"World Agriculture Towards 2030/2050" a major report predicting global agricultural trends (Alexandratos & Bruinsma, 2012). It was produced by the economics division of the UN Food and Agriculture Organization (FAO). In its abstract the FAO authors make a prominent disclaimer. Its projections, they stress (both on p. i and p. 7), are not to be used for normative purposes; that is, their report is not a prescription of how the global food system should develop. It is merely an exploratory model; their most reliable projection of business as usual (Alexandratos & Bruinsma, 2012).
In all probability this disclaimer resulted from the intense global attention that its predecessor (FAO, 2006) received. This “interim report” was cited across the globe as claiming that the world must produce 70% more food by the year 2050. This 70% number (sometimes even adjusted to a “doubling”) was almost invariably recruited to bolster a number of technological modernizing agendas for agriculture, for in- stance, in the promotion of genetically modified crops. Thus the UK’s chief scientist in 2009 predicted an imminent “perfect storm” of climate change and food shortages (Beddington 2009). Similar analyses were repeated in scientific articles, in generalist publications such as the Economist magazine, and by agribusiness (Peekhaus, 2010;Tomlinson, 2011; Stone & Glover, 2011).
Thus FAO’s number was repeatedly taken out of context and presented as a grand challenge requiring special efforts or difficult compromises. That is, it was used nor- matively. Those citing FAO may not have said that:
The battle to feed all of humanity is over. In the 1970s the world will undergo famines—hundreds of millions of people are going to starve to death in spite of any crash programs embarked upon now. (The Population Bomb, Paul Ehrlich, 1968).
However, the implications were comparable.
But this is not merely a simple story of statistics being taken out of context. In 2016, the same FAO department described in more detail their modeling system (in 2012 it was renamed the Global Agriculture Perspectives System, or GAPS) used to derive the original prediction (Kavalleri et al., 2016). Drawing attention, in a clearly norma- tive fashion, to its newest quantitative prediction the authors wrote: “A key finding... is that world food production should increase by some 60% from 2005/07 to 2050” (Kavalleri et al., 2016, p. 1, emphasis added). By contradicting their colleagues’ pre- vious disclaimer, Kavalleri et al. raised the issue, since FAO is a stakeholder in the food crisis narrative, of whether the original disclaimer was sincere and whether more could not have been done to avoid the normative usages of FAO numbers.
The answer given in this chapter supports the analysis of Tomlinson that a “slide” is operating in many of the texts written by FAO, and that this slide is particularly problematic in texts written by FAO leadership (cited in Tomlinson, 2011). Given that FAO oscillates between normative and nonnormative statements, this slide might be better called a “shuffle,” but it embodies perfectly the central paradox of all quantita- tive models of the global food system, whether produced by FAO or by others. This paradox is that, though FAO modeling supposedly exists to “identify challenges in world food and agricultural sectors and to offer strategic policy perspectives” in an unbiased fashion (Kavalleri et al., 2016, p. 1), what GAPS does in practice is quan- tify food. This frames agriculture as primarily a question of production. The focus on production, no matter any disclaimer, is normative, because it marginalizes issues of poverty and access to food, ecological costs, and social costs. These are either unexamined or subsidiary. So while the titles of FAOs reports and models are broad, e.g., World Agriculture: Towards 2015/30, the focus is narrowly on the quantification of production, even though, according to the International Assessment of Agricultural Knowledge, Science and Technology for Development (IAASTD), the truly “key” questions swirling around agriculture are not about productivity (IAASTD, 2009). Productivity, concluded the IAASTD report, is a distraction. As Robert Watson, chair of IAASTD, told the press at its launch, in agriculture, “Business as usual is not an op- tion.” The real question of agriculture is: Can we feed the people without also feeding social and ecological disasters?
But farmer suicides and insect declines, salinization, dead zones, and the pollu- tion of water bodies and other consequences of dysfunctional agriculture are absent from World Agriculture Towards 2030/2050. Moreover, World Agriculture Towards 2030/2050 even fails, ultimately, to show that productivity per se merits specific mod- eling attention. First, because the model it describes predicts that any necessary pro- duction increases will be solved by business as usual. Second, it also predicts “Modest reductions in the numbers undernourished” by 2050 but this reduction is dependent on continued economic growth (i.e., increases in wealth); that is, not on agricultural production (Alexandratos & Bruinsma, 2012).
FAO’s previous iteration, World Agriculture Towards 2015/2030, had reached a vir- tually identical conclusion. It projected that out of a then total of 850 million hungry people, just 120 million would be lifted out of hunger if food production reached its target increase of 70% by 2030 (Bruinsma, 2003).
So, even according to FAO’s own models, increasing production does not solve hunger. This result mirrors the conclusion of Amartya Sen in his celebrated historyPoverty and Famines. When hunger and famine strike, he found, production shortfalls have virtually never been the cause (Sen, 1981). To many food system commentators this is settled beyond question (e.g., Lappé & Collins, 2015). But it is a finding that has nevertheless been disregarded by many, including FAO leaders (Tomlinson, 2011), who have instead commonly cited FAO in support of a scarcity narrative with its con- sequent need for a productivity focus (e.g., Conway, 2012). Thus the scarcity view, whose credibility rests almost purely on the findings of models like GAPS, finds, at best, only equivocal support there.
5.1.1 Global food models and projections
The purpose of this chapter, however, is to provide a quantitative critique of models like GAPS at the level of their underlying assumptions.
Unless otherwise noted the focus will be on FAO’s GAPS. This focus is specifically not intended to validate the quantification of food; indeed, quantification of food as calories and weight is detrimental to a full understanding of food systems. It is rather an acknowledgment that FAO’s work is the most prominently cited and that, encour- aged by FAO’s shuffle, the world has overwhelmingly interpreted this 70% number as normative. In 2012 FAO's prediction was updated to 60%, mainly to reflect a shifting baseline: we are now much closer to 2050 than we were in 2003 (Alexandratos & Bruinsma, 2012). So, at the risk of appearing to validate the general approach, it is on a purely quantitative level that these models are most transparently open to challenge.
Malthus, 1798 is considered to have made the first mathematical model of a food system. His simple projection concluded that exponential population growth would eventually outstrip linear supply growth. The basic form of his model, followed ever since, was to separate food supply from food demand (McCalla & Revoredo, 2001).
Besides FAO, institutions such as the International Food Policy Research Institute (IFPRI) have developed their own models (Robinson et al., 2015). In addition, spe- cial projects such as the Millennium Ecosystem Assessment, the Comprehensive Assessment of Water Management, and Agrimonde (2009) (a joint project of the French Institut Nationale de la Recherche Agronomique and the Centre de Cooperation Internationale en Recherche Agronomique) have extended the general method but with an emphasis on investigating specific questions, such as water constraints, climate impacts, and the effects of specific policy decisions (de Fraiture et al., 2007; Fischer et al., 1988; Rosegrant et al., 1995, 2001; Parry et al., 2004; Chaumet et al., 2009). All are intended to inform decision making. However, those more suited to exploring diverse potential outcomes are often called scenarios. These distinctions, along with some of the strengths and weaknesses of the models, have previously been reviewed by Reilly and Willenbockel (2010) and by Wise (2013).
What these reviewers note, above all, is that there is overall a strong degree of con- sistency among models and scenarios that there is no need for extraordinary measures to enhance production. To quote FAO: “from the standpoint of global production poten- tial there should be no insurmountable constraints” (Alexandratos & Bruinsma, 2012).
None foresees a classic Ehrlich-style crisis, unless they expressly incorporate in their scenarios some form of mismanagement. For instance, the Millennium Assessment has as one of its four scenarios “Order through Strength” (OS) that envisages low cooperation and high trade barriers. Under OS conditions there is no overall global food shortage but there is increased malnutrition and even civil war in parts of Africa.
This is a broadly reassuring conclusion, but it should nevertheless be qualified by the looming shadow of climate change (Battisti and Naylor, 2009; Nelson et al., 2010). Agreement that food production is unlikely to develop into a crisis situation (climate excepted) has not banished the alarmist narrative in the media, however (see, e.g.,Hincks, 2018).
Yet, there are grounds to suppose that even this convergence, which does predict a need for increased production, is excessively pessimistic. In 2011 researchers from the World Bank Institute proposed that the world already produced enough food for 14 billion people (Herren et al., 2011). This number is well above UN population pre- dictions, which are expected to reach 10–11 billion in 2050 and perhaps even decline thereafter (UN, 2017).
Moreover, models also contradict global food price trends. Before the 2007/2008 price spike caused by changes in US and EU biofuel policies (de Gorter et al., 2015), food prices had been declining at approximately 4% per year. Since that spike, prices appear to have returned approximately to that track. This long-term decline, across every sector of agriculture, suggests strongly that food supply significantly exceeds current food demand and that the gap is if anything widening. The exact extent of this excess is not clear but the 2017 FAO estimate for global cereal stocks is 762 million tons. This amount represents approximately one-third of annual global production (FAOSTAT). Thus, independent of any modeling, there are strong grounds for suppos- ing that even the most optimistic models are still pessimistic. They are overestimating demand or underestimating supply, or both. The overarching questions are: How does one reconcile low (and declining) food prices and persistent global commodity gluts with the projections, claimed by GAPS and other models, of the need to produce more food? Are the models flawed? If so, what are those flaws?
5.1.2 How flawed are food system models?
The use of highly complex models always raises many questions of how well they represent reality (Scrieciu, 2007). But food system models are especially complex, seeking as they do to integrate biophysical, social, economic, and institutional com- ponents. Thus a criticism sometimes made of such models is their use of calories as the measure of nutrition (e.g., Herforth, 2015). Both nutritionists and those seeking a more expansive definition of food security have pointed out that calories fall well short of the definition of food security adopted at the 1996 World Food Summit: “Food security exists when all people, at all times, have physical and economic access to sufficient, safe and nutritious food to meet their dietary needs and food preferences” (e.g., Burchi et al., 2011). Thus the achievement of caloric sufficiency may ultimately be irrelevant.
To frame this diversity of critiques the primary issues with quantitative models are sometimes divided into technical uncertainties, methodological uncertainties, and epistemological uncertainties (Funtowicz & Ravetz, 1990).
Narrower technical concerns include the “knowledge gaps and priorities” raised by Reilly and Willenbockel (2010) and also by Wise (2013). On this level these au- thors agree there exist very significant problems with data quality. In many countries that extends to quantifying even the most basic inputs of the models: poverty, GDP, water availability, even simple population. The problem of questionable data is high- lighted by the case of Ghana. Its national statistical agency announced in 2010 that it was revising all future GDP estimates upward by over 60%. This made Ghana into a lower-middle-income country literally overnight (Jerven, 2012). Such difficulties are acknowledged (though ultimately dismissed) in FAO’s Food balance sheets: a hand- book (FAO, 2001).
Problems of comparable magnitude also apply to modeling the relationships be- tween data points. According to Reilly and Willenbockel: “more work on the vali- dation of model components used in integrated assessment studies is required.” But these authors equally caution that validation is a two-edged sword; calibrating models to past experiences, especially in the presence of climate change and other poten- tial abrupt changes, introduces the problem commonly known as “overfitting.” Some of these “technical”-level difficulties are acknowledged by the modelers themselves (e.g., Bruinsma, 2003).
A yet further significant problem is that, because they come from many different nations, datasets in separate parts of the models are based on different scales, time peri- ods, and conceptual schemes (FAO, 2001). Methods to reconcile such disparities are not available, however. Unsurprisingly perhaps, having discussed these limitations, Reilly and Willenbockel conclude that “model outputs should not be misinterpreted as fore- casts with well-defined confidence intervals. Rather they are meant to provide quantified insights about the complex interactions in a highly interdependent system and the poten- tial general order of size effects” (Reilly & Willenbockel, 2010). This comment raises some important issues. The first is that this limitation seems to have generally eluded those who cite these numbers. Second, without confidence intervals no one—including the modelers themselves—knows what this “general size order of effects” is.
We can gain insight into how such uncertainties might affect the quality of predic- tions by examining an extended critique made by Thomas Hertel in his presidential address to the American Agricultural and Applied Economics Association (Hertel, 2011). A major assumption in FAO and other models, notes Hertel, concerns how they relate prices, demand, and supply. Focusing on FAO’s quantitative model (pre-GAPS) he notes that it assumed that agricultural supply hardly responds to higher prices. This assumption was introduced because measurements of how food supply responds to demand in agriculture have mostly been taken over the short term. The remit of these models is the longer term, however. Measurements taken over the long term suggest that the elasticity picture is very different. Hertel contends that if future growth in demand was sufficiently great for food prices to rise, then this would in turn stimulate supply. Thus higher agricultural prices can be expected to favor high yields, raise land prices (protecting existing land and bringing more into production), stimulate agri- cultural research, and reduce waste, and this is indeed what the longer-term evidence shows (Hertel, 2011). Even the declining trend in the growth of global crop yields, which according to FAO is a major determinant of future food availability, may be a function of price. To this end, Hertel quotes economist Robert Herdt: “the economics of substantially higher yields is not attractive” (International Rice Research Institute, 1979). In this connection, Hertel also quotes FAO economist Jelle Bruinsma: “given the right incentives, much of the increased demand for cereals and oilseeds in 2050 could be met using existing technology.” Hertel therefore concludes that the frequently noted long-term “slowing of yield growth may simply be due to a slowing of net de- mand growth.” And he summarizes: “it is not clear that the resulting models are well- suited for the kind of long run sustainability analysis envisioned here.”
To summarize Hertel: there is strong evidence that incentives acting on farmers and other decision-makers are key to explaining agricultural productivity, but since com- modity prices have been in long-term decline, FAO has been modeling a low-incentive system.
With this as background the next section is devoted to analyzing four additional assumptions underlying predictive models and using GAPS as the example.