Main

It is generally accepted that solid fuels, including coal and biomass fuels, are extensively used in rural China, and that they emit large quantities of air pollutants and climate-forcing components1,2,3,4. It has been estimated that 32% of the premature mortality induced by ambient air pollution in China in 2010 could be attributed to residential emissions5. As the quantities and types of rural residential energy consumption in China were not well recorded, any evaluation of its environmental impact based on existing data is associated with high uncertainty.

Moreover, rapid energy transitions in the rural residential sector led to an even greater uncertainty. The results of several local surveys suggested fast changes in the energy mix in rural China because of the rapid socioeconomic transition6,7,8. In contrast, International Energy Agency (IEA) and Food and Agriculture Organization (FAO) statistics suggested that the rural biomass fuel consumption had decreased by only 0.5% annually from 1992 to 20129. Such a slight decline is hardly realistic because the use of electricity and liquefied petroleum gas (LPG) increased by 97 and 7 times, respectively10, whereas the rural population decreased by 32% during the same period11. Despite this opacity, the IEA statistics are used extensively12,13.

To address this issue, we conducted a nationwide rural residential energy-mix survey (the first phase) and a household solid-fuel-weighing campaign (the second phase) in 2012 to compile a bottom-up Chinese rural residential energy consumption database for cooking and heating. In addition to gathering current data for 2012, recalled data for 5 (2007), 10 (2002) and 20 (1992) years previously were also collected to address the energy transition over the two decades. The energy consumption data for the early years are associated with a relatively high uncertainty due to the recalled historical data. The influence of the transition on the emissions of major air pollutants was quantified.

Solid fuel use in rural China

The data collected in the first phase are shares of time during which various energy types were used, including coal (excluding honeycomb briquettes), honeycomb briquettes (compressed blocks of coal powder), straw (crop residues excluding corncobs), corncobs (the dried corncobs with the grain removed), fuelwood, brushwood, charcoal, LPG, biogas, and electricity for kettles, rice cookers, induction cookers, microwave ovens, electric heaters and air conditioners. Among these, LPG, biogas and electricity are categorized as ‘clean fuels and electricity’ (CF&E) and have much lower health impacts than those of solid fuels. Information on the detailed energy consumption activities, which include staple food cooking, subsidiary food preparation, water boiling and heating, were recorded separately. Based on these data, the relative contributions of various types of energy used for cooking and heating in rural China were derived as their shares of time used. The overall time-sharing data for heating were taken directly from the survey, whereas the time-sharing data for cooking were calculated as weighted averages of the data on individual activities, including staple food cooking, subsidiary food preparation and water boiling, and the weighting factors were assumed to be 15%, 80% and 5%, respectively. The uncertainty of the assumed percentages was evaluated by a sensitivity analysis, and the robustness of the approach was well demonstrated (Methods gives the details).

Figure 1 shows the relative contributions of various energy types in 2012. Cooking energy was dominated by electricity, LPG and biomass fuels, whereas coal and biomass fuels were the major sources of heating energy. This pattern was very different to those of 5, 10 and 20 years previously, when solid fuels contributed much higher shares (see the next section). In most previous surveys, the main energy usage was investigated (Supplementary Fig. 1)6,14,15. As more than one type of energy was often used and some energy types, such as coal, are more often used than others as the main energy, there is a tendency to overestimate the main energy. For example, 33% of households cooked staple food using biomass fuels for more than six months, and 84% of them also used at least one type of supplementary energy, which would be underestimated if only main energy use was recorded.

Fig. 1: Percentage share of fuel type in residential energy use in rural China in 2012.
figure 1

a, Overall time-sharing data for cooking (staple food cooking, subsidiary food preparation and water boiling). b, Time-sharing data for heating.

Full size image

The energy mix used by rural residents varied widely. This was particularly true during the fast socioeconomic transition period, when both energy accessibility and affordability changed rapidly. As a result, most households used more than one type of energy. As shown in Supplementary Figure 2, the households using a single type of cooking energy decreased from 43% in 1992 to 17% in 2012, whereas those that used three to five energy types increased. In comparison, the energy diversity was relatively low for heating. There was a slight increase in the number of households using a single energy type and a corresponding decrease in the number of households using two energy types for heating. The stacked energy at the household level would be more obvious when both cooking and heating are taken into consideration.

Change in fuel use for cooking and heating

Temporal trends in rural residential cooking and heating energy uses are shown in Fig. 2a,b as the percentages of time used. From 1992 to 2012, the use of electricity and gases (LPG and biogas) for cooking increased from 3.5% and 5.1% to 34% and 24%, respectively, whereas the use of wood (fuelwood and brushwood) and crop residues (straw and corncobs) declined from 47% and 33% to 20% and 14%, respectively. The transition in the heating energy mix was much slower, as almost no rural households could afford to heat their homes using gas16. Though electric heating increased 7.6 times, its relative contribution to the total energy use was merely 18% in 2012. A plot of the fraction of electric heating against mean heating days (HD, the number of days heating is required) at the provincial level (Supplementary Fig. 3) suggests that increases in electric heating occurred largely in the mid-latitude provinces, in which heating is only needed for a short period, and permanent heating facilities are not needed. In northern provinces with longer and colder winters, electric heating would be too expensive. However, the situation is changing in Beijing and several provinces in the surrounding region. Electric and natural gas heaters are being promoted to replace coal stoves for heating in many villages. The campaign was pushed hard by government, with heavy subsidies for both facilities and operational costs17. It will be very interesting to evaluate the potential influence of this campaign on the environment and health in the future.

Fig. 2: Changes in the relative time usages of residential energy from 1992 to 2012 in rural China.
figure 2

a,b, The results are overall time-sharing data for cooking (staple food cooking, subsidiary food preparation and water boiling) (a) and heating (b).

Full size image

Generally, poor families tend to use more free biomass fuels and move towards coal and then CF&E as their living conditions improve18. The transition trajectories of the rural residential energy mix for cooking and heating are shown in Fig. 3a,b for all provinces. The heating in most provinces followed the general trajectory from biomass fuels to coal and then turned towards electricity. Cooking energy in most provinces skipped the coal stage and jumped to CF&E, probably due to the rapid economic development. Similar patterns can also be seen for 346 municipalities (Supplementary Fig. 4). The direct replacement of cooking biomass fuels by CF&E in rural Chinese households was reported previously8. For heating, the relatively high cost and low affordability, as well as a likely path dependency for solid fuels, limit the transition towards CF&E. On the other hand, evident transitions from biomass fuels to coal for cooking and potential improvements in living conditions in rural China suggest a possible latent demand for clean heating in the future. Supplementary Fig. 5 shows the geospatial distribution patterns of the per-person rural residential consumption of major energy sources in 2012. Data with a high spatial resolution (0.1°) were generated by disaggregation using rural population density as a proxy. The North China Plain and northwest provinces were characterized by relatively high fractions of coal and biomass fuels. Biomass fuels were also widely used in the northeast and southwest regions. In the meantime, CF&E dominated the southeast region.

Fig. 3: The transition trajectories of the rural residential energy mix from 1992 to 2012 in China.
figure 3

The three endmembers of the triangles are biomass fuels (top), coal (bottom right) and CF&E (bottom left). The red lines are provincial means, and the thick blue lines are national averages. a, Cooking. b, Heating.

Full size image

Role of socioeconomic development in fuel use change

To quantify the driving forces that affect the residential energy transition, various socioeconomic factors associated with affordability and accessibility19,20,21 were evaluated using regression analysis. The independent variables tested included per-person income (Icap) and gross domestic product (GDPcap), both with purchasing power parity at the 2005 constant price, the education index, HD, HDD (heating degree days), electricity price, per-person production of coal, grain and corn, forest coverage and road density. As the data on most parameters are only available at the provincial level and the income data for individual households collected in this study may not be accurate, for privacy reasons, the regression was conducted at the provincial level. Therefore, the results only show regional trends in general and cannot be applied to predict the energy mixes of individual households. To assess the transition, fractions of CF&E for cooking (Fc) and heating (Fh) were selected as dependent variables. The models were chosen using a trial-and-error process, and both linear and non-linear functions of various forms were evaluated. A stepwise multivariate regression was performed using the least-squares criteria. It was found that Icap and several other parameters are significant. In the models, Icap, which is correlated with many other similar parameters such as GDPcap, is considered to be a quantitative indicator of socioeconomic status. Figure 4 shows the relationships between the predicted and observed Fc (Fig. 4a) and Fh (Fig. 4b) for all provinces in the four years. For cooking, the following non-linear multivariate function was derived where \({R}_{{\rm{adj}}}^{2}\) is adjusted coefficient of determination of the regression:

$$\begin{array}{lll}{\rm{log}}{F}_{{\rm{c}}} & = & 67.6\left(1-{\rm{\exp }}\left(-1.36\,{\rm{log}}{I}_{{\rm{cap}}}\right)\right)-4.46\times 1{0}^{-5}{\rm{HDD}}-67.5,\,\\ {R}_{{\rm{adj}}}^{2} & = & 0.886\end{array}$$
Fig. 4: Comparison between the observed and model-predicted fractions of CF&E.
figure 4

The results are presented for all provinces in 1992, 2002, 2007 and 2012 in China. 1:1 lines are shown. a, Cooking. b, Heating.

Full size image

The detailed regression results, including the standard errors, the 95% confident intervals and P values for the individual regression coefficients and the loss function are listed in Supplementary Table 1. Icap alone can explain 87.8% of the total variation. In fact, without the HDD term, the results can still be well fitted by a single exponential function. It is expected that, as the living conditions of rural residents continue to improve, this spontaneous transition will continue. HDD is also significant (P < 0.05) in the equation, and its contribution can be explained by the inter-relationship between the cooking and heating activities. Those who use a given energy to heat tend to use the same energy to cook for convenience. Other factors, such as the availability of fuel resources (for example, biomass and coal) could have collinearity with income, both explicitly and implicitly. For cooking, the data from the four surveyed years can be modelled well by a single equation, which suggests that Icap affects both the provincial and annual variations in a similar way. Such a similarity between the factors that affects both spatial and temporal variations of energy use has been demonstrated previously22,23. A jackknife test was conducted to evaluate the robustness of the equation. For 30 repeated regression calculations with a randomly selected case deleted each time, the calculated \({R}_{{\rm{adj}}}^{2}\) values vary between 0.883 and 0.891 with a mean, standard deviation and coefficient of variation of 0.886, 0.0021 and 0.24% (n = 124), respectively, which are indicative of a robust relationship.

For heating, a significantly (P < 0.05) positive influence of Icap was also identified, though with a relatively small contribution (36.9%) to the overall variation. Two other significant (P < 0.05) factors are HDD and coal production (Pcoal) and the detailed regression results are also shown in Supplementary Table 1. The following non-linear multivariate function was derived for heating:

$$\begin{array}{lll}{\rm{log}}{F}_{{\rm{h}}} & = & 0.921\,{\rm{log}}{I}_{{\rm{cap}}}-2.17\times {10}^{-4}{\rm{HDD}}-4.49\times{10}^{-3}P_{\rm{coal}}-4.38,\,\\ {R}_{{\rm{adj}}}^{2} & = & 0.457\end{array}$$

Unlike cooking, a non-linear relationship was not identified for heating. Again, the most important factor is Icap. The negative slope of HDD agrees with the fact that the transition towards electric heating is difficult for provinces with long-term heating needs (Supplementary Fig. 3). Residents in the provinces with large coal production tend to use more coal for heating, which leads to a negative slope of Pcoal. Though the price of electricity could be another important aspect of affordability, a dependence on electricity price in rural China was not found because it is subsidized and almost fixed24,25. Compared with cooking, the \({{\rm{R}}}_{adj}^{2}\) value (0.457) for heating was relatively small, which suggests that other unidentified important factors affect the heating energy transition. Mean, standard deviation and coefficients of variation of \({R}_{{\rm{adj}}}^{2}\) from the jackknife test were 0.457, 0.0155 and 0.34% (n = 124), respectively.

Though the regression models were successfully developed, our statistical method has the following limitations: upscaling data to the provincial scale might cause averaging in some local trends, connecting statistical yearbook data with data from the survey could cause some inconsistencies and the collinearities between income and other factors could mask the influences of other potential factors.

Comparison with data from the literature

Mass quantities of various fuels and electricity consumed for cooking and heating in individual municipalities were calculated based on the energy-mix and daily consumption-quantity data collected in this study. The national total consumptions of various energy types were weighted averages of municipality data using the total number of households of individual municipalities as weighting factors. The results are provided in Supplementary Table 2 as fuel-use quantities and their energy values, and are compared with those from the IEA9 and FAO26 in Fig. 5. More energy was used for cooking than for heating for all types because heating is needed only in northern China during winter. The quantities of commercially supplied coal and LPG were generally similar between our results and those from the IEA. The only exceptions are the LPG consumptions in 1992 and 2002, for which our survey could be overestimated due to uncertainty in the recalled data. The slight difference in coal use was probably due to an extra supply from the local coal mines. The use of honeycomb coal was previously underestimated because a large portion of this material is homemade21. Very large differences were found for biomass fuels, and the previously reported decreasing trends of biomass fuel use were too slow to be realistic. The much faster decrease found in this study agrees with the rapid decrease in the rural population10 and the rapid increase in CF&E use. The finding was also supported by the results of several recent surveys. For example, it was found that the annual mean reduction rate of residential biomass fuel use was 5.2% for 409 rural households in four provinces from 2008 to 20127, which is very close to our estimate of 5.6% from 2007 to 2012. A nationwide longitudinal survey that covered more than 12,000 households in 25 provinces also revealed a rapid decline in biomass fuel use in the rural residential sector6. The striking differences in biogas use can be explained by the fact that many biogas facilities are no longer active because of poor maintenance and the rapid decrease in the number of households that raise hogs at home27. The energy-rebound effect in the residential sector in China has been demonstrated previously28,29. It is likely that rebound may occur during the transition, especially in the replacement of dirty fuels with clean ones. Such a rebound effect cannot be quantified in this study due to a data gap and should be addressed in future studies.

Fig. 5: Changes in residential energy consumption from 1992 to 2012 in rural China.
figure 5

The energy consumptions for cooking and heating are presented as stacked bars. The results (bars) are compared with those from the IEA9 and FAO (blue dots)26. a, Solid fuels. b, CF&E.

Full size image

Fuel use and pollutant emissions in rural China

Annual emissions of CO2, CO, SO2, nitrogen oxides (NOx), PM2.5, PM10 (particulate matter with aerodynamic diameters of less than 2.5 or 10 μm, respectively), black carbon (BC), organic carbon (OC) and BaP (benzo(a)pyrene) from rural residential and the other sources in 1992, 2002, 2007 and 2012 were derived based on the data from this study and total emission data from Peking University (PKU) emission inventories (Methods). Supplementary Fig. 6 and Supplementary Table 3 show the absolute and relative contributions of this sector to the total anthropogenic emissions of these pollutants. The emissions of incomplete combustion products from this sector were reduced by 49–57% due to the energy transition, whereas the emission of SO2 was reduced by only 22%. Such decreases were partially caused by the increase in emissions from other sectors3,30. The contributions of this sector to the total BaP emissions increased from 2002 to 2012, primarily because all beehive coke ovens were phased out31, which resulted in a significant reduction in the total emissions. Though the annual emissions of most pollutants had reduced significantly over the 20-year period, the relative contributions of the rural residential sector to total emissions of incomplete combustion products, especially BC, OC and BaP, remained high in 2012. Contributions of NOx and SO2 to the total emissions were relatively low because they were predominantly from power stations, industry and transportation. Though a similar trend has been reported before, both globally and for East Asia32, changes in China have not previously been quantified based on field-collected energy data at this scale. The collected data in this study provide important information. For example, the total emissions of primarily PM2.5 from the rural residential sector was 3.39 Tg in 2012, whereas the previous estimate was 5.10 Tg (ref. 3), which indicates an approximately 50% overestimation without the updated energy data. Despite the rapid decrease, the rural residential sector still contributed significantly to the emissions of many air pollutants.

Data quality and limitation

Based on the data collected in this study, the sampling procedure was evaluated by calculating the relative error (RE), which is defined as the ratio of the 95% confidence interval divided by the corresponding mean. This indicator is used to describe the uncertainty of the means at various scales, given that the samples collected are representative. For the average fuel-consumption quantities, the calculated REs for various fuels are plotted against sample size in Supplementary Fig. 7a, which shows non-linear (power function) decreases in the REs as the sample size increased. The REs increased rapidly when the sample size decreased below 100. For the nationwide survey, because the sample sizes were relatively large (over 30,000 in total), the RE values for the national averages of all fuels were below 5% (Supplementary Fig. 7b). However, for individual municipalities, the sample sizes were relatively small (97 on average), and the REs were relatively high (Supplementary Fig, 7c). For example, for 70% or 44% of the municipalities in which the sample sizes were lower than 100 or lower than 50, the REs were between 12–14% and 15–20%, respectively, which suggests relatively high uncertainties in terms of the municipality means for individual fuels. A similar evaluation was conducted for the fuel weighing data. The REs of the national means of daily consumption of individual fuels were calculated based on the statistics from this study. Supplementary Fig. 7d shows the dependence of the REs on sample size. The REs that correspond to actual sample sizes are marked as solid dots, which are generally below 6%, except for animal dung (14.5%), which is not commonly used except in some remote areas of Tibet and Xinjiang. If more data on dung can be collected in the future, the overall accuracy of the estimation can be further improved. Based on the results of the analysis, it is generally believed that the uncertainty in the sampling scheme is unlikely to affect the major insights and conclusions of this study.

There are limitations and potential biases in the sampling scheme, identified as follows. First, as the interviewers were indirectly trained by the trained professors, they could bring different interpretations to the questions. Second, there are often different terms in local dialects for energy-use activities and fuel or stove types. For example, meal preparation (‘Zuofan’ in Chinese) can mean cooking in general or rice cooking specifically. Coalballs (‘Meiqiu’ in Chinese) refer to ball-shaped briquettes in southern China, but honeycomb briquettes in Shanxi. Similarly, coal gas (‘Meiqi’ in Chinese) can mean pipelined natural gas or LPG in different regions. Third, though the student interviewers were randomly selected, those who entered university were not totally random. For example, very poor villages often have relatively low admission rates due to poor education conditions. Another important limitation is that the survey was conducted in 2012, and the data for 1992, 2002 and 2007 were recalled. Recalled historical data are often associated with relatively high uncertainties. Fortunately, the results could be validated by comparisons between our results and government statistics for coal and LPG, which are commercially available, and the statistics are relatively reliable. In addition, the weighed solid fuel sample size is rather small, given the high diversity in cooking practices in rural China. Finally, the income data may not be reliable due to privacy concerns. For the same reason, some detailed household address data are missing on some questionnaires. As a result, the data analysis was conducted at the provincial level.

Discussion

Based on the first-hand nationwide survey data from both the household energy-mix survey and household fuel-weighing campaign, it was demonstrated that there is a rapid transition in direct residential energy consumption in rural China. This is particularly true for cooking energy. Contributions of electricity, LPG and biogas to the total energy use increased from 8.6% to 58.8% for cooking and from 2.0% to 15.1% for heating, respectively, which shows that the transition rate is much higher than those reported previously9,26. Rapid socioeconomic development was the primary driving force for such a transition. Accordingly, emissions of incomplete combustion products, which include primary PM2.5, OC, BC and benzo[a]pyrene, from this sector reduced significantly.

Despite the rapid transition, solid fuel use, which still accounted for almost 41% of cooking energy and 85% of heating energy in rural China in 2012, remains a tough challenge for the future. This is particularly true for cold areas, which include the North China Plain, in which heating is required for months every year. A continuous and spontaneous transition towards CF&E was promising for cooking in the past. It is predicted, based on the regression model, that the CF&E for cooking will be 90% when the mean Icap in rural China reaches RMB15,000. If Icap in rural China can increase at an annual rate of 5% as a conservative estimation (9.5% for the past five years)33, this goal can be achieved by 2024. Phasing out solid fuels for heating in northern China will not occur without effective measures. The Beijing municipal government has pledged to bear the high cost of replacing all coal-heating stoves with electricity34, and similar plans have been recently launched in Shandong, Hebei, Shanxi and Tianjin in the region17. However, the sustainability of these programmes needs to be evaluated on a cost–benefit basis, taking into consideration the high cost. Another possible solution is to promote pressed biomass fuels in combination with clean stoves35, which can substantially reduce emissions at a relatively low cost. Relatively clean coal stoves could also be a transitional alternative36. A number of issues, which include cost, the collection and distribution system, the willingness and habits of residents, and government subsidies, need to be addressed.

A full substitution of solid fuels with CF&E would offer significant benefits to the environment, health and climate. In addition to economic drives, government policy can also play a key role. The current campaign for coal-stove replacement is a typical example. The process can be accelerated by incorporating CF&E promotion into the new round of the poverty-alleviation campaign37,38, the new-village development programme39 and the project for clean-stove promotion40. Education and risk communication are critical because the adverse effects of rural residential solid fuel consumption are overlooked36, and the majority of rural residents and policymakers are unaware of the health risks associated with burning solid fuels41. The adverse impacts of residential solid fuels are not limited to China1,2,3,4, and the experience gained in this study is not only valuable to China but can also be used by many countries in South and Southeast Asia, Africa and South America.

It is demonstrated that such systematic surveys are valuable to gather critical information to support the decision-making process in energy and environmental sectors, and thus they need to be conducted on a regular basis in China. In reality, though air pollutants from the residential sector are important, they are not included in previous national census of pollutant sources. Partially based on the information derived from this study, residential emissions of major air pollutants have been added to the Second National Census of Pollutant Sources, which will be conducted in 2018 in China42. We have helped to develop the survey plan and questionnaire and will take the responsibility to lead the survey.

Methods

Energy-mix survey

A stratified random-sampling scheme with four levels (provinces, municipalities, villages and households) that covered 350 municipalities in all 31 provinces in China was adopted to attain a good geographical representation (Supplementary Fig. 8). A modified probability-proportional-to-estimated-size procedure was used. The minimum sampling rate for the provinces was 1/20,000 and power function weighting factors were applied to ensure sufficiently large sample sizes for the small provinces. Given the orders of magnitude difference in the number of rural households among provinces, the sample sizes at the provincial level are designed to be negatively correlated (power function) with the number of households, so that the provinces with small sample sizes would not be underrepresented. The power-function correlation coefficient between the sample sizes and household numbers for the finally collected samples was 0.82. At the municipality level, we covered 346 out of 350 municipalities in China (supplementary Fig. 8). Only four remote municipalities (Kelamayi, Ali, Baotin and Guoluo) with less than 0.1% of the total number of households in China were left out.

Questionnaire design

Questionnaires used by other international programmes were studied and referred to during the questionnaire design stage43; an important consideration is to customize the habits and complexities of residential energy use in rural China. The energy-use habits of Chinese rural residents were incorporated into the questionnaire design (Supplementary Fig. 9), with methods used in international survey programmes referred to43. Three rounds of field tests were conducted during a preliminary study to improve the questionnaire before the survey was formally conducted. The questionnaires were designed to follow a sequence from basic and easy questions (personal and family-background information) to more complex ones (the amount and fraction of energy use). As shown in Supplementary Fig. 9 (in Chinese, with the key terms translated into English), the general personal and family information included address, family size, income, demographics of family members, annual consumption of coal and LPG, expenditures on electricity, coal, and LPG, and ownership by appliances and vehicles. Stove information included the number, location, type and usage. Energy information covered the usage of individual energy types (straw, corncobs, fuelwood, brushwood, coal, honeycomb briquettes, charcoal, LPG, biogas, electric kettles, rice cookers, induction cookers, microwave ovens, electric heaters and air conditioners), which were separately recorded for five major activities of staple food (grains) cooking, subsidiary food (vegetables and meat dishes) preparation, water boiling, animal feed heating (not used in the study) and space heating. The brushwood and fuelwood were separated due to the large difference found between their emission factors (EFs)44. Corncobs were found to be used extensively in rural China and were separated from other crop residues. For energy consumption, in addition to the current (2012) use, the activities from 5 (2007), 10 (2002) and 20 years previously (1992) were also recalled. For each activity, the percentages of time spent using various energy types were recorded on an annual basis. In addition, annual consumption quantities of coal (tons), honeycomb briquettes (pieces), LPG (cans) and electricity (cost and price) were recorded in separated questions, which could be used to check the reliability of individual questionnaires.

Personal interviews

Personal interviews were conducted to achieve a high response rate. More than 100 trained professors from 86 universities or research institutes (Supplementary Table 4) that covered all the provinces in mainland China participated in the survey. Many of the universities that participated in the survey are either agricultural or normal universities at the provincial level, with the majority of students being from rural areas. Each professor randomly selected and trained, on a voluntary basis, a group of students whose hometowns are in rural areas of the provinces. The adequate level of education and intelligence of the interviewers means that the bias caused by illiteracy and misconceptions of complex questions for respondents could be largely reduced by recruiting and training university students as the interviewers. Each trained student visited ten randomly selected households in his/her home village to conduct face-to-face interviews and fill out the questionnaires. This strategy guaranteed that the interviewers were able to understand local dialects and culture. The number of students in each university was decided based on targeted sampling rates. As the enrolments of local universities generally covered all the counties within the provinces45, the randomly collected samples were relatively evenly distributed in space.

Questionnaire examination

The recovered questionnaires were examined for completeness (all key items filled), consistency (the total amount of coal and LPG consumed should be equal to the products of the expenses and prices, and the annual consumption of coal and LPG should be equal to 12 times that of the monthly consumption) and rationality (based on a series of critical ranges, such as the total time using a single energy type should not be larger than 12 months). In total, 34,489 valid questionnaires were identified for data analysis. The data were maintained in a double entry database. The representativeness of the samples was quantitatively evaluated by calculating the REs.

Fuel-weighing and consumption calculations

The daily consumption quantities of biomass fuels and coal with distinct cooking and heating uses were weighed on site for 1,670 households. The surveys were conducted at 23 selected sites spread over mainland China (Supplementary Fig. 8), of which eight were conducted in the winter, when heating is needed. Information on demography, stoves and energy use was also collected using a face-to-face questionnaire survey (Supplementary Fig. 10). Quantities of the various energy types consumed per household per day were calculated based on the time used from the energy-mix survey, the daily consumed quantities from the weighing campaign, the moisture contents of coal (4.7 ± 3.3%, n = 89), wood (16.3 ± 11.2%, n = 48) and straw (8.7 ± 3.7%, n = 65) measured during the study, and local HD. The results were family-size-adjusted statistically. The national total consumptions of various energy types were the weighted averages of the municipality data using the total number of households of individual municipalities as the weighting factors. One limitation of the data was the relatively small sample size despite the high variation in cooking practices in rural China. In addition, possible daily variations in heating energy consumption were not addressed.

Other information

Socioeconomic information, which included Icap, GDPcap, the production of coal, grain and corn, and road densities, was extracted from the China Statistical Yearbooks Database11, whereas the energy consumption in other sectors was collected from the China Energy Statistical Yearbook11. The HD data are from Chen et al.8

Emission estimation

Annual per capita emissions of CO2, SO2, NOx, primary PM2.5, PM10, BC, OC and BaP from the direct energy use by rural residents were calculated as products of the quantities of fuel consumed and EFs1,2,3,4,31,46,47. To estimate the CO2 emissions from fuelwood, the fractions of non-renewable fuelwood were applied48. Electricity consumption was divided into power plants burning coal, petroleum, gas or others11, and only emissions from fossil-fuel-fired plants were considered. Emissions from all other sources were from PKU emission inventories49.

Data analysis

Statistical tests were conducted using SPSS, and a significance level of 0.05 was adopted. For the assumed weighting factors for staple food cooking (15%), subsidiary food preparation (80%) and water boiling (5%), a sensitivity test was conducted by changing the weighting factors within relatively large ranges of 2–30%, 65–93% and 1–20% for the three activities, respectively. It was found that the overall time-sharing data for cooking were rather robust, with 92% of cases varying less than ±10% (Supplementary Fig. 11). One explanation for such robustness is the significant correlations among the individual cooking activities (r = 0.66–0.75, P < 0.001).

Ethics

The study protocol was approved by the Biomedical Ethics Committee of Peking University and oral informed consents were obtained from all participants.

Data availability

The data that support the plots within this paper and other findings of this study are available from the corresponding author upon reasonable request.