Holidays in lights: Tracking cultural patterns in demand for energy services.

Successful climate change mitigation will involve not only technological innovation, but also innovation in how we understand the societal and individual behaviors that shape the demand for energy services. Traditionally, individual energy behaviors have been described as a function of utility optimization and behavioral economics, with price restructuring as the dominant policy lever. Previous research at the macro-level has identified economic activity, power generation and technology, and economic role as significant factors that shape energy use. However, most demand models lack basic contextual information on how dominant social phenomenon, the changing demographics of cities, and the sociocultural setting within which people operate, affect energy decisions and use patterns. Here we use high-quality Suomi-NPP VIIRS nighttime environmental products to: (1) observe aggregate human behavior through variations in energy service demand patterns during the Christmas and New Year's season and the Holy Month of Ramadan and (2) demonstrate that patterns in energy behaviors closely track sociocultural boundaries at the country, city, and district level. These findings indicate that energy decision making and demand is a sociocultural process as well as an economic process, often involving a combination of individual price-based incentives and societal-level factors. While nighttime satellite imagery has been used to map regional energy infrastructure distribution, tracking daily dynamic lighting demand at three major scales of urbanization is novel. This methodology can enrich research on the relative importance of drivers of energy demand and conservation behaviors at fine scales. Our initial results demonstrate the importance of seating energy demand frameworks in a social context.

Introduction

Much attention has been given to the role of technology, economics, and individual behavior in the attenuation of global energy demand [Lijesen, 2007; Stern, 2007; Wilson and Dowlatabadi, 2007; Gillingham et al., 2009; Popp et al., 2009; Stern, 2014], a key component of cost‐effective climate change mitigation scenarios [Clarke et al., 2014]. Empirical research from economics and psychology primarily focuses on energy decision making as a rational, individual process [Wilson and Dowlatabadi, 2007]. Factors such as income, demographics, life stage, and ethnicity have been identified as having a statistically significant impact on energy use [Lutzenhiser, 1993; O'Neill et al., 2010]. However, individual decision models are insufficient to explain energy use [Lutzenhiser, 2002; Shove, 2003]. Energy consumption—both total and peak demands—is also the result of energy behaviors shaped by the routines [Lutzenhiser, 1993], social norms [Schultz, 2002], habits [Steg and Vlek, 2009], and values of a culture [Shove, 2003]. Sociological theoretical frameworks are founded on this distinction [Lutzenhiser, 1993; Wilhite et al., 2000], but measurement of the embeddedness of energy service demand in social and cultural contexts has been empirically elusive [Chester et al., 2014; Romero‐Lankao et al., 2014].

During holidays aggregate energy behaviors change, which in turn affects the timing and location of demand for energy services. For lighting, these changes in demand are now apparent from night‐sky imagery. Here, we examine variations in urban lighting patterns across cultural and social boundaries for cities in the Southern United States (US), Northern Mexico, and the Middle East during two major holiday periods: (1) Christmas and New Year's and (2) the Holy Month of Ramadan.

We processed 36 months of daily night‐lights data, from early 2012 to late 2014, based on remotely sensed estimates acquired from the visible infrared imaging radiometer suite (VIIRS) day/night band onboard the Suomi National Polar Orbiting Partnership (NPP) satellite [Miller et al., 2013; Cao and Bai, 2014]. In previous studies, the magnitude of satellite‐derived nighttime lights have served as a spatial proxy for total electricity use [Elvidge et al., 1997; Townsend and Bruce, 2010], energy consumption [Letu et al., 2010], and greenhouse gas emissions [Elvidge et al., 1997; Doll et al., 2000]. Here, we take advantage of a new algorithm to estimate daily patterns of total lighting electricity usage (TLEU). The approach integrates cloud‐free, atmospheric‐, terrain‐, fire‐, lightning‐, snow‐, lunar‐ and straylight‐corrected radiances to estimate the nighttime luminosity at the time of the Suomi‐NPP satellite overpass. We used daily electric load profiles for different energy service sectors to convert the corresponding luminosity values to daily measures of total outdoor lighting electricity demand. Finally, we aggregated these estimates for all economic sectors to derive daily estimates of TLEU. The multilevel processing approach is described in Appendices B and C, and its quality and validity were established through comparisons with multi‐year time series data. This suite of science‐quality nighttime environmental products provides new information on the magnitude and signature of daily, weekly, and seasonal energy service demand, at the native VIIRS resolution (750 m), and with rigorous quality assurance and uncertainty information.

Patterns of Energy Service Demand During Christmas and New Years

At the national scale, the U.S. cyclical holiday TLEU signal (ΔCS)—obtained using a seasonal‐trend decomposition procedure based on Loess [Cleveland et al., 1990]—is shown to uniformly increase across cities (Figures 1a and 2a–c). Despite being ethnically and religiously diverse, the United States has a holiday increase in TLEU that is present across most communities—indicative that it is a civil holiday, as well as a religious one. Lighting patterns, therefore, track this national shared tradition. Although a larger portion of sampled cities in Northern Mexico (33%) experienced significant changes in TLEU (>5%) compared to the United States (22%) (Figure 3), Northern Mexico's cyclical signal (Figure 1b) was not as ubiquitous as the U.S. signal.

Figure 1

Normalized annual signature of cyclical component of total light energy usage TLEU (solid‐blue line) across sampled cities (filled black circles) in six regions: (a) Southern United States (n = 70); (b) Northern Mexico (n = 30); (c) Saudi Arabia (n = 9); (d) Turkey (n = 26); (e) Egypt (n = 22); and (f) Iraq (n = 21). The corresponding holiday period range is also shown by dashed red lines for comparison.

Figure 2

(Left‐axis) Time series of total lighting electricity usage (TLEU) for a selection of cities in the South Continental U.S. and Northern Mexico. Green and yellow circles indicate snow‐free daily TLEU retrievals using the primary (full‐inversion) and backup (magnitude‐inversion) algorithms, respectively. Red circles indicate poor‐quality spatial‐temporal gap filled values. New Years' day (1 January 2013 and 2014) is shown by solid black lines. (Right‐axis) MODIS‐derived daily snow cover fraction shown in black dotted lines. Summary statistics [E,ϵ ] are provided for each city (see Appendix C).

Figure 3

Relative changes in total light electricity use (ΔTLEU = TLEUHoliday ÷ TLEURest of Year ) during the Christmas and New Years' season for sampled cities in the Southern United States and Northern Mexico.

Variations in urban lighting patterns during the Christmas and New Year holidays also differ across residential contexts. We illustrate this effect in Figures 4 and 5, where holiday activity is shown to peak in the suburbs and peri‐urban areas of major Southern US cities (ΔTLEU > 1.0), where Christmas lights are prevalent. In contrast, most central urban districts, with compact dwelling types affording less space for light displays, experience a slight decrease or no change in energy service demand (ΔTLEU ≤ 1.0).

Figure 4

Regional subset of relative change in TLEU shown for Christmas and New Years in the South Eastern United States.

Figure 5

Regional subset of relative change in TLEU shown for Christmas and New Years in the South Western United States.

Patterns of Energy Service Demand During Ramadan

In contrast to Western Holidays, the Holy Month of Ramadan does not involve a shift in setting or location as much as a shift of daily activities in time. This makes Ramadan in the Middle East an ideal setting for the observation of energy behaviors occurring at the same time of Suomi‐NPP's satellite overpass (01:30 A.M., local time). Electricity usage for lighting is therefore expected to change as prayer and eating during the Holiday shifts later into the night. We show that variations in energy use between cities during Ramadan indicate different religions as well as cultural observance practices. Cities in majority‐Muslim countries have different TLEU signatures than cities in Middle‐Eastern Jewish countries, typified by lighting peaks during and slightly after the 30 days of Ramadan (Figures 6, 7, 8). The cyclical Ramadan signal was also found to vary significantly across Muslim countries. For example, Saudi Arabia's cities consistently peaked during both the Ramadan period and the Hajj (Figure 1c); but light signals in Turkey varied between cities (Figures 1d and 9).

Figure 6

Regional subset of relative change in TLEU shown for the Holy Month of Ramadan in the Middle East.

Figure 7

Stacked histogram of relative changes in total light electricity use (ΔTLEU = TLEUHoliday ÷ TLEURest of Year) for sampled cities in : (a) the United States and Mexico during Christmas and New Years; and (b) Middle Eastern countries during Ramadan.

Figure 8

Relative changes in total light electricity use (ΔTLEU = TLEUHoliday ÷ TLEURest of Year) during Ramadan for sampled cities in the Middle East.

Figure 9

Time series of total lighting electricity usage (TLEU) for a selection of cities in the Middle East. Green and yellow circles indicate snow‐free daily TLEU retrievals using the primary (full‐inversion) and backup (magnitude‐inversion) algorithms, respectively. Red circles indicate poor‐quality spatial‐temporal gap filled values. New Years' day (1 January 2013 and 2014) is shown by solid black lines. (Right‐axis) MODIS‐derived daily snow cover fraction shown in black dotted lines. Summary statistics [E,ϵ] are provided for each city (see Appendix C).

In contrast to daytime satellite measurements and derived products, the VIIRS TLEU patterns not only reveal information about infrastructure use and availability, but also its location. For instance, muslim middle‐eastern cities with no consistent Ramadan signal (ΔTLEU < 1.0) had complicating factors that interrupted energy service demand signals. Negative cyclical trends during Ramadan were prevalent in middle‐eastern cities with highly vulnerable electrical power infrastructure. For example, cities across Lebanon and Iraq have uniform cyclical summer decreases in TLEU (average ΔCS = −15%) that are likely related to increases in demand for air‐conditioning and appliance loads (Figure 1f) [Kharrufa, 2008; Dagher and Ruble, 2011]. In Syria, Iraq, and Palestinian cities, regional conflict also played a major role in changes in energy service supply and demand (Figures 10a, 11, and 12).

Figure 10

Time series of total lighting electricity usage (TLEU) for a selection of cities in the Israel and Palestinian Territories. Green and yellow circles indicate snow‐free daily TLEU retrievals using the primary (full‐inversion) and backup (magnitude‐inversion) algorithms, respectively. Red circles indicate poor‐quality spatial‐temporal gap filled values. New Years' day (1 January 2013 and 2014) is shown by solid black lines. (Right‐axis) MODIS‐derived daily snow cover fraction shown in black dotted lines. Summary statistics [E,ϵ] are provided for each city (see Appendix C).

Figure 11

Time series of total lighting electricity usage (TLEU) for a selection of cities in the Syria. Green and yellow circles indicate snow‐free daily TLEU retrievals using the primary (full‐inversion) and backup (magnitude‐inversion) algorithms, respectively. Red circles indicate poor‐quality spatial‐temporal gap filled values. New Years' day (1 January 2013 and 2014) is shown by solid black lines. (Right‐axis) MODIS‐derived daily snow cover fraction shown in black dotted lines. Summary statistics [E,ϵ] are provided for each city (see Appendix C).

Figure 12

Time series of total lighting electricity usage (TLEU) for a selection of cities in the Iraq. Green and yellow circles indicate snow‐free daily TLEU retrievals using the primary (full‐inversion) and backup (magnitude‐inversion) algorithms, respectively. Red circles indicate poor‐quality spatial‐temporal gap filled values. New Years' day (1 January 2013 and 2014) is shown by solid black lines. (Right‐axis) MODIS‐derived daily snow cover fraction shown in black dotted lines. Summary statistics [E,ϵ] are provided for each city (see Appendix C).

Striking downturns in TLEU occurred in 2012 in the Syrian cities of Damascus (−50%) and Aleppo (−94%), corresponding to the onset of major military confrontations (7/15 and 7/23, respectively) (Figure 11). These declining patterns are consistent with regional assessments performed across Syrian cities using monthly nighttime lights composites acquired from the Defense Meteorological Satellite Program's Operational Linescan System (DMSP‐OLS) [Li and Li, 2014]. However, the VIIRS results indicate a more pronounced decline in energy service demand (−20%) compared to the DMSP‐OLS estimates for the same time period (2012–2014). This shows qualitative similarity in response to the energy declines across Syria, with quantitative differences indicating higher sensitivity of the VIIRS TLEU data to sub‐pixel nighttime lights sources. Iraq cities captured by the Islamic State in Iraq and Syria (ISIS) in 2014 (e.g., Mosul, Tikrit, Tal Afor, Ramadi) also showed marked average decreases in energy service provision (−84% since 1 April 2014) compared to their unoccupied counterparts, for example, Baghdad and Sulaimaniya (+6%) (Figure 12).

Patterns in Energy Behaviors Across Different Sectors in Egypt

In Egypt, communities with similar political affiliations were found to have similar TLEU patterns, demonstrating that energy patterns cluster based on sociopolitical groupings. We examined the correlation between variations in TLEU and the vote share for three major political blocs that participated in the Egyptian parliamentary elections that were held in 2011–2012. These included: (1) the Al‐Nour Party, an ultra‐conservative Islamist party maintaining a strict version of Islam, known as the Salafi methodology; (2) the religiously moderate Freedom and Justice Party (FJP) (better known as the Muslim Brotherhood); and (3) the main liberal parties (Egyptian Bloc and Al‐Wafd). To obtain a better understanding of the impact of the sociodemographic profile of Egypt's electoral constituencies, we used a classification scheme created by Erle et al. [2011], which partitions Egypt's proportional system districts [SJCE, 2011] into clusters based on voting outcomes, sanitation, and illiteracy (Table 1). We then performed a multiple linear regression using the voting and socioeconomic variables to predict average change in TLEU over 21 electoral districts in Egypt during three Ramadan periods (2012–2014, Figure 13). Results in Table 2 show that social clusters with shared voting preferences and socioeconomic standing in Egypt also shared similar lighting patterns during Ramadan (R 2 = 0.5038, P < 2.83 × 10−5). The Ramadan time series of TLEU profiles reveal how different socioeconomic groups use energy differently. For instance, districts classified as “Rich” (Figure 13a) observed an early and sustained Ramadan peak signal, while “Poor” and “Very Poor” voting districts (Figures 13d and 13e) did not show a peak signal until the end of Ramadan, when the “Eid ul‐Fitr” (or Feast of Breaking the Fast) takes place. Patterns of energy service demand are therefore capturing a diverse range of sociopolitical influences; whereas physically driven patterns of infrastructure existence (e.g., urban land cover and impervious area distribution maps) are naturally binary.

Table 1

Egypt Proportional System Districts Organized by Clusters of Electoral Constituencies [Erle et al., 2011], and Percentage of Parliamentary Election (2011–2012) Votes Received by Major Party Blocs Out of Votes Cast in Each District [SJCE, 2011]

Social Cluster	Election District	Al‐Nour (%)	FJP (%)	Egyptian Block and Al‐Wafd (%)	% Illiteracy	% Non‐Public Sanitation
Rich	Alexandria (1)	29.85	34.41	22.42	14	1
	Cairo (1)	15.39	39.45	29.73	22	5
	Cairo (2)	11.31	35.65	31.98	19	1
	Cairo (4)	19.35	40.48	20.46	16	1
Middle One	Daqahliyya (1)	26.91	30.43	14.07	27	5
	Gharbiya (1)	29.73	28.67	21.19	27	40
	Gharbiya (2)	28.90	37.59	22.02	27	30
	Ismailia	27.63	38.19	14.05	23	50
	Port Said	20.69	32.66	23.54	16	5
	Sharqiya (1)	26.55	37.99	17.21	27	25
Middle Two	Alexandria (2)	32.96	35.32	11.70	25	19
	Cairo (3)	14.52	39.94	33.00	30	5
	Damietta	38.58	31.14	7.91	22	12
	Giza (1)	29.10	41.09	18.32	27	45
	Giza (2)	28.12	37.83	14.65	24	10
Poor	Beheira (1)	40.01	35.69	13.16	39	40
	Sharqiya (2)	20.80	36.54	17.76	36	53
Very Poor	Assiut (1)	20.84	39.77	24.04	38	76
	Minya (2)	28.09	35.77	26.27	44	85
	Qena (2)	19.16	26.03	12.62	40	87
	Sohag (2)	8.19	22.70	10.33	44	86

The values in Bold indicate the majority party that won each particular districts.

Percentage of parliamentary election (2011–2012) votes received by major party blocs out of votes cast in each district [SJCE, 2011]. These included: (1) the religiously conservative Al‐Nour Party, where Salafis represented its core constituency; (2) the religiously moderate Freedom and Justice Party (FJP), supported by the Muslim Brotherhood; and (3) the main liberal parties (Egyptian Bloc & Al‐Wafd). Poor districts were identified by a high share of illiteracy and a high share of non‐access to public sanitation. Conversely, the rich cluster of constituencies was characterized by a low share of illiteracy and a high share of access to public sanitation. The overall voter turnout was 54%.

Figure 13

Normalized estimates of TLEU for 21 electoral districts in Egypt for the Ramadan periods of 2012–2014. Districts were divided into five sociopolitical clusters and the most populous six districts in each cluster are shown: (a) “Rich”; (b) “Middle One”; (c) “Middle Two”; (d) “Poor”; and (e) “Very Poor”. Population‐weighted averages for each cluster are also shown (solid black lines). The bottom right figure (f) shows the corresponding quality assurance results for all electoral districts. The latter is based on values associated with retrieval performance—i.e., “Full Inversion”, “Magnitude Inversion”, and “Gap Filled” (see Appendix C).

Table 2

Summary Statistics of Multiple Linear Regression Model for Predicting TLEU During Ramadan Over 3 Years (2012–2014) From Voting and Socioeconomic Variables Available in 21 Electoral Districts in Egypt

Multiple Regression Statistics
R Square	0.5605
Adjusted R Square	0.5038
Standard Error	3.66
F‐statistic	9.884
P value	2.83E‐05

P values for each variable are shown in the last bottom column. Non‐access to public sanitation and conservative‐leaning voting preferences Al‐Nour had negative coefficients, while illiteracy rates and moderate‐leaning voting preferences FJP had positive coefficients.

Conclusions

The latest Intergovernmental Panel on Climate Change Working Group III report states that reducing energy demand is the key to minimize climate mitigation costs [Seto and Dhakal, 2014]. However, recent reviews emphasize a research “blindspot” in the human dimensions of global environmental change, including urban development pathways and public policies as drivers of and responses to climate change [Chester et al., 2014; Romero‐Lankao et al., 2014]. Accurate energy demand forecasting requires attention to the broader social contexts within which people operate, especially as these contexts change with globalization, economic development, and urbanization. We illustrate how the uniformities and variations between nighttime light signatures, as observed by the VIIRS instrument on Suomi NPP, can provide new insights into how energy behaviors, motivated by social incentives, vary across cultural boundaries. Our results also suggest that energy is providing services that enable social and cultural activities, and thus energy decision‐making patterns reflect social and cultural identities.

We demonstrate that individual factors such as poverty and irregular access to infrastructure services also affect energy behavior patterns, alongside macro‐cultural forces. The daily dynamic VIIRS TLEU retrievals are thus capturing the parts of a community's energy portfolio that are not only associated with individual behavior and decision‐making, but also with the energy required to live in a certain cultural context and to participate in the social and economic activities within that context. We conclude that research on energy conservation and energy decision‐making needs to be coupled with research about lifestyles, cultural norms, and demographics to better account for the broad societal forces impacting energy demand across all scales of urbanization.

Beyond the results of this study, the methodology developed in this paper has broader impacts on traditional energy demand research. Effective energy policies are currently limited by a lack of data that is both global in extent and local in resolution for understanding how the drivers of energy use and the available policy levers for reducing energy consumption may vary. A globally systemic assessment of these activity patterns will lead to improved understanding of neighborhood‐level energy intelligence, as well as conservation strategies that can be tailored to specific communities and sociocultural contexts.