Abstract
Mitigating emissions from methane, a potent greenhouse gas (GHG), is a critical task of fossil fuel alternatives in energy generation as well as in other sectors with large environmental impacts such as agriculture. Agricultural methane emissions have not been given sufficient attention in social science approaches to the human dynamics of GHG emissions. Given the importance of methane emissions, the need for renewable energy development, and the relationship between hydropower and agricultural systems, we ask: does hydroelectricity development influence agricultural methane emissions? If so, under what socioeconomic conditions? Using the World Bank's World Development Indicators and FAO data, we present fixed-effects models with robust standard errors to predict agricultural methane emissions from 1975–2015. Our results show that in low middle income nations and across all nations, increased hydroelectricity generation was associated with increased agricultural methane emissions during this period. We suggest hydroelectricity generation and affluence are associated with a suite of agricultural techniques, including the organization of agricultural waterbodies and animal feed, which may contribute to higher levels of agricultural methane emissions. Given the pressing need for alternatives to fossil fuels, we recommend further examination of the economic conditions for implementing alternative fuels to avoid unintended environmental harms, including those which directly counteract the intended emissions-reduction purpose of these alternatives.
Export citation and abstract BibTeX RIS
Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 license. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
1. Introduction
Today, the world faces an unprecedented ecological crisis constituted by a series of intertwined anthropogenic processes such as climate and land-use change, biodiversity loss, ocean acidification, chemical pollution, and a disruption in the nitrogen and phosphorus biogeochemical cycles (Steffen et al 2015, Ceballos et al 2017). Although tackling these aspects through laws, policies, research, and societal action is of utmost importance, addressing climate change is particularly pressing given the urgency to act to meet the 2015 Paris Agreement target of holding the increase in global average temperature to below 1.5 °C above pre-industrial levels. A particularly important driver of climate change is agriculture which, together with forestry and other land use, currently generates almost a quarter of the world's total greenhouse gas (GHG) emissions, more so than, for instance, the industry or transport sectors (IPCC 2022). While most literature on climate change, both in the natural and social sciences, is based on analyzing emissions of the most abundant GHG, carbon dioxide (CO2), fewer studies address the impact of methane (CH4). Thus, more attention should be paid to the social factors influencing agricultural methane emissions, given the increasingly important relevance of this GHG in relation to climate change (Fletcher and Schaefer 2019, Schiermeier 2020).
Environmental macro-sociological theory focuses on the relationships between human social systems and environmental change. Recent empirical work examines to what extent renewable energy development produces pro-environmental outcomes in the energy sector and beyond; this work has highlighted some paradoxes in the aggregate implementation of renewables thus far. To illustrate, it has been shown that renewable energy development may not have the presumed effect of replacing fossil fuel use at a one-to-one rate (York 2012), perhaps because renewable energies act not as a substitute of fossil fuels but are additions to the energy supply in general (York and Bell 2019). In contrast, Nguyen and Kakinaka (2019) show that high income countries lessen their carbon dioxide emissions as renewable energy consumption increases while low income countries do not, consistent with arguments of an environmental Kuznets curve. Thus, it is important to continue examining the relationship between economic development, alternative energy, and environmental change to establish the social conditions under which pro-environmental change does or does not occur. In this paper, we address this debate of the influence of affluence on environmental outcomes through models split by income category.
Despite environmental sociology having produced a wealth of research on GHG emissions, little has examined agricultural emissions. While agricultural exports and practices have been linked to forms of environmental harm (e.g. Shandra et al 2020) and restoration (e.g. Betancourt 2020), measures indicating environmental harm specifically from agricultural operations have been less studied. This follows sociology's historical focus on urban over rural issues and social processes as a discipline (i.e. Newby 1983).
Previous research has investigated the relationship between hydroelectricity generation and irrigation through biophysical factors (Tilmant et al 2009, Lacombe et al 2014, Zeng et al 2017). In this paper, we include social factors of interest to better understand the interaction of social and biophysical factors with respect to the environmental impact of alternative energies. Considering the pervasiveness of hydroelectric power (Ritchie and Roser 2020) and the dearth of research covering agriculturally-induced environmental change, we assess whether hydroelectric power development influences agricultural methane emissions. We ask whether there is a relationship between hydroelectricity generation and agricultural methane emissions, and whether this relationship varies by national income category for a sample of 104 nations from 1975–2015. Using a two-way fixed-effects panel regression model, we find that low middle income nations have a positive relationship between hydroelectricity generation and agricultural methane emission, as well as across all nations. This suggests that national affluence influences agricultural organization intensity with respect to emissions, highlighting how socioeconomic context shapes the impacts of renewable energy development.
2. Literature review
2.1. Environmental and social impacts of dams
Today, hydropower—the harnessing of the kinetic energy of river currents to produce electricity—provides about 16% of the world's electricity (Ritchie 2020). A hydropower facility consists of a power plant (where electricity is produced), a dam (that controls water flow), and a reservoir (that stores water). In absolute terms, the world's current five largest producers of hydropower are China, Canada, Brazil, the United States, and Russia.
While conceptualized as a renewable and low-carbon source of energy and electricity, hydropower nonetheless exhibits manifold deleterious socioecological effects (Nunez 2019). In short, dams alter the metabolism of matter and energy in most riverine ecosystems, by creating a rift in the water flow regime, temperature, salinity, and acidity, sedimentation patterns, energy transfers, and trophic dynamics within them (Dixon et al 1989, Ligon et al 1995, McCully 2001, Aristi et al 2014, Longo et al 2015). The ecological degradation provoked by dams is not restricted to the environs of these facilities, but spreads to both upstream and downstream ecosystems. Upstream, dams may cause flooding, which 'results in the permanent destruction of terrestrial ecosystems through inundation' (McCartney et al 2001:v). Downstream, one dam may affect ecological conditions hundreds of kilometers away from it. Moreover, dams influence global riverine microplastic transport, acting as sinks for these materials at longer timescales (Watkins et al 2019).
Dams impact human life directly in several ways, as their construction has entailed the eviction and displacement of millions of people worldwide, as well as their loss of access to clean water, fishes, grazing land, wood, several crops, the induction of landslides, as well as the emergence of diseases (Dixon et al 1989, McCully 2001, Leslie 2005). These effects fall disproportionately on the poorest members of society, who are already at a disadvantage in terms of access to basic material needs like food, water, housing, healthcare, and employment (Cole and Foster 2001).
Several works have discussed displacement, decision-making, and governance of large dam construction globally (i.e. Tilt et al 2009, Olson and Gareau 2018). Much of the world's hydroelectricity is generated by large scale damming projects, and it is the size of these projects that most exacerbate issues of ecological and social harm (Sovacool and Walter 2019). As hydroelectric dams often serve multiple purposes, such as irrigating marginal land, flood control, river transport control, and electricity production, dams can influence a wide variety of social processes and economic activity. While river development with damming can often encounter a tradeoff between electricity generation and irrigation, for a small number of river basins this relationship can be complementary (Zeng et al 2017).
Large dam construction and multipurpose dam engineering knowledge were exported from the United States and championed by the World Bank during the 20th century. Large dams have also been state projects of 'the Soviet Union during many decades after 1920 (with disasters such as the Aral Sea), through Nehru's India, Franco's Spain, Nasser's Egypt, and the Brazil of the military dictators of the 1970s and 1980s, to China before and after Mao' (Alier 1998:145). Given this entanglement of actors as well as the scale of these projects, dam development has been deployed as a tool of consolidation of political power and neo-colonialism (Pritchard 2012, Swyngedouw 2015, Martínez and Castillo 2016).
2.2. Damming, emissions, and agriculture
Research suggests that hydropower may in fact not be a low-carbon energy source given its ties, both direct and indirect, to GHG emissions. For instance, decaying organic matter in reservoirs created by dams are an important source of GHGs (particularly methane) to the atmosphere and even downstream (Kemenes et al 2007, Deemar et al 2016). In fact, organic carbon sediment trapping by dams is known to have created methane 'hot spots' (Macek et al 2013). After carbon dioxide (CO2), methane is the most abundant GHG on Earth, accounting for 17% of total anthropogenic emissions in 2016 (IPCC 2022), while being some 34 times stronger than CO2 as a radiation-trapping molecule (IPCC 2013). Considering methane's atmospheric lifetime (about 12 years) and its overall impact over 100 years, one ton of methane is equivalent to some 28 tons of CO2 (Schiermeier 2020). Thus, it is no wonder that over the last two decades, several natural and social scientists have argued that, at present, the planet is facing a no-analog epoch known as the Anthropocene, wherein human socioeconomic activity itself has become a geological force that critically endangers the continuity of life on Earth as we know it (Crutzen and Stoermer 2000, Crutzen 2002, Angus 2016, Waters et al 2016). Thus, the direct link between dam reservoirs and methane emissions is cause for careful thinking about dams as a low-GHG electricity source.
In this paper, we address hydropower's indirect contributions to methane emissions through its connection to agriculture. Dam building, particularly of large and costly projects, requires a large source of funding. Often, these projects are financed by a consortium of interested parties, including local and national governments, international capital, and transnational companies (McCully 2001). In order to secure financing from such diverse funders, dams are built for multiple purposes, including electricity production, irrigation, flood control, and river transport (Billington and Jackson 2006). Furthermore, the process of prioritizing a dam's function is often a political struggle between interest groups (Jones 1995). Irrigation from damming is often developed for marginally arable land, which often requires further mechanized intervention to prove productive and profitable, as occurs, for instance, in the United States high desert (Reisner 1993). Thus, dams may indirectly increase methane emissions by sustaining these intensive, industrialized agricultural practices. Specifically, these multipurpose dams may contribute to GHG emissions beyond those created by reservoir flooding and construction as described above.
As noted, industrialized agriculture is an important source of GHGs, including methane (Kirschke et al 2013, Schiermeier 2020). The chief sources of agricultural methane emissions are livestock enteric fermentation, livestock waste, rice cultivation, agricultural water storage bodies, and waste burning (EIA 2011). Changes in the relative and absolute scale of these farming practices can increase total emissions. Methane agricultural emissions have increased in the last decade, following a rise in red meat production, a resource-intensive and polluting enterprise (Fletcher et al 2019, Schiermeier 2020, York 2021).
The intensification of industrial agriculture is often characterized by large petroleum-based inputs of fertilizer and pesticides, machinery, and long supply chains (Smil 2011). This intensification, which has increasingly turned the land into a factory-like system dependent on machinery and external inputs, is associated with several environmental impacts (Perfecto et al 2009). The environmental impacts of industrial agriculture are evident in the nitrogen cycle, methane and nitrous oxide emissions, land use change, waterbody pollution, pesticide use, and environmental health concerns (see Cole and Foster 2001). Globally, industrial agriculture is also associated with the breeding and genetic modification of modern varieties of staple crops to increase yields (Evenson and Gollin 2003). These varieties have been associated with shifting power and property relationships, which may have benefited consumers more than farmers themselves (Shiva 1994).
Through their multiple purposes, dams become embedded in agricultural production, land use patterns, and transport in ways which may result in unintended intensification of resource use, or movement towards qualitatively different agricultural technologies. This may, in turn, counteract emissions reductions from hydroelectricity. We argue that these indirect emissions are important to consider when building new energy regimes to move away from fossil fuels, and we present one approach to begin understanding these impacts in the context of global political economy.
2.3. Environmental sociology and environmental degradation
While hydropower is the most common renewable energy source worldwide (Ritchie and Roser 2020), it has been found to displace fossil fuels only modestly, while creating considerable environmental impacts (York 2012). This occurs because, given the complexity of economic systems and human interactions, technologies implemented with the intention to reduce a certain resource consumption (such as dams), either through substitution or efficiency improvements, often do not attain the intended outcome. In particular, the latter trend, wherein an increase in efficiency is commonly associated with increased resource use (which in principle should decrease)—a phenomenon known as the Jevons paradox—can aptly characterize the relationship between technological improvements and fossil fuel consumption (York 2006, York and McGee 2016). In the context of adding alternative energy sources, such as hydroelectricity, with at least an implicit intention of replacing units of fossil fuel production and thus reducing overall emissions while keeping energy consumption relatively constant, the displacement paradox can describe positive relationships between alternative energies and total emissions (York 2012, McGee 2015, York and Bell 2019). Therefore, it is worth considering whether and how hydropower is associated with agricultural emissions, moving beyond analyses of emissions-related harm from damming at the point of construction or through reservoir life-course to considering linkages with other social systems (e.g. food/agricultural systems). Environmental sociology and structural human ecology have developed methodological and theoretical approaches to address the interactions between agricultural systems and hydropower.
STIRPAT (stochastic impacts of regression on population, affluence, and technology) has proven an effectual modeling system for predicting measures of environmental change. Derived from earlier, accounting-based systems (see Ehrlich and Holdren 1972), STIRPAT offers the ability for hypothesis testing and easier interpretability due to elasticity coefficients (York et al 2003a). The modeling approach is theoretically informed, logging all variables to produce a multiplicative relationship between them.
Early studies using this approach established the salience of population and affluence in predicting carbon dioxide emissions and ecological footprints (Dietz and Rosa 1997, York et al 2003b), showing that population (relative to affluence and technology) and economic growth negatively influence ecological conditions. Breadth and depth have been added to these investigations as well. This has been accomplished by using this approach to assess environmental sociological theories of militarization (e.g. Jorgenson and Clark 2009) and unequal trade relationships (Jorgenson 2006b, Shandra et al 2009). Meanwhile, breaking down population dynamics (Liddle 2011, York and Rosa 2012, Clement and Podowski 2013), showing how economic growth's impact on emissions exhibits directional asymmetry (e.g. Huang and Jorgenson 2018) and functions under certain conditions of renewable energy development (York and McGee 2017), world system position (Greiner and McGee 2018), income inequality (McGee and Greiner 2018), and gender inequality (McGee et al 2020), has added depth and nuance to our understanding of the role of population and affluence in environmental harm.
While much environmental social science research investigates carbon dioxide emissions, less attention has been paid to the second largest anthropogenic GHG: methane. Yet, similar to research on carbon emissions, population size and economic development are associated with more methane emissions (e.g. Jorgenson and Birkholz 2010). In addition, the organization of production and foreign direct investment influences methane emissions (Jorgenson 2006a, Jorgenson and Birkholz 2010). However, possible associations of renewable energy development and methane emissions have not yet been explored. Moreover, STIRPAT has largely been used to assess social drivers of emissions that deal with industrialization and urbanization (see Schnaiberg 1980, Foster 1999), occluding an investigation as to how agricultural emissions come about in predominantly rural areas.
Given the importance of methane emissions, the need for renewable energy development, and the relationship between hydropower and agricultural systems, we ask: does hydroelectricity development influence agricultural methane emissions? If so, under what socioeconomic conditions? We seek to fill an empirical gap in scholarly attention to emissions associated with agriculture and non-urban socioecological interactions, as well as expand discussion interrogating the conditions of pro-environmental outcomes of non-fossil fuel energy development.
3. Data and methods
To test for the effects of hydroelectricity on agricultural methane emissions, measured in thousand metric tons of CO2 equivalent, we estimate three fixed-effects panel regression models using World Bank data on all nations with available information from 1975–2015 (the World Bank does not report agricultural methane emissions beyond 2018). For the purposes of this paper we are most interested in a broad look of dominant patterns in hydroelectric power historically, to interact with other social activities such as agriculture.
We split the models by income categories(low, low middle, upper middle, and high) to investigate variation in effect by affluence. Income category has previously been used as a proxy for capital intensiveness or productivity of domestic economies (Rice 2007). We use 1993 classifications, near the middle of our observations. The thresholds for each category are GNI per capita in 1993 U.S. dollars and are as follows: Low (⩽695), Low middle (696–2785), Upper Middle (2786–8625), High (>8625). Table 2 lists countries included in Model 3 by income category.
All variables are logged, following STIRPAT conventions, therefore coefficients are elastic. See Table 1 for descriptive statistics. As hydroelectricity has a minimum value of 0, we added 0.01 to all values of this variable to allow for logging. Our first model includes indicators associated with the STIRPAT modeling structure. These indicators account for several factors established in previous research as key social structural influences on emissions and are: GDP in constant 2010 U.S. dollars; urbanization measured by percent of population in urban centers; agricultural production as percent of total GDP; agricultural exports as percentage of total merchandise exports; and total population. Many studies which employ a STIRPAT modeling logic include terms in per capita form. This suggests the implicit coefficient for population is one, whereby any increase in population would create a 1:1 increase in emissions. Therefore, it is generally more parsimonious to include other indicators in per capita terms and omit population as its own indicator. We include population as its own indicator to allow for other population relationships, as we are inquiring about an under-investigated GHG emission.
Table 1. Descriptive statistics for independent and dependent variables for all nations included in Model 3.
| Variable | Mean | Median | Max. | Min. | S.D. |
|---|---|---|---|---|---|
| Agricultural methane emissions (tons of CO2 equivalent) | 30 330.6 | 6335.869 | 577 909 | 20 | 75 775.6 |
| Hydroelectricity generation (TWh) (1 Twh = 1 billion kwh) | 24 | 3.72 | 1090 | 0 | 69 |
| GDP (millions of constant 2010 US$) | 363 000 | 73 300 | 8890 000 | 1,470 | 821 0001 |
| Urbanization (% of population in cities) | 58.53 | 60.76 | 100 | 4.83 | 20.78 |
| Agriculture as % of GDP | 11.84 | 8 | 69 | 0.11 | 10.96 |
| Agriculture as % of value added exports | 5.34 | 2.67 | 81.3 | 0.00009 | 8.47 |
| Population (in millions) | 57.7 | 11.1 | 1370 | 1.01 | 181 |
| Cattle (millions of heads) | 12.9 | 2.74 | 215 | .005898 | 32.6 |
| Cereal production (millions of metric tons) | 18.6 | 2.69 | 621 | 0.000273 | 55.5 |
Table 2. Nations included in Model 3. Income category by 1993 per capita GNI. Total nation years = 2650.
| Nation years = 584 | Nation years = 516 | Nation years = 981 | Nation years = 569 |
|---|---|---|---|
| High income | Upper-middle income | Lower-middle income | Low income |
| Australia | Argentina | Algeria | Albania |
| Austria | Belarus | Angola | Armenia |
| Belgium | Brazil | Azerbaijan | Bangladesh |
| Canada | Chile | Botswana | Benin |
| Cyprus | Estonia | Bulgaria | Bosnia and Herzegovina |
| Denmark | Gabon | Cameroon | Cambodia |
| Finland | Greece | Colombia | China |
| France | Hungary | Costa Rica | Eritrea |
| Germany | Libya | Croatia | Ethiopia |
| Ireland | Malaysia | Cuba | Georgia |
| Israel | Mauritius | Dominican Republic | Ghana |
| Italy | Mexico | Ecuador | Haiti |
| Japan | Oman | El Salvador | Honduras |
| Kuwait | Portugal | Guatemala | India |
| Netherlands | Saudi Arabia | Indonesia | Kenya |
| New Zealand | Slovenia | Iraq | Mongolia |
| Norway | South Africa | Jamaica | Mozambique |
| Qatar | Trinidad and Tobago | Jordan | Nepal |
| Spain | Uruguay | Kazakhstan | Nicaragua |
| Sweden | Latvia | Nigeria | |
| Switzerland | Lebanon | Pakistan | |
| United Arab Emirates | Lithuania | Sri Lanka | |
| Morocco | Tajikistan | ||
| Namibia Panama | Togo Zambia | ||
| Paraguay | Zimbabwe | ||
| Peru | |||
| Philippines | |||
| Poland | |||
| Romania | |||
| Russia | |||
| Senegal | |||
| Thailand | |||
| Tunisia | |||
| Turkey | |||
| Turkmenistan | |||
| Ukraine |
Our second model adds hydroelectricity generation in kilowatts per hour (kWh). We use this variable, rather than the percentage of domestic electricity generation that is generated by hydropower, to focus on the absolute instead of the relative size of damming facilities, as we argue this will better capture the effect of hydroelectric generation on agricultural methane emissions. Finally, in our third model, we add cereal production in metric tons and heads of cattle. Cereal production data was gathered from the World Bank, and comprises production of all grains including rice, which is directly associated with methane emissions. Likewise, cattle production is also directly associated with agricultural methane emissions, as stated above. Data for head of cattle were gathered from FAO (2020). We add these two indicators last as they are the most highly correlated with agricultural emissions.
We exclude nations with populations below one million, as is convention for much STIRPAT work. We construct two-way fixed-effects panel regression models with robust standard errors to adjust for clustering of residuals by nation, using the felm command in R (Gaure 2022). We also include dummy variables for nation and year to control for period effects that influence all nations and nation-specific time invariant factors (e.g. geographical matters). We include both the R2 for the model as a whole and the R2 for the projected model. The latter addresses the inflation of the R2 due to the large number of dummy variables for nation and time period. All coefficients and standard errors are rounded to the fourth decimal place. The analytical dataset is available upon request.
We include all available observations with full data in our analytical dataset. Like much cross-national data, all of our key variables have a long right-hand tail—excluding these points would exclude important cases such as high hydroelectric producing nations (95th percentile and above: Brazil, Canada, China, India, Norway, Russia), and high agricultural methane emitters (95th percentile and above: Australia, Brazil, China, India, Indonesia, Pakistan). We natural log these predictors and the outcome variable in part to address this skew. We also ran a Hausman test to assess whether a random or fixed effects model would be appropriate. The p-value for this test was 0.0000, indicating that fixed effects provide more reliable estimates relative to random effects.
To check for multicollinearity, we examined the variance inflation factors for all subsections of Model 3. We found that the VIF for GDP was high, above ten across all nations. We re-ran the model without GDP (table S3 in supplemental information), and found the substantive results remained the same. We chose to keep GDP in the below models as it is a key predictor in social science investigations of emissions.
4. Results
Results for Model 1 are displayed in Table 3. Model 1 includes STIRPAT indicators, and all variables are natural logged. Across all nations, there are no statistically significant predictors. Across low, low middle, and upper middle income nations, only population is statistically significant, and the estimated coefficient decreases from low, to low middle, to upper middle income. For high income nations, GDP, agriculture as a percent of GDP, and population are statistically significant. GDP has a relatively large estimated coefficient, indicating that a 1% increase in GDP is associated with a 2.1% decrease in agricultural methane emissions, suggesting economic growth has a suppressing effect on these emissions. However, across all nations GDP is not distinguishable from zero in Model 1, suggesting that economic growth is not associated with decreased agricultural methane emissions across all nations.
Table 3. Model 1 for all income categories, examining the association of conventional STIRPAT predictors with agricultural methane emissions. Two-way fixed effects panel regression with robust standard errors, all variables logged, predicting agricultural methane emissions measured in tons of CO2 equivalent, 1975–2015.
| All | Low | Low middle | Upper middle | High | |
|---|---|---|---|---|---|
| GDP | −0.0834 (0.0877) | 0.1910 (0.12904) | −0.0367 (0.1005) | −0.2280 (0.14830) | −2.0957** (0.8011) |
| Urbanization | 0.3001 (0.3072) | 0.0798 (0.2560) | −0.0762 (0.4291) | 0.0831 (0.6725) | 0.4361 (0.8110) |
| Agriculture as % GDP | 0.3515 (0.1997) | 0.1373 (0.0942) | 0.0063 (0.0978) | 0.0707 (0.0781) | 0.3061* (0.1288) |
| Agriculture as % exports | −0.0010 (0.0180) | −0.0118 (0.0212) | 0.0164 (0.0285) | −0.0003 (0.0468) | −0.1291 (0.1369) |
| Population | 0.442885 (0.3784) | 1.4525*** (0.2803) | 1.0104*** (0.1672) | 0.7525* (0.3383) | 1.2149** (0.3885) |
| R2 | 0.9839 | 0.9867 | 0.9827 | 0.9956 | 0.9843 |
| R2 proj model | 0.205 | 0.2237 | 0.1588 | 0.2764 | 0.4759 |
| Nation-years | 3358 | 913 | 1149 | 612 | 684 |
p-value thresholds: *** = 0.001; ** = 0.01; * = 0.05; robust standard errors in parentheses.
Agriculture as a percent of GDP is also positive and statistically significant for high income nations, suggesting that the size of the agricultural sector adds to agricultural methane emissions for high income nations. The estimated coefficient for population is over one, such that a 1% increase in population is associated with a 1.21% increase in agricultural methane emissions. This may be due to changes in consumption patterns that correspond to income level, perhaps an increase in meat consumption.
Results for Model 2 are displayed in Table 4. Model 2 adds hydroelectricity generation, the variable of interest, to our STIRPAT model. Among income categories, only low middle income nations have a statistically significant relationship between hydroelectricity and agricultural methane emissions. This coefficient is significant and positive across all nations, suggesting that holding other structural factors constant, a 1% increase in kWh of hydroelectricity generation is associated with about a 0.01% increase in agricultural methane emissions. All other predictors across all nations remain non-significant, as in Model 1. Departing from previous work using STIRPAT modeling, these indicators do not seem to be strongly associated with agricultural methane emissions. This suggests that agricultural methane emissions are not generally associated with economic growth or modernization processes across nations, but are rather centered on other social processes which seem to vary across nations and over time. This may include qualitative differences in the organization of the agricultural industry, rather than a quantitative increase of scale, as most modernization processes are conceptualized. We discuss some specific agricultural practices that may explain this in the Discussion section. We also split our sample by decades to more closely examine the role of time across all nations, see table S1 in the supplement section for this model.
Table 4. Model 2 for all income categories, including conventional STIRPAT predictors and hydroelectricity generation. Two-way fixed effects regression with robust standard errors predicting agricultural methane emissions measured in tons of CO2 equivalent, 1975–2015. All variables logged.
| All | Low | Low middle | Upper middle | High | |
|---|---|---|---|---|---|
| GDP | −0.1503 (0.0924) | 0.0261 (.1058) | −0.0288 (0.1094) | −0.1847 (0.1234) | −2.1103** (0.7940) |
| Urbanization | 0.3104 (0.3726) | −0.0221 (0.2256) | −0.3503 (0.3290) | −0.2577 (0.6278) | 0.3719 (0.9677) |
| Agriculture as % GDP | 0.3682 (0.1982) | 0.1017 (0.0876) | 0.0774 (0.0774) | 0.0362 (0.0715) | 0.3021* (0.1253) |
| Agriculture as % exports | −0.0010 (0.0197) | −0.0299 (0.0265) | 0.0062 (0.0275) | 0.0096 (0.0453) | −0.1335 (0.1368) |
| Population | 0.4078 (0.3880) | 1.358*** (0.2772) | 0.9034*** (0.1790) | 0.9029** (0.3362) | 1.2169** (0.3914) |
| Hydroelectricity generation (kwh) | 0.0120* (0.0058) | −0.00008 (0.0044) | 0.0089* (0.0041) | 0.0423 (0.0595) | 0.0030 (0.0092) |
| R2 | 0.9853 | 0.9886 | 0.9864 | 0.9957 | 0.9843 |
| R2 proj model | 0.2369 | 0.2359 | 0.226 | 0.3059 | 0.4761 |
| Nation-years | 3091 | 687 | 1128 | 592 | 684 |
p-value thresholds: *** = 0.001; ** = 0.01; * = 0.05; robust standard errors in parentheses.
Results for Model 3 are displayed in Table 5. Model 3 adds cattle and cereal production to the STIRPAT model. Hydroelectricity retains its sign and significance from Model 2: across all nations and in low middle income nations it has a positive and significant effect on agricultural methane emissions. It is also noteworthy that in high income nations, the coefficient is positive and nears significance (p = 0.082). Heads of cattle is positive and significant for all nations and all income categories, with an effect size of less than one in each. This indicates that adding cattle production does not increase agricultural methane emissions on a 1:1 scale, likely indicating that the methane impact of the cattle industry is both due to scale (number of cattle) and organization (specific farming practices). Meanwhile, cereal production is positive and significant across all nations and upper middle income nations, indicating that the scale and intensity of agricultural production is associated with agricultural emissions for some nations For example, cattle is statistically significant across all nations, with coefficients all estimated below one. Cereal production is non-significant for low and low middle income nations, though this may be due to the mix of cereals produced in these nations, as the variable includes rice as well as other cereals.
Table 5. Model 3 for all income categories. Two-way fixed effects regression with robust standard errors predicting agricultural methane emissions measured in tons of CO2 equivalent, 1975–2015. All variables logged.
| All | Low | Low middle | Upper middle | High | |
|---|---|---|---|---|---|
| GDP | −0.1345* (0.0665) | −0.0868 (0.1505) | −0.1542 (0.0857) | −0.1736 (0.0897) | −0.3900 (0.4256) |
| Urbanization | −0.1889 (0.1415) | 0.1946 (0.2189) | −0.5503*** (0.1068) | −1.1591*** (0.3479) | −0.3888 (0.6749) |
| Agriculture as % GDP | 0.0017 (0.0347) | −0.0505 (0.1164) | 0.1057* (0.0465) | −0.0640 (0.0554) | −0.0919 (0.0767) |
| Agriculture as % exports | 0.0024 (0.0151) | −0.0031 (0.0232) | 0.0105 (0.0138) | 0.0465 (0.0296) | −0.0855* (0.0413) |
| Population | 0.3556** (0.1286) | 0.5068 (0.3183) | 0.0328 (0.1507) | 0.5827*** (0.1281) | 0.8056*** (0.1355) |
| Hydroelectricity generation (kwh) | 0.0059* (0.0024) | −0.0015 (0.0060) | 0.0077** (0.0025) | 0.0035 (0.0255) | 0.0115 (0.0066) |
| Cattle | 0.5619*** (0.0549) | 0.5736*** (0.124) | 0.7009*** (0.0540) | 0.5953*** (0.1229) | 0.3390** (0.1293) |
| Cereal Production | 0.0663** (0.0248) | 0.0503 (0.0793) | −0.0063 (0.0268) | 0.1289*** (0.0322) | 0.0573 (0.0361) |
| R2 | 0.9938 | 0.9906 | 0.9936 | 0.998 | 0.9934 |
| R2 proj model | 0.4472 | 0.3146 | 0.6096 | 0.6334 | 0.3762 |
| Nation-years | 2650 | 569 | 981 | 516 | 584 |
p-value thresholds: *** = 0.001; ** = 0.01; * = 0.05; robust standard errors in parentheses.
Population has a non-unitary but positive statistically significant effect on emissions for high, upper middle income nations and across all nations, but is non-significant for low and low middle income nations in Model 3. A one percent growth in population is associated with a 0.36% increase in agricultural methane emissions across all nations. Agriculture as a percent of export value is negative and statistically significant at the 0.05 level for high income nations, suggesting agricultural goods made for export may not contribute significantly to methane emissions. We also ran a model (table S2 in the supplement) with a quadratic term for GDP to test for a non-linear relationship with economic growth, but the quadratic term was not statistically significant across all nations.
5. Discussion and conclusion
In this paper, we addressed an open empirical question in human ecology research on the drivers of agricultural methane emissions. Our results suggest these drivers are distinct from those with robust evidence in critical human ecology, particularly the non-unitary relationship between population and agricultural methane emissions, GDP, and urbanization. While many studies find positive associations between economic development, urbanization, and carbon dioxide emissions or ecological footprint (York et al 2003b, Jorgenson and Clark 2012), in our examination of agricultural methane emissions these commonly-found associations do not seem as pertinent. Our focus on an understudied dimension of emissions, agriculture, highlights that theoretical tools in environmental social science may be built for explaining many, but not all, sources of degradation, and that agricultural emissions may be produced from distinct sets of structural factors relative to other sources of emissions.
We linked agricultural methane emissions and hydroelectricity development and generation to ask whether hydroelectricity is associated with agricultural methane emissions, and whether this relationship is moderated by national affluence. We found that hydroelectricity is associated with higher levels of agricultural methane emissions from 1975–2015 in low middle income nations, all nations, and some evidence among high income nations (p < 0.10), holding key socioeconomic and agricultural characteristics constant. We do not necessarily suggest this is a direct or causal association, but rather emphasize that there may have been a suite of associated agricultural strategies over this period, which include damming technology, influencing changes in these emissions over time. Consistent with prior work highlighting the paradoxes of capitalism for the implementation of renewable energies or 'green' substitutes (e.g. York 2006, 2012, McGee 2015), there appears to be evidence for unforeseen downstream effects from hydroelectric power generation vis-a-vis agricultural methane emissions. Given the deep ties interconnecting various functions of dams, including diversifying financial sources for new dams (e.g. political, mining, and agricultural interests), this hypothesized association is worthy of deeper examination as we work towards mitigating the effects of climate change via fossil fuel alternatives. Existing alternative energies have given insight into the influence of social organization of new technologies being implemented in the present and near future on their pro-environmental impacts.
There are many possible reasons hydroelectricity may contribute to increased methane emissions. Based on existing literature, we suggest a few indirect connections, including the production of agricultural waterbodies, management of irrigation and power generation functions of multipurpose dams, and irrigation of marginal agricultural land which requires several (often mechanized) inputs.
First, the agricultural methane emissions' measure we employ includes anaerobic decomposition from small agricultural waterbodies, which may be associated with damming. These models may be capturing differential reservoir size and management, which is one way in which dams contribute to methane emissions. Second, the roles of multipurpose dams, including irrigation and power generation, can conflict or be administered in different ways in different places. Previous research suggests that technocratic, market-oriented, or state-run damming projects have variable outcomes as to who benefits from dams: communities or corporations (Sayan and Kibaroglu 2016). It is possible that certain income groups may emphasize the use of hydroelectric dams for irrigation purposes in different ways than other income nations during this period. This brings us to our third suggestion: reservoir irrigation can bring land into production or grazing that would not otherwise be arable and can contribute to land use change contributing to methane emissions (Ritchie 2019). More land under cultivation can increase agricultural methane emissions. Also, land which otherwise may not be considered arable or is of lower quality, may require more mechanized and industrial inputs to grow viable and profitable crops. We ran an additional set of models including the saturated predictors in Model 3, and added a control for total arable land. The substantive results of the association between agricultural methane emissions and hydroelectricity generation were unchanged. This suggests that general increases in the amount of arable land due to irrigation of agriculturally marginal land may not be the main driver of this relationship. See table S4 in the supplemental information for this model and more discussion on total arable land.
Dams have long been associated with highly technocratic and centralized state systems due to their great cost (Billington and Jackson 2006, Smil 2017). Hydroelectricity production was tied to increases in agricultural methane emissions while controlling for the total production of cattle and cereals, the two main contributors to agricultural methane emissions for all nations, and low middle and perhaps high income nations. The years when hydroelectricity production increased in the period between 1975–2015, may also be the period in which more methane-intensive agricultural activity grew. This outcome can be due to the management of manure changes in the kinds of uses of marginal arable lands and increases in flooded agricultural land due to the construction of irrigation systems linked to dam reservoirs. These activities are characteristic of capital-intensive agriculture and situate hydroelectricity as one mechanized input among several, associated with high GHG emissions.
Our aim is to draw attention to the ways in which existing alternative energy infrastructure is embedded in our existing production system, with ties to intensive agriculture and other practices which may dampen the ability of hydroelectricity to meaningfully address GHG emissions in aggregate. Future research into the underlying mechanism of the association we highlight here, particularly case studies or life-cycle assessments of hydroelectric dams with broad parameters, will help unveil the many connections our energy systems have to our productive processes.
Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: https://databank.worldbank.org/source/world-development-indicators.
Supplementary data (<0.1 MB PDF)