Effects of Climate and Environment in North and South America on sustainable protein production: An exploratory model
Spencer Evans, Graduate Student studying Agroecology, firstname.lastname@example.org
Ryan Jansen, Undergraduate studying Biology, email@example.com
Brandon Kahen, Undergraduate studying Neurobiology, firstname.lastname@example.org
We are the Commission on Sustainable Agriculture and Climate Change, which is a part of the Consultative Group on International Agricultural Research (CGIAR), a global agricultural research group. We have been contracted by the FAO to determine the sustainability of specific protein sources in the western hemisphere. Our goal is to establish locations in which protein sources can most sustainably be developed, in an attempt to feed the world with a smaller carbon footprint. The data produced via our research will be implemented in the new millennium development goals of 2020. Our research focuses on North and Latin American protein production and the future sustainability of these various systems. This data can then be used to guide global protein consumption habits to better-fit sustainable production potential.
This website attempts to approach the problem of sustainable protein production systems by analyzing significance of climatic, geographic and economic variables through a multiple linear regression model. This work will be an exploration of the correlation between these factors on the sustainability of various protein production systems, as well as the correlation between the variables themselves. This model or the information therein could be used at the national or international level to help coordinate efforts to support sustainable food production systems. Although multiple shortcomings exist in such an approach, we feel that looking at new ways to discuss the problem of sustainability in protein production is always important. Our hope is that such a model could be critiqued and improved upon so as to prove significantly useful in sustainability efforts around the world.
The global community faces an unprecedented increase in world population. In just 12 years, 1999 to 2011, the world population grew by one billion people. This growth rate has shown no signs of slowing down. By 2050, the world population is estimated to be nearly 10 billion people. Overpopulation is driven by two factors: increasing the number of newborns per year and an overall extension of life expectancy. Life expectancy has increased by 35 years since 1900, placing a heavier burden on global population. With continued population growth, the global community must find ways to feed itself sustainably and effectively. Therefore, to avoid such effects and allow for a healthy and happy world population, the world will need to produce at least 504 billion grams of complete protein per day (504,000 metric tons) assuming a world population of 9 billion by 2050 (Godfray et al, 2010). This equates to about 184 million metric tons of protein per year. The good news is that 229 million metric tons of meat alone was produced in 2000, and that production is predicted to increase to 465 million metric tons by 2050 (Aiking, 2011). Therefore, the problem may not be if we can produce enough protein, but how to do so more sustainably, as the environmental effects of protein production are numerous and severe. Another consequence of overpopulation is climate change due to human-caused green house gas(GHG) emissions. GHGs like carbon dioxide, are being emitted faster than the natural processes that sequester them. Global warming, caused by the green house effect, has resulted in higher commodity prices, loss of species, collapse of fisheries and natural disasters. Meat production in particular has been shown to produce significant environmental impacts including loss of biodiversity and land-use change (Machovina et al, 2015), as well as soil erosion from overgrazing and ground and surface water contamination due to waste concentrations in animal operations (Bradford, 1999).
As the human population continues to grow so does the demand for protein, a component of the human diet. Humans consume protein to obtain amino acids, which are essential building blocks of the proteins produced in our cells; proteins and enzymes are the key structural and functional parts of cells. Amino acids are classed into three categories as they pertain to the human diet:
- Essential(Those that cannot be synthesized in our body)
- Conditionally essential(Those that become essential under certain pathological or physiological conditions)
- Non-essential (Those that can be synthesized in the body)
The recommended protein intake for an adult is 0.8 grams per kilogram of body weight. The quality of a protein source depends on it digestibility and amino acid composition. Incomplete proteins are lacking certain essential amino acids, while complete proteins provide all nine essential amino acids. In general, animal protein sources contain 30-35 percent more quality amino acids than plant sources. As a result, animal-derived protein accounts for almost 40% of humanity's total protein consumption, and can be expected to increase substantially by 2050 if the trend goes on uninterrupted.
As it pertains to protein production, GHG emissions can vary considerably depending on the protein produced (Eshel et al, 2014) as well as the means of production (Herrero et al, 2013). Further, what functional unit is used to measure relative GHG emission as well as system boundaries also makes significant differences in how sustainable protein production systems are relative to each other (Carlsson-Kanyama, 1998). In general, beef is the least sustainable protein source in terms of CO2-e emissions per kilogram of protein produced, followed by pork, dairy, poultry, eggs and legumes (Eshel et al, 2014). However, current models and LCA lack the resolution to see what specific factors may be causing some protein sources to be produced more sustainably in specific locations than others. Our research will therefore explore such variables using multiple linear regression analysis. When looking at future protein production, our research will consider where protein sources can most easily and sustainably be produced in North America and South America and what factors significantly contribute to each source’s sustainable production. Food miles will not be considered a significant part of the carbon footprint of protein sources, as the majority of GHG emissions associated with protein production (80-90%) are created at the farm gate.
There is a diversity of approaches to analyzing the sustainability of protein production and consumption. Life Cycle Assessment (LCA) has become a popular and critical tool in analyzing sustainable production systems. De Vries utilized LCA to observe overall environmental impacts of livestock products while Nijdam used LCA to observe land use change and carbon footprints (de Vries et al, 2010). Head even used LCA to create a consumer friendly app that can be used to make sustainable decisions in the grocery store (Head et al, 2014), while Baroni utilized LCA to observe the sustainability of different diets in the EU (Baroni et al, 2007). LCA isn’t the only approach to analysis of sustainability however. Looking closer at the consumption side of the food system has also become a very popular analysis tool. De Boer predicted the sustainability of certain proteins through consumption, supply and expenditure figures (De Boer et al, 2010). Davis approached the problem by calculating the impact of meals with different protein sources (Davis et al, 2015), while Schosler looked at the feasibility of changing diets to be less protein-based (Scholser et al, 2012). Further, there are plenty of meta-analyses that show meat consumption being the main driver behind sustainable protein production (Goodland, 1997; Machovina et al, 2015). Other approaches include looking at diet and energy (Eshel et al, 2006), protein production and land use change (Elferink et al, 2007), impacts of food-miles on sustainability (Weber et al, 2008), and economic analysis on food and diet choices (Drenowski et al, 2005). The most common approach to discussing and analyzing the sustainability of protein production, however, is the meta-analysis (Bradford, 1999; Tilman et al, 2002; Robertson et al, 2005; Aiking, 2011; Boland et al, 2013). However useful these prove to driving a discussion, ultimately where they fall short is in providing a good answer to the problem. We will try to alleviate this shortcoming by creating a model that shows relationships between geographical and economic data and GHG emissions. Although papers have attempted to create models to show links between economic and GHG data (Abdullah, 2015) as well as predict future GHG emissions in livestock production (Fiala, 2008), there are no papers that we are currently aware of that have attempted to create this more comprehensive model.Given the substantial economic, social and geographical differences between North and South America, we predict that beef and pork production, which can require more land and technology, will be a more sustainable production option in North America while dairy, poultry and plant protein will be more sustainable in South America.
Where in the Western Hemisphere can each of the below protein sources be most sustainably produced? Given the diversity of climates, landscapes and protein production systems of the western hemisphere, we sets out to explore links between certain climatic, geographical and economic characteristics with the sustainability (herein defined in terms of CO2-e) of various protein production systems. The most prudent approach to reduce the impacts of protein production on climate change is to find ways to reduce CO2-e emissions per kilogram of protein produced. Given current governmental and life cycle analysis (LCA) data, as well as various consumption projections, we will determine what factors other than on-farm efficiency improvements are most impactful on the sustainability of protein sources in South America and North America?
Given the substantial economic, social and geographical differences between the two continents, we predict that beef and pork production, which can require more land and technology, will be a more sustainable production option in North America while dairy, poultry and plant protein will be more sustainable in South America.
What metric should be used to measure sustainability?
There are many ways to approach answering our research question. We are relying heavily on the data present in the literature, and the metrics in the literature (kg protein/land area, kg protein/energy input, kg protein/nitrogen out, kg protein/ water input…etc) vary widely, which makes sense; each of the metrics found were selected to best fit the research question at hand. For clarity and simplicity, we decided that we should only use one metric when measuring sustainability in our project. It is for this reason that the metric of CO2 equivalents/unit protein out is used in our research. The literature had a high quantity of papers using this metric, which aided in the decision making process. Lastly, most of the other metrics can be converted to CO2-equivalents easier than the other way around. It is important to note that ‘food-miles’ will not be considered a significant part of the carbon footprint of protein sources, as the majority of GHG emissions associated with protein production (~90%) are created at the farm gate (Weber et al, 2008). Therefore, when looking at future protein production, this paper will consider where certain protein sources can most easily and sustainably be produced in North America and South America.
What pillar of sustainability does this metric address?
The metric we chose heavily focuses on the environmental pillar of sustainability; this decision will be addressed more in the discussion section.
The CO2 equivalents associated with the production (inputs and outputs) of these protein sources will vary in relation to where in the Western Hemisphere they are produced. For this reason we needed a method that could do more for us than a simple literature review would. This is outlined below.
We will explore the idea of creating six multiple linear regression models with CO2-e as the dependent (response) variable and all countries of interest as our observations to be predicted by three explanatory variables. The amount of explanatory variables we could include in our analysis is almost infinite. In order to guide our decision-making, we looked at multiple climatic, geographical and economic differences between countries. Ultimately, we wanted to use variables that emphasized the differences between countries in North and South America in a meaningful way. Therefore, we decided on the following explanatory variables: GDP, area and population. There will be a qualitative variable dictating whether the country is in North or South America (the Panama/Colombia border being the cutoff). Each model will correspond to one of the following protein sources: dairy, beef, poultry, soybean, pork and eggs and will be driven by data collected from the FAOstat website on a per country basis. This will ensure the data has been measured using a consistent metric. Each model will follow the general form Yi = β0 + β1x1i + β2x2i + β3x3i + β4x4i + β5x5i + β6x6i + β7x7i + εi for each observation within the dataset. Variable explanations follow:
Each general model will be run once per protein source. T-tests were run to parse out significance against the null hypothesis H0: βi = 0|all other βs. When the T-test return significant results, further sequential ANOVAs were run in order to see what effect the presence of some factors had on the significance of others. Scatterplots were observed and variance inflation factors calculated to protect against collinearity. Once potentially significant variables were considered, R2adj and BIC were used in a stepwise sequential model selection to validate the findings. Each model could then, ideally, be used to predict the sustainability of various protein products in any particular country in the western hemisphere.
Utilizing total CO2-e from agriculture per country as a substitute for our proposed functional unit as CO2-e/kg data was unavailable (see data in Appendix 1), the models were run in R and results are below. The same steps that are outlined below would be taken using CO2-e/kg as the functional unit once data becomes available. Significant differences can be seen between continents within all three variables. Additional Sum of Squares tests were run to validate significance of these differences between models. P-values of 0.004, .0000017 and 4.3e-9 were returned when testing population, GDP and land area respectively. The full model was run and significant (t>.05) T-tests were returned for population (X3, p-value = .0093), GDP/continent interaction (X6, .021) and population/category interaction (X7, .008), indicating that, when all other factors were in the model, these variables offered significant changes when added. This information was unpacked further by looking at each continent individually.
Interestingly, none of the three variables contributed significant differences to the South American model once the other two variables were entered. This fact was validated by sequential ANOVA for this dataset. Three ANOVAs were run with each variable entered first. Whichever variable was entered first showed significant effects on the response variable while the other two were insignificant. This was cause for concern as it indicates potential collinearity between variables. A scatterplot matrix of the variables are shown below. One can see that there is significant collinearity between factors, which can severely affect ANOVA outcomes. Variance Inflation Factors (VIFs) were calculated and all were well above 1 (all were larger than 100 in fact), validating the pattern seen below.
To offset collinearity, the three variables were combined into two variables—GDP/capita and population density. A model was created similar to the one above, with the exception of one less variable and one less interaction (i.e. only 5 βs). VIFs were calculated and all variables returned a VIF close to 1. These models were run in the same manner as before. The graphs below seem to show very clear differences between the two continents regarding CO2-e and the relationship to population density and GDP/capita. However, due to the relatively low power, these differences aren’t as statistically significant as it appears. Testing for differences between continents for GDP/capita returned an F-statistic of .0097 and for population density .077. Therefore, there is strong evidence that the relationship of GDP/capita to CO2-e is different between the two continents, while there is weak evidence for population density and its relationship to CO2-e and continent. T-tests were run on all factors and none showed significance once all others were considered. ANOVA results show moderate evidence that population density and the GDP/continent interaction effect the CO2 response variable, validating the F-tests above and graphs below. We conclude that there is a difference between agricultural emissions between North and South America as it pertains to GDP/capita and population density. However, the effects of GDP/capita and population density within continents are not significant.
In South America, one can see that CO2-e varies greatly despite little variation in dependent variables, while in North America, the dependent variable varies greatly with little variation in CO2-e. Thus, although one can’t conclude that the above variables have any significant effect on agricultural emissions, the continental effect does have a significant effect on agricultural emissions, it is merely a matter of finding which variables are causing such variation. Given that population, GDP and land area have little statistical impact on agricultural emissions, one is left to conclude that these differences most likely derive from climatic variations. South America in general has a warmer climate with more moisture, although this can vary greatly within the continent as well as within countries. This higher average temperature can lead to more volatilization of methane and other GHG emissions, but can also prevent or hinder certain protein production systems from being successful here. As an example, beef and dairy cattle have very low tolerance for heat. These facts are parsed out in figures 5 & 6 below—most beef and dairy production occurs in temperate climates while simultaneously producing fewer GHG emissions per kilogram of protein. Although no substantial conclusions came out of our simplistic regression analysis, the ‘white noise’ we see in the data is most likely associated with these production trends and GHG emissions variability. One could argue, then, that beef and dairy is more sustainably produced in North America. Although this doesn’t fully address the question we originally set out to answer, it does offer insight into the shortcomings of the modeling process and what factors may be important when producing sustainability models.
The three pillars of sustainability
As mentioned before, our methods only produce an answer to where these sources of protein can be produced most sustainably from an environmental perspective. And even within this pillar of sustainability, we reduce all the data we can find into a single metric, squishing some of the dimensionally out from our results. This is not because we deem the other two pillars (Social and Economic) unworthy of considering or even that the environmental pillar is the most important. Instead, we were asked to conduct research on where protein can sustainably be produced in the Western Hemisphere. With our limited time to conduct research, we decided that the approach we used.
Nothing is ever as black and white as research makes it
We group protein sources into groups like Dairy, Poultry, and Beef, each of which could be broken up into groups like, milk, cheese, yogurt, chicken, turkey, duck, Holstein, and Jersey- all of which undoubtedly vary in environmental impact. The resolution of our groups of protein we are studying is a result of the time commitment involved in hashing out each of our groups into smaller groups. We also decided that the groups we chose to study will yield results with enough detail to aid in decision making by policy makers and stakeholders.
There are protein sources we are not considering
We are only looking at 5 sources of protein and the majority of which are animal sources (with the exception of soybeans). We considered adding insect or fungal protein sources into our model, but due to their relative popularity, we decided that they lacked the aspects of social sustainability (especially in the US) that we hope to achieve with our results. We also are excluding most plant sources due to the fact that animal protein contains ~35% more essential amino acids than plant sources, as well as containing higher quantities of vitamins and other trace nutrients.
Other factors that play a role production efficiency
In our model we look at factors like GDP, Area, and Population, which are easy numbers to quantify and find within the data. What our model does not factor into it’s methods are factors that affect protein production efficiency (CO2-e/kg protein) that are harder to quantify, or turn into a simple number or input. Namely, we recognize that climate must be a big driving force behind production efficiency.
In the literature, efficiency is often cited as being an extremely important number with respect to production. The data for individual sources of protein though are illusive in the literature. Once these numbers are produced or tracked down, they can be entered into our model.
While our model has shortcomings, our main intention is for it to act as a conversation starter for policy makers. Most research we found in the literature suggests a decrease in meat consumption or an increase in production efficiency; our model steps away from these common suggestions and looks at how to find where sources can be produced most efficiently. It looks at geo-political boundaries and characteristics which commonly are ignored in the search for decreasing emissions stemming from protein production. Some might doubt the applicability of this research and the results of this model, stating that the cooperation among nations is unfeasible, but climate change is a global issue that can only be solved via global actions. Our model can also be used within a country, aiding local, small scale policy to be formed to help incentivise sustainable production locations. Using a smaller range for comparison (within a country rather than comparing countries), allows for a higher resolution output with more meaningful results.
There is little doubt that the next generation of food production systems must be vastly different from what we see today. This paper can hopefully help steer the current discussion in a new direction, allowing for better outcomes. Although the implications are controversial, countries must show a more coordinated effort in stemming the systemic problems surroundings climate change and agriculture. We hope that our model provides as kindling to a new discussion on environmental sustainability as it relates to food production. Often the topic of discussion is centered around increasing outputs while decreasing inputs (efficiencies), or decreasing the demand for certain food sources that have high carbon emissions (vegetarianism). This is a good discussion, and has led to some impressive improvements in the modern food industry; unfortunately, we can only increase efficiencies to a certain point. In the future we hope that policy makers will incentivize the production of protein, and other food, toward geo-political boundaries that decrease the average emissions per product.
We would like to thanks Professor Michel Wattiaux and Professor Erin Silva for their support and input into this project. We would also like to thank Alisha Bower for her consistent and helpful feedback. There were no conflicts of interest in the execution of this research.
Link to Annotated BibAnnotated Bibliography