Energy & Buildings 244 (2021) 111050

Contents lists available at ScienceDirect

Energy & Buildings

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

Impact of global warming on heating and cooling degree days in major

Indian cities
Rahul Ukey, Aakash C. Rai ⇑
Department of Mechanical Engineering, Birla Institute of Technology and Science, Pilani, Rajasthan 333031, India

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

Article history: Global warming induced rise in ambient temperatures would significantly impact the space heating and
Received 4 December 2020 cooling (H/C) energy requirements in buildings. This investigation assessed the impact of global warming
Revised 31 March 2021 on space H/C energy requirements in eight major Indian cities covering all of the county’s climate zones.
Accepted 21 April 2021
By using historical weather records and general circulation model outputs, we quantified the historical
Available online 27 April 2021
(1969–2017) and future (2018–2100) trends in annual mean temperatures together with heating and
cooling degree days (HDDs and CDDs), which are well-known metrics for quantifying buildings’ H/C
Keywords:
requirements. The investigation demonstrated that annual temperatures would be higher by 0.1–
Climate change
Building energy demand
1.1 °C in the 2020 s, by 0.6–2.8 °C in the 2050 s, and by 1.0–4.6 °C in the 2080 s, depending on the city
Degree days, general circulation model and the emission scenario. Due to rising temperatures, CDDs would also increase by 2.9–22.9% in the
Representative concentration pathways 2020 s, by 8.3–54.1% in the 2050 s, and by 11.89–83.0% in the 2080 s; thus, increasing the cooling require-
Time-series analysis ments by a similar amount. In contrast, HDDs would decrease by 8.1–30.3% in the 2020 s, by 17.6–83.3%
in the 2050 s, and by 19.3–97.1% in the 2080 s, thereby reducing the heating requirements.
Ó 2021 Elsevier B.V. All rights reserved.

1. Introduction global warming may substantially increase the energy consump-

tion in buildings and the associated CO2 emissions.
In 2018 energy consumption in buildings accounted for around The impact of global warming on the H/C energy consumption
30% of the total energy consumed worldwide, of which space heat- in buildings can either be estimated by conducting hourly energy
ing and cooling (H/C) were responsible for 34% and 6%, respectively simulation [7–9] or by using the degree-day method [5]. In the for-
[1]. A similar situation can also be seen in India, where in 2018, mer approach, simulation programs (like EnergyPlus) are used to
buildings accounted for about 33% of the country’s total energy predict the energy requirements for space H/C of archetype build-
consumption [2]. Of the energy consumed by Indian buildings, ings by modeling their thermal performance under current and
space cooling accounts for about 5% [2,3], while the heating energy future climatic conditions. The latter approach uses heating and
consumption is likely to be relatively small (exact estimates cooling degree-days (HDDs and CDDs), which are essentially sum-
unavailable) since most of the country experiences a warm cli- mations of temperature differences (ambient temperature minus a
mate. Thus, space H/C in buildings constitutes a large portion of base temperature) over time, as indicators for predicting energy
energy consumption globally and in India. The energy consump- requirements for space H/C. Although degree-days do not consider
tion in buildings is also responsible for about 28% of energy- the effect of variables like relative humidity and wind speed that
related CO2 emissions worldwide, which is the primary reason can influence buildings’ H/C energy requirements, nevertheless,
for global warming [2]. Due to global warming, the global mean they remain excellent indicators for the same [10]. The degree-
surface temperatures for 2081–2100 are projected to be 0.3– days approach stands out due to its simplicity, transparency, and
4.8 °C higher (depending on the emission scenario) than those in repeatability that hourly energy simulations may not provide [11].
1986–2005 [4]. As global temperatures rise, the energy required
for space cooling will increase, while the space heating require-
ments will decrease [5]. Thus, in India, where energy requirements 1.1. Historical HDD and CDD trends
for space cooling are much larger than those for space heating [6],
Several researchers have investigated past trends in degree-
days in different regions worldwide by estimating the HDD and
⇑ Corresponding author. CDD values based on historical weather records [12–17]. For exam-
E-mail address: [email protected] (A.C. Rai). ple, Rosa et al. [16] studied the changes in HDDs and CDDs in six

https://doi.org/10.1016/j.enbuild.2021.111050
0378-7788/Ó 2021 Elsevier B.V. All rights reserved.
R. Ukey and A.C. Rai Energy & Buildings 244 (2021) 111050

cities in Italy from 1978 to 2013, covering all its climate zones. much larger than heating requirements, global warming would
Their study found that HDDs decreased (by 2.6–17.1%, depending substantially increase the building energy consumption. Therefore,
on the city) while CDDs increased (by 7.4–170%, depending on this investigation assessed the impact of global warming on space
the city) in 2000–2013, as compared to their values in 1980– H/C energy requirements in eight major Indian cities covering all of
1993. Similarly, Christenson et al. [12] estimated the impact of cli- the county’s climate zones. By using historical weather records and
mate warming on degree-days in Switzerland’s four major cities general circulation model outputs, we quantified the historical
during 1901–2003. Their investigation also showed that HDDs (1969–2017) and future (2018–2100) trends in annual mean tem-
reduced by 11–18% while CDDs increased by 50–170% during the peratures, HDDs, and CDDs.
study period, depending upon location and the base temperature
used. A similar conclusion was obtained by Ortizbeviá et al. [14],
who studied the trends in degree-days at 31 weather stations in 2. Data and method
Spain during 1970–2005. They reported decreasing trends (2.3–
6%/decade) in HDDs at 21 stations, while no significant HDD trends To assess the impact of global warming on the H/C energy
were detected in the remaining stations. On the other hand, CDDs requirements in India, this investigation first selected eight repre-
had increasing trends (5–40%/decade) at 23 stations, while no sig- sentative Indian cities, for which the historical temperature
nificant CDD trends were detected in the rest. On similar lines, records (from 1969 to 2017) were obtained from the Indian Mete-
Jiang et al. [13] studied variations in HDDs and CDDs at 51 weather orological Department (IMD). Since the IMD temperature data had
stations in Xinjiang, China, from 1959 to 2004. They reported that missing values, we filled those values using other datasets. Next,
HDDs exhibited decreasing trends (from
25.1 °Cd/decade to this study obtained the future temperature projections (from

312.1 °Cd/decade) in 49 stations, while no significant trend was 2018 to 2100) for the selected cities from the National Aeronautics
detected in the remaining stations. The CDD trends ranged from and Space Administration (NASA) earth exchange (NEX) global

21.5 °Cd/decade to 38.4 °Cd/decade, with 29 stations showing daily downscaled projections (GDDP) dataset after identifying a
positive trends, 18 showing no trends, and the remaining four GCM suitable for each city. Subsequently, we calculated the annual
showing negative trends. mean temperatures, HDDs, and CDDs for each city from 1969 to
2100 and estimated their linear trends. The complete methodology
1.2. Future HDD and CDD trends is summarized in Fig. 1 and elaborated in the following sub-
sections.
In addition to studying the historical degree-day trends, several
investigations have also made predictions of future degree-days
using future weather data obtained from climate simulation mod- 2.1. Selection of representative cities
els [18–20]. For example, Isaac and Vuuren [21] estimated that due
to climate change (3.7 °C increase in global temperature over the We selected eight representative cities covering all the climate
pre-industrial age value), global HDDs would reduce by 34%, while zones of the country, as shown in Fig. 2. The selected cities
CDDs will increase by 70% in 2100, as compared to their values in included India’s major metropolitan hubs, covering a total area of
2100 assuming no climate change. A similar conclusion was 4,812 km2 and supporting a population of 103 million [24]. A
obtained by Warren et al. [22], who reported that a 1–5 °C increase detailed description of those cities is given in Table 1. We investi-
in global temperatures would decrease global HDDs by 14–18% gated the historical and future CDDs for seven cities since one city
and increase CDDs by 18–99%. Spinoni et al. [18] studied the (Srinagar) did not have any cooling requirements. In contrast,
impact of global warming on degree-days in Europe from 1981 HDDs were only studied for two cities (Delhi and Srinagar) since
to 2100 under two different representative concentration path- the remaining did not have any heating requirements.
ways (RCPs): RCP4.5 and RCP8.5. The RCPs are defined by the
increase in ‘radiative forcing’ in the year 2100 relative to 1750
2.2. Calculation of HDDs and CDDs
due to increasing greenhouse gas concentrations in the atmo-
sphere, i.e., 4.5 W/m2 for RCP4.5 and 8.5 W/m2 for RCP8.5 [4]. Their
This investigation calculated the annual HDDs and CDDs for the
study showed that in Europe from 1981 to 2100, HDDs would
selected cities, as per the definition of the American Society of
decrease by 49–84 °Cd/decade while CDDs would increase by 8–
Heating, Refrigerating, and Air-conditioning Engineers (ASHRAE)
20 °Cd/decade, depending on the RCP. On similar lines, Petri and
[25]:
Caldeira [19] studied the variations in degree-days in 25 of the Uni-
ted States’ most populous cities under the RCP8.5 emission sce-
X
N
nario. Their study showed that depending on the location, HDDs HDD ¼ ðT base
T Þþ ð1Þ
would decrease by 28–99% while CDDs would increase by 12– i¼1
650% in 2080–2099, as compared to those in 1981–2010. Ramon
et al. [20] also obtained a similar conclusion by studying the XN
impact of climate change on future degree days in Belgium under CDD ¼ i¼1
ðT
T base Þþ ð2Þ
the RCP8.5 scenario. They reported that there would be a 27%
decrease in HDDs and a 140% increase in CDDs in 2070–2098, over where T is the daily mean ambient temperature, Tbase the base tem-
those in 1976–2004. perature, and N the number of days in the year. The ‘+’ superscript
denotes that only positive values contribute to the degree-day cal-
1.3. Research gap and objectives culations. In the above equations, the ambient temperature (T) was
approximated as the arithmetic mean of the daily minimum and
The above-mentioned studies have shown that HDDs in the last maximum temperatures, and the base temperature (Tbase) was
century decreased while CDDs increased almost universally due to taken as 18 °C. We used 18 °C for calculating both HDDs and CDDs
global warming. Similar trends in HDDs and CDDs are projected to since this base temperature value was recommended for India by
continue in the future, leading to a decrease in the heating energy Bhatnagar et al. [26], based on detailed energy simulations of sev-
demand and an increase in the cooling energy demand of build- eral building types located across the different climate zones of
ings. Thus, in India, where buildings’ cooling requirements are the country.
2
R. Ukey and A.C. Rai Energy & Buildings 244 (2021) 111050

Fig. 1. Flow chart of the methodology.

Table 1
Description of the selected cities.

Climate City Description

zone
Hot–Dry Ahmedabad Ahmedabad is a large metropolis located in the
western part of India. Winters are short and mild,
with the lowest monthly temperature being
12 °C, whereas summers are long and harsh, with
maximum monthly temperatures reaching up to
42 °C.
Warm- Chennai Chennai, Kolkata, and Mumbai are all coastal
Humid Kolkata cities and the major metropolitan hubs of the
Mumbai country. The cities experience a very mild winter
season with temperatures generally exceeding
15 °C throughout the day. Summers are hot and
humid with maximum monthly temperatures
ranging between 34 and 38 °C.
Composite Delhi Delhi is a major metropolitan area and the capital
city of India. Winters are cold, with temperatures
dropping routinely below 10 °C at night, while
summers are hot with temperatures exceeding
40 °C during the day.
Hyderabad Hyderabad is a major metropolitan city located
on the Telangana plateau in the southern part of
the country. Winters are mild, with temperatures
generally ranging between 15 and 30 °C, while
summers are hot with temperatures exceeding
40 °C.
Temperate Bengaluru Bengaluru is another large metropolitan city in
India. The city experiences a moderate climate
Fig. 2. The selected cities covering all the climate zones of India. The climate
with temperatures generally ranging between 16
classification map was taken from National Building Code of India [23].
and 34 °C throughout the year.
Cold Srinagar Srinagar is one of India’s most populous city in
the cold climate zone and lies in a valley region
2.3. Historical weather data (1969–2017) surrounded by the Himalayan mountains.
Winters are cold, with temperatures dropping
The degree-days for the years 1969–2017 were calculated from below 0 °C at night, while summers are mild,
with the maximum daytime temperature being
the daily values of minimum and maximum ambient temperatures
around 30 °C.
obtained from IMD. IMD has been monitoring meteorological
parameters in >500 stations in India for over a century, and
releases weather data following stringent quality tests and pre-
processing based on World Meteorological Organization (WMO)
protocols [27]. However, IMD’s temperature data had some miss- els (GCM) under the ‘historical’ scenario (see details in Section 2.4).
ing values (0.2–4.5%, depending on the city). Those missing values Since the historical scenario ends in 2005, the missing values from
were filled using the data obtained from the National Oceanic and 2006 to 2017 were filled by data from TuTiempo.net, a website that
Atmospheric Administration (NOAA), which maintains an archive provides access to WMO data. However, one or two temperature
of global historical climate data. We chose IMD as the primary data values were still missing for 2006–2017. Those values were filled
source and NOAA as the secondary source since IMD data had by using linear interpolation since the consecutive missing values
lower missing values than the NOAA data. Even after combining were less than five [28,29]. Finally, if the endpoint (31st December
both datasets, there were still some missing values (0.1–2.6%, 2017) of the data was missing, it was filled by the previous day’s
depending on the city). The missing temperature values for the value. Thus, by following this procedure, we obtained the complete
years 1969–2005 were taken as the arithmetic means of the tem- ambient air temperature records for all eight cities from 1969 to
perature predictions given by 21 different general circulation mod- 2017.

3
R. Ukey and A.C. Rai Energy & Buildings 244 (2021) 111050

To assess the suitability of the procedure used for filling the temperature from 1969 to 2005, respectively, obtained from the
missing values, we calculated the RMSE in the daily mean ambient ordinary least square (OLS) analysis given in Section 2.5.1.
temperatures between the primary dataset (from IMD) and other Note that RMSE values quantify the errors in the daily mean
datasets (from NOAA, mean GCM, or TuTiempo), as reported in temperatures predicted by each GCM, thus quantifying the model’s
Table SM1. The RMSE values were calculated using all common performance for short time-periods, whereas the AELS values
data-points available between the primary and secondary datasets, assess the model’s long-term performance. To assess the model
except for the TuTiempo data, for which we randomly selected 30 performances relative to each other, we defined ERMSE and EAELS
data points for RMSE calculation since the TuTiempo data had to be metrics, as suggested by Gleckler et al. [33]:
manually recorded for each day. Table SM1 shows that the RMSE
values ranged between 0.46 and 2.67 for the different datasets, RMSE
RMSE
ERMSE ¼ ð5Þ
depending on the city. The RMSE values compared reasonably well RMSE
with other imputation algorithms reported in the literature [30].
We did not study the performance of the linear interpolation tech- AELS
AELS
EAELS ¼ ð6Þ
nique for missing value imputation since only 1–2 data-points per AELS
city were filled by this method.
where ERMSE and ERMSE are the relative errors in the RMSE and AELS
values, respectively; RMSE and AELS are the median values and
2.4. Future weather data (2018–2100) quantify the ‘typical’ model errors. By using ‘median’ rather than
the ‘mean,’ we clear out the possible problems arising from outliers
For calculating future degree-days, daily predictions of ambient in the dataset. The smaller the E value is for a model, the better is its
air temperatures were obtained from the NASA-NEX-GDDP dataset. performance compared to others. For example, if ERMSE =
0.1 for a
This dataset (available at https://cds.nccs.nasa.gov/nex-gddp/) was particular model, it demonstrates that the model’s RMSE is 10%
developed from simulations conducted under the 5th phase of lower than that of the typical (median) model, which means that
the coupled model inter-comparison project (CMIP-5) using 21 dif- the model outperforms the typical model.
ferent GCMs, as given in Table SM2. The dataset contains statisti- Figure SM1 shows the relative error values (ERMSE and EAELS
cally downscaled projections of the minimum and maximum denoted by the color scale) for the 21 GCMs (names given in the
ambient temperatures at a geospatial resolution of 0.25°  0.25° center) for all eight cities. Figure SM1 a) shows that for most of
(about 25 km  25 km) for RCP4.5 and RCP8.5 scenarios from the cities, the MIROC_ESM_CHEM model had the least ERMSE values
2005 to 2100. Note that RCP4.5 is a stabilization scenario with (best short-term performance) while the IPSL_CM5A_MR model
the total radiative forcing stabilizing at 4.5 W/m2 in 2100, relative had the highest ERMSE values (worst short-term performance). On
to 1750 [31]; while RCP8.5 is a scenario with very high greenhouse the other hand, Figure SM1 b) shows that the GFDL_ESM2G model
gas emissions leading to an increase in radiative forcing of 8.5 W/ generally had the least EAELS values (best long-term performance)
m2 in 2100, relative to 1750 [32]. From the NEX-GDDP dataset, we while the IPSL_CM5A_LR model had some of the highest EAELS val-
obtained the daily temperature predictions (minimum and maxi- ues (worst long-term performance). Thus, the models that had
mum values) at four closest grid-points for a particular city, and good short-term performance did not necessarily capture the
then used bilinear interpolation for estimating the city’s tempera- long-term temperature trends. It is also evident from Figure SM1
tures [8]. that a single GCM could not consistently perform well for all the
Since the NEX-GDDP dataset contained future temperature pre- eight cities, and different models seem suitable for different cities.
dictions from 21 different GCMs; therefore, choosing an appropri- Thus, to choose an appropriate model for a particular city, we
ate GCM for making future temperature predictions was a non- first ranked the 21 GCMs based on their short-term and long-
trivial exercise. So, we first evaluated the performance of different term performances, i.e., each GCM was ranked from 1 to 21 corre-
GCMs by comparing their predictions under the ‘historical’ sce- sponding to ERMSE and EAELS metrics. Subsequently, we summed
nario with the corresponding observations (1969–2005) for all both the ranks to obtain the final model rankings for each city;
the eight cities. The historical scenario provides the modeled his- thus, a lower rank represented a well-performing model and
torical evolution of temperature using reconstructed historical vice-versa. The models with the best rankings (see Table 2) were
forcing. Thus, we evaluated different GCMs by comparing their selected to make future temperature predictions. Note that
temperature predictions under the historical run with their corre- Ahmedabad had three models (BNU-ESM, MIROC-ESM, and
sponding observed values for all eight cities. MIROC5) while Mumbai had two models (BCC-CSM1-1 and
This investigation calculated the root mean square error (RMSE) MIROC-ESM) with equal summed ranks. Those ties were broken
values for each model for all cities by using the following equation: based on the model performances in predicting the historical
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi CDD slopes, as discussed in the following paragraph.
u N
uX ðT GCM
T Obs Þ2 This investigation assessed the performances of the GCMs by
RMSE ¼ t ð3Þ
i¼1
N comparing the historical CDD and HDD trends (the OLS slopes
described in Section 2.5.1) with their corresponding model predic-
where T GCM and T Obs are the predicted (by a particular GCM) and tions obtained from the historical runs. Figure SM2 shows the box
observed values of the daily mean temperature on the ith day,
respectively, and N is the number of days (N = 13514 days for years
Table 2
1969–2005). Selected general circulation models (GCMs) for making future temperature predic-
We also calculated the absolute error in predicting the historical tions in major Indian cities.
temperature trends (linear slope) for each GCM by using the fol-
City Selected GCM
lowing equation:
Ahmedabad BNU_ESM
 
 
AELS ¼  b
b GCM
b
Bengaluru IPSL_CM5A_LR
b Obs  ð4Þ Chennai BCC_CSM1_1
Delhi MIROC_ESM_CHEM
where AELS is the absolute error in the linear slope, b b
b GCM and b Obs
Hyderabad, Kolkata, and Mumbai MIROC_ESM
Srinagar GFDL_ESM2G
are the GCM predicted and observed trends (slopes) in annual mean
4
R. Ukey and A.C. Rai Energy & Buildings 244 (2021) 111050

plots of the CDD and HDD trends from 1969 to 2005 predicted by After prewhitening, we applied the Mann Kendall (MK) test
the different GCMs together with the observed trends and those [42,43] on the time-series. The MK test (details in Supplementary
predicted by the selected GCMs given in Table 2. Figure SM2 shows Material) is a rank-based method to detect trends, which is less
large differences in model predictions; however, the selected GCMs sensitive to outliers than the OLS method and does not require
can reasonably capture the historical HDD and CDD trends with the residuals to be normally distributed. Although the MK test
prediction errors ranging between 1 and 39%. Thus, those GCMs can detect the existence or non-existence of a trend, it cannot pro-
were deemed suitable for forecasting future temperatures (2018– vide its magnitude. Thus, we used the Theil-Sen slope [44,45] esti-
2100) and calculating the future HDDs and CDDs. Note that since mator ( bb TS ) to calculate the magnitude of the linear trend that is
there were ties in model rankings for Ahmedabad and Mumbai, given by:
we selected those GCMs for making future projections that best  
captured the observed CDD trends in those cities. b xj
xi
b TS ¼ Median for all j > i ð10Þ
j
i
2.5. Linear trend analysis of temperatures and degree-days where xj and xi are the ith and jth values in the time-series (j > i).

This investigation quantified the trends in annual mean temper-

3. Results
atures and degree days for the past (1969–2017) and the future
(2018–2100) by using two approaches: 1) applying the ordinary
This study analyzed the annual mean temperatures, CDD, and
least square regression to the time series (OLS analysis) and 2)
HDD variations in eight major Indian cities for 1969–2100. We first
prewhitening the time series followed by conducting the Mann-
discuss the historical (1969–2017) and future (2018–2100) trends
Kendall (MK) test and estimating the Theil-Sen (TS) slope
in annual temperatures, followed by a similar discussion for CDDs
(PWMKTS analysis).
and HDDs.
2.5.1. Ordinary least square (OLS) analysis
OLS regression is a widely used method to quantify trends in 3.1. Temperature trends (1969–2100)
climatological time-series because of its simplicity and broad
acceptance [17,34–38]. In this technique, a straight line is fit to 3.1.1. Historical temperature trends (1969–2017)
the time-series modeled as: Fig. 3 shows the historical and future projections of annual
mean temperature anomalies for Delhi from 1969 to 2100. The
xi ¼ a þ bt þ ei ; i ¼ 1; 2;    n ð7Þ temperature anomalies were calculated as differences between
th
where xi is the i observations in the time series, t the time, a and b the annual mean temperatures and a baseline temperature (mean
are the linear intercept and slope values, respectively, and ei is the temperature from 1969 to 2017). Fig. 3 shows that the annual
error term. The OLS method provides an estimate of the linear trend mean temperatures increased (positive anomaly values) during
b OLS ) by minimizing the sum of squared errors (Pn e2 ). The statis- 1969–2017 in Delhi at a rate of 0.12 °C/decade (p-value = 0.009
(b i¼1 i and 0.03–0.21 is the 95% confidence interval), as estimated from
tical significance of the trend was tested using the Student’s t-test at
the OLS slope given in Table 3. We obtained a similar conclusion
a significance level of 0.05. from the PWMKTS analysis and the corresponding TS slope values
(see Table 3), i.e., the observed temperature trend was significant
2.5.2. Prewhitening, Mann Kendall test, and Theil-Sen slope (PWMKTS) for Delhi (p-value = 0.042), and its TS slope was 0.10 °C/decade
analysis (0.01–0.19 is the 95% confidence interval). The TS slope was
Although the OLS technique is popular for quantifying linear slightly lower than the OLS slope since the prewhitened time-
trends, it assumes that ei is normally distributed and there is no series was used for TS slope calculations, whereas the original
autocorrelation (correlation of a variable with itself at differing
time lags) among the observations, which is seldom true for clima-
tological time-series. Thus, we employed another model that
assumes that the time-series is the sum of an AR(1) autoregressive
process (current observation depends on the immediately preced-
ing value) and a linear model [39], given by:
xi ¼ axi
1 þ a þ bt þ ei ; i ¼ 1; 2;    n ð8Þ
where a denotes the lag-1 autocorrelation between xi and xi
1.
The above equation was transformed to Eq. (7) by a ‘prewhiten-
ing’ process (substituting yi ¼ xi
axi
1 in Eq. (8) converts it into
Eq. (7) form) that removes the lag-1 autocorrelation from the
time-series without affecting the ‘true’ trend coefficient (b). The
lag-1 autocorrelation was estimated ( a b ) as the correlation coeffi-
cient between two N–1 long subsamples of the time-series lagged
by one-time step using the following equation:
Pn
1
i¼1 fðxi
xÞ  ðxiþ1
xÞg
1
ab ¼ n
1 Pn 2
ð9Þ
i¼1 ðxi
xÞ
1
n

where x represents the mean value.

The lag-1 autocorrelation ( a
b ) was significant (significance level
of 0.05) for the time-series of annual temperatures, CDDs, and
HDDs and ranged between 0.3 and 0.9 for the different time-
series, as shown in Figure SM3. Thus, the time-series were prewhi-
tened before conducting additional statistical tests [40,41]. Fig. 3. Historical and future anomalies in annual mean temperatures in Delhi.

5
R. Ukey and A.C. Rai Energy & Buildings 244 (2021) 111050

time-series (without prewhitening) was used to calculate the OLS 3.1.2. Future temperature trends (2018–2100)
slope. The OLS slope possibly overestimates the annual tempera- Fig. 3 also shows that Delhi’s annual temperatures will gradu-
ture trends since there is significant positive autocorrelation in ally increase from 2018 to 2100 under both the emission scenarios
the original time-series. In contrast, the TS slope likely underesti- (RCP4.5 and RCP8.5). Under the RCP4.5 scenario, the annual tem-
mates the trend since prewhitening could have reduced the magni- peratures in Delhi are projected to be 0.6 °C higher in the 2020 s
tude of the trend [46]. (2005–2034), 1.4 °C higher in the 2050 s (2035–2064), and 2.2 °C
Similarly, for all the other seven cities, the annual temperatures higher in the 2080 s (2065–2094) when compared to the baseline
increased in the past 49 years, as shown in Figures SM4 a–g, with (mean temperature from 1969 to 2017), as shown in Fig. 4. On the
the OLS slopes ranging from 0.13 to 0.23 °C/decade, as given in other hand, under the RCP8.5 scenario, temperatures will be 0.6 °C
Table 3. The PWMKTS analysis also reached a similar conclusion, higher in the 2020 s, 2.0 °C higher in the 2050 s, and 4.5 °C higher
i.e., the observed temperature trends were statistically significant, in the 2080 s than the baseline temperature. Similarly, for all the
and the magnitudes of TS slope ranged between 0.08 and 0.14 °C/ other cities, the annual temperatures will increase from 2018 to
decade, as given in Table 3. Once again, the OLS slopes were higher 2100 under both the emission scenarios, as shown in Fig. 4 and Fig-
than their corresponding TS slopes due to the reasons discussed in ures SM4 a–g.
the preceding paragraph. In the 2020 s, the projected increase in annual temperatures
We also compared the temperature trends in the selected cities ranged between 0.1 °C–1.1 °C and was almost equal in both the
with those previously reported for India during the periods 1969– scenarios for all cities. However, in the 2050 s and 2080 s, the
2005 [47] and 1971–2003 [48]. During 1969–2005, Basha et al. annual temperature rise is projected to be significantly lower
[47] reported that the country’s annual temperatures increased under the RCP4.5 scenario than under the RCP8.5 scenario. The
by 0.081 °C/decade or by 0.168 °C/decade based on datasets OLS analysis detected statistically significant trends in the pro-
obtained from IMD, India, or Climate Research Unit, UK, respec- jected annual temperatures for all the eight cities (see Table 3),
tively. Similarly, Kothawale and Kumar [48] also reported that with magnitudes ranging from 0.10 to 0.38 °C/decade under the
India’s annual temperatures increased by 0.22 °C/decade during RCP4.5 scenario and from 0.35 to 0.78 °C/decade under the
1971–2003. Our investigation also found increasing temperature RCP8.5 scenario. The estimated temperature trends are in qualita-
trends for the selected cities, with magnitudes ranging between tive agreement with those reported by Basha et al. [47], who found
0.03 and 0.23 °C/decade during 1969–2005 and between 0.04 that the annual mean temperatures in India would increase by
and 0.28 during 1971–2003. Thus, our results are in qualitative 0.52 °C/decade in the 21st century under the RCP8.5 scenario.
agreement with those reported previously. Note that the tempera- The PWMKTS analysis also detected statistically significant
ture rise in Indian cities may not be solely due to increasing green- trends for all cities in both the emission scenarios with TS slopes
house gas (GHG) emissions since other factors such as ranging from 0.03 to 0.07 °C/decade under the RCP4.5 emission
anthropogenic aerosols, natural forcings, and land-use/land-cover scenario and from 0.03 to 0.06 °C/decade under the RCP8.5 sce-
(LULC) changes also impact the surface air temperatures nario. The exceptions were Delhi, Hyderabad, and Kolkata under
[47,49,50]. In India, increasing GHG emissions and LULC changes the RCP4.5 scenario, and Bengaluru under the RCP8.5 scenario,
have contributed significantly towards increasing temperatures for which the PWMKTS analysis did not detect statistically signifi-
during the 20th century, while changes in anthropogenic aerosol cant trends (p-value > 0.05). The discrepancies between the OLS
emissions and natural forcings provided some cooling effect [47]. and the PWMKTS analysis arise because we used the original
Thus, we also anticipate a similar contribution of those factors time-series for the former, while the latter used the prewhitened
towards the estimated temperature trends in the selected cities. time-series. This likely led to an overestimation of the trend by

Table 3
Historical and future trends in annual mean temperature.

Time period City OLS analysis PWMKTS analysis

b
Slope (°C/decade) p-value TS Slopeb (°C/decade) MK test p-value
Historical (1969–2017) Ahmedabad 0.13 (0.02, 0.23) 0.019 0.09 (0.00, 0.19) 0.042
Bengaluru 0.22 (0.15, 0.27) < 0.001 0.13 (0.06, 0.19) 0.001
Chennai 0.16 (0.10, 0.20) < 0.001 0.08 (0.01, 0.14) 0.015
Delhi 0.12 (0.03, 0.20) 0.009 0.10 (0.01, 0.19) 0.028
Hyderabad 0.21 (0.14, 0.28) < 0.001 0.10 (0.03, 0.18) 0.007
Kolkata 0.23 (0.17, 0.28) < 0.001 0.12 (0.06, 0.19) 0.001
Mumbai 0.13 (0.06, 0.18) < 0.001 0.01 (0.02, 0.18) 0.013
Srinagar 0.20 (0.08, 0.32) 0.001 0.14 (0.00, 0.26) 0.044
RCP4.5 (2018–2100) Ahmedabad 0.18 (0.14, 0.22) < 0.001 0.05 (0.01, 0.08) 0.010
Bengaluru 0.19 (0.16, 0.22) < 0.001 0.03 (0.00, 0.05) 0.020
Chennai 0.12 (0.09, 0.15) < 0.001 0.04 (0.019, 0.08) 0.002
Delhi 0.27 (0.22, 0.31) < 0.001 0.03a (
0.01, 0.06) 0.176
Hyderabad 0.38 (0.33, 0.43) < 0.001 0.01a (
0.02, 0.03) 0.725
Kolkata 0.21 (0.16, 0.25) < 0.001 0.03a (0.00, 0.06) 0.060
Mumbai 0.19 (0.16, 0.22) < 0.001 0.04 (0.01, 0.07) 0.010
Srinagar 0.10 (0.05, 0.14) < 0.001 0.07 (0.02, 0.11) 0.006
RCP8.5 (2018–2100) Ahmedabad 0.53 (0.49, 0.56) < 0.001 0.06 (0.02, 0.09) 0.006
Bengaluru 0.52 (0.50, 0.54) < 0.001 0.03a (0.00, 0.05) 0.073
Chennai 0.34 (0.31, 0.37) < 0.001 0.04 (0.00, 0.08) 0.044
Delhi 0.78 (0.73, 0.83) < 0.001 0.05 (0.02, 0.09) 0.001
Hyderabad 0.72 (0.69, 0.76) < 0.001 0.03 (0.01, 0.06) 0.013
Kolkata 0.65 (0.62, 0.68) < 0.001 0.03 (0.00, 0.06) 0.023
Mumbai 0.51 (0.49, 0.54) < 0.001 0.04 (0.00, 0.07) 0.041
Srinagar 0.54 (0.49, 0.58) < 0.001 0.06 (0.01, 0.11) 0.022
a
Not a statistically significant result (p-value > 0.05).
b
The values in parenthesis represent the 95% confidence interval.

6
R. Ukey and A.C. Rai Energy & Buildings 244 (2021) 111050

Fig. 4. Increase in annual mean temperature in 2020 s (2005–2034), 2050 s (2035–2064), and 2080 s (2065–2094) for eight major Indian cities under the a) RCP4.5 and b)
RCP8.5 emission scenario.

the OLS analysis and an under-estimation of the trend by the

PWMKTS analysis, as discussed previously.

3.2. CDD trends (1969–2100)

This section discusses the historical and future trends in CDDs

for all cities, except Srinagar, which has a cold climate and there-
fore no cooling requirements.

3.2.1. Historical CDD trends (1969–2017)

Fig. 5 shows the historical (1969–2017) and future projections
(2018–2100) of CDDs for Delhi, as well as the baseline value
(2809 °Cd), which is the average CDD value from 1969 to 2017.
The CDDs in Delhi increased significantly during 1969–2017, i.e.,
by 40.8 °Cd/decade (14.7–67.3 °Cd/decade at 95% confidence level),
as quantified by the OLS slope given in Table 4, which corresponds
to a 1.5% CDD increase per decade. The PWMKTS analysis also
detected a statistically significant trend in CDDs during 1969–
2017 of magnitude 34.2 °Cd/decade (6.6–60.5 at 95% confidence
level), as shown in Table 4, i.e., a 1.2% CDD increase per decade.
Similarly, for the other six cities with cooling requirements, CDDs
increased from 1969 to 2017 (see Figures SM5 a–f). The OLS slopes
for CDDs were statistically significant and ranged between 40.8
and 86.7 °Cd/decade (1.3–3.8% CDD increase per decade), as given Fig. 5. Historical and future cooling degree days (CDDs) in Delhi.
in Table 4. The PWMKTS analysis also detected statistically signif-
icant trends in CDDs for all cities except Ahmedabad (MK test p-
value = 0.056), with the corresponding TS slopes ranging between investigations have reported CDDs to increase by about 3.7–85.4%/
27.9 and 48.4 °Cd/decade (0.7–2.3% CDD increase per decade) in decade in Italian cities from 1980 to 2013 [16] and by 0–40%/dec-
Table 4. ade in Spanish cities during 1970–2005 [14].
Once more, OLS slopes were higher than the corresponding TS
slopes since prewhitening was used before calculating the TS 3.2.2. Future CDD trends (2018–2100)
slopes. Overall, it is clear that the CDDs have increased for all the Fig. 5 also shows the future projections of Delhi’s CDDs from
cities from 1969 to 2017; thus, increasing the cooling energy 2018 to 2100 under the RCP4.5 and RCP8.5 scenarios. Depending
requirements in buildings. The CDD trends in Indian cities are gen- on the emission scenario, CDDs in Delhi will be higher by 6.4–
erally in qualitative agreement with previous investigations con- 7.1% in the 2020 s, by 15.2–20.8% in the 2050 s, and by 24.3–
ducted in other countries (CDD trends for India are unavailable 52.7% in the 2080 s, compared to the baseline (see Fig. 5), with
in the existing literature). However, significant differences can be the largest increase happening under RCP8.5 (high emission sce-
seen in the quantitative values due to the differences in the cli- nario) in the 2080 s. Similarly, CDDs will increase during 2018–
matic conditions, research approach, and the base values used for 2100 for all the other cities under both emission scenarios, as pre-
calculating CDDs. For example, we found that during 1969–2017, sented in Fig. 6 (see Figures SM5 a–f). Fig. 6 shows that in the
the CDDs increased in Indian cities by 0.7–3.8%/decade, while other 2020 s and 2050 s, the CDD increase will be roughly equal under
7
R. Ukey and A.C. Rai Energy & Buildings 244 (2021) 111050

Table 4
Historical and future trends in cooling degree days (CDDs).

Time period City OLS analysis PWMKTS analysis

b
Slope (°Cd/decade) p-value TS Slopeb (°Cd/decade) MK test p-value
Historical (1969–2017) Ahmedabad 45.7 (7.6, 83.9) 0.020 31.9 (
0.1, 66.3) 0.056
Bengaluru 78.2 (57.7, 98.7) < 0.001 48.4 (22.0, 68.5) < 0.001
Chennai 56.9 (39.1, 74.6) < 0.001 27.9 (6.0, 49.7) 0.015
Delhi 40.8 (14.7, 67.3) 0.003 34.2 (6.6, 60.5) 0.017
Hyderabad 77.8 (53.4, 102.3) < 0.001 38.5 (9.5, 67.1) 0.005
Kolkata 86.7 (67.3, 106.1) < 0.001 45.6 (22.1, 69.6) 0.001
Mumbai 47.1 (23.4, 70.8) < 0.001 35.4 (7.5, 63.4) 0.016
RCP4.5 (2018–2100) Ahmedabad 66.8 (56.4, 77.2) < 0.001 16.5 (4.1, 29.7) 0.012
Bengaluru 69.9 (38.8, 101.0) < 0.001 11.5 (1.7, 19.9) 0.022
Chennai 44.5 (33.7, 56.3) < 0.001 16.4 (6.3, 27.8) 0.002
Delhi 86.0 (71.4, 98.5) < 0.001 7.7a (
2.9, 17.2) 0.132
Hyderabad 140.1 (120.7, 159.4) < 0.001 1.6a (
8.1, 12.2) 0.731
Kolkata 76.8 (59.9, 93.7) < 0.001 10.6a (
0.7, 21.6) 0.063
Mumbai 71.0 (61.1, 80.8) < 0.001 14.3 (3.8, 25.9) 0.009
RCP8.5 (2018–2100) Ahmedabad 193.3 (196.5, 190.2) < 0.001 20.4 (6.0, 33.3) 0.008
Bengaluru 191.6 (260.6, 122.7) < 0.001 8.8a (
0.4, 19.8) 0.058
Chennai 126.7 (116.4, 136.9) < 0.001 13.6 (0.2, 27.2) 0.046
Delhi 265.1 (246.4, 283.7) < 0.001 18.0 (7.6, 28.8) 0.001
Hyderabad 265.5 (252.1, 278.8) < 0.001 11.9 (2.3, 20.7) 0.017
Kolkata 239.7 (229.4, 279.9) < 0.001 12.0 (1.4, 23.2) 0.031
Mumbai 189.2 (179.4, 198.9) < 0.001 12.7 (0.4, 23.9) 0.044
a
Not a statistically significant result (p-value > 0.05).
b
The values in parenthesis represent the 95% confidence interval.

Fig. 6. Increase in cooling degree days (CDDs) in 2020 s (2005–2034), 2050 s (2035–2064), and 2080 s (2065–2094) for seven major Indian cities under the a) RCP4.5 and b)
RCP8.5 emission scenario.

both the emission scenarios, with CDD increasing by 2.9–22.9% in 4 °Cd/decade (0.4–0.6% CDD increase per decade) under RCP4.5
the 2020 s and by 8.3–54.1% in the 2050 s for the selected cities. and RCP8.5 scenarios, respectively. The discrepancies in detecting
However, by the 2080 s, the CDD increase under the RCP8.5 sce- trends between the OLS and PWMKTS analysis arise due to the
nario will be much higher (by 12–29 percentage points depending prewhitening process applied in PWMKTS analysis.
on the city) than that under the RCP4.5 scenario.
The OLS slopes of CDDs were also statistically significant for all 3.3. HDD trends (1969–2100)
the six cities, with magnitudes ranging from 44.5 to 140.1 °Cd/dec-
ade (1.1–4.5% CDD increase per decade) and 126.7–265.5 °Cd/dec This section discusses the historical and future trends in HDDs
ade (3.3–9.4% CDD increase per decade) under the RCP4.5 and for Delhi and Srinagar since only those cities have heating
RCP8.5 scenarios, respectively, as shown in Table 4. The PWMKTS requirements.
analysis also detected statistically significant CDD trends for most
cities, except Delhi, Hyderabad, and Kolkata under the RCP4.5 sce- 3.3.1. Historical HDD trends (1969–2017)
nario and Bengaluru under the RCP8.5 scenario. The corresponding Fig. 7 shows the historical (1969–2017) and future (2018–2100)
TS Slopes (when statistically significant) ranged between 11.5 and projections of HDDs for Delhi, along with its baseline value
16.5 °Cd/decade (0.1–0.6% CDD increase per decade) and 11.9–20. (277 °Cd). The HDDs were almost constant during 1969–2017,
8
R. Ukey and A.C. Rai Energy & Buildings 244 (2021) 111050

ments will also reduce, as HDDs decrease by 8.1–10.4% in the

2020 s, by 17.6–23.9% in the 2050 s, and by 19.3–38.1% in the
2080 s, depending on the emission scenario. The OLS analysis
found significant linear trends in future HDDs for Delhi. Delhi’s
HDDs decrease by –14.8 °Cd/decade (–5.5% per decade) and by –
21.2 °Cd/decade (–7.9% per decade) under the RCP4.5 and RCP8.5
scenarios, respectively.
On the other hand, the PWMKTS analysis detected a significant
HDD trend of magnitude –4.6 °Cd/decade (–1.7% per decade) under
the RCP4.5 scenario, but not under the RCP8.5 scenario. For Srina-
gar, both the OLS and PWMKTS analysis detected significant
decreasing trends in future HDDs. The OLS slopes were –29.2 °Cd/
decade (–1.3% per decade) and –102.4 °Cd/decade (–4.6% per dec-
ade) under the RCP4.5 and RCP8.5 scenarios, respectively; while
the TS slope was –19.2 °Cd/decade (–0.9% per decade) under both
scenarios. Once more, we postulate that the ‘‘true” HDD trend is
likely overestimated by OLS analysis as it does not account for
autocorrelation in the time-series, while the PWMKTS analysis
underestimates it since prewhitening can partly remove the trend.

4. Discussion
Fig. 7. Historical and future heating degree days (HDDs) in Delhi.
4.1. Implications of research findings and policy actions

which was also established from the OLS and PWMKTS analysis (no This investigation is the first to quantify the historical trends in
significant linear trend), presented in Table 4. In contrast, HDDs in annual temperatures, HDDs, and CDDs in major Indian cities,
Srinagar decreased significantly during 1969–2017 (see Fig- together with their future projections under two different green-
ure SM6), with an OLS slope of
68.9 °Cd/decade (
100.0,
37.8 house gas emission scenarios. The study found that rising temper-
°Cd/decade at 95% confidence level), as presented in Table 5, atures have and would significantly increase the space cooling
which corresponds to a 3.1% HDD decrease per decade. The requirements in major Indian cities, which becomes especially
PWMKTS analysis also obtained a similar conclusion, i.e., there problematic since India is one of the world’s most populous and
was a significant downward HDD trend in Srinagar (see Table 5). hottest regions. A warming climate coupled with rising household
Srinagar’s HDDs decreased by 42.5 °Cd/decade (5.6–77.2 °Cd/dec incomes and built-up areas could lead to a staggering increase in
ade is the 95% confidence level), which amounts to 1.9% HDD air-conditioning demand in India, which has about four times the
decrease per decade. Thus, building heating requirements population and more than three times as many CDDs per person
decreased in Srinagar while they remained constant in Delhi dur- as the United States [51]. Thus, the study highlights the urgent
ing 1969–2017. The decrease in HDDs during 1969–2017 was need to adopt energy-efficient building practices in India and glo-
<3.1%/decade in Indian cities, whereas HDDs decreased by 1.3– bal action on climate change.
8.6%/decade during 1980–2013 and by 2.3–6.0%/decade during Addressing India’s rising space cooling requirements in a sus-
1970–2005 in Italian [16] and Spanish [14] cities, respectively. tainable fashion would require synergy between energy policies
Therefore, our estimated results are also in qualitative agreement and actions. One of the key inputs for this would be to ensure data
with previous investigations. availability concerning the cooling energy demand across different
building sectors, air-conditioning sales, and energy-efficiency
3.3.2. Future HDD trends (2018–2100) statistics as a part of the Indian government’s open data platform
Fig. 8 shows that HDDs will decrease in the future in Delhi and (www.data.gov.in). Future policy actions should focus on improv-
Srinagar under both the emission scenarios when compared to ing the energy performance of the building sector and air-
their corresponding baseline values. For Delhi, HDDs decrease by conditioning systems through regulations and incentive schemes.
28.4–30.3% in the 2020 s, by 71.6–83.3% in the 2050 s, and by A two-pronged approach that firstly reduces active space cooling
83.5–97.1% in the 2080 s, depending on the emission scenario. needs in buildings by promoting energy-efficient construction
Thus, the building heating requirements will become very small and retrofitment practices and then meets the remaining cooling
(HDDs < 80 °Cd) in Delhi by the 2050 s under both emission sce- needs through efficient air-conditioning systems can help fulfill
narios (see Fig. 7). Similarly, Srinagar’s building heating require- the country’s growing cooling demand.

Table 5
Historical and future trends in heating degree days (HDDs).

Time period City OLS analysis PWMKTS analysis

Slopeb (°Cd/decade) p-value Slopeb (°Cd/decade) MK test p-value
a a
Historical (1969–2017) Delhi
3.2 (
13.6, 7.2) 0.541
3.8 (
13.2, 7.7) 0.472
Srinagar
68.9 (
100.0,
37.8) < 0.001
42.4 (
77.2,
5.6) 0.020
RCP4.5 (2018–2100) Delhi
14.8 (
18.6,
11.0) < 0.001
4.6 (
7.6,
1.2) 0.010
Srinagar
29.2 (
39.8,
18.5) < 0.001
19.2 (
30.7,
7.8) 0.002
RCP8.5 (2018–2100) Delhi
21.2 (
2.3,
18.1) < 0.001
1.3a (
3.7, 0.0) 0.065
Srinagar
102.4 (
15.2,
89.7) < 0.001
19.2 (
34.9,
4.1) 0.013
a
Not a statistically significant result (p-value > 0.05).
b
The values in parenthesis represent the 95% confidence interval.

9
R. Ukey and A.C. Rai Energy & Buildings 244 (2021) 111050

Fig. 8. Decrease in in annual heating degree days (HDDs) in 2020 s (2005–2034), 2050 s (2035–2064), and 2080 s (2065–2094) for two major Indian cities under the a) RCP4.5
and b) RCP8.5 emission scenario.

4.2. Study limitations and future research directions estimated CDD trends between 0.7 and 2.3%/decade. In the two
cities with heating requirements, the historical HDD trends were
This investigation used HDDs and CDDs to quantify the space H/ either insignificant (in Delhi) or decreasing (by 1.9% or 3.1% per
C energy requirements of buildings, which only depend on the out- decade in Srinagar, depending on the analysis). We also detected
door air temperatures. However, other factors such as the build- statistically significant trends in most cities’ future temperatures,
ings’ thermal characteristics, relative humidity, and solar with magnitudes ranging between 0.10 and 0.78 °C/decade (OLS)
radiation substantially impact the H/C energy requirements, which or 0.03–0.07 °C/decade (PWMKTS), depending upon the city and
should be included in future investigations. Further uncertainties the emission scenario. Future CDDs also had an increasing trend
in results can arise due to missing values present in the historical in cities with cooling requirements, with magnitudes equal to
weather data and the imputation algorithm employed for filling 1.1–9.4%/decade (OLS) or 0.1–0.6%/decade (PWMKTS). In contrast,
them, as well as the inherent limitations of the GCMs in making cities with heating requirements generally displayed a decreasing
reliable temperature predictions. This research also did not charac- trend in future HDDs with magnitudes of 1.3–7.9%/decade (OLS)
terize the inter-model variabilities in future climate predictions or 0.9–1.7% (PWMKTS).
since a single GCM, selected based on its historical performance, We also estimated that annual temperatures would be 0.1–
was used for each city. Further research can provide multi-model 1.1 °C higher in the 2020 s, 0.6–2.8 °C higher in the 2050 s, and
projections of the future HDDs and CDDs in India to characterizing 1.0–4.6 °C higher in the 2080 s in the studied cities. Due to increas-
the inter-model variabilities. It is also recommended to explore the ing temperatures, CDDs will also increase by 2.9–22.9% in the
effects of increasing population and incomes in India on the build- 2020 s, by 8.3–54.1% in the 2050 s, and by 11.89–83.0% in the
ing energy demand and the associated CO2 emissions. 2080 s, depending on the city and the emission scenario. Thus,
increasing the space cooling energy requirements in buildings by
a similar amount. In contrast, HDDs will decrease by 8.1–30.3%
5. Conclusions in the 2020 s, by 17.6–83.3% in the 2050 s, and by 19.3–97.1% in
the 2080 s, thereby reducing the space heating requirements.
This research studied the impact of global warming on the his-
torical (1969–2017) and future (2018–2100) annual temperatures,
HDDs, and CDDs in India’s eight major cities, covering all the cli- Declaration of Competing Interest
mate zones. We estimated the linear trends in the temperature,
HDD, and CDD time-series by using two different techniques: i) The authors declare that they have no known competing finan-
applying ordinary least square regression on the time-series (OLS cial interests or personal relationships that could have appeared
analysis) and ii) prewhitening the time-series followed by applying to influence the work reported in this paper.
the Mann-Kendall Test and computing the Theil-Sen slope
(PWMKTS analysis). Acknowledgment
Both techniques showed that all cities witnessed a significant
increase in temperature during 1969–2017. Depending on the city, The authors thank Mr. Rishabh Gupta, Department of Mechanical
the temperature trends were between 0.13 and 0.23 °C/decade or Engineering, BITS-Pilani, for his help in obtaining the weather data.
0.08–0.13 °C/decade, obtained from the OLS or PWMKTS analysis, Climate scenarios used were from the NEX-GDDP dataset, prepared
respectively. Due to risings temperatures, the historical CDD trends by the Climate Analytics Group and NASA Ames Research Center
were statistically significant and positive in all the seven cities using the NASA Earth Exchange, and distributed by the NASA Cen-
with cooling requirements (Ahmedabad was a slight exception). ter for Climate Simulation (NCCS). We also acknowledge the Indian
The historical CDD trends ranged between 1.3 and 3.8%/decade Metrological Department (IMD) for providing the historical tem-
obtained from the OLS analysis, whereas the PWMKTS analysis perature datasets.
10
R. Ukey and A.C. Rai Energy & Buildings 244 (2021) 111050

Funding [23] BIS (Bureau of Indian Standards), National building code of India 2005, Bureau
of Indian standards, New Delhi, 2005..
[24] UN (United Nations), The world’s cities in 2018, Department of Economic and
This research did not receive any specific grant from funding Social Affairs, Population Division, 2018..
agencies in the public, commercial, or not-for-profit sectors. [25] ASHRAE (American Society of Heating Refrigerating and Air-conditioning
Engineers), Handbook HVAC fundamentals, 2009..
[26] M. Bhatnagar, J. Mathur, V. Garg, Determining base temperature for heating
Appendix A. Supplementary data and cooling degree-days for India, J. Build. Eng. 18 (2018) 270–280, https://doi.
org/10.1016/j.jobe.2018.03.020.
[27] L.S. Rathore, S.D. Attri, A.K. Jaswal, State level climate change trends in India,
Supplementary data to this article can be found online at Ministry of Earth Sciences, Goverment of India, New Delhi, 2013.
https://doi.org/10.1016/j.enbuild.2021.111050. [28] G. Tardivo, A. Berti, A dynamic method for gap filling in daily temperature
datasets, J. Appl. Meteorol. Climatol. 51 (2012) 1079–1086, https://doi.org/
10.1175/JAMC-D-11-0117.1.
References [29] W.P. Kemp, D.G. Burnell, D.O. Everson, A.J. Thomson, Estimating missing daily
maximum and minimum temperatures, J. Clim. Appl. Meteorol. 22 (9) (1983)
[1] IEA (International Energy Agency), Global status report for buildings and 1587–1593, https://doi.org/10.1175/1520-0450(1983)022<1587:
construction: towards a zero-emissions, efficient and resilient buildings and EMDMAM>2.0.CO;2.
construction sector, Paris, 2019.. [30] C. Dunis, V. Karalis, Weather derivatives pricing and filling analysis for missing
[2] IEA (International Energy Agency), Data and statistics for India, (2018). https:// temperature data, Deriv. Use Trading Regul. (2003) 1–19. http://cwis.livjm.ac.
www.iea.org/data-and-statistics/data-tables?country=INDIA&energy= uk/AFE/AFE_docs/cibef0103.pdf..
Balances&year=2018 (accessed October 15, 2020).. [31] A.M. Thomson, K.V. Calvin, S.J. Smith, G.P. Kyle, A. Volke, P. Patel, S. Delgado-
[3] S. Kumar, S. Kachhawa, A. Goenka, S. Kasamsetty, G. George, Demand analysis Arias, B. Bond-Lamberty, M.A. Wise, L.E. Clarke, J.A. Edmonds, RCP4.5: a
for cooling by sector in India in 2027, Alliance for an Energy Efficient Economy, pathway for stabilization of radiative forcing by 2100, Clim. Change. 109
New Delhi, 2018.. (2011) 77–94, https://doi.org/10.1007/s10584-011-0151-4.
[4] IPCC, (Intergovernmental Panel on Climate Change), Climate change 2013: the [32] K. Riahi, S. Rao, V. Krey, C. Cho, V. Chirkov, G. Fischer, G. Kindermann, N.
physical science basis: working group I contribution to the fifth assessment Nakicenovic, P. Rafaj, RCP 8.5–a scenario of comparatively high greenhouse gas
report of the Intergovernmental Panel on Climate Change, Cambridge emissions, Clim. Change. 109 (1-2) (2011) 33–57, https://doi.org/10.1007/
University Press, 2014. s10584-011-0149-y.
[5] M. Santamouris, Cooling the buildings – past, present and future, Energy Build. [33] P.J. Gleckler, K.E. Taylor, C. Doutriaux, Performance metrics for climate models,
128 (2016) 617–638, https://doi.org/10.1016/j.enbuild.2016.07.034. J. Geophys. Res. 113 (2008) 1–20, https://doi.org/10.1029/2007JD008972.
[6] MoFF (Ministry for Environment Forest and Climate Change), India cooling [34] M.S. Desai, A.G. Dhorde, Trends in thermal discomfort indices over western
action plan, Ministry of Enironment and Forests, Government of India, New coastal cities of India, Theor. Appl. Climatol. 131 (3-4) (2018) 1305–1321,
Delhi, 2018.. https://doi.org/10.1007/s00704-017-2042-8.
[7] P. Shen, Impacts of climate change on U.S. building energy use by using [35] S.G. Meshram, V.P. Singh, C. Meshram, Long-term trend and variability of
downscaled hourly future weather data, Energy Build. 134 (2017) 61–70, precipitation in Chhattisgarh State, India, Theor. Appl. Climatol. 129 (3-4)
https://doi.org/10.1016/j.enbuild.2016.09.028. (2017) 729–744, https://doi.org/10.1007/s00704-016-1804-z.
[8] H. Wang, Q. Chen, Impact of climate change heating and cooling energy use in [36] P. Singh, V. Kumar, T. Thomas, M. Arora, Basin-wide assessment of
buildings in the United States, Energy Build. 82 (2014) 428–436, https://doi. temperature trends in northwest and central India, Hydrol. Sci. J. 53 (2008)
org/10.1016/j.enbuild.2014.07.034. 421–433, https://doi.org/10.1623/hysj.53.2.421.
[9] X. Wang, D. Chen, Z. Ren, Assessment of climate change impact on residential [37] P. Sonali, D. Nagesh Kumar, Review of trend detection methods and their
building heating and cooling energy requirement in Australia, Build. Environ. application to detect temperature changes in India, J. Hydrol. 476 (2013) 212–
45 (7) (2010) 1663–1682, https://doi.org/10.1016/j.buildenv.2010.01.022. 227, https://doi.org/10.1016/j.jhydrol.2012.10.034.
[10] M. Sivak, Potential energy demand for cooling in the 50 largest metropolitan [38] H. Tabari, P. Hosseinzadeh Talaee, Recent trends of mean maximum and
areas of the world: implications for developing countries, Energy Policy. 37 minimum air temperatures in the western half of Iran, Meteorol. Atmos. Phys.
(2009) 1382–1384, https://doi.org/10.1016/j.enpol.2008.11.031. 111 (3-4) (2011) 121–131, https://doi.org/10.1007/s00703-011-0125-0.
[11] CIBSE (Chartered Institutiion of Building Services Engineers), Degree-days: [39] K.H. Hamed, Enhancing the effectiveness of prewhitening in trend analysis of
theory and application (TM41: 2006), London, 2006.. hydrologic data, J. Hydrol. 368 (1-4) (2009) 143–155, https://doi.org/10.1016/
[12] M. Christenson, H. Manz, D. Gyalistras, Climate warming impact on degree- j.jhydrol.2009.01.040.
days and building energy demand in Switzerland, Energy Convers. Manag. 47 [40] M. Gocic, S. Trajkovic, Analysis of changes in meteorological variables using
(6) (2006) 671–686, https://doi.org/10.1016/j.enconman.2005.06.009. Mann-Kendall and Sen’s slope estimator statistical tests in Serbia, Glob. Planet.
[13] F. Jiang, X. Li, B. Wei, R. Hu, Z. Li, Observed trends of heating and cooling Change. 100 (2013) 172–182, https://doi.org/10.1016/j.gloplacha.2012.10.014.
degree-days in Xinjiang province, China, Theor. Appl. Climatol. 97 (3-4) (2009) [41] J.D. Salas, J.W. Delleur, V.M. Yevjevich, W.L. Lane, Applied modeling of
349–360, https://doi.org/10.1007/s00704-008-0078-5. hydrologic time series, Water Resource Publications, Colorado, 1980.
[14] M.J. OrtizBeviá, G. Sánchez-López, F.J. Alvarez-Garcìa, A. RuizdeElvira, [42] H.B. Mann, Nonparmetric tests against trend, Econometrica. 13 (1945) 245–
Evolution of heating and cooling degree-days in Spain: trends and 259.
interannual variability, Glob. Planet. Change. 92–93 (2012) 236–247, https:// [43] M.G. Kendall, Rank correlation methods, fourth edition., London, Charles
doi.org/10.1016/j.gloplacha.2012.05.023. Griffin, 1948.
[15] A.R.M.T. Islam, I. Ahmed, M.S. Rahman, Trends in cooling and heating degree- [44] P.K. Sen, N. Carolina, C. Hill, Estimates of the regression coefficient based on
days overtimes in Bangladesh? An investigation of the possible causes of Kendall’s tau, J. Am. Stat. Assoc. 1459 (1968).
changes, Nat. Hazards. 101 (3) (2020) 879–909, https://doi.org/10.1007/ [45] H. Theil, A rank-invariant method of linear and polynomial regression analysis,
s11069-020-03900-5. Part I, Royal Netherlands Academy of Art and Scinces (1950), https://doi.org/
[16] M. De Rosa, V. Bianco, F. Scarpa, L.A. Tagliafico, Historical trends and current 10.1007/978-94-011-2546-8_20.
state of heating and cooling degree days in Italy, Energy Convers. Manag. 90 [46] S. Yue, P. Pilon, B. Phinney, G. Cavadias, The influence of autocorrelation on the
(2015) 323–335, https://doi.org/10.1016/j.enconman.2014.11.022. ability to detect trend in hydrological series, Hydrol. Process. 16 (9) (2002)
[17] I. Livada, A. Synnefa, S. Haddad, R. Paolini, S. Garshasbi, G. Ulpiani, F. Fiorito, K. 1807–1829, https://doi.org/10.1002/(ISSN)1099-108510.1002/hyp.
Vassilakopoulou, P. Osmond, M. Santamouris, Time series analysis of ambient v16:910.1002/hyp.1095.
air-temperature during the period 1970–2016 over Sydney, Australia, Sci. [47] G. Basha, P. Kishore, M.V. Ratnam, A. Jayaraman, A.A. Kouchak, T.B.M.J. Ouarda,
Total Environ. 648 (2019) 1627–1638, https://doi.org/10.1016/j. I. Velicogna, Historical and projected surface temperature over India during
scitotenv.2018.08.144. the 20th and 21st century, Sci. Rep. 7 (2017) 1–10, https://doi.org/10.1038/
[18] J. Spinoni, J.V. Vogt, P. Barbosa, A. Dosio, N. McCormick, A. Bigano, H.M. Füssel, s41598-017-02130-3.
Changes of heating and cooling degree-days in Europe from 1981 to 2100, Int. [48] D.R. Kothawale, K. Rupa Kumar, On the recent changes in surface temperature
J. Climatol. 38 (2018) (1981) e191–e208, https://doi.org/10.1002/joc.5362. trends over India, Geophys. Res. Lett. 32 (18) (2005) n/a–n/a, https://doi.org/
[19] Y. Petri, K. Caldeira, Impacts of global warming on residential heating and 10.1029/2005GL023528.
cooling degree-days in the United States, Nat. Publ. Gr. (2010) 1–14, https:// [49] R.C. Hale, K.P. Gallo, T.W. Owen, T.R. Loveland, Land use/land cover change
doi.org/10.1038/srep12427. effects on temperature trends at U.S. Climate Normals stations, Geophys. Res.
[20] D. Ramon, K. Allacker, F. De Troyer, H. Wouters, N.P.M. van Lipzig, Future Lett. 33 (2006) 2–5, https://doi.org/10.1029/2006GL026358.
heating and cooling degree days for Belgium under a high-end climate change [50] K.P. Gallo, J.O. Adegoke, T.W. Owen, C.D. Elvidge, Satellite-based detection of
scenario, Energy Build. 216 (2020), https://doi.org/10.1016/j. global urban heat-island temperature influence, J. Geophys. Res. Atmos. 107
enbuild.2020.109935 109935. (2002) ACL 16-1-ACL 16-6. 10.1029/2002JD002588..
[21] M. Isaac, D.P. Van Vuuren, Modeling global residential sector energy demand [51] M. Sivak, Will AC put a chill on the global energy supply?, Am. Sci. 101 (2013)
for heating and air conditioning in the context of climate change, 37 (2009) 330–333, https://doi.org/10.1511/2013.104.330.
507–521. 10.1016/j.enpol.2008.09.051..
[22] R. Warren, N. Arnell, R. Nicholls, L. Peter, J. Price, Understanding the regional
impacts of climate change, Tyndall Centre for Climate Change Research
Working Paper 90, 2006..

11