Heating and lighting: understanding overlooked energy-consumption activities in the Indian residential sector

To predict information for the non-surveyed districts, multivariate stepwise linear regression analysis was performed using demographic, housing, and meteorological data (from 49 districts) as predictors. Secondary information sources for housing and demography were Census 2011 and India National Family Health Survey (NFHS 2016), considering these are the latest and nation-wide available surveys. Precipitation and minimum temperature data were obtained from Climate Research Unit TS4.03 dataset (Harris et al 2020). A similar exercise, using multi-year census data, has also been previously done on global scales (Lam et al 2012).

A stepwise multivariate regression model was generated for each target variable by adding and removing predictors based on the significance of their contribution toward the overall ${{{\rm{R}}}_{{\rm{adj}}}}^{2}.$ Akaike information criterion (AIC) was used to prevent overfitting and ensure a balance between model complexity and predictability. Additionally, each variable included in the final model was checked for statistical and logical connection with the target. The generalizability of these regressions was assessed by the leave-one-out cross-validation (LOOCV) method, where a model is rebuilt as many times as the number of data points with the same predictors but with one data point removed. The predicted estimates for the removed data point are compared with actual values. While building the regression models, the number of predictors is restricted to avoid overfitting. In general, we have kept one predictor per ten observational data points to ensure lesser complexity with higher predictability. In this study, some variables in kerosene lighting were taken as constants due to unpredictability, whereas some were normalized based on the prediction of the dominant variable. All fuels were normalized after prediction to have a total user fraction of one.

The model was also tested using the following: residual diagnostics, such as the visual trend in residual distributions, White test and Engle ARCH test for heteroscedasticity (Engle, 1982), Durbin-Watson test for auto-correlation (Durbin and Watson, 1950), p-value of each predictor from t-statistic test, and root-mean-square-error.

When consumer information is available, a bottom-up approach is considered useful for arriving at estimates. Many earlier studies have used this approach to estimate fuel consumption (Gao et al 2019, Gately et al 2013, Habib et al 2004, Zhao et al 2011, Lam et al 2012, Pallavidino et al 2014, Pandey et al 2014, Sadavarte et al 2019). Pandey et al (2014) and Habib et al (2004) used specific energy (energy per capita) to evaluate fuel usage, while Lam et al (2012) and Sadavarte et al (2019) used specific fuel consumption (fuel consumed per household). In the present study, we use the COALESCE survey data to estimate specific fuel consumption at the district level, using the equation below (figure S3).

$$\begin{eqnarray}&&Fuel\,Consumption\,\left({F}_{ij}\right)=\displaystyle \sum _{k=1}^{n}\displaystyle \sum _{m=1}^{n}\displaystyle \sum _{l=1}^{12}{N}_{ij}\times \widehat{{f}_{ijklm}}\times B{R}_{km}\times D\times \widehat{{T}_{ij}}\times \widehat{{A}_{ij}}\end{eqnarray} \tag{ 1 }$$

This bottom-up approach equation (equation (1)) estimates fuel consumption ${F}_{ij}$ over ith district and jth population type (urban or rural), where ${N}_{ij}$ is the maximum number of available consumers (total households [HHs] for water heating, space heating, and kerosene lighting, total number of security personnel for nighttime heating), $\widehat{{f}_{ijklm}}\,$is the fraction of users performing a non-cooking residential activity using mth fuel with k^th device in l^th month, $D$ is the number of days (fixed at 30 for simplicity) per month, $\widehat{{T}_{ij}}$ denotes the duration of the activity per day that varies with district (nighttime heating for security personnel assumed to be 7 ± 1 h per day), and $\widehat{{A}_{ij}}$ indicates activities performed per day (number of members per HH for water heating, rooms per HH for space heating, and kerosene lamps per HH for kerosene lighting). Predicted variables are indicated by a cap. Because COALESCE survey does not cover urban populations, the urban user fraction for heating activities was estimated by multiplying rural user fractions with the ratio (between 0 to 1) of urban to rural biomass users for cooking, derived from the latest NFHS 2016. Similarly, for kerosene lighting, a ratio of non-electricity users was considered. Population estimates for 2017–18 were obtained from data shared by the Ministry of Health and Family Welfare, India (MHFW 2019), and the 4 km spatial grid pattern was taken from Ravishankara et al (2020).

Many assumptions made by earlier studies, owing to non-availability of non-cooking residential data, are refined or eliminated in the present study. For instance, data on secondary kerosene users, which has been considered in this study, were missing from the estimation by Lam et al (2012) or were assumed to be constant across India by Pandey et al (2014). Earlier studies used an end-use-energy approach for water heating, whereby 4.87 MJ HH⁻¹ day⁻¹ was assumed as the total required energy (Pandey et al 2014, Sadavarte et al 2019). Further, the monthly user fraction for heating was considered a constant if the temperature fell below 20 °C, else it was zero (Pandey et al 2014). However, the regression model developed in this study associates user fractions for heating with temperatures specific to the district. Pandey et al (2014) assumed the non-cooking fuel user fractions to be the same as cooking fuel user fractions given in Census 2011, but our fuel-user fractions are based on survey data and verified against demographic and housing data sources. However, it must be mentioned that this study manages nighttime heating separately due to its partly-residential and urban dominant features. In addition, the study uses a temperature threshold for nighttime heating due to the unavailability of monthly information in the COALESCE surveys.

An uncertainty estimate is included for each variable, mainly from the regression model at 5–95th confidence interval (90% confidence level), while standard deviation is obtained from a t-statistic table. This uncertainty or error propagates into the final estimate as multiple variables interact with each other, as shown in equation (1) above. Refsgaard et al (2007) support uncertainty estimation in environment models, and a similar approach is used in the present study: the square root of the sum of squared error is taken for the first-order operations (add or subtract), and a similar method is used but with the ratio of error to mean (relative uncertainty) for the second-order operations (multiply or divide).

Data from around 6000 household surveys conducted across 49 districts along with the newly developed sectoral methodology and regression models have allowed this study to go beyond previously held assumptions and provide first-time estimates of the total fuel and energy consumed across India for non-cooking residential sector activities. The survey results present new and vital information about fuels and devices used in the non-cooking residential sector.

The traditional stove dominates the devices used for water heating in India, with almost negligible (∼3%) use of the improved stove. Similarly, simple wick lamps are used by 78 [±29] % of the households, while the rest rely on lamps such as hurricane and petromax. Such high heterogeneity in kerosene lighting devices has not been reported previously (Lam et al 2012). Further, an earlier study (Pandey et al 2014) assumed a constant duration of power outages in the evening (4 h) across India for 365 days, however, the current study corrects that assumption by reporting an average power outage of 1.7 h a day for one-third of a year (table 1). Such information was not available earlier, therefore, Pandey et al (2014) only considered wick lamps to estimate kerosene consumption for lighting, which in turn could have misled the climate impact assessment of residential emissions. With regard to space heating, this is one the few studies that examine fuel consumption from regional devices such as kangari (firepot), Hamaam and bukhari (space heater), which are widely used in the Jammu and Kashmir region for warmth.

Our results offer new information about user fractions for different fuels utilized for space and water heating. Firewood, agriculture residue, and dung cake are the major biomass fuels used by 71%, 7%, and 22% of the households, respectively, for water heating. The use of charcoal for space heating is significant (12%), especially in Jammu and Kashmir. Interestingly, although earlier studies have reported the regional usage of charcoal, they have not been able to quantify the fractional use of biomass fuels (Akhtar, 1992, Bhat and Rubab 2008, 2009). With negligible use of commercial fuels like LPG for heating, the share of biomass fuel users is large, significantly higher than that for cooking. Surveys showed that 4% of households use kerosene as the primary fuel for lighting, which is significantly lower than the 31% reported in Census 2011. Our finding is supported by data from the Ministry of Statistics and Programme Implementation (2021), which has shown a sharp (4 times) decline in kerosene consumption since 2011, possibly due to massive electrification projects undertaken in India (Jairaj and Kumar, 2019, Saubhagya, 2020).

Other variables estimated from the survey, to calculate energy use for heating, include heating time, number of rooms heated per household, and members per household using heated water. The time required for heating water per person is 22 [±9] minutes, while the duration of space heating varies from 1 to 14 h a day depending on the location of the households. While earlier studies have explained the seasonality of heating activities by the temperature threshold method (T < 20 °C) (Pandey et al 2014, Sadavarte et al 2019), we found that water heating in the western and southern parts of India is a round-the-year activity. In contrast, space heating is more sensitive to temperature (Pawar and Sinha 2022). Further, to the best of our knowledge, none of the previous studies have reported seasonality in kerosene consumption for lighting, however, we observed a reasonable increase in kerosene users during summer, possibly due to the rainy season that often causes power disruptions.

Regression models were used to understand and estimate the surveyed variables over non-surveyed districts, where secondary data from Census 2011, NFHS 2016, and meteorological data were used as predictors (see table S3). The final predictive models shown in table 2 feature a logically explainable association between the target and predictors, with reasonable prediction strength (R_adj ² > 0.3). Fuel and device-related variables rely on socioeconomic, housing, and demographic data, while heating frequency and time variables are mainly temperature driven. It is important to mention that some variables—user fraction for wick lamps, primary kerosene users, number of kerosene lamps, monthly user fraction for water heating, and monthly fraction of kerosene users—are not predicted because of spatial variation or unavailability of predictor data, thus, they are considered constant (table 2).

Table 2. Prediction models details. (N is the number of observations and meaning of each predictor variable is provided in the table S3).

Target	Description	${{{\rm{R}}}_{{\rm{adj}}}}^{2}$	RMSE	N	${{{\rm{R}}}_{{\rm{Loocv}}}}^{2}$
WH_TS	1.02 + 0.0005 (DC_R_11)—0.0120 (4W_R_11)—0.0005 (DC_R_11:4W_R_11)	0.77	0.04	44	0.53
WH_FW	0.0928 + 0.0098 (FW_R_11)	0.45	0.25	44	0.43
WH_CR	0.7762–0.0078 (Elec_R_16)	0.63	0.09	44	0.58
WH_DC	−0.1644 + 0.2211 (North) + 0.0024 (NoLat_R_11) + 0.0158 (DC_R_11)	0.70	0.15	44	0.66
WH_Mem	7.9908 + 0.8611 (West) −0.0574 (HH1–3_R_11)—0.1001 (Tmin_CRU)	0.46	0.97	44	0.41
WH_Time	42.7845–7.3 (Tmin_CRU)	0.33	7.17	44	0.20
SH_FS	−0.2461 + 0.0158 (AgriLab_R_11)	0.35	0.31	20	0.28
SH_FW	0.8069 + 0.0101 (FW_R_11)	0.56	0.26	21	0.42
SH_CR	1.0288–0.0102 (Elec_16)	0.66	0.14	21	0.62
SH_DC	0.6044–0.0068 (FW_R_11)	0.40	0.22	21	0.32
SH_Char	0.8254–0.0406 (Tmin_CRU)	0.79	0.13	21	0.72
SH_Room	2.7744–0.0801(Tmin_CRU)	0.40	0.60	20	0.26
SH_Time	13.2904–0.5280 (Tmin_CRU)	0.65	2.37	21	0.57
SH_Mon	0.4155–0.0181 (Tmin_CRU)	0.47	0.17	252	0.43
KL_Sec	0.1510 + 0.5398 (East) −0.0288 (Coal_R_11)	0.52	0.19	26	0.21
KL_Time	6.2397–0.1044 (Tmin_CRU)—0.0228 (Elec_R_16)	0.36	0.88	49	0.27
WH_Mon	Averaged over 4 regions and 12 months.
KL_Mon	Averaged over 4 regions and 12 months.
KL_Prim	Averaged over 4 regions.
KL_Lamp	Averaged over India [σ ≤ μ]
KL_WL	Averaged over India [σ ≤ μ]
WH_IS, SH_OT, and KL_OT are calculated by normalizing them with the paired device.

Note: DC_R_11: % rural HHs uses dung cake for cooking, Census 2011, 4W_R_11: % rural HHs have 4 wheeler, Census 2011, FW_R_11: % rural HHs uses firewood for cooking, Census 2011, Elec_R_16: % rural HHs uses electricity for lighting, NFHS 2016, North: Categorical variable represent North, NoLat_R_11: % rural HHs do not have latrine, Census 2011, West: Categorical variable represent West, HH1–3_R_11: % rural HHs with 1 to 3 members, Census 2011, Tmin_CRU: Minimum temperature (averaged 2011–2018) (°C), AgriLab_R_11: % agriculture labour in rural India, Census 2011, East: Categorical variable represent East.

Regression models for fuel and device users show that usage is mainly driven by users' choice of cooking fuel. Higher user fractions for cooking with dung cake (DC_R_11) are associated with more users of traditional stoves and dung cake for water heating. Similarly, more firewood users for cooking (FW_R_11) at the district level indicate higher consumption of firewood for heating. Further, we also note that more members per household use warm water for bathing in low-temperature regions (Tmin_CRU). Regions with a higher fraction of agricultural labor (AgriLab_R_11) have more users of firespace for space heating, which possibly suggests that firespace (FS) use is associated with a low socioeconomic status. Interestingly, electricity use (Elec_16) is inversely linked to users of agriculture residue for heating activities. Secondary users of kerosene lighting are higher in eastern India (East), which is consistent with data from Census 2011. However, primary users have become secondary users in the same region with the introduction of the National Electrification Scheme (Kundu and Bhattacharya, 2018). High use of coal/charcoal for cooking (Coal_R_11) is a proxy for less kerosene usage, it indicates the possibility of coal availability in the district and/or proximity to thermal power plants (NTPC, 2019). Regions with low temperatures such as north and north-eastern India experience higher power outages, which is consistent with power supply reports (CEA, 2020).

Heterogeneity in fuel consumption helps understand the dominant fuel and device combinations. The quantity of firewood used for water heating, space heating, and nighttime heating (NH) for security personnel is 22773, 22343, and 3129 Gg y⁻¹, respectively (table 3). Heating activities consume 8619 and 9123 Gg y⁻¹ of agriculture residue and dung cake, respectively. This study finds that the consumption of charcoal for space heating (4232 Gg y⁻¹) is significantly more than that of other biomass fuels, which has not been reported previously as fuel fractions in space heating have not been examined (Pandey et al 2014, Sadavarte et al 2019). About 85% of the charcoal is consumed in hilly northern states, especially in Jammu and Kashmir, mainly during winter, and it affects the region's BC emissions (Bhat et al 2017). These fuel consumption estimates are 50% less than previous values (Pandey et al 2014) mainly because the temperature thresholds considered earlier resulted in overestimation of seasonal usage. We find that residential lighting uses 439 Gg y⁻¹ of kerosene, with wick lamps accounting for 50% of the consumption. This highlights the need to consider other types of kerosene lamps when assessing fuel consumption for climate impact. We also notice a four-fold decline in total kerosene consumption for lighting since 2011 (MSPI 2021, Pandey et al 2014), owing to massive electrification of regions undertaken as part of the Saubhagya Yojana (Saubhagya 2020). However, kerosene continues to be used as a secondary fuel, mainly during disruptions in electric supply in some regions, despite a sharp decline in primary users. An earlier study showed that replacing kerosene-based subsidies with solar subsidies for users, manufacturers, and stakeholders can be a more sustainable option (Laan et al 2019).

Figure 1 shows regional heterogeneity in fuel consumption. Firewood consumption is higher in the western and southern regions, which can be attributed to the relatively higher use of biomass-based fuels for water heating. Taken together, the northern and eastern regions of India use ∼80% of agricultural residue and dung cake for heating purposes: the use of dung cake (60%) is dominant in the east whereas the northern region is dominated by agricultural residue (35%). Space heating in the northern hilly regions of India accounts for nearly 90% of the charcoal consumed. Similarly, eastern India consumes 70% of the total kerosene for lighting because of higher secondary user fractions (figure 1). Pandey et al (2014) reported a higher prevalence of primary kerosene users in the eastern part of India, however, recent policy interventions under Saubhagya Yojana have enabled many users to transition from primary to secondary users (Saubhagya 2020). Additional efforts toward ensuring uninterrupted electricity supply to the eastern region will likely help eliminate kerosene lighting and its associated climate impact.

Zoom In Zoom Out Reset image size

The fuel consumed by different residential non-cooking activities is compared by converting it to energy using the calorific value given in table S1. The use of biomass for water heating is higher in the western and southern regions, while hilly northern regions, such as Ladakh, show lesser dependency on biomass, possibly because of more than 100 installed solar water heating facilities that cover a total area of 15000 m² (LREDA 2011, Lohan and Sharma 2012, Lohan et al 2012). However, hilly northern regions of India use biomass (especially charcoal in kangari) for the extended hours (>10 h) for space heating. A few states—Arunachal Pradesh, Himachal Pradesh, Jammu and Kashmir, Sikkim, and Uttarakhand—use more than 25 GJ y⁻¹ HH⁻¹ of energy, of which >90% is utilized for space heating (figure S4) and >30% comes from charcoal (figure S5). In contrast, a negligible amount of energy is consumed by the southern states for space heating (figure 2) owing to the warm tropical conditions in the region. Nighttime heating is a dominant activity in the metro cities because of the high demand for security personnel in places such as Delhi and Chandigarh. However, nighttime heating for security personnel is also equally dominant in rural parts of India, which is supported by the observations of bonfire activity during night in rural locations of Gujarat (Gautam et al 2018). Almost 5% of the biomass energy is used by security persons for heating. Eastern parts of India and western Maharashtra show significant energy consumption by kerosene lighting, mainly because of a high number of secondary users and relatively frequent power outages.

Zoom In Zoom Out Reset image size

Overall ∼1.1 [±0.2] exajoules (1 EJ = 10¹⁸ J) of energy is consumed by the non-cooking residential sector in India, with Uttar Pradesh, Maharashtra, and Madhya Pradesh accounting for ∼40% of the share (figure 3) owing to their large populations. Half of the total energy consumption occurs in the winter season, however, northern hilly regions consume significantly higher levels of energy throughout the year because of cold weather conditions. This finding is also supported by the low seasonal variation in charcoal consumption. Northern and eastern parts of India have clear seasonal variation: consumption is the least during summer and higher in winter. Southern India does not have any seasonal extremes in consumption of energy (figure 3). Similarly, kerosene consumption for lighting shows low seasonal fluctuation, but 40% is consumed during summer.

Zoom In Zoom Out Reset image size

Determining the spatio-temporal distribution of fuel consumption is vital to developing efficient region-specific mitigation plans. The spatial pattern of cooking and non-cooking residential activities is almost identical, except in hilly northern India where non-cooking residential energy consumption is high and in West Bengal where cooking is the dominant energy-consuming activity (Kumari et al unpublished). The Indo-Gangetic plain shows higher energy consumption per unit area owing to high population density (figure 4). Around 20% of the biomass and kerosene-based energy consumption is associated with non-cooking activities, which highlights the importance of such highly overlooked activities (Kumari et al unpublished). Figure 4 shows that rural and urban parts of India account for 89% and 11% of the total non-cooking energy consumption, respectively, and 94% of it is used for residential heating activities. About 2% of the total non-cooking energy is fueled by kerosene, however, the BC emission factor is 100 times higher with kerosene lighting than with biomass burning. Thus, kerosene lighting can contribute heavily to BC emissions and have severe adverse consequences for the climate (Lam et al 2012). Firewood, agriculture residue, dung cake, and charcoal make up 70%, 12%, 10%, and 6% of the non-cooking energy, respectively (figure 4).

Zoom In Zoom Out Reset image size

Future policies addressing clean energy in the residential sector, would need to account for energy consumption for heating and lighting activities. India has launched various schemes in the last decade targeting the residential activities, such as Saubhagya Yojana (Electrification), Ujjwala Yojana (LPG promotion for cooking) and Ujala LED program for lighting. However, our estimates show a significant contribution from the heating sector, which addresses previously present underestimation in the residential sector. Thus, amendment of policies to mitigate emissions from the biomass dominated heating sector is necessary. In addition, implementation of policies can be optimized through the region-specific fuel-device energy consumption details provided in the present study. Hiremath et al (2010) showed that individual districts could be self-reliant with decentralized planning, and areas with large wastelands can produce biomass feedstock for power generation that can meet electric heating needs. States such as Rajasthan, Gujarat and Karnataka, with high solar potential, can create a conducive environment for the adoption of solar powered devices, involving incentives for solar device manufacturing and use, boosting India's pace toward achieving sustainable development goals (Veeraboina and Ratnam, 2012). For example, solar powered centralized heating systems in the Ladakh region significantly lower the dependency on biomass (LREDA 2011, Lohan and Sharma 2012, Lohan et al 2012). Hybrid solar and wind energy systems (Adomavičius and Watkowski 2008) are viable in hilly northern parts of India and western India, where wind power density is high (Charles Rajesh Kumar et al 2019). Regions of high wind energy density in India (Charles et al 2019) could also benefit from wind power alone, similar to other world regions, where wind energy has been successfully harnessed (∼5 MW wind farm) to supply almost all of heating energy demand of small communities (Hughes 2010). Thus, successful implementation of programs is needed focusing on heating and lighting sectors.