Past and present biophysical redundancy of countries as a buffer to changes in food supply

The potential agricultural production with available renewable water resources (WA) was calculated at the national level using six combinations of water availability, water productivity and water reserves for environmental flow requirements (table A.1, equation (A.1)). Using the AQUASTAT database (FAO 2015b), we subtracted the total water withdrawal as sum of municipal, agricultural and industrial withdrawal—(TWW) from the total actual renewable water resources of each country (TARWR). TARWR includes internal and external (coming from other countries) surface and groundwater resources. In some scenarios we additionally subtracted the amount of water reserved for environmental requirements (EFR) and the water that is difficult to access due to high variability of precipitations (S). EFR was represented as either 36% or 57% (Gerten et al 2013) of TARWR. For S we calculated two scenarios, the first one assuming that the variability of precipitation, including concentration of rainfall in one season, has no influence for water accessibility. Thus, water availability was not reduced in this scenario. In the second one, this term has values of 10%, 20% or 30% of (TARWR–TWW) depending on the average seasonal precipitation variability of the country. This means that in this scenario we assume that, in regions with highly variable precipitation, water is more difficult to store, manage and access, and water availability is accordingly reduced by 10%, 20% or 30% depending on the coefficient of variation (see section A.1.1 of the SI for more details).

The resulting water availability (in m³) was then multiplied by the area-weighted (rainfed and irrigated) water productivity (WP in kcal m⁻³) in the country for each year as simulated by the agro-ecosystem model LPJmL (Bondeau et al 2007, Rost et al 2008, Fader et al 2010, 2015, Schaphoff et al 2013, Waha et al 2013) for the main groups of crops worldwide. This yields the potential caloric production with available water resources. In order to account for the possibility that WP of unused areas is lower than in cultivated areas, we also calculated an alternative scenarios assuming that WP is 30% lower (see section A.1.1 of the SI for more details).

The potential agricultural production of fertile, spare land (LA) was calculated at national level using six combinations of land available and agricultural productivity associated with it (table A.2, equation (A.2)). From the total area of the country (TA) we subtracted unsuitable land (NS) and land estimated for settlements and infrastructure (NAG). TA, NS and NAG were extracted from the GAEZ (Fischer et al 2002). The result represents the extent of very suitable, suitable, moderately suitable and marginally suitable land, taking into account climate, soil, elevation and terrain constraints. We then subtracted the agricultural area (LU) from HYDE 3.2 (Klein Goldewijk, personal communication, 2015). We calculated two scenarios, one with LU including only cropland (i.e. not comprising managed grasslands), and one with LU including cropland and managed grasslands (see more details in section A.1.2 of the SI).

In some scenarios we additionally exclude protected areas (IUCN classes I and II, from UNEP-WCMC (United Nations Environment Program-World Conservation Monitoring Centre) 2007) and areas worthy of protection (the union of Greenpeace's Intact forest landscapes and WRI's frontier forest, see Greenpeace International 2005, and Bryant et al 1997).

The resulting land availability (in ha) was multiplied by the average yield of the country ($\bar{Y}$ in t ha⁻¹) reported by FAOSTAT (2015). This yields the potential caloric production with available water resources. In order to account for the possibility that actual yields are lower on unused areas, we also calculated an alternative scenarios assuming that $\bar{Y}$ is 30% lower (see more details in section A.1.2 of the SI).

Potential calorie production due to productivity increases on cultivated areas (YG) was calculated by the difference between potential and actual yields, divided by the number of crops grown in a country, and multiplied by the harvested area (see equation (A.3) and section A.1.3).

Potential productivity increases on unused areas (YG_E) were calculated by multiplying spare land (see section 2.2) by the difference between potential and actual yields (section A.1.4 and equation (A.4)).

Actual yields and harvested area were taken from FAOSTAT (2015), potential yields from Mueller et al (2011). Mueller et al (2011) estimated potential yields by using data from other regions with similar precipitation and growing degree day characteristics (i.e. comparable climatic conditions). In some scenarios we assumed potential yields to be as high as on used areas (and thus equal to the values in Mueller et al 2011). In other scenarios we assumed them to decrease as a function of actual yields, i.e. potential yields are lowered by 30% of actual yields. The assumption in the latter case is that countries with high actual yields are already using the most productive areas, while countries with low actual yields might have high productive areas as part of their spare land (see table A.3 for more details on the yield gap scenarios).

The resulting potential food production from the former sections was divided by population numbers from FAO (2015a). The results were then divided by the amount of food production need per capita for one year (3000 kcal cap⁻¹ d⁻¹ including food waste; see Rockström et al 2009a, Kummu et al 2012, FAO 2012).

To summarize our data, we use the following classification: countries that have redundant resources for producing the caloric nutritional needs for at least 50% of its population for one full year were considered to have high water, land or productivity redundancy (see table 1). Countries with values between 25% and 50% were considered to have limited water, land or productivity redundancy. And countries that have redundant resources for producing the caloric nutritional needs for less than 25% of its population have very low water, land or productivity redundancy. Note that this refers to potential calories production in addition to current production.

We quantified the overall biophysical redundancy in three scenarios (R_low, R_middle, R_high) that combine different scenarios of the scaled values of LA, WA, YG and YG_E (see tables 1 and 2).

All three scenarios follow the calculation scheme of equation (1):

$$\begin{eqnarray}&&{R}_{x}\;=\;\displaystyle \frac{{\rm{min}}({\rm{LA}},\;\;{\rm{WA}})+{\rm{YG}}+{\rm{YG}}\_{\rm{E}}}{nInd},\end{eqnarray} \tag{ 1 }$$

where nInd is the number of terms in the nominator of the equation for which there is data in the different datasets and scenarios used. For countries with a full set of data nInd is equal to 3.

The minimum values of LA and WA were taken under the assumption that water and land are complementary factors, i.e. the lack of one of them would render the presence of the other one useless in terms of potential food production.

The R_low scenario is designed to represent very constrained redundancy for all countries: managed grasslands are considered as used and thus not available, protected areas and areas worth of protection are assumed to be not available for agricultural production. Moreover, precipitation and seasonality of precipitation are assumed to reduce accessibility of water resources by 10%–30% (depending on the variation coefficient of precipitation) and environmental flow requirements are assumed to be high and met by every country. Additionally, water productivity on unused areas is assumed to be lower than on used areas (see table 2).

On the other end, the R_high scenario is meant to represent the least constrained redundancy: managed grasslands are considered available, protected areas and areas worth of protection are assumed to be available for production, potential yields of unused areas are considered to be as high as in used areas, precipitation and seasonality of precipitation are assumed not to affect the accessibility of water resources, water productivity on unused areas is assumed to be as high as on used areas, and no water is reserved for environmental flow requirements.

Finally, R_middle represents a scenario in the middle of the range: managed grasslands are considered available, protected areas and areas worthy of protection are assumed to be unavailable for agricultural production, spare areas are considered to be 30% less productive than used areas, potential yields are assumed to be lower on unused areas than on used areas, precipitation and seasonality of precipitation are assumed to reduce accessibility of water resources by 10%–30% (depending on the variation coefficient of precipitation), environmental flow requirements are assumed to be lower than in R_low but they are assumed to be met.

Results shown in the next sections correspond to the R_middle scenario, with a discussion on the differences between R_low, R_middle and R_high in section 3.4.

Figure 1 shows water, land and productivity redundancies, and the combined biophysical redundancy for the year 2012. Overall redundancy changes across different geographical areas affected by the different specific characteristics. In Asia redundancy ranges from limited to very low (figure 1(e)), with very low productivity redundancy being a major factor in South East Asia. In the Middle East very low water redundancy has a strong impact instead. Europe is particularly affected by limited productivity redundancy. Some American countries have very low biophysical redundancy due to both very low productivity and land redundancy. The African and South American tropics have mostly high overall redundancy due to high land, water and productivity redundancy on spare land. This is notable since in the medium scenario (R_middle) protected areas and areas worth of protection are considered unavailable for food production.

Zoom In Zoom Out Reset image size

Overall there are 75 (48) countries that could not feed at least 50% (25%) of their population during a year with redundant resources (figure 1(e)). Those 75 countries are home to 4.8 billion people or around 70% of the world population.

Biophysical redundancy has decreased over the last two decades in most countries (figure 2). Most of the exceptions to this pattern are former Soviet Union countries, where large agricultural areas were abandoned after the end of the Cold War.

Zoom In Zoom Out Reset image size

Not all decreases in biophysical redundancy are linked with category changes, i.e. changes in the redundancy classes. Figure 3 shows only countries with time series that have sustained negative trends in biophysical redundancy and underwent category changes. While eleven countries, many of them in Europe, have made a transition from high to limited biophysical redundancy (figure 3(A)), nine other countries, most in Asia and South America, went from limited to very low biophysical redundancy (figure 3(B)).

Zoom In Zoom Out Reset image size

These changes are caused by the change in the drivers that affect the subcomponents of redundancy (see equation (A.1)–(A.4)). The role of yield improvement is especially complicated because it diminishes the yield gap, and thus, the productivity redundancy. At the same time, it increases the potential production of food on spare land, which is also true for an improvement in water productivities. Figure 4 offers a more detailed look into the drivers of redundancy change for some selected countries with changes in categories towards very low redundancy. In all cases population growth was an important factor. Additionally, Vietnam, Ecuador, Mongolia and Honduras have a decrease in redundancy due to improvements in yields that caused a smaller productivity redundancy on used areas. This is combined with a complete depletion of spare land, and thus, no increase in the productivity redundancy on available areas (compare also figure 1). In Pakistan, depletion of water resources also contributes to the trend. Canada, despite a less significant improvement in yields, shows a positive trend in the last decade that is combined with high variability of land use.

Zoom In Zoom Out Reset image size

LIEs present an interesting case to examine because they have very limited financial resources to react to external or internal disruptions of food supply, due to their very low capacity to pay a higher price for agricultural commodities in the world market. Thus, redundant resources might be their only option for counteracting a disruption in food supply.

The World Bank designates a country as a LIE if it had a Gross National Income per capita of $1045 or less in 2014 (World Bank 2015). Currently, 28 countries of the world, mainly situated in Africa, are LIEs, with a total population of about 558 million people. Our results show that four LIEs (Haiti, Rwanda, Niger, Eritrea) currently have very low biophysical redundancy, and five LIEs (Burundi, Afghanistan, Nepal, Uganda and Somalia) have limited biophysical redundancy. Of these countries, Eritrea and Uganda went through a sustained transition from high to limited (Uganda) and to very low redundancy (Eritrea). These nine countries together have a total population of around 158 million people.

The rest of the LIEs consistently had high redundancy over the last two decades. While most of them have sustained high redundancy values, Sudan, Ethiopia, Chad, Malawi, Togo and Tanzania have negative trends in biophysical redundancy. Should these trends continue in future, they would surpass the threshold to limited redundancy in the next 50 years.

It is important to note that a decreasing redundancy could be an indication of productivity gains or expansion of agriculture and, thus, of a better caloric nutritional situation today compared to two decades ago. Indeed, the LIEs as a group have made a transition in food production from ∼2300 kcal cap⁻¹ d⁻¹ to ∼3400 kcal cap⁻¹ d⁻¹. Currently, there are 13 LIEs that produce less than 3000 kcal cap⁻¹ d⁻¹, as opposed to 18 of them in 1992. However, the picture at national level is heterogenous. Cambodia, Malawi, and Sierra Leone had considerable increases in domestically produced food, surpassing 3000 kcal cap⁻¹ d⁻¹ during the last two decades. In contrast, many other LIEs show no significant change in food availability per capita from domestic production, with Eritrea, Uganda, Sudan, Burundi, Zimbabwe and Somalia having negative trends in production per capita.

The overall redundancy of China, Canada, Russia, Japan, and some countries in Latin America and Southeast Asia is higher in the R_high scenario, mainly due to the consideration of protected areas and areas worthy of protection as available for food production (figure 5). Also, some Sub-Saharan countries and countries with monsoon precipitation have higher water redundancy in the R_high scenario due to the assumptions that precipitation variability and seasonality does not constrain the accessibility of water resources and that water productivity is as high on unused areas as on used areas (figure A.3). While both precipitation variability and water productivity have an influence on water redundancy, assuming lower water productivities has a stronger effect than assuming constraints in accessibility due to precipitation variability (data not shown).

Zoom In Zoom Out Reset image size

Not considering managed grasslands as redundant resources (R_low) has by far the largest influence on overall biophysical redundancy. It decreases the biophysical redundancy of Australia and most American countries to the limited redundancy level, and it drives Northern African, Asian, and European countries to very low redundancy levels (figure 5).

Comparing the water redundancy from the R_low and the R_middle scenario highlights the influence of assuming high instead of middle high environmental flow requirements (36%–57%) in addition to precipitation variability. This pushes countries of Eastern and Southern Africa as well as Niger and Nigeria to lower redundancy categories (figures 1 and A.3).

There are almost no differences in LIEs when comparing the R_midlle and R_high scenarios. Only Somalia shows a significant change, having high redundancy in the R_high scenario and limited redundancy in the R_middle scenario. However, there are substantial differences when comparing the R_middle and R_low scenarios: from nine countries in the limited and very low redundancy categories in the R_middle scenario, to 17 countries in the R_low scenario. These 17 countries are home to around 380 million people. This result indicates that the consideration or not of managed grasslands as redundant resources is very important in poor countries.

The percentage of world population living in 2012 in countries with very low and limited redundancy diminishes from 70% to 60% when going from the R_middle to the R_high scenario. However, moving to the R_low scenario increases the total to 92%, with 70% of the world population in countries with very low redundancy (35% in R_middle).

The world as a whole shows a decreasing trend in biophysical redundancy in the last two decades, with considerable differences among the R_low, R_middle and R_high scenarios (figure 6). Simple global means (i.e. average regardless of country size, population and agricultural area) have a much smoother trend than those means calculated by weighting country values by their agricultural area or population. In the R_middle scenario the simple mean and the mean weighted by agricultural area start in the high redundancy class and are approaching the 0.5 threshold of the limited redundancy category. When the global mean is calculated by weighting national redundancies by country population, the global redundancy of R_middle starts and stays within the limited redundancy class and has a slightly steeper trend. This indicates that countries with higher populations have lower redundancies and steeper decreases in redundancy through time. In the R_low scenario, the global values calculated by weighting with population just transgressed the threshold from limited redundancy to very low redundancy. This indicates that the role of managed grasslands in redundancy calculations is especially important in highly populated countries. It is worthwhile noting that in the present study we restrict attention to the last two decades. If this period were to be extended, greater variability and non-linearities in the biophysical redundancy trend may appear, possibly as a result of historical events such as population decimation through pests, rapid irrigation expansion and transitions connected to the green revolution.

Zoom In Zoom Out Reset image size

Our results are in line with former studies. Namely, our calculations suggest closing the yield gap on used areas could increase production by 4.32 + E15 kcal in 2011, an estimate very close to the 5 + E15 kcal from Foley et al (2011).

Our estimates of spare land are 2476 Mha, 1492 Mha, and 303 Mha for the R_high, R_middle and R_low scenario, respectively. The latter two are in very good agreement with the medium (1376 Mha) and low (317 Mha) estimates of Eitelberg et al (2015). However, they have higher values for the high estimate (3783 Mha), possibly because they included barren and sparsely vegetated land. Potential food production was not calculated in that study and, thus, cannot be compared with our estimates.

Our calculation of available water resources is 51 096 km³ (R_high), 26 655 km³ (R_middle) and 17 909 km³ (R_low). The high estimate is in good agreement with LPJmL simulations of water availability (48 292 km³, Fader et al 2013) and the value 55 375 km³ from Gleick (2000). The lower estimate agrees very well with the value of Gerten et al (2013) for 'accessible blue water resources': 16 300 km³ and is not far away from 12 500 km³, the estimate by Rockström et al (2009b).

Our results are the outcome of a flexible methodology that can be applied using other production data, different datasets of resources use and alternative values for caloric nutritional needs, including changes in the composition of diets. It is worthwhile noting that our use of a reference diet of a 3000 kcal cap⁻¹ d⁻¹ from crop products does not account for the fact that most human diets include animal products such as meat, milk, or egg, which are partly produced using feed, and require several calories of feed crops to be produced. Even though the use of animal calories can be easily accounted for in our framework (Davis et al 2014), here we refer only to staple food and other food crops because the reliance on animal calories is likely to drop in conditions of severe food shortage. Thus, the reference to a plant base diet is here used as a baseline to evaluate the number of people that could be fed in each country by the redundant biophysical resources.

Our study focuses on biophysical redundancy and thus intentionally disregards the influence of imports on food security. However, water and land use as well as the intensity of the agricultural management are influenced by the status and evolution of the agricultural and non-agricultural trade balance of each country. This indirect influence was implicitly included in our study by using historical data on production, resource use and agricultural yields. Additionally, in exporting countries, the area and water used for the production of export goods were not considered redundant resources. However, in situations of food shortage, countries might choose to stop exporting and use their natural resources for securing the food supply of their population (Jones and Hiller 2015). This has not being taken into account in this study, mainly because we consider that the revenues of exports are necessary for the national economies.

The relation of international trade, and especially trade liberalization, with food security, is a complex one. Not only due to the challenge of measuring food security, but also to the difficulty of attributing socio-economic developments to trade policies. While a full analysis of these issues is certainly out of the scope of this study, it is important to note that there is an ongoing debate both in political and academic spheres that determines policies that will, in turn, directly affect the biophysical redundancy level. And this is a good reason to include some lines on these issues here.

The neoclassical trade theory assumes a world with perfect competition, where no country or firm is capable of influencing prices, and full internalization of external costs is in place, thus ruling out environmental externalities by construction. Under this theory, trade promotes long-term economic growth, maximizes the total potential economic welfare and ensures efficient resource use, by allocating production to sites where the resources needed for that production are abundant (keywords here are 'comparative advantage' and the Heckscher–Ohlin theorem). FAO (2003) lists some studies that show empirical evidence on the relationship between economic growth and free trade, and other authors show evidence that international trade avoids higher levels of land use change (e.g. Fader et al 2011). FAO (2003) also points to the unrealistic character of the assumptions of the neoclassical theory and makes a strong effort to summarize the risks connected to liberal trade policies. This book concludes that there is a need for protection of vulnerable groups and of understanding the complex relationships between income, income distribution and access to food availability, at national and household level. Lambin and Meyfroidt (2011) contribute to the debate pointing to processes amplified by globalization that may increase deforestation and land scarcity, changing future biophysical redundancy. These processes include displacement of environmental costs to other territories, indirect land use change, increase in resource use due to new technologies that make that use more efficient and thus cheaper, and the influence of remittance of migrants on the change, intensification or expansion of agricultural activities (Lambin and Meyfroidt 2011).

However, trade does allows increasing food supply in times of national crisis by increasing imports (in countries that can afford them) and allows countries to sell products in moments of food surplus, increasing the financial resilience of some economic groups to a degree. But it also facilitates the propagation of shocks through higher connectivity, the coupling with the energy market, financial speculation in the international food commodities market, and a potentially harmful dependency on food produced in other countries that may, out of necessity or convenience, restrict food supply (Fader et al 2013, Suweis et al 2013, 2015).

Finally, foreign land investments, also called 'grabbed' land, could be used as a mechanism for artificially increase biophysical redundancy of the investor countries, decreasing correspondingly the biophysical redundancy of host countries. Currently, there are almost 37 Mha contracted land deals, of that, 17 Mha is in Africa (Land Matrix 2015). There is a clear need for better data on these deal's conditions, the destiny of products and the local impacts on accessibility and availability of natural resources.

Collectively, these results demonstrate that there is a clear need to go beyond the biophysical aspects. In this sense, this study paves the way for integrating biophysical redundancy with trade dynamics and socio-economic development indicators in order to quantify the overall resilience of countries to food supply changes.

Lambin (2012) presents an interpretation of Malthusian and Ricardian views over global productive land that is useful for the debate on resources scarcity and human technological capacities to overcome physical constraints. From a Malthusian perspective, land and water availability are hard constraints for population growth. As population grows over the carrying capacity of the Earth, violent conflicts and food insecurity are expected to increase, and welfare is expected to decrease. In contrast, a Ricardian view would propose that limits can be sequentially upgraded when marginal land and water use become increasingly profitable (Lambin 2012). Neither of these views would consider that resources could be intentionally excluded from production. Our findings suggest, however, that the time is ripe for asking whether there is a 'desirable' level of redundancy in biophysical resources that play a fundamental role in food production. That is especially true because eminent climate and socio-economic changes are likely to modify food supply and demand in the near future.

In this study, redundancy of critical resources for agricultural production has been introduced as a mirror concept to redundancy in engineering and ecology. As such, our concept of biophysical redundancy is equivalent to the 'stand-by redundancy' in engineering, i.e. equipment that does not take part in the process and is most of the time idle, until a failure of a system component occurs (King 1990). There are at least three points worthwhile mentioning in connection with this 'stand-by' character.

First, spare land and available water resources are not 'idle', in fact they play a major role in the provision of ecosystem services that are also fundamental for agricultural production (e.g. provision of pollinators, water purification, etc) (Godfray et al 2010). Also, agricultural intensification, on the one hand, might avoid conversion of natural areas to agricultural production, increasing 'spare' land, and thus, land redundancy. On the other hand, intensification increases actual yields, diminishing the yield gap and, thus, reducing the productivity redundancy. But most importantly, high or low redundancy does not imply environmentally friendly or unfriendly practices. When intensification is linked to mechanization and increases in chemical inputs and irrigation, it may have detrimental effects on both land suitability (salinization, water logging, soil degradation, soil compaction, etc) and ecosystem functioning (nutrient leaching, pollution, etc) (Montgomery 2008). Extensification, on the other side, is linked to increased greenhouse gas emissions, biodiversity loss and hydrological alterations (Gibson et al 2011, Tilman et al 2011).

Second, this study does not analyze the feasibility of using redundant resources in response to food supply changes nor recommends using them in cases of food supply shortage. It is worthwhile to note that the use of redundant resources can be difficult under some circumstances. For example, Alexandratos (2005) mentions the remoteness and prevalence of highly infectious diseases in some LIEs as impediments for land conversion. Also, some studies point out that yields gaps, besides being connected to physical factors, like slope and irrigation, are highly linked to market-related issues, management practices, labor force and inequality (Neumann et al 2010, Allouche 2011, Dietrich et al 2012). These factors might be as important as the availability of redundant resources in shaping the response to long-term food supply changes.

Third, in this study, countries with large areas of fertile, spare land, untapped freshwater resources, and a combination of low actual yields and high potential yields are assumed to have high redundancy. The reason for this would be that they have a buffer with the potential to compensate—to certain extent—internal or external supply shortcomings by increasing domestic production. However, it has to be noted that we refer to medium and long term changes, since the stand-by character of the resources discussed here does not allow for a response to sudden disruptions, shocks or price spikes. In this context, other factors like the diversity of domestic production before the shock and the availability of financial means for increasing imports may play a much more decisive role in maintenance of food security (Porkka et al 2013).