Global Food Demand and Carbon-Preserving Cropland Expansion under Varying Levels of Intensification

I. INTRODUCTION

Two related challenges confront humanity in the twenty-first century: how to meet the food demands of a global population that will surpass 9 billion by 2050, and how to minimize losses in human welfare from climate change. Agricultural production is responsible for over 25% of global anthropogenic green-house gas (GHG) emissions; about half of the ice-free land area of Earth is used as cropland or for pasture (Tilman and Clark 2014). Converting land from natural vegetation to cultivation reduces the amount of carbon that is stored in the landscape, contributing to increased levels of carbon dioxide (CO₂) in the atmosphere and magnified climate change (Millennium Ecosystem Assessment 2005). We address these challenges in a combined framework to assess how to minimize carbon storage loss while producing enough agricultural output to meet demand in 2050.

The United Nations Food and Agriculture Organization (FAO) has forecast that global food demand will grow by some 70% from 2005 to 2050 due to increases in population and incomes (Alexandratos and Bruinsma 2012). Other predictions are as high as 100% to 110% (Tilman et al. 2011). Here we assume conservatively that a 70% increase in global calories will be needed (Pardey et al. 2014). There are two basic ways the global food production system can meet this increased demand: (1) increase the yield of existing cropland through expanded use of fertilizers and other chemicals, water for irrigation, improved seeds, farm labor, mechanical capital, and informatics such as geographic information system–guided planting and harvesting (intensification); and (2) increase the amount of land in agriculture by converting natural areas to cropland (extensification). Technological optimists, including much of the agricultural research establishment and most commercial suppliers of farm inputs, argue that the world can achieve the food needs of 2050 largely through intensification. Even more optimistically, others argue that “peak land” occurred in 2009, such that the percentage of land in agriculture will henceforth decrease over time (Ausubel, Wernick, and Waggoner 2013).

Many prominent agricultural researchers, however, are not as optimistic. Grassini, Eskridge, and Cassman (2013) found a widespread deceleration in the relative rate of average yield increases of the major cereal crops between 1990 and 2010 in countries that were the largest producers of these crops. They found strong evidence of yield plateaus, and in some cases an abrupt drop in yield gains. Average annual global yields rose for wheat by 2.92% between 1961 and 2007, for rice by 1.91%, and for maize by 2.47%. However, the FAO projects yield increases of only 0.86% for wheat, 0.63% for rice, and 0.83% for maize between 2005/2007 and 2050 (Alexandratos and Bruinsma 2012). Future yield in-creases may also decline due to climate change (Lobell, Schlenker, and Costa-Roberts 2011; Schlenker and Roberts 2009).

With smaller yield increases, the FAO estimates that global cropland will need to expand 20% overall, and by 30% in developing countries (Alexandratos and Bruinsma 2012). Grassini, Eskridge, and Cassman (2013, 9) conclude that overestimating yield trajectories “leads to estimates of land requirements that are too low and diminish [the] capacity for effective strategic planning and research prioritization to ensure future food security and conservation of natural resources.” They are supported by Ray et al. (2013), who project yield trends “far below” what will be needed to meet rising demands. We base our arguments in this paper on the argument that intensification alone will not be sufficient to meet demand and that cropland expansion will be necessary to meet future demand.

Here we analyze how extensification can be located to meet increased caloric demand in a way that minimizes carbon losses. We address this question using a global, geospatial optimization technique (Johnson et al. 2014) that we update and expand to consider varying levels of intensification. We use high-resolution global data to estimate the gain in calories from expanding crop production relative to the loss in carbon storage. The analysis uses grid-cell-based, spatially explicit production functions to define a carbon cost-minimization problem. We focus our analysis on the carbon storage that is lost when land changes from uncultivated (natural vegetation) to cultivated use. Emissions from this type of land use transition constitute a large impact on climate change (Pachauri et al. 2014, 123). We do not consider other potential environmental costs such as the climate impact from nitrogen fertilizer or changes in solar reflectivity (albedo).

We expand on prior literature in several important ways. First, rather than identify a priori what portion of future food (caloric) needs will be met by intensification or extensification, we consider a range of possible intensive/extensive combinations along a continuum, from 95% intensive–5% extensive to 50% intensive–50% extensive. Second, we create a new spatial dataset to identify the change in carbon storage as a function of new hectares cultivated. Instead of basing carbon change on potential natural vegetation, as in existing research, we combine higher-resolution carbon storage data with cropland extent data to calculate marginal land natural vegetation carbon storage. Third, we use improved data for calculating losses in soil carbon. Fourth, we include an analysis of the policy implications of our findings, absent in much of the prior scientific literature.

II. DATA

Our primary data source is EarthStat (IonE and Ramankutty 2014), a global high-resolution spatial dataset for agriculture.¹ In each 5 by 5 minute grid-cell (roughly 10 by 10 km at the equator) we use data on existing cropland (Ramankutty et al. 2008) and crop yields for 175 crops (Monfreda, Ramankutty, and Foley 2008). The global cropland extent dataset developed by Ramankutty et al. (2008) combines two satellite-based land cover datasets with agricultural inventory data to calibrate an algorithm that specifies what proportion of the grid-cell is cultivated. Monfreda, Ramankutty, and Foley (2008) utilized national, state-, and county-level census data to calculate yield and production statistics for 175 different crops on each grid-cell by spatially allocating the production reported in the census data to the cropland extents reported by Ramankutty et al. (2008). We converted these data to aggregate caloric content, using the nutrient content of each crop from the FAO (FAO 2014, documented by Johnson et al. 2014).

Existing assessments of the change in carbon storage rely on estimated potential natural vegetation carbon (PNVC) to assess how much carbon is lost when expanding agriculture onto new lands (Ramankutty and Foley 1999; West et al. 2010). These assessments are based on assumptions about what natural land cover would have existed in the absence of human activity. Use of these data is problematic when assessing current land use choices because natural land near cultivated areas is not pristine and does not store as much carbon as the potential natural vegetation (Thompson et al. 2013; Putz and Redford 2010; Sasaki and Putz 2009). Thus, using PNVC to define the loss in carbon from agricultural expansion may overestimate the impact.

To avoid this type of overestimation, we developed a new dataset that combines the proportion cultivation in each grid-cell (Ramankutty et al. 2008) with a higher-resolution map of carbon storage (30 by 30 arc seconds; 100 grid-cells of the higher-resolution data fit into one of the lower-resolution, 5 by 5 arc minute grid-cells) (Reusch and Gibbs 2008). We defined the proportion of the 5-minute grid-cell that is natural, n∈[0,1], by subtracting the proportion cultivated from one. Next, we ranked each of the 30-second grid-cells by carbon storage in ascending order and defined the n-proportion highest carbon storage cells as the subset that is natural land. We then calculated the mean carbon storage present in the natural land subset and labeled this as the marginal-land natural vegetation carbon, MNVC.

This method of calculating MNVC may underestimate the amount of carbon storage on natural lands if the 30-second data contain many cells that have both agriculture and natural land. We tested this possibility by running a series of sensitivity analyses where we defined MNVC with a range of higher values for n up to n=1.0, thus increasing the value of MNVC that we calculate. Rerunning the full cost-minimization with these definitions did not change our results by more than 0.1%. However, running the minimization with PNVC resulted in approximately 10% more carbon storage being retained, suggesting that using PNVC may overestimate the effect.

We further improved existing methods by using newer, higher-resolution data on soil carbon (Hengl et al. 2014), which we aggregated from 1 by 1 km resolution to match our 10 by 10 km grid-cells. We calculated the loss in soil carbon from extensification as 25% of existing soil carbon, SC_xy, following the method of West et al. (2010). Here and throughout, subscripts x∈Xand y∈Y denote the latitude and longitude of specific gridcells, and X and Y represent all latitudes and longitudes considered. We converted all pergrid-cell values to per-hectare values, while adjusting for the size of each grid-cell at different latitudes. The soil carbon did not have a large impact on the results, primarily because in locations where extensification occurred, the change in soil carbon was at least one order of magnitude smaller than the change in aboveground carbon.

We calculated the per-hectare loss in carbon storage from expanding cultivation, ΔC_xy, by taking the loss in soil carbon perhectare, SC_xy, and the loss in marginal-land natural vegetation carbon storage per-hectare, MNVC_xy, and subtracting the amount of carbon stored in agricultural crops per hectare: where CC_ixy measures the per-hectare carbon stored in crop i, assuming the proportion of each individual crop remains constant in gridcell xy.

Using these data, we calculated the marginal carbon impact per calorie produced by expanding cultivation in grid-cell xy, MCC_xy, by dividing the change in carbon by the caloric production potential of cultivating the land, multiplied by the social cost of carbon: [1] where p_c is the social cost of carbon, ΔC_xy is the per-hectare loss of carbon storage as defined above, and F_xy is the per-hectare caloric yield of the grid-cell.

The accuracy of our analysis depends on the quality of the datasets we used. Some of our datasets, particularly those dealing with soils, are not as accurate as detailed observations from field studies and suffer from relatively coarse data resolution and interpolation errors. These are problems typical of research based on remotely sensed data for global analysis and suggest that conclusions should not be made for specific grid-cells. Rather, this paper focuses on the broader spatial patterns across regions that encompass hundreds of thousands of grid-cells, where noise in measurement of individual grid-cells matters far less. Given the global, scenario-based assessments in this paper, these datasets are useful for analyzing the underlying phenomena and general locations of trade-offs. Table 1 summarizes the data used and discusses the resolution, scale, assumptions, and methods used to create them.

III. MODEL AND METHODS

In this section we present our model and methods that incorporate the improvements discussed above. We framed our model in economic terms rather than the biophysical terms of Johnson et al. (2014). The main goal in this reframing is to show how biophysical, spatial optimization problems can be expressed in economic terms (e.g., with production functions and cost minimization), which are particularly useful for analyzing tradeoffs.

Our model solves for the number of hectares brought into cultivation on each gridcell, H_xy∈H, in order to minimize carbon storage loss while producing enough calories to meet the required global caloric increase where H={H_xy:x∈X, y∈Y}. Specifically we solved [2]

In this problem, the cost function, c(H_xy), measures the dollar value of lost carbon storage, c(H _xy)=p_cH_xyΔC_xy, as a function of cultivation expansion. The right-hand side of the constraint represents the caloric production goal that needs to be met, defined as the proportional increase d that is required above current aggregate caloric production, P_current.

The production function in each grid-cell is defined as [3] where is the initial number of hectares in cultivation, Y_ixy is the per-hectare yield of the ith crop before intensification, k is the factor by which we assume yields increase, and (F_ixy)/(Σ_jF_jxy) scales the crop-specific caloric yield, F_ixy, by the proportion that crop contributes to the grid-cell’s total caloric yield, Σ_jF_jxy, observed in the data. We assume for simplicity that the future crop mix is proportionally the same as the current crop mix. This assumption can be relaxed to address changing crop mixes as climate and demand conditions change. The key input for each grid-cell’s production function is Y_ixy, or per-hectare yield, based on work by Monfreda, Ramankutty, and Foley (2008), converted to calories as by Johnson et al. (2014).

To solve our optimization problem, we rank all grid-cells by their MCC_xy value. Starting with the lowest-valued cell, we increase the amount of hectares cultivated until it reaches the maximum level of cultivation in that cell (defined below). We then repeat this process for the next lowest valued cell, con-tinuing until we satisfy our production goal.

It is not possible in nearly all grid-cells to cultivate 100% of the land. Even grid-cells that generally have good soils and are highly productive will have some area that cannot be brought into cultivation (e.g., contain built structures or have very steep slopes). To account for this, we limit the maximum hectares of expansion in grid-cell xy, , to be a function of the number of hectares and the proportion of the grid-cell currently cultivated: where p_xy is the current proportion of gridcell xy cultivated, β∈(0,1) is a parameter (we set β = 0.5 for the results in this paper), and is the initial number of hectares cultivated in grid-cell xy. We set to zero for p_xy<0.05 and p_xy>0.95 so that we do not expand in areas where there is little or no cultivation or in areas where the grid-cell is near complete cultivation (discussed in more depth below). We ran sensitivity analyses on these constraints and found that the results did not change by more than 0.1% when minimum and maximum were modified by ± 0.05.

To account for how different levels of intensification affect our results, we ran the model with a range of values for k in the equations above, from 50% to 95%. We solved the model multiples times and interpolated the relationship between the model results to identify a continuum of solutions.

In addition to solving for the carbon-loss minimizing solution (defined as the “carbon-preserving scenario” hereinafter), we also defined a “proportional increase scenario” in which cultivation expands proportionally across grid-cells, subject to the feasibility constraints discussed above. We compare the carbon-preserving scenario with the proportional increase scenario in the following results section. For both scenarios we assume that the current mix of 175 crops in each grid-cell stays the same.

IV. RESULTS

Figure 1 plots results of the model, showing the difference in carbon storage in metric tons between the carbon-preserving scenario and the proportional increase scenario. The proportion of the 70% increase in calories met by cropland extensification is reflected on the horizontal axis, from 5% to 50%. The amount of carbon lost is shown on the vertical axis. The difference between the two curves—between the carbon-preserving approach and the proportional increase approach—reflects the carbon storage preserved by following a selective expansion policy.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1

Carbon Storage Loss by Scenario

As one moves from left to right in Figure 1, and extensification becomes a larger part of the way caloric demand is met, both scenarios have greater total carbon loss because more natural land must be brought into cultivation to meet the caloric production goal. The difference in carbon storage preserved between the proportional increase and carbon-preserving scenarios widens as the proportion of caloric demand met through extensification grows. When 20% of the calorie target increase is met through extensification and 80% via intensification, consistent with FAO estimates, approximately 7 billion tons of stored carbon can be saved by following the carbon-preserving expansion scenario. These results demonstrate an important finding: if yield increases slow and extensification expands, the carbon costs of the proportional increase versus carbon-preserving scenario grow. How we will produce more calories while assessing the carbon storage costs of alternative land use strategies will substantially determine the future of agriculture as a contributor to climate change. Moreover, if yield increases are slowing, choosing where to extensify becomes even more important.

In addition to identifying aggregate levels of carbon preservation under different levels of extensification, we also developed spatial maps of our results. We subtracted the baseline data (circa 2000) from the proportional increase and carbon-preserving scenarios, thus expressing our results as changes from the baseline (rather than absolute values) to enable clearer visual display. Figure 2 shows where new cultivation is located under our carbon-preserving scenario, expressed in hectares.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2

New Hectares Cultivated in Carbon-Preserving Scenario

To compare the carbon-preserving scenario with the proportional increase scenario, we next calculated the difference between the number of new hectares cultivated in the carbon-preserving scenario and the number in the proportional increase scenario. Because it is difficult to present both positive and negative values on a single gray-scale map, we split this map into two panels for positive and negative values, respectively.² Figure 3a shows a global map of hectares conserved in their natural state under the carbon-preserving scenario but newly cultivated under the proportional increase scenario. Figure 3b shows the hectares that are newly cultivated under the carbon-preserving scenario but would not have been cultivated under the proportional increase scenario. Because these maps plot the difference between the scenarios, they can be interpreted as where the carbon-preserving scenario conserves more or less of the land than if we had followed the proportional increase scenario.

These maps show that under the carbon-preserving scenario, most of the extensification occurs on the fringes of the productive and intensively farmed regions of the world, including the U.S. Midwest, northern Europe, and northeastern China. In these areas, carbon-preserving extensification still results in losses of carbon storage, but that loss is minimized relative to caloric comparative advantage. There is increasing pressure to plant “fencerow to fencerow” in these regions, including the U.S. Corn Belt, where farmers have pulled land from the Conservation Reserve Program in response to high corn prices (Cook 2015). But compared to the proportional increase approach, vast areas of the Earth are protected from cropland expansion under the carbon-preserving approach (the shaded areas in Figure 3b) that would be lost.

It should be noted that large areas of the Earth’s surface, appearing white in Figure 3a and b, did not differ between the scenarios because they were precluded from extensification by our constraints, which did not allow expansion where the proportion cultivated was either very low or very high: p_xy<0.05 and p_xy>0.95. Those grid-cells with less than 5% cultivation included regions not suited to agriculture, with latitudes too far north and south, high elevations, and/or arid areas without irrigation. For this reason, northern Canada, the Sahara Desert, the Tibetan Plateau, and the Mongolian Desert are white in Figure 3. Other reasons a grid-cell might have less than 5% cultivation include remoteness, lack of infrastructure and market access, or status as a protected area. Such grid-cells might be suitable for agriculture from a biophysical perspective, but given these other reasons, they are not cultivated in either scenario. Examples of such regions, also in white in Figure 3, include the Amazon and Congo River basins.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 3

a. Hectares Conserved; b. Hectares Lost

Having looked at global estimates, we now consider particular regions: the U.S. Midwest and Southeast Asia, in Figure 4. Like the global map, the regional maps of the Midwest and Southeast Asia show a pattern of cropland expansion within and at the edges of currently intensively cultivated areas. While preserving more carbon on balance, this expansion is hardly environmentally benign. The expansion of cultivation in Nebraska and western Kansas is into more arid areas, most of which will require irrigation and some of which overlay the Ogallala Aquifer of fossilized water (see Hornbeck and Keskin 2014). Hence, carbon storage changes will come with implications for groundwater stores and may not be sustainable over longer periods of time. In Southeast Asia, expanded carbon-preserving extensification will occur mainly in existing river deltas, such as the Mekong in Vietnam, but may also require greater cultivation of uplands and associated erosion, as well as diverting water and loss of natural ecosystems.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 4

a. U.S. Midwest Hectares Conserved; b. U.S. Midwest Hectares Lost; c. Southeast Asia Hectares Conserved; d. Southeast Asia Hectares Lost

Table 2 shows various global alternatives of mixing extensification and intensification to meet a 70% increase in calories by 2050. The table shows the implications of a proportional increase versus a carbon-preserving approach to cropland expansion (columns 3 and 4) in terms of the millions of metric tons of carbon lost on a global scale. Net carbon saved through carbon-preserving expansion, based on MCC, compared to proportionally increased extensification, is shown in columns (5) and (6), first in millions of metric tons and then in monetary terms.

The monetary figures in column (6) were calculated using a social cost of carbon of $181 per metric ton of carbon (or roughly $50 per ton of carbon dioxide). This valuation was based on the mean value of the fitted distribution of Tol (2009) using a 1% pure rate of time preference, adjusted to 2012 dollars. If the pure rate of time preference is assumed to be 0%, the social cost would be $221 per ton, and at 3%, $75. At $181 per metric ton, the value of carbon saved by pursuing carbon-preserving rather than proportionally increased extensification from 2000 to 2050 amounts to $1.32 trillion, using the FAO’s assumption of a 20% expansion in global cropland by 2050 (Alexandratos and Bruinsma 2012). This figure represents the present value of all future damages from a one-time conversion in land use in the present period (following Pearce 2003).

V. DISCUSSION

The application of geographic and spatial elements in economics has a long but uneven history. Von Thünen (1826) laid the groundwork for a theory of “central place,” later developed by Hoyt (1939) and other economic geographers. Marshall was especially interested in the role of geographic concentrations of specific industries, or “industrial zones” (Henderson 2003). Hotelling (1929) was interested in the economic quandaries of physical space and the location of firms. Krugman revisited Marshall’s analysis of location in the context of trade theory (Fujita and Thisse 2009). Recent advances in remote sensing and data collection have vastly increased the amount of spatially explicit biophysical data (e.g., data on deforestation from Hansen and Naughton 2013). Such data are now being used by economists (e.g., Plantinga 2015). In short, there has long been an affinity in economic theory and practice linking space and geography to economic outcomes.

The capacity to bring “big data” to bear from remote sensing satellites now allows analysis that is both global and has fine-scale resolution to “zoom in” on more specific regions and areas of interest (see Cukier and Mayer-Schoenberger 2013). Although high-resolution geospatial data are now used frequently in economic analysis, they often are used only to generate parameters or nonspatially indexed variables that are then applied in nonspatial or regional models. The approach we define preserves the high-resolution nature of the data throughout the calculations by defining cost and production functions specific to each grid-cell.

The cost and production functions we defined have several limitations. The cost function includes only carbon storage loss and does not include other environmental costs. For example, we do not include nitrous oxide (N2O) emissions from agriculture, another major contributor to climate change. After deforestation for cropland expansion, N₂O is the single largest contributor by farming to GHG emissions, releasing an estimated 2.8 billion metric tons on a CO₂-equivalent basis into the atmosphere in 2005 from synthetic nitrogen fertilizer and animal manure (Bellarby et al. 2008). We lacked global data on nitrogen emissions of sufficient resolution and quality to include in the analysis. We also do not include other environmental impacts (e.g., biodiversity loss, option value of preserved areas, water- and soil-related services such as sediment retention).

Our cost minimization model uses calories produced rather than a measure of welfare or profit. We lacked spatially explicit global data on production costs, such as for machinery, labor, or transportation, required for measures of profit or welfare. Unlike nitrogen, however, part of the information on input costs is already included in our optimization framework because we base future agricultural possibilities on observed yields and cultivation locations. The farmer-level decisions that affect yield and cultivation location are in part driven by input prices, and thus their choices partially reflect these economic linkages. Limiting our analysis to implicit input costs reduces our ability to consider counterfactual situations in which future input costs are markedly different from those that led to observed behavior. Yet we believe that identifying the general locations and distribution of carbon and agricultural trade-offs remains important to land use decisions even with these omissions. We hope that future work will show how trade-offs change when underlying input costs change or when the carbon-preserving scenario is contrasted with a “business as usual” scenario based on robust land use change modeling with endogenously updating prices (e.g., Lawler et al. 2014).

Our model does not consider the environmental costs of intensification, but intensification will itself involve substantial additional GHG emissions. Increasing yields, especially on less productive croplands, or “closing the yield gap” (Foley et al. 2011) would require greatly expanded use of nitrogen fertilizer. Only 17% of the 100 million tons of nitrogen fertilizer applied in 2005 was actually taken up by plants (Scherr and Sthapit 2009). Nitrogen leaches into groundwater and surface water and causes “dead zones” at river mouths, as well as being released into the atmosphere as N₂O, a GHG with 310 times greater warming potential than CO₂ per unit (Conway 2012). To the extent that yield increases can be achieved through other methods, such as improved seed varieties or precision farming, these intensification-related damages will be less substantial.

Some of the assumptions made in the definitions of our scenarios are necessarily simplified to allow for a global analysis, such as that a proportional increase represents a useful counterfactual scenario. For these reasons, we do not make policy conclusions specific to particular grid-cells, but instead analyze the broader geographic trends of the results. Our assumptions are warranted insofar as they enable identification of the general locations of agriculture and carbon trade-offs, which are robust to a wide variety of alternate specifications (see the Supplemental Information section by Johnson et al. [2014] for an indepth discussion of this point).

VI. POLICY IMPLICATIONS

From an institutional and policy perspective, the question is how global society can better conserve carbon while expanding cultivation to meet the food needs of the future. Well-designed and carefully implemented international and national policies and programs have the best chance of influencing outcomes over land use choices. This global collective action problem can be solved only by internationally, multinationally, and nationally coordinated actions, requiring institutional innovations and leadership at each level, as well as careful and targeted land conversion choices by individual farmers and landowners (Dietz, Ostrom, and Stern 2003). This is a very tall order.

One thing is clear: there are consequential trade-offs in achieving agricultural expansion to meet increased food demand. At the high-extensification end of the spectrum, burdens will be placed on croplands and water resources, accompanied by losses of forest and grassland habitat, animal and plant species, and carbon storage capacity as well as other ecosystem services. At the high-intensification end of the spectrum, by contrast, more fertilizer and other agrochemicals will be applied to existing lands, along with increases in irrigation, seeding, and agricultural equipment in order to raise yields. These loadings on the land will raise levels of surface water pollution, groundwater depletion, and N₂O emissions from nitrogen fertilizer.

Thus, at any point on the spectrum, we confront choices in which losses in human welfare occur due to either caloric shortfalls or carbon losses and climate change or other environmental issues (see Calabresi and Bobbit 1978). The trade-off can be mitigated by carbon-preserving extensification but will be severe with proportionally increased extensification. By being more selective in these choices and expanding onto lands where carbon losses are lowest, policies and incentives can have an effect. These would encourage land use for agriculture in ways that are more consistent with agricultural production and carbon storage in different parts of the globe, and reduce the worst excesses of agricultural disruption in relation to carbon loadings and climate change.

There are reasons for guarded optimism with respect to carbon storage. In Brazil, deforestation in the Amazon over the last 40 years has cleared almost 20% of the rainforest, an area twice the size of Germany (Johnson and Jelmayer 2014). An area equal to Massachusetts was cleared in 2004. Since 2004, when it reached a peak, the rate of deforestation has been reduced by a remarkable 83%. Brazil has succeeded by creating new protected areas, relying on the rule of law and aggressive enforcement of existing land use regulations. In particular, the enforcement of Brazil’s 1965 Forest Code, which sets strict limits on deforestation particularly in the Amazon, played a major role (Tollefson 2013). As Hansen and Naughton conclude (2013, 713) in their study of the Brazilian Amazon: “Ecologically, agricultural strategies based on farming the most productive land could have positive effects for both the maintenance of tropical biodiversity and atmospheric carbon sequestration.”

Brazil’s success at drastically slowing deforestation in the Amazon was enhanced by the Amazon Fund in 2008. Norway was the major donor to the fund, with a commitment of $1 billion. Since 2008, Norway has also committed $1 billion to Indonesia to prevent deforestation. A broad program based on the concept of such payments could be used to direct cropland expansion toward areas with low MCC values, where the carbon losses are low relative to caloric gains.

At the international level, a broad framework was established in 2008 by the United Nations for a program on Reducing Emissions from Deforestation and Forest Degradation in Developing Countries, or REDD (United Nations 2010). Funds come from developed nations to developing countries to compensate them for preserving forested lands. REDD payments are essential for ecosystems services, especially the storage of carbon by tropical forests (see Robalino and Pfaff 2013). Busch (2013) found that expanding REDD payments to cover biodiversity preservation, as well as carbon storage, has the potential to provide greater climate benefits (p. 655). Yet, many of the details remain to be worked out, including whether transfers are to be between governments or through some type of carbon-offset market. Such details are likely to be clarified only in the negotiations of a “hypothetical future treaty on climate change” (Tollefson 2013, 6). The concept of MCC can be used in frameworks such as these to locate where incentives or payments for ecosystem services should be targeted to increase conservation efficiency.

In conclusion, the need to move agricultural land use decisions in the direction of greater sustainability is arguably one of the greatest challenges facing the future of humankind. Understanding the marginal carbon costs of land conversions allows us to incorporate one aspect of sustainability into land expansion. We have shown what agricultural increases and carbon storage protection are possible if such decisions select appropriate land areas for expanded cultivation. But translating what is possible into what actually occurs involves economic incentives, institutions, and choices by landowners, farmers, and governments. The real challenge is to use the data and conclusions that we have developed to structure these new incentives, institutions, and individual decisions.