Adaptive Responses to Climate Change: The Effects of Temperature Levels on Residential Electricity use in China

: Rising temperatures are likely to boost residential demand for electricity in warm locations 537 due to increased use of air-conditioners, fans, and refrigeration. Yet the precise effect of temperatures 538 on residential electricity use may vary by geographical area and with socio-economic conditions. 539 Knowledge on this effect in developing countries is limited due to data availability and reliability issues. 540 Using a high-quality provincial-level monthly dataset for China and fixed-effect panel methods, we find 541 a U-shaped and asymmetrical relationship between ambient temperature and monthly household 542 electricity use. An additional day with a maximum temperature exceeding 34°C on average results in a 543 1.6% increase in monthly per capita household electricity use relative to if that day’s maximum 544 temperature had been in the 22–26°C range. The effect of an additional cold day is smaller. There are 545 differences in effects for the south and the north of China and in urban versus rural areas. We estimate 546 that temperature increases associated with climate change will lead to about a 3–5% increase in annual 547 household electricity consumption by the end of the century under different carbon emission trajectories 548 according to the projections in the 2021 IPCC report. The estimated effect is larger for summer months.

This paper focuses on the response of residential electricity use to temperatures in China. 559 A key likely adaptation strategy to higher temperatures is increased used of electricity for air-560 conditioning, fans, and refrigeration (Barreca et al. 2016; Park et al. 2020). In addition, people are often 561 more likely to stay at home on days of extreme heat. Household electricity consumption is likely to be 562 affected by temperature increases along both an intensive margin (residents tend to consume more 563 electricity from existing appliances) and an extensive margin (residents tend to purchase additional 564 household appliances such as air conditioners and fans). Many residents in developing countries such 565 as China are yet to install air-conditioning but are likely to do so over coming years (Wolfram et al. 566 2012). 567 The response of household electricity consumption to temperature may well vary across China given 568 the country's large latitudinal span and its urban-rural dualism. Differences in climate and in residential 569 energy use across China are large. Residents in both the north and south are inclined to rely on electricity 570 for cooling in summer, using either electric fans or air conditioners. However winter fuels vary. Since 571 the 1950s, coal has been widely used for centralized heating in urban areas north of the Qinling-Huaihe 572 Line (Ebenstein et al. 2017). In recent times, some cities such as Beijing have used an increasing amount 573 of natural gas for heating. In the south, heating is primarily powered by electricity and there is little 574 centralized heating. Rural residents throughout the country are typically not connected to centralized 575 heating, instead relying on the combustion of biomass and scattered coal plus some use of electric 576 heaters. 577 Electricity consumption in China's household sector has increased thirtyfold since 1990. As of 2017, 578 the residential sector accounted for 14% of China's total final electricity consumption (NBS 2019a). 579 The number of residential air conditioners in urban areas of China far exceeds that in rural areas, with 580 more than 200 air-conditioning units per 100 households in some urban areas often less than 100 units 581 in rural areas (NBS 2019b). 582 There is a body of research focusing on factors that influence electricity consumption such as electricity 583 prices, incomes, population, gender, urbanization level, and air pollution levels ( consumption to high and to low temperatures. Berkouwer (2020) used micro data from South Africa to 600 explore the relationship between temperatures and household electricity consumption, concluding that 601 residential electricity consumption is likely to decrease by 6.2% per household relative to the 602 counterfactual if there is a 3.25°C increase in temperature by mid-century. This is largely due to reduced 603 heating needs in winter. 604 Most studies for China have examined the response of electricity to temperature at a fairly aggregated 605 level. Fan et al. (2015) and Asadoorian et al. (2008) used data by sector and region. There is a notable 606 gap in terms of research on rural areas. Evidence using household data is also rare due to data constraints. 607 One exception is the study of Li et al. (2019), who used daily household electricity consumption data 608 for Shanghai over 2014-2016. They observed a U-shaped temperature-electricity response curve. 609 In this paper we explore the relationship between monthly residential electricity consumption and 610 temperature at the provincial level in China, conducting various heterogeneity analyses. The use of 611 relatively high frequency and geographically-disaggregated data is more suitable for identifying 612 temperature response functions than more aggregated approaches such as using national and/or annual 613 data. 1 We also use the results to understand the likely effects of different climate scenarios on residential 614 electricity use in China by 2100. Residential electricity consumption is particularly relevant for human 615 welfare and may be more sensitive to temperature than the electricity use of other sectors such as heavy 616 industry given the importance of temperature control at the residential level. 617 2 Data and method 618

619
Accurate provincial temperature data are needed for the study. Station-averaged or area-weighted 620 temperature data would not be ideal, as it is the temperatures where people live that should matter for 621 residential electricity use. Some provinces are geographically large and have clustered populations. We 622 thus use population-weighted temperature data. 623 Daily temperature variables were constructed as follows. First, we obtained daily maximum and daily 624 average temperature from the National Meteorological Information Center (NMIC) for around 700 625 weather stations. The data are believed to be highly accurate, with quality tests having been undertaken 626 by the NMIC. County-level daily maximum and daily average temperatures were then calculated using 627 an inverse distance weighting procedure (Chen et al. 2017). This was done using average data from 628 weather stations located within 200 km of each county's centroid. We then constructed provincial 629 variables by weighting the county data by each county's population in China's 2010 census. Provincial-630 level measures of relative humidity were calculated in the same way. 631 The empirical estimations will use "temperature bin" variables so as to focus on the effects of daily 632 weather fluctuations in a monthly specification. The approach also allows for potential non-linearity in 633 the effect of daily temperatures on monthly electricity use (Bessec and Fouquau 2008). First, all months 634 were normalized to 30 days. The daily maximum temperature data were then divided into ten bins, 635 ranging from below 2°C to above 34°C. We then calculated the number of days in each temperature bin 636 in each province in each month. The dataset covers 2008-2017 and 30 provinces. The weather data were 637 processed using ArcGIS and Python. 638 There is a huge gap in January maximum temperatures between the north and the south: the average 40 daily maximum temperature is below 0°C in the northeast and northwest and above 20°C in some 646 provinces in the south. Southern provinces thus have much smaller heating requirements. In contrast, 647 there are relatively minor differences in the July average daily maximum temperature across China. This 648 temperature is below 30°C in most northern provinces and above 32°C in some southern provinces. provinces, per capita household electricity consumption is slightly higher in summer than winter, likely 658 due to use of air conditioners and electric fans. Electricity consumption in advanced provinces such as 659 Beijing exceeds that in the northwest, consistent with variation in the number of air conditioners. The 660 number of air conditioners in some southeast coastal provinces is above 130 per 100 households. This 661 number is below 10 in some northeast and northwest provinces. 662

Insert Fig. 3 663
We obtained data on average annual per capita household incomes and the annual urbanization rate from 664 the China Statistical Yearbooks. The available statistics switch from per capita rural net income to per 665 capita rural disposable income as of 2013. We used a fixed rate of change method to unify the two series 666 into a single annual per capita disposable income measure. 2 667

Method
668 The basic regression model is shown in equation (1): 669 is monthly per capita household electricity consumption divided by the number of days per 671 month (in consideration of month-to-month differences in the number of days).
, , , represents 672 the number of days for which the daily maximum temperature was in each temperature bin j (with each 673 month normalized to 30 days). We set 22-26°C as the base range. To focus on extremes, we initially 674 consider the daily maximum temperature. We then present estimates using the daily average temperature. 675 , is provincial per capita disposable income, and , , is an error term. i, y, and m represent 676 province, year, and month.
is a vector of coefficients showing how monthly household electricity 677 consumption varies as a result of an additional day in each temperature bin. An additional day in each 678 bin will result in an increase in household electricity consumption of 100( The specification controls for province fixed effects, , to capture time-invariant unobserved factors 680 at the province level such as the degree of isolation of a province and other aspects of its geography. 681 Geographical differences in electricity prices are partly captured by these province fixed effects 3  Lagged temperatures may affect current electricity usage, for example if temperatures are relevant for 696 decisions in an earlier month regarding the purchase of an air conditioner (Auffhammer 2014). We thus 697 also consider a model with one-month lagged temperature bin terms as in specification (2). The total 698 effect of an extra day in a temperature bin over the current plus lagged month is + 1 : 699 To examine potential heterogeneity as a result of climatic and other differences, we explore the 702 importance of the north-south regional divide according to the Qinling-Huaihe Line. 5 Specifically, we 703 interact north/south regional dummy variables with the temperature bin variables to obtain separate 704 estimates for the north and south: 705 We also estimate separate responses for urban and rural areas. To do so we use separate urban and rural 709 residential electricity consumption measures by province, while controlling for urban and rural measures 710 of per capita disposable income. We use the same temperature measures as in the main analysis. 711 In robustness checks we control for interactions between the province and month-of-year fixed effects 712 so as to account for unobserved geographically-varying seasonal factors. We also control for month 713 fixed effects (i.e. year by month-of-year) to take common time-varying factors into account and explore 714 potential non-linear effects of per capita income by including a quadratic term. In addition we pursue 715 estimations using temperature bins based on the daily average rather than the daily maximum 716 temperature, with 14-18°C as the base group when doing so. Ten temperature bins are again used, 717 varying from above 30°C to below -2°C. 718 Temperatures are likely to be highly exogenous to any province's residential electricity use (Wang et al.  Table 1 shows the baseline results. Column (1) is an initial specification. Province fixed effects are 733 controlled for in column (2), year and month-of-year fixed effects in column (3), and the additional 734 variables in (4). The results are similar across the columns. 735 Insert Table 1  736 A key finding in Table 1 is that the response of household electricity to hot temperatures is larger than 737 the response to low temperatures. The likely reason is that cooling is currently more electricity-intensive 738 than heating in China. An additional day in the >34°C maximum temperature range on average results 739 in around a 1.2-1.6% increase in monthly per capita household electricity consumption compared to if 740 that day had a maximum temperature in the range 22-26°C. The effect for an additional day in the 30-741 34°C maximum temperature range is 1.1% (column (4)). For cold days, each additional day with a 742 maximum temperature of only 2-6°C on average leads to monthly per capita electricity consumption 743 increasing by only about 0.6% relative to if that day's maximum temperature had been in the base 744 category (column (4)). The hot and 2-6°C effects on residential electricity use are statistically different 745 from one another at the 1% significance level. 746 The estimates in Table 1  of residential electricity use to hot weather is larger in the north than the south (a difference that is 762 significant at the 1% level). This may be for two key reasons. First, households in the south are likely 763 to be more familiar with higher temperatures and tend to live in dwellings that are better equipped with 764 fittings such as external window shadings. 6 Second, households in the south on average consume more 765 6 The average number of days for which the maximum temperature exceeds 34°C in southern provinces is around 22 per annum, five times the average for northern provinces. electricity relative to the north. 7 As a result, a smaller proportional response to an additional hot day 766 should be expected. 767 Insert Fig.5  768 The coefficient for the number of <2°C days in the south is not significantly different from zero ( Figure  769 5). It is also imprecisely estimated given that it is rare to have extremely cold weather in the south. For 770 the north, the effect for the <2°C temperature bin is significant at the 5% level and is slightly smaller 771 than that for the 2-6°C temperature bin. This is perhaps because some residents switch to using solid 772 fuels on extremely cold days. 773 Figure 6 shows separate response curves for urban and rural residents. For urban residents, the curve 774 rises steeply for days with a maximum temperature over 26°C as well as days below 10°C. An additional 775 day >34°C in a month tends to result in a 1.6% increase in monthly per capita electricity consumption 776 in urban areas, but in rural areas this figure is only 1.0%. A probable contributor to this difference is 777 that there are fewer air conditioners in rural China. Each additional day per month with a maximum 778 temperature in the range 2-6°C tends to make urban monthly per capita electricity consumption increase 779 by about 0.9%. The effect of days in this temperature bin is just 0.5% for rural residents. 780 Insert Fig. 6  781 For rural areas, the estimated effect for the 6-10°C temperature bin is larger than for the <2°C 782 temperature bin, though the latter effect is not statistically significant. This is perhaps due to some 783 dependence on electricity instead of coal for heating in the rural south, where maximum temperatures 784 do not drop too low. Heating in the rural north remains dominated by solid fuels rather than electricity. 785 Solid fuels exhibit low combustion efficiency and lead to serious air pollution problems (Chafe et al. 786

2014). 787
As shown in Table 2, electricity consumption appears to be quite insensitive to per capita income in 788 urban areas, while in rural areas the income elasticity is above 3. It makes sense that residential 789 electricity use is more closely tied to incomes. Rural residents remain at an earlier stage in the process 790 of energy transition, with residents typically transitioning to cleaner energy sources such as electricity 791 as their incomes rise (Barnes et al. 1997). 792

Robustness checks
794 Table 3 presents results for alternative specifications. Columns (7)-(8) control for the alternative sets of 795 fixed effects: province by month-of-year fixed effects and month fixed effects, respectively. We take 796 non-linear effects of income into consideration in column (9), observing an inverted U-shaped effect of 797 per capita disposable income on per capita residential electricity consumption. The turning point is at 798 around 83,000 yuan per capita (in year-2008 terms), which will be reached after 2050 according to the 799 Economist Intelligence Unit (EIU). 8 800 Insert Table 3  801 Column (10) of Table 3 uses temperature bins based on the daily average instead of the daily maximum  802 temperature. An additional day per month with an average temperature >30°C on average results in a 803 2% increase in per capita electricity consumption in the month relative to the reference average 804 temperature of 14-18°C. The trough-to-peak magnitude is slightly higher than when using the daily 805 maximum, perhaps because the daily average is more closely related to day-long electricity use needs. (4) 820 The results are shown in Figure 7. Under all three scenarios, annual per capita household electricity 821 consumption in 2081-2100 is expected to be about 3-5% higher in 2010 due to the expected temperature 822 rise alone. This is a small overall effect when considered against the large increase in residential 823 electricity use that is likely to result from future income growth and ongoing electrification of residential 824 energy services. The effect is highly seasonal. Under the SSP2-4.5 scenario, per capita household 825 electricity consumption in summer will increase by around 8% due to the warming effect. In winter, 826 electricity consumption will decrease by about 1% as heating needs decline. Under SSP5-8.5, residential 827 electricity use in China is expected to increase by 14% from June to August relative to 2010 levels. 828 Demand for residential electricity in spring and autumn would also increase. temperature-electricity response relationship. They found that an additional day with an average 855 temperature >32°C per month would result in a 5.7% increase in the monthly household electricity 856 consumption, a larger effect than our results for urban areas. This is perhaps explained by the fact that 857 Shanghai is one of the wealthiest cities in China, with a high penetration of air conditioners. 858

859
Understanding key responses to temperature extremes is important for modeling of the effects of climate 860 change (Tol and Yohe 2006), for electricity system planning, and for other purposes. In this study we 861 assembled a dataset of daily temperatures using an inverse distance weighting method and explored the 862 effect of daily temperatures on monthly household electricity consumption data in China. The paper 863 contributes to understanding the responses to climate change in a key country in the developing world. 864 We draw conclusions and implications as follows: 865 866 We found that the temperature response function for residential electricity use is U-shaped. Responses 867 to high temperatures are larger than responses to low temperatures, for the likely reason that residential 868 demand for air conditioning, electric fans, and refrigeration increases on hot days. Electricity is typically 869 less vital as an energy source on cold days, with biomass, coal, and natural gas all quite commonly being 870 used for space heating in China. 871

The response function is U shaped, with heterogeneity by region
The electricity consumption of residents is less sensitive to high temperatures in southern provinces than 872 in the north (in proportional terms), likely because people and residences are more accustomed and 873 adapted to heat and because the underlying residential electricity usage level is higher in the south. 874 Urban residents are more likely to increase their electricity use in hot or cold weather than rural residents, 875 likely at least in part due to greater availability of air conditioners. There is still huge growth potential 876 53 in both air-conditioner adoption and use in China, especially in rural areas. 877 878 Our estimates are useful for understanding the effects of future climate change on residential electricity 879 consumption in China. We estimate the effects by 2100 relative to 2010 under different carbon emission 880 scenarios based on the IPCC projections. As the climate warms, residents will face greater cooling needs 881 in summer and reduced heating needs in winter. Our estimates suggest that total residential electricity 882 consumption will likely increase by about 3-5%, with an increase that could be as high as 14% in 883 summer months. Residential electricity demand in China tends to peak in summer, so additional pressure 884 on summer demand will add to China's investment needs in electricity generation, transmission, and 885 distribution. Future research may be able to explore the effect of temperature on adoption of air-conditioners in China, 896

Effects of future warming
a key topic from a climate change adaptation point of view. It would also be of interest to explore the 897 effects of extreme temperature days -both hot and cold -on the use of other types of energy at the 898 residential level, such as coal and natural gas. 899 Declarations 900    Note: Columns (4) and (7)