Tobler’s First Law of Geography asserts that while everything is interconnected, proximity plays a key role, with closer entities exhibiting stronger relationships than those farther apart. In the context of food consumption and carbon footprint, factors such as economic ties, population movement, and cultural interchange between adjacent areas can lead to interactions impacting food consumption and carbon footprint (Zambrano-Monserrate et al., 2020). For example, neighboring regions with extensive economic integration often develop similar consumption patterns and market demands, consequently shaping comparable food consumption structures and carbon footprint. Furthermore, dietary practices from one region may diffuse to neighboring areas through population mobility, influencing their food preferences and carbon footprint. As a result, Hypothesis 1 is posited.

The STIRPAT model provides a robust theoretical framework for analyzing environmental impact factors. It assumes that environmental impacts are a function of factors such as population, affluence, and technology. More importantly, the STIRPAT model has the flexibility to add and modify relevant influences, thereby allowing it to be extended to include more variables. This theoretical model has become a well-recognized technique and is widely used to reveal the determinants of carbon emissions (Wu et al., 2021; Qi et al., 2023; Yu et al., 2023).

Meanwhile, it is worth noting that in the context of accelerating regional integration and increasingly frequent food trade, the resource allocation effect of trade impacts FCCF through market mechanisms. If the trade factor is ignored, it is difficult to fully and accurately analyze the reasons for changes in FCCF. Besides, it is necessary to consider the impact of the food consumption structure. The carbon intensity of different food groups, including animal and plant foods, varies significantly (Xu and Lan, 2016). Rising incomes drive dietary transitions toward high-carbon options, such as increased red meat consumption, leading to an increased carbon footprint (Cao and Hao, 2018; Zhu et al., 2021). Therefore, this study modifies and extends the STIRPAT model, drawing on prior literature, to meet the research needs. Specifically, when analyzing the drivers of FCCF, this study not only focuses on the impacts of population, affluence, and technology but also considers the impacts of trade and the structure of food consumption.

First, population factors encompass household size and household structure. On the one hand, a larger household size can induce scale effects, thereby reducing the energy consumption per unit of food purchased, stored, and cooked, and thus lowering the carbon footprint; conversely, a smaller household size increases the carbon footprint due to diseconomies of scale (Underwood and Zahran, 2015). On the other hand, as the aging process intensifies, the food consumption structure of the elderly evolves, with a preference emerging for low-carbon footprint foods such as poultry, eggs, and aquatic products over livestock products (Li et al., 2024), thereby also contributing to the reduction of FCCF. Consequently, Hypothesis 2 is proposed.

Second, affluence factors encompass per capita disposable income and the urban–rural income gap. Generally, higher incomes are likely to lead to an increase in the consumption of high carbon footprint foods, particularly meat, thereby augmenting FCCF (Li et al., 2021; Zhu et al., 2021). As the urban–rural income gap narrows, the food consumption patterns of rural residents gradually converge with those of urban residents (Yuan et al., 2019; Ren et al., 2021). Accordingly, a reduction in the urban–rural income gap and an increase in per capita rural disposable income are likely to drive an increase in FCCF. Therefore, Hypothesis 3 is formulated.

Third, technology factors, which reflect the level of agricultural production technology, encompass the rate of agricultural mechanization and the per capita output value of agriculture, forestry, animal husbandry, and fishery. Innovations in agricultural production technologies have significantly enhanced the yield, quality, and accessibility of low-carbon foods (e.g., plant-based foods and alternative proteins), facilitating a consumer transition away from high-carbon foods (e.g., red meat) toward low-carbon alternatives (Xu et al., 2015; McClements et al., 2021). Therefore, Hypothesis 4 is proposed.

Fourth, trade factors encompass the level of agricultural trade and the food retail price index. Agricultural trade facilitates the optimal allocation of resources across regions, a process that enhances the diversity of food supplies available to households (Xu et al., 2020). Such enhanced diversity in food supplies significantly drives the diversification of household food consumption (Dithmer and Abdulai, 2017; Krivonos and Kuhn, 2019). A more diversified consumption structure is inclined to reduce dependence on high-carbon foods (e.g., red meat), thus contributing to lower FCCF (Dou and Liu, 2024). Besides, the food retail price index can be interpreted as the cost of consumer purchases. An increase in the food retail price index usually indicates an increase in the cost of consumer purchases and a decrease in consumer purchases of food, thus reducing FCCF. Consequently, Hypothesis 5 is formulated.

Fifth, the structure of food consumption encompasses the proportion of livestock and poultry consumption, as well as the proportion of eggs, dairy, and aquatic products. It has been widely demonstrated that elevated meat consumption significantly increases FCCF; conversely, an increase in the consumption of foods such as eggs leads to a decrease in the consumption of other animal foods, thereby reducing FCCF (Xu and Lan, 2016; González et al., 2020). Accordingly, the structure of food consumption is also a key factor influencing FCCF. Therefore, Hypothesis 6 is proposed.

Furthermore, it is crucial to recognize that, due to the significant disparities in economic development levels, resource endowments, dietary customs, and policy environments across different regions, regional heterogeneity inevitably manifests in the extent and manner in which factors such as population, affluence, technology, trade, and food consumption structure influence FCCF. Therefore, we further propose Hypothesis 7.

The FCCF includes direct and indirect carbon footprint. Among them, the direct carbon footprint is calculated using the carbon conversion factor method, which determines the carbon footprint by converting the consumption of various food items into their corresponding carbon emissions based on specific conversion factors (Khan et al., 2024). It is founded on residents’ consumption of various food types and their respective comprehensive carbon conversion factors. The specific calculation formula is as follows.

In Equations 1, 2, C F i denotes the per capita direct carbon footprint of food category i. Q i is the per capita consumption; R i is the integrated carbon conversion factor; r i is the direct carbon conversion factor; m i is the meat conversion ratio; r 1 is the direct carbon conversion factor of cereal consumption. C F d is the sum of the per capita direct carbon footprint of 12 categories of food consumption.

Following Yang (2022), we quantify the indirect carbon footprint from food storage and cooking. In the storage segment, the carbon footprint of food refrigeration’s electricity consumption is calculated using the following equation, which is derived from the number of refrigerators in the household and their average annual electricity consumption.

In Equation 3, FS is the per capita carbon footprint of electricity consumption of refrigerated food in residential households. EC is the average annual electricity consumption of each refrigerator, which is 507.37 kWh according to the Maximum Allowable Values of the Energy Consumption and Energy Efficiency Grade for Household Refrigerators. SF is the coefficient of standardized coal conversion of electricity, which is taken as the value of 0.1229 kg ce/kWh. ER is the carbon footprint coefficient of electricity consumption, taking the value of 2.2132 kg C/kg ce. N is the number of refrigerators owned by each 100 households. H is the average number of permanent residents per household.

The carbon footprint of cooking is calculated by multiplying the carbon conversion factor for cooking and processing by food consumption. It should be noted that since fruits are mostly consumed directly without additional cooking and processing, the carbon conversion factor for cooking and processing is recorded as zero. The same is true for sugar.

In Equation 4, FC is the per capita carbon footprint from the energy consumption of food cooking and processing in rural households. k i is the carbon conversion factor of cooking and processing of food type i. In Equation 5, C F i n d , F C and F S are the per capita indirect carbon footprint from food consumption, per capita carbon footprint from energy consumption for food cooking and processing, and per capita carbon footprint from electricity consumption for food refrigeration, respectively.

Accordingly, the per capita FCCF equals the sum of the per capita direct carbon footprint and the per capita indirect carbon footprint. The expression is shown in Equation 6.

Spatial correlation reveals the spatial cluster or dispersion of spatial unit attributes. To investigate the spatial pattern of FCCF among rural Chinese residents, global Moran’s I and local Moran’s I are used to identify local clusters and estimate the effects at each location. The global Moran’s I is expressed as follows.

In Equations 7, 8, n is the number of research regions, indexed by i and j ; x is the variable of interest; x ¯ is the mean of x ; x i and x j denote the per capita FCCF among rural Chinese residents in the research regions i and j ; and w i j is an element of the binary spatial weight matrix. Moran’s I ranges from −1 to 1. At a given significance level, the larger the absolute value of global Moran’s I, the higher the degree of spatial correlation. Specifically, when Moran’s I is close to 1, there is a significant positive spatial correlation among provinces; when Moran’s I is close to −1, there is a significant negative spatial correlation. Moreover, it shows no spatial correlation if Moran’s I is equal to zero, which means the FCCF on the province level is randomly distributed. The significance of the spatial correlation is usually judged using the normalized statistic z with the following formula.

In Equation 9, E ( I ) is the expected value of I, and V A R ( I ) is the variance of I.

However, the global Moran’s I can only reflect the generally spatial correlation, which cannot provide information about the degree of spatial correlation on the provincial level. To solve this problem, we use local Moran’s I to identify the local spatial clusters and estimate the effects among provinces. The local Moran’s I can be expressed as follows.

In Equation 10, I i is Moran’s I of the i space unit, and the other variables are identical to those in Equations 7, 8. The meaning of the local Moran’s I is similar to the global Moran’s I. A positive I i means that the high (low) values of region i are surrounded by the high (low) values of the surrounding region; a negative I i means that the high (low) values of region i are surrounded by the low (high) values of the surrounding region.

Given that the FCCF of provincial administrative regions may exhibit spatial spillover effects (e.g., cross-regional trade), these effects must be captured through a spatial weight matrix to reflect the influence of neighboring areas. Therefore, subsequent analysis will use spatial econometric methods to analyze the drivers of FCCF. Compared with traditional econometric models, spatial econometric models incorporate spatial interaction terms, which can reflect the spatial effects among different regions. Typically, spatial econometric models primarily consist of three types: the Spatial Error Model (SEM), which examines the impact of errors in neighboring areas on regional observations and accentuates the spatial lag effect of omitted variables; the Spatial Lag Model (SLM), which explores whether variables display a diffusion phenomenon within the same region, concentrating on the spatial lag of the dependent variable; and the Spatial Durbin Model (SDM), which takes into account both SEM and SLM and effectively captures the spatial spillover effects of variables. The general form of the spatial econometric model is as follows.

In Equations 11, 12, C F i t represents the dependent variable FCCF; X i t represents the independent variable; w i j is an element of the binary spatial weight matrix; α is the constant term; β stands for the spatial regression coefficient of the independent variable; ρ stands for the spatial regression coefficient of the dependent variable; θ stands for the spatial regression coefficient of control variable; φ stands for the spatial error regression coefficient; and ε i t stands for the random error term. When ρ = 0 , θ = 0 , φ ≠ 0 , this model is SEM; when ρ ≠ 0 , θ = 0 , φ = 0 , this model is SLM; when ρ ≠ 0 , θ ≠ 0 , φ = 0 , this model is SDM.

From 2001 to 2023, the total FCCF in rural China generally decreased, while the per capita FCCF exhibited a fluctuating upward trend (Figure 2). Specifically, the per capita FCCF increased from 160.74 kg CO₂ eq/person in 2001 to 237.51 kg CO₂ eq/person in 2023, indicating an increasing environmental impact of food consumption by rural residents. This underscores the need for attention and appropriate measures to mitigate FCCF. The total FCCF remains relatively stable, with slight fluctuations, indicating that the increase in per capita FCCF does not result in a significant change to the total FCCF. This may be attributed to rural population migration driven by rapid urbanization and industrialization.

To delve deeper into the changes in the per capita FCCF across provinces, cross-sectional data from representative five-year intervals (2001, 2006, 2011, 2016, 2021, and 2023) were analyzed (see Figure 3). The results indicate that provincial trends generally aligned with the national average. Despite this consistency, the spatio-temporal analysis reveals two distinct patterns: (1) coastal-inland disparities in per capita FCCF, and (2) “high-high” and “low-low” clustering confirmed by Moran’s I. The figure visualizes the east–west gradient in FCCF, with hotspots concentrated in economically developed coastal provinces (e.g., Guangdong, Shanghai).

The global Moran’s I, Z-value, and p-value for the per capita FCCF of rural Chinese residents are presented in Table 3. The global Moran’s I for the per capita FCCF was predominantly greater than zero and passed the significance test. This result suggests that the distribution of FCCF exhibited significant positive spatial correlation, characterized by spatial dependence and agglomeration effects in neighboring regions. That is, provinces with high carbon footprint values tended to neighbor provinces with similarly high carbon footprint values, while provinces with low carbon footprint values tended to be adjacent to provinces with similarly low carbon footprint values. It is notable, however, that despite being greater than zero, Moran’s I did not pass the significance test between 2011 and 2020. This indicates that the spatial agglomeration state is relatively unstable, characterized by a “significant—insignificant—significant” development pattern.

Further research was conducted on the local spatial distribution of clustering. Moran’s I scatter plots of the FCCF of rural residents in China for 2001, 2006, 2011, 2016, 2021, and 2023 were plotted (see Figure 4). The x-axis represents the FCCF of rural residents in each province, while the y-axis represents the spatial lag value. The four quadrants correspond to different local spatial correlations. Specifically, the first quadrant (upper right) represents a high-high agglomeration type, indicating that the region is a high-value center surrounded by similar high-value areas. The second quadrant (upper left) represents a low-high agglomeration type, indicating that the region is a low-value center surrounded by dissimilar high-value areas. The third quadrant (lower left) represents a low-low aggregation type, while the fourth quadrant (lower right) represents a high-low type, exhibiting analogous meanings to the first two quadrants.

The spatial agglomeration of FCCF for rural residents in China is primarily concentrated in high-high agglomeration areas and low-low agglomeration areas. Compared with other regions, areas with rapid economic development and abundant resources are more likely to form high-high agglomeration areas. For example, Jiangsu, Shanghai, Zhejiang, Fujian, and Guangdong fall into this category. Areas with relatively underdeveloped economies are more prone to forming low-low agglomeration areas. Shanxi, Henan, Ningxia, and Gansu are typical examples.

To further identify the primary factors driving the FCCF of rural residents, this study integrates the extended STIRPAT model with the spatial econometric model to examine the effects of population, affluence, technology, trade, and food consumption structure. Descriptive statistics for the specific variables are presented in Table 4.

Pre-diagnostic tests, including the Lagrange Multiplier (LM) test, the Likelihood Ratio (LR) test, the Wald test, and the Hausman test, were performed before spatial econometric modeling. Results presented in Table 5 validate the appropriateness of the SDM with a fixed-effects specification for subsequent analysis.

Table 6 presents the estimation results for the drivers of rural residents’ FCCF. It can be observed that the coefficients of the spatial lag terms of SDM under time-fixed and spatial-fixed effects are significant, at least at the 5% significance level with a positive direction, except in the case of both-fixed effects. This suggests a positive spatial spillover effect of rural residents’ FCCF across regions.

Furthermore, the SDM under the both-fixed effects has the smallest AIC and BIC; therefore, the subsequent analysis mainly focuses on this model. Due to the existence of the spatial lag term, the regression coefficients in the model no longer indicate the actual effect of the explanatory variables on the explained variables (Elhorst, 2014). Thus, the spatial effects of the variables are further decomposed using the partial differential method, as shown in Table 7.

In terms of direct effects, spatial regression identifies three dominant drivers: aging population, income growth, and meat consumption share. Among them, the proportion of the rural population aged 65 and above in the total population and the proportion of pork, beef, mutton, and poultry in total food consumption have the most significant impacts, with coefficients of 0.011, signifying that an increase of one percentage point in these two proportions increases the per capita FCCF of local rural residents by 1.1 percentage points. Furthermore, the coefficient of per capita rural disposable income is 0.285. Specifically, every 1% increase in per capita rural disposable income can drive the per capita FCCF of local rural residents to increase by 0.285 percentage points. The coefficients of average family size ( P o p 1 ) and agricultural trade level ( C o m 2 ) are negative and significant at the 10% significance level, indicating that the above factors negatively impact the FCCF of local rural residents. For every unit increase in average household size, the per capita FCCF of local rural residents decreases by 1.1 percentage points. Moreover, it can be seen that an increase in the level of agricultural product trade helps to mitigate the growth of the per capita FCCF of local rural residents.

Regarding indirect effects, the coefficients of the proportion of rural residents aged 65 and over in the total population ( P o p 2 ) and urban–rural income gap ( A f f 2 ) are both positive and reach the 1 and 10% significance levels, respectively. This indicates that the two above variables have positive spillover effects on rural residents’ FCCF in neighboring provinces. An increase of one percentage point in the local proportion of rural residents aged 65 and over in the total population will increase the per capita FCCF among rural residents of neighboring provinces by 1.2 percentage points. Every unit increase in the local urban–rural income gap will increase the per capita FCCF among rural residents of neighboring provinces by 8.4 percentage points. These results underscore the importance of being vigilant about the environmental impacts of population aging and urban–rural income inequality, particularly their spillover effects on neighboring regions. The coefficients of per capita output value of agriculture, forestry, animal husbandry, and fishery ( T e c h 2 ) and the proportion of eggs, dairy, and aquatic products in total food consumption ( S t r 2 ) are significantly negative, at least at the 10% level. A high per capita output value of agriculture, forestry, animal husbandry, and fishery means that the region has advanced technology, which can drive the upgrading of the production structure of neighboring areas through a technology diffusion mechanism, thereby helping to improve food consumption structures and reduce the per capita FCCF. Additionally, due to the geographical proximity between regions, an increase in the proportion of consumed eggs, dairy, and aquatic products can generate a demonstration effect on the food consumption structure of neighboring provinces, which in turn can have a negative spillover effect on their FCCF.

As for total effects, the proportion of rural residents aged 65 and over in the total population ( P o p 2 ), per capita rural disposable income ( A f f 1 ), as well as the proportion of pork, beef, mutton, and poultry in total food consumption ( S t r 1 ) positively influence rural residents’ FCCF, with population aging playing the most significant role. The per capita output value of agriculture, forestry, animal husbandry, and fishery ( T e c h 2 ) negatively affects FCCF.

To enhance the reliability of regression outcomes, robustness checks were conducted using two approaches: 1% tail truncation and time horizon truncation. Specifically, all variables were winsorized at the 1% level, and regression analysis was performed again. Meanwhile, given that 2020 was the first year of major public health emergencies, during which the economy and people’s livelihoods were severely affected, the data for 2020 were removed, and regression analysis was performed once more. The regression results are presented in Table 8. A comparison with the original regression results indicates that the coefficients are generally consistent in size, direction, and significance level, thereby demonstrating the robustness of the above findings.

Given the differences in resource endowment and economic development among provinces, it is challenging to prescribe targeted measures for managing FCCF in different regions based solely on the regression results of the whole sample. Therefore, this study undertakes a comprehensive analysis of the heterogeneous drivers of rural FCCF. Thirty provinces are categorized into four regions: eastern, central, western, and northeastern, according to the official regional classification of the National Bureau of Statistics (see Table 9).

As shown in Table 10, there are variations in the drivers of the per capita FCCF across different regions. The analysis begins by examining the role of population factors. The direct effect of household size is significantly negative in the eastern and western areas and significantly positive in the central and northeastern regions. Moreover, there are positive spillover effects in the central and northeastern areas. These results reflect the heterogeneous impact of household size on the per capita FCCF in different regions. The direct effect of the aging level is significantly positive in the western and northeastern areas, and there is a positive spillover effect in the northeast region. However, they are not significant in either the eastern or central areas. This suggests that the contribution of aging to the per capita FCCF is limited to the western and northeastern rural regions.

The analysis of affluence effects follows. The direct impact of per capita rural disposable income is positive in the central and western regions, yet there is no significant promotion in the eastern and northeastern areas. Rural disposable income has a positive spillover effect in the eastern region and a negative one in the western and northeastern areas. Furthermore, the influence of the narrowing of the urban–rural income gap on rural residents’ FCCF in different places is complex. In the eastern and northeastern regions, narrowing the urban–rural income gap contributes to the increase in the FCCF of rural residents. Conversely, it has a dampening effect in the central area. Such changes also impact FCCF in other regions through inter-regional spillover effects.

The level of agricultural mechanization exerts a significant dampening effect on the per capita FCCF of rural residents in the northeast region. Also, it has a positive spillover effect in the east and northeast regions and a negative spillover effect in the central region. The growth in per capita output value of agriculture, forestry, animal husbandry, and fishery has an inhibitory effect on the per capita FCCF of rural residents in the western region. However, it is not conducive to carbon reduction in the central and northeastern areas. In addition, the per capita output value of agriculture, forestry, animal husbandry, and fishery has a positive spillover effect in the western and northeastern regions. This indicates significant differences in the impacts of agricultural mechanization and per capita output value of agriculture, forestry, animal husbandry, and fishery on the per capita FCCF of rural residents in different regions. These differences may be influenced by the region’s stage of economic development.

The retail price index of food commodities has a significantly positive direct effect in the northeast. This suggests that in the northeast, an increase in the food retail price index could contribute to an increase in the per capita FCCF of local rural residents. Furthermore, the rise in the food retail price index would be transmitted to adjacent regions via market mechanisms. Nevertheless, this positive spillover effect in the western region is not apparent, probably due to differences in economic conditions and consumption habits. Additionally, it is noteworthy that the level of agricultural product trade has a contributory influence on the per capita per capita FCCF of local rural residents in the northeast region, which diverges from the results of the benchmark regression. A possible reason for this is that, although the demand for food consumption increases as living standards rise, the food access conditions of rural residents in the northeast region are restricted, making it challenging to fully meet the consumption demand. With the development of agricultural trade, food availability has increased, which has contributed to the growth of the per capita FCCF of rural residents in the northeast region.

Lastly, the analysis focuses on the effects of food consumption structure. Increases in livestock and poultry consumption shares in central and western rural areas significantly augment per capita FCCF. However, it is interesting to note that in the rural areas of the eastern region, the increase in the share of livestock and poultry consumption reduces the per capita FCCF. Moreover, there is a negative spillover effect associated with the share of livestock and poultry consumption. With a relatively stable food supply, livestock and poultry consumption in local and neighboring areas showed a seesaw trend. It was also found that changes in the consumption share of eggs, dairy, and aquatic products had different impacts on the per capita FCCF in various regions. In the western region, an increase in the share leads to a rise in the per capita FCCF, while in the northeastern region, it has a negative effect and contributes to a decrease in the per capita FCCF.