Analytical model, variable justification and data

The analytical framework is based on the STIRPAT structural model, first formulated by Ehrlich and Holdren (Reference Ehrlich and Holdren1971), initially known as the ‘Impact by Population Affluence and Technology (IPAT)’. IPAT quantified the effects of population-induced economic activity on the natural environment. It unfolded the impact on the environment through three underlying variables: the size of the population, affluence and technology or the effect per unit of economic activity. IPAT was later reformulated to STIRPAT to enable the testing of hypotheses with greater rigor and strength, as shown by Dietz and Rosa (Reference Dietz and Rosa1994), Zhang and Lin (Reference Zhang and Lin2012) and Ji and Chen (Reference Ji and Chen2017). The general formulation of the STIRPAT model is shown in Equation (1):

(1)$$I_{i, t} = \varphi P_{i, t}^{\lambda 1} \;A_{i, t}^{\lambda 2} T_{i, t\;}^{\lambda 3} \varepsilon _{i, t}$$

Here, I, P, A and T represent environmental impact, population size, affluence and technology. φ is the constant, λ ₁, λ ₂ and λ ₃ are the exponents of P, A and T, respectively, and ε denotes the error term.

Previous scholars (Grossman and Krueger, Reference Grossman and Krueger1991; Holtz-Eakin and Selden, Reference Holtz-Eakin and Seldeon1995; Ozturk and Acaravci, Reference Ozturk and Acaravci2011; Ozturk et al., Reference Ozturk, Al-Mulali and Saboori2016; Sarkodie and Strezov, Reference Sarkodie and Strezov2019; Xiong, Chen and Huang, Reference Xiong, Chen and Huang2019) advocated that there exists an inverted ‘U’-shaped relationship between a countries per capita income as well as income-squared and its environmental performance, commonly known in the literature as the environmental Kuznets curve (EKC) effect. The EKC is an empirical phenomenon that reveals a hypothesized relationship between various contributors to environmental deterioration and per capita income. The EKC hypothesis states that as the emission of various pollutants (carbon dioxide, sulfur and nitrogen oxides) increase, the environmental quality worsens for nations in their early stages of economic development. However, when nations per capita incomes increase due to economic growth, their demand for cleaner environment increases as governments enact and implement environmental legislation for a cleaner environment; hence, the environmental quality will improve. This suggests that the environmental impacts or emissions are an inverted ‘U’-shaped function of per capita income. Although it is difficult to show a perfect EKC in two dimensions, there appears to be a crude inverted ‘U’ shape as depicted in Appendix Figures 1 and 2 for the sample of high-income countries forming part of the analysis in this study.

The EKC effect can be tested using the STIRPAT modeling framework. Given that the core variable of interest in this study is food production incorporating this together with the EKC measure and other potential influences on GHGs gives the extended STIRPAT model of the general form given in Equation (2):

(2)$$GHG = f( {Food\;production, \;\;EKC, \;\;Control\;variables} ) $$

Here, GHG captures the environmental quality measured by methane and nitrous oxide emissions. Food production includes total food production and two forms of food production: cereal and livestock production. The EKC measure is income and its quadratic term. The main control variables are the industry's size, the area under forests, trade, population density, education, government effectiveness and political stability. Incorporating these individual variables in Equation (2) gives the estimable form represented by Equation (3):

(3)$$\eqalign{GHG_{i, t} = & \beta _0 + \beta _1 food\;production_{i, t} + \beta _2income_{i, t} \cr & \quad + \beta _3income^2_{i, t} + \beta _4industry_{i, t} + \beta _5trade_{i, t} \cr & \quad + \beta _6forests_{i, t} + \beta _7population\;density_{i, t} \cr & \quad + \beta _8education_{i, t}+ \beta _9government\;effectiveness_{i, t} \cr & \quad + \beta _{10}political\;stability_{i, t} + \varepsilon _{i, t}} $$

The coefficients β ₁ to β ₁₀ are the elasticities of GHGs with respect to the right-hand side variables. A finding that β ₂ > 0 and β ₃ < 0 would indicate an EKC exists. The next section describes the variable measures and data.

Variable measures

The sample includes 132 countries. Following the World Bank classification of countries by per capita incomes, the 132 countries are grouped into three categories: 26 low-income, 79 middle-income and 27 high-income (Organization for Economic Cooperation and Development) countries.

The sample of low-income countries is Afghanistan, Benin, Burkina Faso, Burundi, Central African Republic, Chad, Democratic Republic of Congo, Eritrea, Ethiopia, The Gambia, Guinea, Guinea Bissau, Haiti, Liberia, Madagascar, Malawi, Mali, Mozambique, Nepal, Niger, Rwanda, Sierra Leone, Tajikistan, Togo, Uganda and Yemen Republic.

The sample of middle-income countries is Albania, Algeria, Angola, Argentina, Azerbaijan, Bangladesh, Belarus, Belize, Bhutan, Bolivia, Bosnia and Herzegovina, Botswana, Brazil, Bulgaria, Cabo Verde, Cambodia, Cameroon, China, Colombia, Comoros, Congo Rep., Costa Rica, Cote d'Ivoire, Cuba, Dominica, Dominican Republic, Ecuador, Egypt, El Salvador, Eswatini, Gabon, Georgia, Ghana, Grenada, Guatemala, Guyana, Honduras, India, Indonesia, Iran, Iraq, Jamaica, Jordan, Kazakhstan, Kenya, Kyrgyz Republic, Lao PDR, Lesotho, Malaysia, Maldives, Mauritania, Mauritius, Mexico, Morocco, Myanmar, Namibia, Nicaragua, Nigeria, North Macedonia, Pakistan, Paraguay, Peru, Philippines, Romania, Russian Federation, Senegal, South Africa, Sri Lanka, Thailand, Timor-Leste, Tunisia, Turkey, Ukraine, Uzbekistan, Vanuatu, Venezuela, Vietnam, Zambia and Zimbabwe,

The sample of high-income countries is Australia, Austria, Belgium, Canada, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Japan, Netherlands, New Zealand, Norway, Poland, Portugal, Spain, Sweden, Switzerland, United Kingdom and the United States of America.

The data used are annual, covering the years 2002–2016. Table 1 summarizes the variable measures and their sources.

Table 1. Variable measures

Source: Authors compilation.

Findings

The empirical procedure involves the estimation of Equation (3) in two stages. The first-stage estimation investigates the effect of total food production on GHG emissions. In the second stage, Equation (3) is estimated with the food production type. It includes testing the impact of cereal yield and livestock production on GHG emissions. The variables and units of measures tested, reported in Tables 2–4, are the same as those listed in Table 1. Several diagnostic tests are conducted as part of the estimation, and the results are considered satisfactory, given the use of the balanced panel data.

Table 2 includes the first-stage estimation results, where the prime variable of interest is food production. According to Table 2, the food production coefficient has the expected positive effect (β = 0.265 for methane and 0.455 for nitrous oxide) and statistically significant (ρ > 0.01) but only for the high-income group of countries. For middle- and low-income countries, food production coefficient has a negative sign and statistically insignificant for methane emissions. However, food production coefficient has the expected positive and significant effects (β = 0.072 and ρ > 0.01) on nitrous oxide emissions in the middle-income countries. When the impact of food production on nitrous oxide emissions is compared to the high-income countries, the magnitude of this is smaller in middle-income countries (β: 0.072 < β: 0.455) with ρ > 0.01. Overall, the empirical results of total food production provide reliable and sufficient confirmation that in the high-income countries, it contributes significantly to the degradation of environmental quality.

Table 2. Effect of total food production (measured by food production index, 2004–2006 = 100) on methane and nitrous oxide emissions

Numbers in parentheses are t-statistics.

*P < 0.1, **P < 0.05 ***P < 0.01.

Having established the effect of food production on GHG emissions, the second stage in the estimation phase involved testing the impact of food production type on GHG emissions: Tables 3 and 4 present the estimations of cereal and livestock production on GHG emissions, respectively.

In Table 3, the prime variable of interest is cereal production. All other control variables and their measures are the same as listed in Table 1. According to the results in Table 3, the cereal production coefficient is as expected positive and statistically significantly correlated with methane emissions in the high-, medium- and low-income group of countries with β = 0.001 to 0.006 and ρ > 0.01 in all three income group of countries. Interestingly, the sign of cereal production coefficient effect on nitrous oxide emissions is as expected, positive and statistically significant for the middle- and high-income countries (β = 0.002 and 0.001 with ρ > 0.01 for both). These results provide reliable confirmation that cereal production in the high-, medium- and low-income countries contributes significantly to methane emissions. At the same time, its effect on nitrous oxide emission is strong only in high- and middle-income countries.

Table 3. Effect of food type—cereal production (kg per hectare) on methane and nitrous oxide emissions

Numbers in parentheses are t-statistics.

*P < 0.1, **P < 0.05 ***P < 0.01.

Having established the effects of cereal production on GHG emissions, an attempt was made to test the impact of livestock production (as part of food production) on GHG emissions. The results are presented in Table 4. In Table 4, the prime variable of interest is livestock production. All other control variables and their units of measure remain unchanged and are the same as those listed in Tables 1–3. According to the results in Table 4, livestock production coefficient is as expected positive and statistically significantly correlated with methane and nitrous oxide emissions only in the high-income group of countries with β = 0.565 and 0.556 respectively with ρ > 0.01. This result provides reliable confirmation that it is livestock production in high-income countries that contribute significantly to GHG emissions.

Table 4. Effects of food type—livestock production (measured by livestock production index, 2004–2006 = 100) on methane and nitrous oxide emissions

Numbers in parentheses are t-statistics.

*P < 0.1, **P < 0.05 ***P < 0.01.

According to Table 4, livestock production is negatively and statistically significantly correlated with methane emissions in low- and middle-income countries. A possible line of explanation for these outcomes is that many low- and middle-income countries have adopted improved animal husbandry and livestock manure management practices since the late 1990s (United Nations Environment Programme, 2021). According to IDDRI (2019–2020), some of the low-income countries of Africa revealed success in methane efficiency of animal manure (transformation into biogas) that reduced emissions and provided carbon-free energy to sedentary pastoralists. Fernández-Amador et al. (Reference Fernández-Amador, Francois, Oberdabernig and Tomberger2020) also state that between 1997 and 2014, low- and middle-income countries engaged in methane efficiency in the agricultural sector through improvements in manure management as well as bringing about dietary changes to ruminants, small-scale farming and adopting farming systems with low-emission intensity such as poultry and egg production. In addition, Caro et al. (Reference Caro, Kebreab and Mitloehner2016) noted that livestock producers increased the efficiency of feed utilization by introducing small amounts of protein meal and providing molasses and urea blocks and improving the genotypes and better management and disease control.

Looking at the effects of control variables across Tables 2–4, the findings for total food, cereal and livestock production strongly confirm the EKC effect. An examination of data for 2002–2016 indicates that a wider group of high-income countries have shown signs of divergence: increasing per capita GDP while reducing emissions and managed to transit to a low-emission path. For example, the United States, United Kingdom, France, Spain and Italy have reduced GHG emissions while increasing GDP. This observation is also consistent with a study by Mikayilov et al. (Reference Mikayilov, Hasanov and Galeotti2018), who reported evidence favoring relative decoupling in eight out of the 12 European countries, providing evidence of European nations determination in adopting GHG reduction and compliance policies. Data for the low-income countries generally reveal low levels of per capita GDP and low levels of emissions. However, there are exceptions, such as the large developing economies such as Brazil, China and India, revealing rising GHG emissions with increasing per capita incomes.

The estimation results show that industrial expansion is strongly correlated with either methane or nitrous oxide emissions in all three income categories. There is no doubt that industrialization is essential for nations economic prosperity. However, industries also use substantial quantities of energy, primarily fossil fuels, the burning of which worsens environmental quality, consistent with the findings of Sueyoshi et al. (Reference Sueyoshi, Yuan and Goto2017) and Shahbaz et al. (Reference Shahbaz, Sarwar, Chen and Malik2017).

The results in Tables 2–4 reveal that forest cover coefficient is strongly inversely correlated with GHG emissions in the high-, medium- and low-income countries, consistent with the expectations. However, the findings for forest cover do suggest that nations maintaining and expanding on forest cover such as through afforestation is essential from a biodiversity perspective as trees absorb GHGs, sustaining the environment, consistent with Khuc et al. (Reference Khuc2018).

The population density coefficient consistently has negative sign and statistically significant in the high-, middle- and low-income countries in all cases tested, as revealed in Tables 2–4. However, these findings do not lend support to the adverse effect of population density on environmental degradation.

Similarly, trade coefficient is negative and statistically significant across most of the estimations in Tables 2–4, indicating that trade openness does not worsen environmental degradation. Thus, much of the empirical literature on the effect of trade on the environment is inconclusive. This finding is consistent with Frankel and Rose (Reference Frankel and Rose2005) and Jebli et al. (Reference Jebli, Youssef and Ozturk2016) as these researchers confirmed that openness to trade brings in environmental technology from developed countries which can improve environmental quality instead of having an adverse effect.

Education is found to have the expected strong effects only in the middle-income countries for methane emissions. In contrast, a negative but statistically insignificant impact is found for low-income countries. This finding confirms the importance of education as it is vital in curbing environmental degradation, given that schooling promotes environmental awareness and values. The outcome of this variable is also consistent with the results of Post and Meng (Reference Post and Meng2018) and Chankrajang and Muttarak (Reference Chankrajang and Muttarak2017).

In terms of institutional effect, only political stability is negatively correlated with GHG emissions in middle-income countries, suggesting that an improved and stable political environment can reduce GHGs.

GHG emissions from food production contribute to the deterioration of environmental quality, but human-induced climate change resulting from agricultural emissions (among other sectoral emissions) is also adversely impacting agricultural production and food supply. Climate change resulting from agricultural GHG emissions is already influencing agricultural yields. It threatens the sustainability of traditional and small scale agriculture and food supply, particularly in the low- and middle-income countries as millions of people continue to remain hungry. Porfirio et al.'s (Reference Porfirio, Newth and Finnigan2018) study indicates that cereal yields are likely to decline under high GHG emission scenario. Although other factors are contributing to the current state of hunger (with over 820 million people remaining hungry in 2018) other than food availability, meeting the nutritional needs for millions of hungry people globally is threatening the aspirations of the United Nations Sustainable Development Goals of ending world's hunger and achieving food security by 2030. The timely achievement of the Sustainable Development Goal of zero hunger will require mitigating climate change inputs on agricultural yields of cereals and animal sources of protein.

The long-term scenario is that given that the world's population is expected to reach 9.1 billion by 2050, agricultural products will continue to fulfill as much as possible the nutritional demands of the world's population. Hence, GHG emission mitigation policies have to be in place to balance the shift of agricultural production to a low carbon sector while raising agricultural productivity and output. A transition to a low GHG food production system means embarking on strategies to improve crops and grazing land management, water usage for cereal such as rice, livestock and manure management, agro-forestry, use of bio-energy for agricultural production and more distributed world food trading network that can sustain the growing nutritional demand while absorbing climate shocks.

Summary and concluding comments

The effect of food production type on GHG emissions is tested for the high-, middle- and low-income group of countries from 2002 to 2016 using the STIRPAT modeling framework. The findings showed that total food production is strongly correlated with methane and nitrous oxide in high-income countries and nitrous oxide emissions in middle-income countries. After disaggregating food production into food production type, the findings of this study revealed that cereal production is positively and statistically significantly correlated with methane emissions in the high-, middle- and low-income group of countries and with nitrous oxide emissions in the high- and middle-income countries. The findings also confirmed that livestock production positively and statistically significantly correlated with methane and nitrous oxide emissions in high-income countries. Other than confirming the EKC effect, industrial expansion, forest cover and education are other strong common determinants of GHG emissions in all three income categories of countries. Thus, the findings of this study confirm that food production is an essential source of GHG emissions. In addition, the results have some policy implications.

There is ample scientific evidence that methane breaks down quickly and effects are almost immediate, and mitigating actions taken by the food producers and consumers can reduce the GHG build-up and slow the rate of global warming and climate change. Markets have a fundamental role, and emissions from agricultural production can be addressed through market mechanisms. Food producers worldwide can transit from resources that contribute to GHGs, such as cleaner and sustainable inputs that can minimize detrimental effects on the environment. On the supply side, by boosting yields of crops and livestock and innovative farming technologies such as feed additives that minimize enteric fermentation among livestock can mitigate methane emissions. Introduction of lower emission food crop varieties that can reduce nitrous oxide emissions from chemical fertilizers improved livestock management in large developing countries with high food supply and demands such as China, India and other countries in the African and Asian continents can reduce GHG reductions and benefit the environment.

Food producers need to adopt sustainable production systems through fewer chemical inputs such as fertilizer, pesticides, insecticides and fossil fuels. Suppliers of food should transit toward lower industrialized food production systems and emphasize ecological production practices. Protection of the environment from the effects of GHGs requires awareness and understanding of human actions and their impact on the environment.

On the consumption side, behavioral changes among the population through reducing food losses and wastage and transiting to a diet from emission-intensive products such as red meat toward plant-based foods means reducing livestock production and methane emissions.

Food producers can also consider instituting an appropriate balance between cereal and livestock production from an environmental and ecological viewpoint. Increases in livestock stocking can have adverse impacts on the environment. Higher animal stocking numbers not only contributes to GHG emissions but also damages ecological conditions. The results obtained in this study have direct relevance to the debate about how food production can be used in developing efficient food-environment systems that minimize emission levels.

It is known that the agricultural sector is profoundly associated with the world's food security need. At the same time, agriculture has direct implications for the world's adaptation to climate change.

The international community has to initiate a collective action of the governance of agricultural food production that is not biased toward the ideology of free trade in food but effectively addresses problems that agricultural food production imposes on the environment, as shown by the evidence of food production GHG emissions in this study. Valuable international policy action is for countries to adhere to the norms of best practices of environmental governance, such as all parties to the Paris Agreement to commit to agricultural adaptations within the agricultural sector fully and broader climate action frameworks relating to agricultural production.

The protection of the environment from GHGs must be recognized in a market-driven production and management system. Hence, the adoption of sustainable agricultural production systems that mitigate GHGs can be rewarding in the long run and closely comply with the broader objectives of the United Nations Sustainable Development Goals 12, 13 and 15, relating to sustainable production, climate action and life on land, respectively.