.34 - - -
NW Façade Area -0.000146 0.806 49.89 - - -
NE Façade Area -0.00017 0.892 553.6 - - -
SW Façade Area -0.000118 0.849 148.17 - - -
SE Façade Area 0.000571 0.471 101.53 - - -

Figure 7 (a) is the histogram of standardized residual and frequency and 1 outlier may exist in the data, which
needs to be proved in other analysis. Figure 7 (b) of normal probability plot shows an approximately linear pattern
consistent with a normal distribution. The point in the upper-right corner is an outlier (row 33), which could be read
from the plot. The plot of residuals versus the fitted values shows that the variance of the residuals are constant with
a mean of zero.

Fig. 7. (a) Standard residual and frequency plot; (b) normal probability plot.

Fig. 8. (a) Standard residual and fitted value; (b) model validation and error percentage.
 Chao Yang and Joon-Ho Choi / Procedia Engineering 118 (2015) 842 – 852 851

To validate the regression model, datasets are divided into two samples for training and validation. The samples
number is 10% of original datasets for validation. In this case, 5 data from original 45 valid datasets are randomly
selected as validation samples. The R2 of training samples model is 91.02% while D-W statistic is 2.04. Error rates
for the 5 validation samples are 5.94%, 9.12%, 4.57%, 6.36% and 8.74%, which are at the accepted level.

4. Results and discussion

To predict building energy use and compare the accuracy, both simple multiple linear regression model and
stepwise regression model are used and the results show that stepwise is more reliable to predict EUI than MLR. The
comparison of estimation results and reported site EUI are illustrated in Figure 9. Compared with general national
average baseline, building EUI estimated by basic façade features is more specific which considers the individual
building attributes as well as local climate condition. The result is dynamic along with different features input which
is better than one constant and median baseline from CBECS. In addition to assist to EUI benchmarking for
improving building energy efficiency, the research potential outcomes could also be applied for new construction to
provide a more accurate baseline and energy reduction target at the predesign stage and to evaluate basic façade
design decisions. On the other hand, it can help building designers to estimate EUI when there is no detailed building
information available for deep simulation. The reasonable energy consumption rate is achievable by inputting a
minimum amount of data.

Fig. 9. Regression model results and site EUI comparison

References

[1] “Program, Buildings Technologies, and Energy Efficiency. 2012. “2011 Buildings Energy Data Book.”
[2] Shan, Rudai. 2014. “Optimization for Whole Building Energy Simulation Method in Façade Design”: 1–9.
[3] Andrews, Clinton J., and Uta Krogmann. 2009. “Technology Diffusion and Energy Intensity in US Commercial Buildings.” Energy Policy 37
(2) (February): 541–553.
[4] EPA. 2011. “DataTrends Benchmarking and Energy Savings”.
[5] Architecture 2030. 2011. “The 2030 Challenge”
[6] “About Commercial Buildings Energy Consumption Survey.” 2012. http://www.eia.gov/consumption/commercial/about.cfm.
[7] Energy Star. 2014. “U . S . Energy Use Intensity by Property Type.”
[8] Energy Star. 2014. “The New ENERGY STAR Portfolio Manager.”
[9] Milliken, Rebecca, and Betony Jones. “Office Building Benchmarking Guide Engaging the Hard-to-Reach.”
[10] IMT. 2014. “U.S. Benchmarking Policy Landscape.” http://www.buildingrating.org/graphic/us-benchmarking-policy-landscape.
[11] Cox, Matt, Marilyn a Brown, and Xiaojing Sun. 2013. “Energy Benchmarking of Commercial Buildings: A Low-Cost Pathway toward
Urban Sustainability.” Environmental Research Letters 8 (3) (September 1): 035018.
[12] GGBP. 2013. “2013 NEW YORK CITY LOCAL LAW 84 BENCHMARKING REPORT” (September).
[13] Council on Tall Buildings and Urban Habitat. “CTBUH Height Criteria.”
http://www.ctbuh.org/TallBuildings/HeightStatistics/Criteria/tabid/446/language/en-US/Default.aspx.
[14] IBM. 2014. “SPSS Statistics.” http://www-01.ibm.com/software/analytics/spss/products/statistics/.
[15] MathWorks. 2014. “MATLAB.” http://www.mathworks.com/products/matlab/.
852 Chao Yang and Joon-Ho Choi / Procedia Engineering 118 (2015) 842 – 852

[16] Minitab. 2014. “Minitab 17.” http://www.minitab.com/en-us/.

[17] Minitab. 2013. Getting Started with Minitab 17.
[18] Support Minitab. 2014. “Basics of Stepwise Regression.”
http://support.minitab.com/en-us/minitab/17/topic-library/modeling-statistics/regression-and-correlation/basics/basics-of-stepwise-regression/.
[19] U.S. DOE. 2014. “Building Energy Codes Program.” https://www.energycodes.gov/adoption/states/new-york.
[20] NYCECC. 2011. “LOCAL LAWS OF THE CITY OF NEW YORK A LOCAL LAW”. Vol. No. 1.