Modelling Biofuel Production as an Infectious Growth on Food Production

Biofuel capacity or production as a fraction of food supply for three different cases, along with sigmoidal (ie logistic) projections, 1998-2018. Plum curves show US corn ethanol processing capacity in service or under construction as a fraction of ethanol potential of entire US corn crop. Brown curve shows actual production of US ethanol as a fraction of ethanol potential of US corn crop. Violet curve shows global biofuel production as a fraction of estimate of biofuel potential of entire global human food supply. Sigmoidal curves all have K = 1/3 (infection doubling time of three years), and cross the 50% line at 2008, 2010.8 and 2014.2 respectively. Sigmoids are scenarios, not forecasts. Actual biofuel production growth will depend heavily on oil prices and policy responses to increasing food prices. See text for sources and methods.

(Ed note: Stuart has been an important part of this team, but no, he is not “back.” It has just been more than six months since he wrote this article, and it seemed like it might be a good time to revisit it.)
[break]

Many people are aware that food-based biofuel production has had an influence on food prices. Many people also know that US ethanol production is growing rapidly and now using a noticeable fraction of the total corn supply. However, I’m going to argue that the situation in the near term is potentially more serious than is generally realized.

I will use a mixture of existing data, analysis of biofuel profitability, and simple modeling of biofuel production as an infection or diffusion process affecting the food supply, to demonstrate that there are reasonably plausible scenarios for biofuel production growth to cause mass starvation of the global poor, and that this could happen fairly quickly – quite possibly within five years, and certainly well within the life of the existing policy regimes. It doesn’t have to be this way, but unless we start doing things differently soon, the risks are significant.

This piece is very long, and I apologize for that. But I think it’s important – I’m coming to the view that biofuel growth is by far the greatest near-term challenge arising from the plateauing of global oil supply that we have experienced over the last two years.

I’m going to focus a lot on the US corn ethanol situation, because it’s where the pattern has developed the furthest, and it’s also where we have the best data. Then I will broaden out to look at the global situation where I think the same pattern is developing, but a few years behind. Let’s first look at global biofuel production just to orient ourselves.

This next graph shows (in the lower panel) the annual production numbers for ethanol and biodiesel from 1975-2006. Here, and throughout, I am going to use volumetric units of millions of (42 US gallon) barrels/day, for the convenience of those of us used to thinking about oil supply. However, be aware that the energy content of biofuels is lower than that of fossil fuels (ethanol has only about two thirds of the energy content per gallon that gasoline has, for example).

Major components of global biofuel production, 1975-2006 (bottom), with oil prices (top). Sources: Worldwatch Institute for biofuel production through 2005, and various sources for 2006 (1, 2, 3, 4). Oil prices are sourced from BP and are expressed in 2006 US dollars.

This graph shows the major facts. Beginning from a very small base in the 1970s, by 2006, biofuel production reached about 1 million barrels/day, a little over 1% of the roughly 85mbd of liquid fuel production in that year. Ethanol is the largest flow by far, but biodiesel has started to become important. Growth of both products in the last few years was explosive – that is a key fact at the core of the problem. Furthermore, that growth is correlated with oil prices (shown in the panel at top). When oil prices are high, biofuel production increases rapidly, but when oil prices are low, biofuel production grows more slowly or stagnates.

This next graph shows US ethanol production (almost all of which comes from corn) as a fraction of the global biofuel supply:

US ethanol versus global biofuel production, 1980-2006. Sources: US figures are from Renewable Fuels Association. Global figures from Worldwatch Institute for biofuel production through 2005, and various sources for 2006 (1, 2, 3, 4).

As you can see, the US is a major biofuel producer, but both the US and non-US supplies have been growing rapidly. The reason the US corn ethanol supply is growing so rapidly is a large number of new ethanol plants being built (as well as expansions of existing ones). Here are the numbers from the Renewable Fuels Association (they have a very helpful list of every plant in production or under construction):

US ethanol plants in production or under construction at year end, 1998-2007. 2007 numbers may be a month or two earlier than the end of the year. Sources: Renewable Fuels Association.

As you can see, there has been a huge construction boom in the last few years. Looking at 2006, when we had 317kbd of ethanol produced in the US, that was coming from about 100 ethanol plants. Thus the average ethanol plant only produces 3kbd of ethanol – miniscule by oil industry standards. The blue lines indicate the average time for the population under construction to become the population in production at various points along the way. That lag is only 1-2 years, very short by oil industry standards. I think these are important points – these plants are very small, and they can be ramped up quickly at quite modest levels of investment. Thus the biofuel supply can respond quickly to changes in profit incentives of ethanol plant operators. There’s quite a lot of these plants now, and a lot more on the way.

For me, always the most important thing about some flow is to understand how it relates proportionately to other relevant measures in the situation. And what I want to do now is relate US ethanol production capacity to the US corn crop. That requires converting bushels of corn (which the USDA National Agricultural Statistics Service keeps track of), to ethanol. The National Corn Growers Association has statistics for that, as follows:

Conversion efficiency of corn to ethanol over time. Sources: National Corn Growers Association.

As you can see, the efficiency of ethanol plants is rising slowly, but appreciably. Note that the NCGA has a theoretical estimate for 2014. I figure they are doing their jobs well, which is to be a trade group and aggressively promote the interests of their members, so in my extrapolations later, I used the linear extrapolation of the historical trend, rather than something based on their 2014 estimate (the red dot).

Let’s just pause a moment and figure out how much food we are talking about when we discuss bushels of corn, or gallons of ethanol. A bushel of corn is 56 lb (or 25.4kg) of corn. At about 8000 btu/lb we get 113120 kCal/bushel. Given the average human diet globally contains 2800 kCal/day (see figure below), 1 bushel represents 40 days worth of calories for a person (if that person eat only corn!). Thus at current conversion efficiencies of about 2.8 gal/bushel, the corn in a gallon of ethanol represents a shade over two weeks worth of food (again, all corn). A 15 gallon fuel tank of ethanol is thus 7 months worth of corn calories for one person. Of course, the American corn crop is mainly fed to animals, and after conversion to meat, eggs, or dairy at efficiencies in the range of 1/10 – 1/3, the 15 gallon tank of ethanol is more like 1-2 months worth of food calories for a person.

Anyway, given the USDA corn production statistics, NCGA conversion efficiencies, and RFA data for the amount of ethanol production capacity on stream or under construction, I made the following graph:

Capacity of ethanol plants at year end, in production and under construction (stacked on each other), together with total ethanol potential of the entire US corn crop (not stacked). Expressed in millions of barrels/day of ethanol. Sources: USDA National Agricultural Statistics Service for corn production, National Corn Growers Association for conversion efficiencies, and Renewable Fuels Association for ethanol plant capacities.

This graph completely floored me from the moment I saw it, and immediately suggested the conclusion of this piece. Let me try to develop the argument in stages though. The first thing to note is that the corn crop generally increases over time. This is due both to ongoing improvements in the yield/acre of modern industrial agriculture, but also increasing acreage planted to corn versus other crops (especially in 2007). The corn crop also fluctuates, both due to acreage decisions and the vagaries of the weather.

More importantly, the amount of ethanol processing capacity is growing much faster than the corn crop. Whereas in 2006, the US produced 317kbd out of a global total of 982kbd, once all the construction under way is complete, the US will be able to produce almost 1mbd by itself. Let’s focus in on the ratio of the ethanol processing capacity (both finished and under construction) to the corn crop. That looks as follows:

Capacity of ethanol plants at year end, in production and under construction, as a percentage of total ethanol potential of the entire US corn crop in that year. Sources: USDA National Agricultural Statistics Service for corn production, National Corn Growers Association for conversion efficiencies, and Renewable Fuels Association for ethanol plant capacities.

Starting out from 7% in 1998, the percentage of the corn crop covered by ethanol plant capacity in progress has now reached 37-38% of the corn crop.

Furthermore, the general shape of this graph is very familiar to me, and strongly suggests thinking of the process of ethanol conversion of corn as an infection process (or spread process, or diffusion of innovations process, to use various terms from different disciplines for more or less the same kind of thing). There are a large variety of processes in both the natural and social sciences which have the general flavor of something spreading exponentially in a finite setting, and then slowing down as it saturates the finite available capacity. An infection spreading through a vulnerable population of plants or animals is one classic example. The diffusion of a new product or innovation through a marketplace of potential users of that product is another. An invasive plant or animal species spreading through an ecosystem new to it is another case.

In the context of corn ethanol plants, the general idea is that if the existing plants are doing well and making money, there is a basis for building more of them. Because there are plants already operating successfully, there are a set of skilled employees, managers, and contractors that know how to build and operate ethanol plants. There are investors who are comfortable enough with the industry to risk their capital and are excited about the returns that building more plants might offer. There are farmers who are aware of the possibility of selling corn to ethanol plants if there was one close enough (or forming co-operatives to start their own). And there are marketing and distribution channels that know how to get ethanol sold to final consumers.

The larger the industry is currently, the more new plants it could potentially implement next year. (Its desire to do so will be heavily influenced by current profitability, but let’s return to that point in a few paragraphs). So when things are going well, a young industry naturally grows exponentially – the amount of new capacity each year is proportional to the existing size of the industry. Eventually, however, any industry tends to mature – something or other limits further growth. In most cases, it’s lack of further customers interested in the product. However, the corn-ethanol growth process faces another obvious limit, which is that it cannot convert more than 100% of the corn crop to ethanol.

To try to help your intuition for growth/spread processes, I’m including the following short video. The spread of a computer worm or virus through a population of vulnerable computers is another example of exponential growth in a finite situation that I am particularly familiar with, having being involved heavily in research on it a few years back. This example, made by collaborators of mine, shows the progress of the Code Red worm spreading across the globe in 2001 (the size of the red balls depend on the number of infection cases in each city):

Animation of spread of Code Red computer worm on July 19th, 2001. Source: CAIDA.

Notice the way nothing much seems to happen for a while, then the infectious agent seems to infect lots of cities at large scale very quickly, then slows down as it runs out of vulnerable computers to infect. That’s a classic property of exponential-spread-in-a-finite-system situations.

Looked at globally, computer worms infect cities, because that’s where the computers are. By contrast, ethanol plants infect areas with a lot of corn:

Location of ethanol plants onstream and under construction. Sources: USDA: Ethanol Expansion in the United States, plotting data from the Renewable Fuels Association.

Hopefully this suggests to you, as it does to me, the visual metaphor of little dots of red and pink mold growing in a Petri dish (yet another case of exponential spread in a finite system).

The simplest model of exponential spread in a finite system is called the logistic equation, which gives a simple sigmoid (S-shaped) curve. It’s called the SI model in epidemiology. I’m going to spare you the math, since it’s well discussed elsewhere. At the time of the Code Red computer worm, I happened not to be familiar with that piece of math, and I rederived the equation in the middle of the night as the worm was spreading and I was trying to predict how long it was going to take before it saturated (“saturated” meaning that it ran out of vulnerable computers to infect). I ended up with a graph like this:

Rate of infection attempts at one location on the Internet due to Code Red worm on August 1st, 2001. Blue line is data, and red line is logistic model. Sources: S. Staniford, V. Paxson, N. Weaver, How to 0wn the Internet in your Spare Time.

That particular infectious agent was a fairly simple-minded thing, and it followed the simple model very well indeed. Note again the pattern of a long period of very little sign of growth, then the rapid rise in the graph when most of the infections occur, and then the tail off as the worm struggles to find the last few uninfected computers amongst a sea of already infected ones. Once you are in the middle of that graph, things are going pretty fast. It’s this that leads, in infectious disease control, to the huge emphasis on quarantining the early cases. It’s so much easier to put a stop to an infection that hasn’t got a grip yet, versus one that has already gotten a good grip on a sizeable fraction of the vulnerable population and is now making new infection attempts in all directions at a huge rate.

Which is the perspective that I bring to 40%, as the fraction of this years corn crop that could be processed by the ethanol capacity under construction. 40% is well into the steep part of a sigmoid. Let’s take a look again at that graph of the ratio of ethanol capacity (producing and under construction together) to the ethanol potential of 100% of the corn crop. This time I’m going to add a sigmoid model extrapolated out into the future.

Capacity of ethanol plants at year end, in production and under construction, as a percentage of total ethanol potential of the entire US corn crop in that year, together with sigmoid model with K = 1/3 centered on 2008. Sources: USDA National Agricultural Statistics Service for corn production, National Corn Growers Association for conversion efficiencies, and Renewable Fuels Association for ethanol plant capacities.

Ok, the fit is a bit rough – clearly this ethanol plant spread process is a little more complex and noisy than the computer worm I just showed you. Still and all, I think this graph should set off pretty serious alarm bells. The fit does look like it’s capturing some of the important dynamics of this process, and it suggests that we’ll be using almost all of our corn crop for ethanol in 5-7 years. That’s not very far off. Should we believe it?

Let’s investigate further. One of the major departures from a straightforward logistic spread model is that the doubling time (or equivalently the growth rate) has varied significantly through the life of the process. Let’s look more closely at the changes in the growth rate of this ratio, along with oil prices again.

Bottom panel: capacity of ethanol plants at year end, in production and under construction, as a percentage of total ethanol potential of the entire US corn crop in that year (left scale), together with year on year change in that percentage (right scale). Top panel: oil prices (annual average in $2006). Sources: USDA National Agricultural Statistics Service for corn production, National Corn Growers Association for conversion efficiencies, and Renewable Fuels Association for ethanol plant capacities. Oil prices are sourced from BP.

I suggested earlier that the growth rate has a lot to do with oil prices, and I’ve made that more explicit in the graph above with the green lines. When oil prices spike up, a year or so later we have a new burst of ethanol capacity under construction (which then comes on stream 1-2 years after that).

(Note that the drop in the growth rate of the ratio in 2007 is largely a result of a 20% increase in the acreage put under corn from 2006 to 2007, due to the high demand for corn – this increase came almost entirely from reducing the acreage under soybeans and cotton. See p 18 here for details.)

You might argue that correlation isn’t causation, and this suggests that it’s important for us to assess the profitability of ethanol plants more carefully – clearly growth of the industry will have a lot to do with perceived profitability of ethanol plants, but do they actually get more profitable when oil prices go up, or is the low energy return of the ethanol process such that they don’t actually do any better?

Let’s start with the price of ethanol. I can’t find raw data in the public domain, but I did find this graph of rack prices in various locations over the last ten years. (The rack price is basically the wholesale price at regional distribution terminals).

Rack ethanol prices at various points in the country May 1997- May 2007. Source: California Energy Commission.

It helps to understand the relationship between ethanol prices and gasoline prices. I took the graph above and made the contents of it the background to my own graph of gasoline prices (on the same scale). That gave this:

Retail and rack gasoline prices, national US averages, and ethanol prices at various points in the country (background). Source: California Energy Commission for background ethanol rack prices, and EIA for gas prices (which are all grade, all formulation national averages)

The purple curve is a national average retail price (average across all grades and formulations), while the blue curve is the rack gasoline price. Clearly, ethanol and gasoline prices correlate fairly well (as one might expect, given that the main end use of ethanol is to mix it into gasoline). However, wholesale ethanol prices are often higher than wholesale gasoline prices. This is possible for two reason. Firstly, gasoline formulators are effectively required in many states to include ethanol in gasoline for oxygenation (to reduce tailpipe emissions of carbon monoxide). In particular, the spike of ethanol prices above gasoline in mid 2006 is likely due to the phaseout of MTBE (a groundwater polluting oxygenator that Congress decided not to shield oil companies from liability for). Secondly, formulators receive a 51c tax credit for each gallon of ethanol included in retail gasoline. This allows them to pay more for ethanol than for gasoline and still make money.

The fact that ethanol prices tend to strongly correlate with gasoline prices is suggestive, but we also need to understand the costs of making ethanol. I have relied here on the outstanding USDA 2002 Ethanol Cost-of-Production Survey. (We are only looking now at operating costs, not capital costs, ie the costs of running the plant and making ethanol, not the costs of building the plant in the first place – which at that time averaged about $1.50 for each gallon/year of capacity). Let me summarize the operating costs from that survey in the next graph.

Ethanol operating margin analysis for 2002. Source: USDA 2002 Ethanol Cost-of-Production Survey for cost data and and DDG and CO₂ revenue. California Energy Commission for ethanol prices.

This is rather complex, but let me try to explain the highlights. The left column represents a breakdown of the average costs per gallon of making ethanol. The largest item by far (blue) is the cost of the corn. The second largest item (yellow) is fuel to provide process heat in the plant. Generally, this has been natural gas in the past, but there is currently an ongoing shift towards using cheaper coal instead. The rest of the bars are all smaller – administrative expenses, enzymes, maintenance, etc, etc.

The right column of the graph represents the revenues for that gallon of ethanol. The bottom (pink item) is the revenue from selling the distiller’s dry grain (DDG) residue left over from fermentation, which is used as animal feed. The brown column takes us up to the lowest ethanol price obtaining at any time in 2002 (from the earlier California Energy Commission graph). The blue column takes us up to the highest price of the year. 2002 was not a great year to be making ethanol, with operating margins ranging from a scanty 25%, to negative (selling the ethanol for slightly less than the operating costs of producing it). This was because oil and ethanol prices were relatively low in 2002.

This explains why capacity growth in 2003 and 2004 fell back to essentially zero.

To extend this analysis further, we cannot rely on survey data, which has not appeared since the 2002 survey. However, with just a little modeling, we can get close. What I did was to take the 2002 cost structure and divide it into three components: the corn, the natural gas, and everything else. The corn and natural gas components I extended to other years by using corn and natural gas price data. The “everything else” component I assumed to be more slowly changing and I just inflated it at a fixed 2 1/2% annual rate. I think this will get us fairly close. My cost structure model then looks like this:

Ethanol production operating cost model, Jan 1997-October 2007. Source: USDA 2002 Ethanol Cost-of-Production Survey for 2002 cost data. Corn prices came from USDA NASS, with conversion efficiencies from National Corn Growers Association. Natural gas costs were indexed from 2002 using price data from EIA. Other costs were computed from 2002 data by inflating at 2 1/2%/year.

As you can see, the main impact on the cost structure of ethanol producers is the price of the corn, which is quite volatile – varying by a factor of two over the course of the last ten years. Natural gas prices have been less important as a factor, and I assume they will get moderated further over time by switching to the use of coal.

We can now take this cost model and look at the rack price of ethanol against it:

Ethanol prices at various points in the country, along with operating cost model (with DDG revenue subtracted from costs). Source: California Energy Commission for background ethanol rack prices, and USDA 2002 Ethanol Cost-of-Production Survey for 2002 cost data and DDG revenues. Corn prices came from USDA NASS, with conversion efficiencies from National Corn Growers Association. Natural gas costs were indexed from 2002 using price data from EIA. Other costs were computed from 2002 data by inflating by 2 1/2%/year.

In my mind, this makes pretty clear what is going on. Making food into biofuel was profitable in 2000-2001, with oil/gas prices high, so the industry started to expand. It stopped being very profitable in 2002, so the industry stopped growing. Then it became hugely profitable in 2004-2006, and we had an enormous wave of expansion which is still coming to fruition. However, that additional demand has backed up into corn prices, which have now increased. Thus margins are falling, and we will probably see a drop in the growth rate of corn ethanol capacity for a while. However, if oil prices go up much further, then there will be another big growth wave. This one will be starting from around 35% or 40% of the corn crop and going up from there. Clearly, that will drive another big round of corn price increases.

So at this point, corn prices are indexed to oil prices via biofuel arbitrage. There are lags and imprecisions in that linkage, but corn prices cannot fall too far below gasoline prices, or biofuel production will become very profitable and the industry will quickly grow to the point that corn prices are bought back into relationship with oil prices. Furthermore, the large displacement of soybean and cotton acreage to corn in 2007 suggests that this arbitrage is quickly extending to other agricultural commodities. I by no means think that last process is complete, but it has started.

That’s bad news because demand for oil is extremely inelastic, and the world is struggling to grow the supply of it at present, so over the medium term it seems fairly plausible that there will be further rises in oil prices. As we will see shortly, one can throw the entire global food supply at our fuel problems and still only make a modest impact on them.

Before we turn to the global situation, I want to make one last graph on corn ethanol. Taking the same cost model I just showed you, I made a graph that shows the cost of making ethanol as a fraction of the (retail) cost of a gallon of gasoline. In both cases, I subtracted the DDG revenue from the ethanol cost, but in one of the lines I also subtracted the 51c/gallon blending tax credit.

Ethanol production operating cost model as a fraction of retail gasoline cost, Jan 1997-October 2007. Green curve is cost net of DDG revenue. Plum curve is also net of 51c/gallon blending tax credit. Source: USDA 2002 Ethanol Cost-of-Production Survey for 2002 cost data. Corn prices came from USDA NASS, with conversion efficiencies from National Corn Growers Association. Natural gas costs were indexed from 2002 using price data from EIA. Other costs were computed from 2002 data by inflating by 2 1/2%/year.

As you can see making ethanol has been getting steadily more profitable, and the unsubsidized margins are recently getting comparable to the subsidized margins back in 1997. (The fit lines are exponential just to guide the eye to the trend – I have no great confidence in the extrapolation).

Let’s now turn to the global picture.

Last year, President Fidel Castro of Cuba alleged that plans by developed countries to power cars with biofuels risked starving up to 3 billion people. While I am no fan at all of communist dictators, I fear he might have a point here. I established above that biofuel profitability/growth creates an arbitrage between oil prices and corn prices. We will see that the same trends are going on globally. They aren’t as advanced, but the basic mechanism are going to be the same, and the growth rates are comparable. With fuel prices and food prices linked together, then the dinner tables of the poor are in a competition with the gas tanks of the global middle and wealthy classes. And we already figured out that a 15 gallon tank of ethanol is 7 months worth of corn calories for one person.

Let’s start with the UN Food and Agriculture Organization’s statistics on energy in the global diet:

Global food energy intake per capita, 1960-2002. Source: FAO.

As you can see, at least until 2002, the world has been getting better and better fed. This comes despite the global increase in population over the period:

World population, 1960-2005. Source: US Census.

So far, so good. Multiplying the food intake by the population, and noting that 1 kilocalorie is 4.184 kilojoules, we can derive the total energy content of the global human food supply:

Total energy in global food supply, 1960-2002. Source: US Census for population, and FAO for food intake.

Since the amount of land in use has been fairly constant, most of the increase in food energy over this period has come from steadily increasing crop yields.

A petajoule is 10¹⁵ joules, or about 278 million kilowatt-hours. Ok, so is 70 odd petajoules a lot, or a little? To answer that question, I’m going to compare the food supply to the global supply of liquid fuel via a notional conversion to biofuels. For the cereal portion of the human diet, I can straightforwardly apply the exact same conversion factors as for corn above (on the theory that a calorie of rice or wheat can be induced to make about the same amount of ethanol as a calorie of corn – a little more than 50% of the calories in the corn make it into the ethanol).

The rest of the diet is more complicated – it ranges from things like lettuce and celery that are probably poor prospects for biofuel feedstock, through things like potatoes and cassava which would probably do about as well on a per-calorie basis as cereals, and then to meat, eggs and dairy products which have, in many cases, been converted at low efficiencies from cereals. I’m not in a position at the moment to make a precise accounting of this, so I just assume as a rough calculation that these things will cancel out, and I directly translate 1 non-cereal food calorie to 1 ethanol calorie. That assumption could be off by a few tens of percent, but it wouldn’t make any difference to the overall conclusion if it was.

Given all that, I can estimate the volume of ethanol equivalent of the global food supply, and compare it to the actual liquid fuels. (Again, remember in these volumetric calculations that the ethanol barrels are really only 2/3 as good as the oil barrels).

Ethanol equivalent of human food supply compared to global liquid fuel supply, 1965-2002. Cereal and non-cereal portions of food supply are stacked, but fuel is not stacked on food. Source: US Census for population, and FAO for food intake. Liquid fuel numbers are from BP.

You can immediately see the problem here. The biofuel potential of the entire human food supply is quite a small amount of energy compared to the global oil supply – somewhere between 15-20% on a volumetric basis, so 10-15% on an energy basis. If you look at the rate of growth from the mid 1980s to 2000 (and it would be similar to 2005 but the graph doesn’t go that far), we were requiring about an additional 10mbd per decade. So if we continue to try to drive more at historical rates of growth, eg as the middle class in China, India, and other developing countries continue to build roads and get cars, while our oil supply is stagnant, we can only get about a decade or thereabouts from converting our entire food supply to fuel.

However, just because it’s not a very good idea globally, doesn’t mean it wouldn’t be profitable to the folks doing the conversion. Let’s look at the growth rates in global biofuel production, and compare them to oil prices.

Annual change in biofuel production, 1975-2006 (bottom), with oil prices (top). Sources: Worldwatch Institute for biofuel production through 2005, and various sources for 2006 (1, 2, 3, 4). Oil prices are sourced from BP and are expressed in 2006 US dollars.

Again, we see a mirror of the US situation – when oil prices are high, biofuel production growth rates respond very dramatically in a short time. When oil prices are low, biofuel production almost stops growing. With the increasing oil prices of recent years, global biofuel production is up by a compound annual growth rate (CAGR) of 23.8%/year from 2001-2006. This next graph shows extrapolations of global food supply (expressed as mbd of ethanol), extrapolated on the highly linear trajectory it’s been following, with biofuel production continuing to grow at 23.8%:

Biofuel production and energy equivalent of food supply, 1975-2018. Food is extrapolated linearly. Biofuel production is extrapolated at the CAGR of growth from 2001-2006 (23.8%/year). Sources: as above.

It appears that the biofuel production will be catching up to the food supply very quickly. Clearly, we are in the same exponential-growth-in-a-finite-box situation, again. It’s just earlier in the process than with the US corn ethanol situation.

To make this comparison clearer, this next graph shows three things. Firstly, I repeat the same ratio I showed you earlier (US ethanol capacity in production and under construction divided by total corn crop ethanol potential). I also repeat the same sigmoid I showed you before. For the global case, we don’t have capacity under construction estimates, just actual biofuel production, which I show as a ratio to the biofuel potential of the global human food supply. To help make the connection, I have put the US ethanol production on the same graph, as a ratio of the US corn crop ethanol potential again. The latter is just a couple of years behind the capacity build-out curve.

With the idea that the dynamics are roughly the same here, I’ve put the same sigmoid with the same basic doubling time as a projection for all three cases. I just shifted the time offset – the global production curve is offset about 3 1/2 years behind the US corn ethanol production curve.

Biofuel capacity or production as a fraction of food supply for three different cases, along with sigmoidal (ie logistic) projections. Plum curves show US corn ethanol processing capacity in service or under construction as a fraction of ethanol potential of entire US corn crop. Brown curve shows actual production of US ethanol as a fraction of ethanol potential of US corn crop. Violet curve shows global biofuel production as a fraction of estimate of biofuel potential of entire global human food supply. Sigmoidal curves all have K = 1/3 (infection doubling time of three years), and cross the 50% line at 2008, 2010.8 and 2014.2 respectively. Sigmoids are scenarios, not forecasts. Actual biofuel production growth will depend heavily on oil prices and policy responses to increasing food prices. See text for sources and methods.

The underlying idea is that both oil and cereals are global commodity markets. If it’s profitable to make food into fuel in the US, even without a subsidy, then it’s profitable elsewhere also – possibly more so given lower labor costs. So the basic growth dynamics are the same. The infection just hasn’t got as strong a grip on the whole globe yet, but it’s growing at similar rates.

I want to stress something here about the implications of the recent growth rates for the timing of the problem. Something growing at 25%/year growth doubles in three years. So in both the last two graphs with different extrapolations, you see global biofuel production hitting half the global food supply within about six or seven years. We’ll discuss in a moment what factors could stop that from happening, but first I just want to point out that these time constants render cellulosic ethanol irrelevant to the issue.

Cellulosic ethanol is what most most advocates of biofuels assume that the future will belong to. It is ethanol made out of the cellulose in various kinds of agricultural waste, fast-growing grass or tree crops, etc. In an abstract, in-principle, kind of way, it might indeed be possible some day to produce a lot of fuel this way, since current global consumption of about 8 gigatons of fossil fuel carbon is an order of magnitude smaller than the roughly 60 gigatons of carbon fixed by the world’s plants (net primary productivity). However, cellulosic ethanol is not commercially practical today, and there are reasons to wonder whether the transportation and material handling issues will be overcome soon. At the moment, there is a single pilot plant operating in the world at a non-commercial scale, and otherwise the technology is in the lab.

Let’s grant, for the purpose of discussion, that all the problems will get solved and cellulosic ethanol will get off the ground commercially a couple of years from now. It won’t have any meaningful impact on what happens with food-based ethanol. Remember the Code Red video, and the shape of the sigmoid curve? Remember how the infectious agent spends a long time quietly multiplying below the radar screen till it gets into the sharply rising part of the curve and seems to take everything over all at once? Cellulosic ethanol is at the very beginning of that long growth process. Food based ethanol is on the steep part of the curve already.

Ok. So this is all incredibly bad news. What could stop this process from continuing?

Well, I think there are three major possibilities worth mentioning. Again, the key point is that the spread rate of biofuel plants is controlled by the profitability of those plants. That in turn is mainly set by the difference between oil prices and food prices.

So, for possibility number one, if oil would go back down to $20 a barrel, that would certainly do the trick. There are people who continue to believe that the current stagnation in oil supply will end soon, and allow prices to fall. I’m not going to spend a lot of time on that possibility: we’re still waiting, alas. Those who would claim oil will go back to $20, or $35, or $40, or $60, are getting quieter and quieter as it passes the $100 mark. My own view is that we are on the bumpy plateau of global oil supply. I do not expect either large increases or large decreases in oil supply any time soon, though small increases and decreases are certainly possible. If that is correct, I expect oil prices to increase in the medium term, though certainly they could go down in the short-term if the credit crunch affects the global economy enough.

The second way that biofuel conversion of food could sharply slow is when food prices get high enough. This is certainly going to happen before 100% of the food gets turned into fuel. The question is, at what point? When we have a bidding war between the gas tanks of the roughly one billion middle class people on the planet, and the dinner tables of the poor, where does that reach equilibrium?

This is not an easy question to answer. The situation is unprecedented enough that it’s not easy to find good data with which to project the situation. Significant uncertainty remains, but I have found a couple of ways of making rough estimates, both of which produce similar answers.

One thing that probably puts a lower bound on the number of persons affected by large food price increases is the number of people who were already chronically hungry. The UN, as part of its Millenium Development Goals effort, has statistics on how many people currently cannot meet minimum dietary energy guidelines. Throughout the 1990s, that hovered around 800 million people:

Global population unable to meet minimal dietary energy requirements according to UN Millenium Development Goals Indicators.

Presumably, it remained a similar number, at least until the major food commodity price increases of the last couple of years. I wouldn’t claim to be very knowledgeable on this, but I struggle to imagine how someone who wasn’t meeting minimum dietary guidelines already can continue to exist on half as much food, or a quarter as much food, as food prices come into equilibrium with the current oil price level, or perhaps double again should oil prices double again. I would imagine that if you are hungry all the time you would already be devoting most of the skills and resources available to you to the problem of eating, and you would have limited ability to increase that in the face of large increases of food prices.

This still leaves the question of how many people who were able to meet their minimal dietary needs at historic food price levels might not be able to do so at doubled or quadrupled prices.

I managed to find some data for food consumption elasticities across a broad range of countries in a USDA study Cross-Country Analysis of Food Consumption Patterns by Regmi et al. The most important graph is the price elasticities:

Food income elasticity by country income according to Regmi et al..

The definition of the price elasticity is that it’s the ratio of the percentage change in quantity consumed as a result of a certain percentage change in price. For the low income countries in the sample, price elasticity is about -0.7. Thus a 10% increase in price would be expected to result in about a 7% reduction in food intake. It’s not clear that elasticities can safely be scaled up to very large changes in price, but if they could, a 100% price increase would imply a 70% decrease in food consumed, which would presumably create severe hardship or death by starvation for most people in poor countries (unless their income derived from growing food, and they had secure title to their land).

The definition of low-income country in the Regmi et al study is that it has less than 15% of US per-capita income. Per capita income in the US in 2000 was just a hair less than $30,000, so 15% of that is $4500. According to this global income distribution data,

Global income cumulative distribution according to Chotikapanich et al..

for which I’ve blown up the low end here,

Low end of global income cumulative distribution according to Chotikapanich et al.. Pink line represents the $4500 income level.

about two-thirds of the world’s population would fall into the low income category, and thus would apparently be extremely vulnerable to doubling or quadrupling of food prices. For another approach to the same thing, we can look at income elasticities (the ratio of the percentage change in food consumption produced by a certain percentage change in income):

Food income elasticity by country income according to Regmi et al..

Here the value for the lower-income 2/3 of the world’s population is about +0.7. What this means is that a 10% reduction in income has about the same effect on food consumption as a 10% increase in food prices. This suggests that we can use the global income distribution (shown above) to roughly estimate the impact of a doubling or quadrupling of food prices. We noted earlier that according to the UN about 800 million people are unable to meet minimal dietary energy requirements. That is 12% of the world population. On the income distribution (one graph back), the 12% mark corresponds to $1020/year in income (shown as the lowermost green dot). By looking at the $2040 level (36% of the global population – second green dot up), and the $4080 level (61% of the global population – third green dot up), we can estimate that a doubling in food prices over 2000 levels might bring 30% or so of the global population below the level of minimal dietary energy requirements, and a quadrupling of food prices over 2000 levels might bring 60% or so of the global population into that situation.

These estimates should be regarded as quite uncertain. Still, it seems hard to make a case that food price increases will cause a cessation of biofuel profitability before a significant fraction of the global population is in serious trouble. The poor will not be able to bid up food prices by factors of two and four and keep eating. In contrast, the quadrupling of global oil prices, and tripling of US gasoline prices, over the last five years has had very minimal impact on driving behavior by the middle classes.

The core problem is that gasoline price elasticity in the US is about -0.05, versus the -0.7 price elasticity for food consumption by poor consumers. This makes clear who is going to win the bidding war for food versus biofuels in a free market.

This brings me to the final thing that could stop runaway biofuel growth: public policy. So far, there has been a fairly broad coalition in favor of increasing ethanol production. This encompasses agricultural interests, environmentalists hoping to reduce carbon emissions and rely on a renewable fuel, and many citizens concerned about reliance on Middle Eastern oil supplies. The Renewable Fuels Association reported recently that 3/4 of Americans believe we should increase our reliance on ethanol. This kind of thinking has led to subsidies and mandates for biofuel production in the US, in Europe, and even in a number of developing countries.

My conclusion in this analysis is that this broad agreement is in fact mistaken. It is based on a failure to appreciate the speed with which high oil prices and profitable biofuel operations can fuel a very rapid growth of the industry up to the point that it consumes a sizeable fraction of global food production. This will have only modest benefits for global fuel supply, but will cause massive abrupt global hardship in poor countries. Many unforseeable consequences may follow from that.

I suggest we reconsider our policy.

Greeen (left scale) monthly spot price of West Texas Intermediate crude oil, expressed in $2005 (CPI deflated) per barrel. Plum (right scale), number of barrels of WTI crude purchasable by forty average hours of private industry wages, pre-tax. Source: EIA for crude prices, BLS for CPI index, and BLS via Alfred for average hourly wages. Dashed lines are extrapolations of exponential fit from Jan 2002 on for illustration of trends only. These are not predictions, and the basis for assuming future trends will be similar to past ones is weak.

[break]

On August 23rd, 2005, shortly after the publication of Twilight in the Desert, New York Times columnist John Tierney announced a bet with author Matt Simmons on the future price of oil:

I don’t share Matthew Simmons’s angst, but I admire his style. He is that rare doomsayer who puts his money where his doom is.

After reading his prediction, quoted Sunday in the cover story of The New York Times Magazine, that oil prices will soar into the triple digits, I called to ask if he’d back his prophecy with cash. Without a second’s hesitation, he agreed to bet me $5,000.

His only concern seemed to be that he was fleecing me. Mr. Simmons, the head of a Houston investment bank specializing in the energy industry, patiently explained to me why Saudi Arabia’s oil production would falter much sooner than expected. That’s the thesis of his new book, “Twilight in the Desert: The Coming Saudi Oil Shock and the World Economy.”

I didn’t try to argue with him about Saudi Arabia, because I know next to nothing about oil production there or anywhere else. I’m just following the advice of a mentor and friend, the economist Julian Simon: if you find anyone willing to bet that natural resource prices are going up, take him for all you can.

After reprising the history of the famous bet between Paul Ehrlich and Julian Simon, the actual terms of this new Simmons-Tierney bet were detailed further down the column:

I proposed to him a bet using what Julian considered the best measure of a resource’s value: how it compares with the average worker’s wage. I offered to bet that the price of oil would not rise faster than the average wage, meaning that future workers would be able to afford oil more easily than they could today.

Mr. Simmons said he favored a simpler wager, based on his expectation that the price of oil, now about $65 per barrel, would more than triple during the next five years. He said he’d bet that the price in 2010, when adjusted for inflation so it’s stated in 2005 dollars, would be at least $200 per barrel.

Remembering a tip from Julian, I suggested that we use the average price for the whole year of 2010 instead of the price on any particular date – that way, neither of us would be vulnerable to a sudden short-term swing as the market reacted to some unexpected news. Mr. Simmons agreed, and we sealed the deal by e-mail.

We are now close to the half way point on this bet, so how is it looking?

To assess this, I constructed two measures – one is the measure on which the bet will actually be decided – oil prices in 2005 dollars. Tierney’s column doesn’t define exactly which oil price, or how to deflate it, but simple choices are to use West Texas Intermediate (WTI) oil prices (from the EIA) and correct for inflation with the BLS’s CPI-U index.

In addition to this, I also looked at a metric to measure what Tierney was originally trying to propose – how much oil can be bought with a given unit of wages, which he said should increase over time (people should become worth more and more, relative to oil). My implementation of that was to take average hourly wages in private industry, multiply by forty, and then see how many barrels of oil that would buy (ie how many barrels of oil does a week’s worth of gross pay buy).

Furthermore, I noted in June 2006 that the run up in oil prices since the beginning of 2002 was exponential in form, and this is still roughly true. So I fit an exponential to both metrics and projected it out for a few years. The average doubling time in price is a shade over three years at present for nominal prices, and about 3 1/2 years for real prices. Obviously, an exponential increase in oil prices cannot continue forever (too many doublings and people would be spending all their income on oil), and I have no real idea when it will stop. Thus these extrapolations are just to be taken as what happens if current trends continue, not an unconditional assertion that they will continue.

With those caveats out of the way, here’s the data:

Greeen (left scale) monthly spot price of West Texas Intermediate crude oil, expressed in $2005 (CPI deflated) per barrel. Plum (right scale), number of barrels of WTI crude purchasable by forty average hours of private industry wages, pre-tax. Source: EIA for crude prices, BLS for CPI index, and BLS via Alfred for average hourly wages. Dashed lines are extrapolations of exponential fit from Jan 2002 on for illustration of trends only. These are not predictions, and the basis for assuming future trends will be similar to past ones is weak.

So the most important observation is that, right now, Simmons is on track to lose the bet. The current trajectory of oil prices do not take us to $200 (in 2005 dollars) until sometime in 2012, assuming the trend continues. So things will need to hurry up and deteriorate faster in order for Simmons to win.

However, it seems to me important to look a little deeper. In a sense both men look wrong in light of the data of the last few years. Simmons looks too pessimistic – at least so far, oil prices are not increasing as fast as he presumably expected them to. On the other hand, if we look not at the final terms of the bet, but rather at what Tierney initially proposed, then Tierney looks much too optimistic. The oil value of a week’s work has not gone up, but instead has continued to fall rather sharply (real wages having been roughly flat, while oil is increasing rapidly). And while Simmons is quantitatively wrong, Tierney’s original proposal would seem to be qualitatively wrong – things are moving in the opposite direction from what he predicted.

Of course, there are still two years, nine months, and a couple of weeks to go before the end of 2010 when the bet will be settled for sure. Who knows what will happen in the intervening time. But the trends right now point to Simmons losing the bet by being right on the big picture, but overstating his case somewhat.

Added in Press

After I had written the piece to this point, on Saturday, I sent it to Matt Simmons and John Tierney to see if they had any comment. Only Simmons responded:

Good piece. This is also first time I re-read John’s column in a long
time. Here are a few observations I would add. At the time when Tierney
called, I obviously had no certain idea where crude prices would be in
2010, but thought the likelihood they would rise a great deal was very
high. To make the story simple, I picked $200.

If you take your same chart and ignore the slow rise until mid 2005, and
then take the times it shot up, or start trend line in 2007, the
trajectory gets you there in fine shape.

We obviously talking about far more than a fun $5,000 bet. If oil has
peaked, and the world stays in denial, there could be such social chaos
that it might be hard to even define what the price of crude even is.

More important is the question “Is $110 oil now priced right?” Answer is
also easy. No since this is still only $.17 a cup!

Simmons in essence is arguing that there’s still hope for him to win along the kind of trajectory I’ve marked in orange here:

Greeen (left scale) monthly spot price of West Texas Intermediate crude oil, expressed in $2005 (CPI deflated) per barrel. Plum (right scale), number of barrels of WTI crude purchasable by forty average hours of private industry wages, pre-tax. Source: EIA for crude prices, BLS for CPI index, and BLS via Alfred for average hourly wages. Dashed lines are extrapolations of exponential fit from Jan 2002 on for illustration of trends only. These are not predictions, and the basis for assuming future trends will be similar to past ones is weak.

It’s true the recent run-up is very rapid. I also think it has a somewhat different cause. During much of the 2002-2007 timeframe, we were approaching or in a plateau of global production. On the plateau, increases that would have occurred in demand (due to economic growth) had to be countered by increases in price. Since the price increases were about 25% annually, that suggests that the elasticity ratio (income elasticity/price elasticity) for oil had to be in the range of 5-7, so that 4%-5% global economic growth and flat oil supply could turn into 25% annual oil price increases.

Now, however, we are in a somewhat different world. It looks like there is at least a small bump in supply at the end of 2007, and the prospect of more in 2008 – maybe as much as a couple of million barrels/day, though it’s hard to be sure, still less precise. Given similar to recent trend GDP growth, this wouldn’t require as large a growth in oil price. And given a global recession, it might be expected to lead to much lower price growth, maybe even price falls.

However, what is happening instead is, as the credit bubble deflates rapidly, we have sharp falls in the dollar and negative real interest rates, sparking a rush to commodities. How long this trend will continue is probably anyone’s guess. The best hope for Simmons was perhaps raised by Paul Kasriel in a very important analysis last week, concerning the possibility of the failure of the currency pegs of the Saudi riyal and Chinese yuan

But, in our opinion, what could turn a walk on the dollar into a sprint would be a decision by
the Chinese and/or Saudi central banks to eliminate the pegs of their currencies to the
greenback. Now, what would motivate these central banks to sever the peg? The desire to rein
in their domestic inflation. In an environment in which the dollar is under downward pressure,
the by-product of pegging one’s currency is higher inflation in the economy whose central
bank is pegging.

The inflation mechanics are as follows. The pegging central bank has to buy U.S. dollars in
the foreign exchange market in order to prevent the dollar from falling against its currency.
The dollar-buying central bank purchases dollar with its own currency. The dollar-buying
central bank gets its own currency the same way all central banks get their own currency – it
figuratively “prints” it. The dollar-purchasing central bank therefore floods its economy with
its own base money, resulting in inflation – inflation in the prices of goods/services and
inflation in the prices of assets.

Recent trends in Saudi Arabian and Chinese consumer inflation. Source: Northern Trust

And certainly, the extraordinary events of this weekend, with the emergency acquisition of Bear Stearns by JP Morgan, with guarantees provided by the Federal Reserve, will have put further pressure on the dollar. Personally, I was already assuming that a number of large US financial institutions were going to end up insolvent as a result of the end of the credit bubble. Thus I found the news very pleasing, because it suggested a Federal Reserve able to take very decisive and rapid action to do what was necessary to maintain the functioning of the financial system (I believe a significant amount of nationalization of insolvent institutions is going to be required before we are through). However, judging by stock market prices – Bear was worth $50/share as recently as Thursday and sold for $2/share by Sunday – the market as a whole did not share that perception and has been surprised to the very negative by what just happened. This will further weaken the dollar, and put greater pressure on foreign central banks to pull their pegs.

And if the Saudi and Chinese pegs come undone, then maybe oil could get expensive enough in dollars for Simmons to win his bet after all.

Average United States yields per unit area for various crops, 1900-2007. Yields are expressed as a multiplier of the 1900-1935 average. Source: National Agricultural Statistics Service.

[break]

This post continues an exercise I began a month or so ago of trying to figure out how civilization could be moved to a mostly sustainable footing by 2050, while still being recognizable as civilization, and in particular allowing some continued level of economic growth between now and then, especially in the developing countries. Let me remind you of the parameters of the exercise:

• Population: The global population is able to grow and go through its demographic transition with death rates continuing to go down. No die-offs.
• Economy: The world economy is able to grow on average over the period – modestly in developed countries, faster in developing countries.
• Carbon emissions: The global energy infrastructure will be mainly replaced with non-carbon-emitting energy sources by the end of the period, and residual emissions will be rapidly diminishing.
• Fossil fuels: I assume that peak oil is here about now but that declines will be governed by the Hubbert model (and thus will be gradual). I assume natural gas and coal are globally plentiful enough that climate policy is required to prevent their full use.
• Technology: I do not assume any massive breakthroughs – no technological miracles that solve problems in ways completely unknown or untested today. However, where technological sectors have long established rates of progress in key metrics, I extrapolate the metric to continue improving at the historic rate (eg the economics of solar power, or the yields/acre of agriculture are assumed to keep improving on the historical trajectory).
• Impact on wild ecosystems. Developed countries are assumed to maintain the protections they currently have in place (for national parks, wildernesses etc). Developing countries are assumed to exploit their unused land up to the point of best current practices for developed countries. Whatever impact on ecosystems arises from climate change due to past carbon emissions and the tail of emissions to 2050 is viewed as unavoidable.
• Conservatism Other than the above, I use the overarching principle of trying to assume as little change in the way the world works as possible – I assume it remains a more-or-less free market world, in which national governments regulate their own countries to temper the worst excesses of the free market and periodically enter into treaties on the more pressing global problems. I assume it remains full of highly imperfect humans mostly struggling to improve their own circumstances. I assume people are willing to come together and take collective action for the common good, but only when the need for that action has become so overwhelming and immediate as to be irrefutable.

In Powering Civilization to 2050 I argued it was potentially feasible to transition to power civilization with a mix of solar, wind, and nuclear energy, with the transition well on the way to completion by 2050. (Luis de Sousa made a broadly similar argument in Olduvai Revisited 2008). This would require a period of belt tightening and conservation in the next couple of decades, but once the transition had overcome the critical threshold (as solar energy in particular became cheap), I suggested energy in general would get cheap again. I adopted the UN medium population projection which has population at about 9 billion by 2050, with growth slowing sharply. Making plausible assumptions for economic growth between now and 2050 if energy was available, we got to a world GDP of about $350 trillion in 2050 (in 2006 purchasing power parity dollars), versus about $70 trillion in 2007

If the average global citizen was significantly wealthier in 2050, they would undoubtedly want to drive more. The switch to primarily electrical energy sources for civilization would preclude doing this with all liquid fuels. In Four Billion Cars in 2050? I argued that, given that the average citizen will be living in a dense third world city by 2050, we can assume rates of ownership typical of the most car-free corners of western Europe at the moment (Holland), which gives rise to a few billion cars in 2050. I further argued that it seems feasible that this many plugin-hybrids could be built – there appears to be enough lithium for the batteries – and run on less than 10mbd of liquid fuels.

In this piece I want to look at another area that many people think is likely to be a critical bottleneck to civilization continuing – the area of food, agriculture, and soil. I am of course not an expert in these areas, but happily there is a lot of excellent scholarship and scenario building that I can lean on. My task is reduced to reporting of the existing science, with some modest adjustments to reflect where my assumptions differ from those of published scenarios (most especially the assumption of a near-term peak in oil supply, and a full-speed effort to convert society to carbon-free energy sources.)

Let’s begin with two very helpful UN Food and Agriculture Organization reports: World agriculture: towards 2015/2030, and the sequel World Agriculture: Towards 2030/2050. What these reports do is basically look at projections for population and economic growth and then estimate how much food people would want in the future, and what quantity of agricultural commodities would be required to fulfill that demand. The first report focusses a lot more on the supply-side factors of how this could be done, while the second report extends the analysis out further in time but confines itself much more to demand side considerations.

The input assumptions about population and world GDP are slightly different than mine, but close enough that I am just going to adopt their food scenario wholesale, rather than trying to construct my own from first principles. The differences would be small – much smaller than the other uncertainties in the problem. Let me first summarize their scenario, and then we will start to explore the potential bottlenecks that might prevent achievement of this much food production. (However, I strongly encourage readers that care about where their food is going to be coming from in the future to take the time and read the FAO reports themselves.)

Let’s start with a look at what the FAO scenario has for average nutrition. This next graph shows both history and projections to 2050 for daily dietary energy (in Kilocalories/day/person) in various regions of the world, as well as the global average.

Per capita food availability 1970-2050 for various regions, together with world average. Values for 2000 and before are data (left of the vertical red line), 2010 onwards are projections (right of vertical red line). Source: Table 2.1 of UN Food and Agriculture Organization, World Agriculture: Towards 2030/2050.

As you can see, the history is that most regions of the world have been getting more and more food. The exceptions are some of the formerly communist countries which suffered a partial collapse of their societies as they attempted to transition to a different economic system. The FAO projects that as the developing countries continues to grow faster than the developed world, they will be able to afford more food, and thus they will continue to approach, but not completely achieve, developed world levels of (over)feeding.

I could quibble with a few things here – I might guess that wealthier developing countries will get closer to current developed country averages by 2050, and I wonder about the sharp trend break between the past and the projections in the developed world. Still, these are minor issues – I think this has to be in the right ballpark for any scenario that assumes continued improvement of economic conditions in the developing world, and no major societal collapses (which is what we are trying to figure out how to avoid).

If we take the FAO’s scenario breakout of food groups (which they give by weight on a per-capita basis) and multiply by population, we get the following for total food demand:

Total food requirement 1970-2050 by major food types. Values for 2000 and before are data (left of the vertical red line), 2010 onwards are projections (right of vertical red line). Source: Table 2.7 of UN Food and Agriculture Organization, World Agriculture: Towards 2030/2050 and UN Medium Population Scenario for population figures. Note that I did not include “Other food”, which is only given in calorific terms in the table, and constitutes less than 10% of calories. Fruits and green vegetables would be included under that category.

As you can see, by 2050, the world would need to be producing about 50% more food than it is today (by weight – somewhat more in terms of energy in crops, since the meat component grows more than 50%). This contrasts with roughly doubling the planetary food production over the last 40 years. However, it’s still an awful lot of extra food to produce – the required absolute increase in food production is similar in size to what has been achieved in the last forty years.

Let’s now consider a variety of potential bottlenecks to achieving this kind of increase in food production. One major area of concern (water) I will reserve for its own future piece, but I address the other big potential constraints that I am aware of.

Land Use and Crop Yields

The doubling of global food production since the 1960s has not come about because of expanding cropland. The world has about 14.8 billion hectares of land area, and the uses of it over the last few decades are as follows:

Major classes of global land use 1961-2003. Source: FAO.

As you can see, the areas of cropland and pasture have increased slightly, at the expense of forests and other land, but the shifts are small percentage-wise. Instead, increased food production for the planet’s extra billions of humans has largely come from big increases in agricultural yields.

I’m going to start with some yield data for the US, where we have long time series on yields for a number of crops. After that, we’ll discuss the global situation. I have taken National Agricultural Statistics Service data on average US yields and reexpressed them on a common basis as a multiplier of the 1900-1935 average (or for those crops were the series doesn’t start till after 1900, from whenever it does start until 1935).

Average United States yields per unit area for selected crops, 1900-2007. Yields are expressed as a multiplier of the 1900-1935 average. Source: National Agricultural Statistics Service.

All the series show a roughly similar pattern. They were all fairly flat (with noise) until sometime in the late 1930s or 1940s. Then they all took off and began growing roughly linearly (again with noise). Modern yields are anywhere from 2.3 to 6.5 times greater than yields in the early twentieth century. Although some series have had periods of lagging for a decade or two (eg peanuts after 1983, dried beans – garbanzos and the like – after 1990), on the whole most of the series look like they are still increasing – there is no obvious pattern of yields flattening off yet. I encourage you to stare at this remarkable data for a long time. It’s really worth thinking about the implications of it. Here are a few conclusions I draw.

Firstly, mechanization (and fossil-fuel powered machinery) are not the main cause of modern yields. Steam tractors were in widespread use in the late 1800s and early 1900s:

Steam Tractor in action in Ontario, 1916. Source: Ontario Govt Photo Archive.

The first gasoline powered tractor to be mass produced was introduced by Ford in 1917. Yet the yield take-off doesn’t begin until 1940, and is almost certainly due to the agricultural innovations that comprise the green revolution. As The Future of Crop Yields and Cropped Area explains it:

The Green Revolution strategy emerged from a surprising confluence of different lines of agricultural
research (Evans, 1998) – the development of cheap nitrogenous fertilizers, of dwarf varieties of major
cereals, and of effective weed control. Nitrogenous fertilizers increase crop production substantially, but
make plants top-heavy, causing them to fall over. The development of dwarf varieties solves this problem,
but at the cost of making plants highly susceptible to weeds, which grow higher than the dwarf plants,
depriving them of light. The development of effective herbicides removed this problem. Further Green
Revolution development focused on crop breeding to increase the harvest index – the ratio of the mass of
grain to total above-ground biomass.

Secondly, anyone who wants to suggest that the world can be fed other than through industrial agriculture has some explaining to do about this data. Every crop shows yields prior to the green revolution that were flat and a small fraction of modern yields. If we returned to yields like that, either a lot of us would be starving, or we’d be terracing and irrigating most of the currently forested hillsides on the planet for food. While shopping for locally grown produce at your nearest organic farmer’s market, stop and give a moment of thanks for the massive productivity of the industrial enterprise that brings you, or at least your fellow citizens, almost all of your calorie input.

Which raises a third important point. Food = Area Cropped x Average Yield. If average yields had not increased like this, humanity’s impact on natural ecosystems would be much greater. It’s true that industrial agriculture has a lot of impacts (nitrogen runoff and the like). However, the alternative would probably have been worse, since it would have required us to intensively exploit enormous areas of fragile, and currently less intensively exploited, land.

Fourthly, the period of greatest global warming, since 1950, coincides with the explosion of yields. I do not suggest that global warming caused increased yields. But at any rate, it would be hard to argue that industrial agriculture yields cannot grow rapidly in the face of the kind of warming we have seen to date: they just did

Well, is the global situation the same, or is this US data unrepresentative? I don’t have access to as much data, but roughly, yes, it’s the same:

Average global cereal yields, 1961-2000. T. Dyson: World Food Trends: A Neo-Malthusian Prospect?, compiled from FAO data.

As you can see, global cereal yields are on the same roughly linear upward trajectory since 1961. Cereals are by far the most important food crop since not only do people eat a lot of them directly, but they also account for much of the input to the meat and dairy food groups that people eat, and thus are the base for the bulk of human calorie intake.

So obviously the critical question is whether or not yields can continue to increase in this manner? If we can just project out the linear increase than clearly a linearly increasing amount of food from a roughly constant amount of land is feasible, and humanity will be able to feed itself without having too much further impact on other ecosystems. On the other hand, if yields fail to increase, then we will be faced with unpleasant tradeoffs like trying to farm fairly unsuitable regions (think tropical rainforests, or the hilly parts of the western US), or not have enough food. So are we near some kind of theoretical yield limit?

Some people seem to think so. Lester Brown, who has been issuing alarming prognostications about food for several decades now, writes in Chapter 4 of his book Outgrowing the Earth

Although the investment level in agricultural research,
public and private, has not changed materially in recent
years, the backlog of unused agricultural technology to
raise land productivity is shrinking. In every farming
community where yields have been rising rapidly, there
comes a time when the rise slows and eventually levels
off. For wheat growers in the United States and rice growers
in Japan, for example, most of the available yield-raising technologies are already in use. Farmers in these
countries are looking over the shoulders of agricultural
researchers in their quest for new technologies to raise
yields further. Unfortunately, they are not finding much.

From 1950 to 1990 the world’s grain farmers raised the
productivity oftheir land by an unprecedented 2.1 percent
a year, slightly faster than the 1.9 annual growth of world
population during the same period. But from 1990 to 2000
this dropped to 1.2 percent per year, scarcely half as fast.

The argument in the second paragraph doesn’t hold water to me. Population has been increasing pretty much linearly in recent decades, and agricultural yields have also been increasing pretty much linearly – I don’t see any break from that pattern in the 1990-2000 decade. Of course, a linear rise will look like a dropping exponential growth rate, but Brown is careful to only point out the slowing in the yield growth rate. What he doesn’t tell you is that world population growth had also dropped to only 1.4% during 1990-2000. In general, food prices until very recently were in a multi-decade secular decline, indicating that food production was not under serious supply-side constraint until the last few years:

Ratio of crude food/feed producer price index to all US consumer prices, Jan 1969-Dec 2007. Source: St Louis Fed.

And the argument in the first Brown paragraph I quoted doesn’t seem to be how the agricultural scientists themselves are feeling. For example, Science reported last week:

A decade ago, sequencing the maize genome was just too daunting. With 2.5 billion DNA bases, it rivaled the human genome in size and contained many repetitive regions that confounded the assembly of a final sequence. But last week, not one but three corn genomes, in various stages of completion, were introduced to the maize genetics community. In addition, researchers announced the availability of specially bred strains that will greatly speed tracking down genes involved in traits such as flowering time and disease resistance. These resources are ushering in a new era in maize genetics and should lead to tougher breeds, better yields, and biofuel alternatives. “We’re sitting on very exciting times,” says Geoff Graham, a plant breeder at Pioneer Hi-Bred International Inc.

The geneticists are well on the way to having complete genome sequences for thousands of corn varietals from all over the world. If I was a corn geneticist, I’d be pretty excited too.

A more grounded attempt to estimate the issue seems to be the FAO’s discussion in World agriculture: towards 2015/2030:

The slower growth in production projected for the next 30 years means that yields will not need to grow as rapidly as in the past. Growth in wheat yields is projected to slow to 1.1 percent a year in the next 30 years, while rice yields are expected to rise by only 0.9 percent per year.

Nevertheless, increased yields will be required – so is the projected increase feasible? One way of judging is to look at the difference in performance between groups of countries. Some developing countries have attained very high crop yields. In 1997-99, for example, the top performing 10 percent had average wheat yields more than six times higher that those of the worst performing 10 percent and twice as high as the average in the largest producers, China, India and Turkey. For rice the gaps were roughly similar.

National yield differences like these are due to two main sets of causes:

Some of the differences are due to differing conditions of soil, climate and slope. In Mexico, for example, much of the country is arid or semi-arid and less than a fifth of the land cultivated to maize is suitable for improved hybrid varieties. As a result, the country’s maize yield of 2.4 tonnes per ha is not much more than a quarter of the United States average. Yield gaps of this kind, caused by agro-ecological differences, cannot be narrowed.

Other parts of the yield gap, however, are the result of differences in crop management practices, such as the amount of fertilizer used. These gaps can be narrowed, if it is economic for farmers to do so.

To find out what progress in yields is feasible, it is necessary to distinguish between the gaps that can be narrowed and those that cannot. A detailed FAO/IIASA study based on agro-ecological zones has taken stock of the amount of land in each country that is suitable, in varying degrees, for different crops. Using these data it is possible to work out a national maximum obtainable yield for each crop.

This maximum assumes that high levels of inputs and the best suited crop varieties are used for each area, and that each crop is grown on a range of land quality that reflects the national mix. It is a realistic figure because it is based on technologies already known and does not assume any major breakthroughs in plant breeding. If anything, it is likely to under-estimate maximum obtainable yields, because in practice crops will tend to be grown on the land best suited for them.

The maximum obtainable yield can then be compared with actual national average yield to give some idea of the yield gap that can be bridged. The study showed that even a technologically progressive country such as France is not yet close to reaching its maximum obtainable yield. France could obtain an average wheat yield of 8.7 tonnes per ha, rising to 11.6 tonnes per ha on her best wheat land, yet her actual average yield today is only 7.2 tonnes per ha.

For example:

Gap between actual national yields and estimated yield with best currently known varietals and inputs. Source: FAO report, World agriculture: towards 2015/2030

And so,

Similar yield gaps exist for most countries studied in this way. Only a few countries are actually achieving their maximum obtainable yield. When real prices rise, there is every reason to believe that farmers will work to bridge yield gaps. In the past, farmers with good access to technologies, inputs and markets have responded very quickly to higher prices. Argentina, for example, increased her wheat production by no less than 68 percent in just one year (1996), following price rises, although this was done mainly be extending the area under wheat. Where land is scarcer, farmers respond by switching to higher-yielding varieties and increasing their use of other inputs to achieve higher yields.

It seems clear that, even if no more new technologies become available, there is still scope for increasing crop yields in line with requirements. Indeed, if just 11 of the countries that produce wheat, accounting for less than two-fifths of world production, were to bridge only half the gap between their maximum obtainable and their actual yields, then the world’s wheat output would increase by almost a quarter.

Another way to try to get at the issue is to look at how current yields compare to the theoretical potential of photosynthesis. This is generally expressed as net primary productivity (NPP) – the amount of carbon that plants can fix, exclusive of that used to power their own respiration. The net primary productivity is the photosynthetic product that is available to be eaten by people and other animals, rot into the soil, etc. Here is a map of the fraction of net primary productivity appropriated by humans published by Haberl et al last year in the Proceedings of the National Academy of Sciences, which I take to be a decent representative of the state-of-the-art in this kind of calculation:

Global distribution of fraction of potential net primary productivity appropriated by humans. Source: Haberl et al: Quantifying and mapping the human appropriation of net primary production in earth’s terrestrial ecosystems

You might look at the red – 60%-80% appropriation of NPP in many of the world’s key crop growing areas, and think there wasn’t enough head room for another 50%+ increase in yield in those areas. However, it’s important to understand exactly how the accounting in these calculations is done. Let’s consider a piece of the US midwest that used to be tall-grass prairie and is now under corn. What Haberl et al would do is first use a vegetation model (specifically, this one) to establish that it would be a prairie there absent human intervention, and figure out how much carbon the prairie would have fixed as NPP. That quantity they call NPP₀ (for that particular area – they compute NPP₀ for every cell in a global grid). So this is an estimate of the theoretical carbon fixation in the absence of any human influence. In particular, this is with the rainfall that falls naturally – carbon fixation in actual use could potentially exceed this if the crop was irrigated.

Then they would run the model again, but constrained to have cornfields rather than prairies. The carbon fixed by the model in that scenario would be NPP_act. Thus a model estimate of the actual carbon fixation in the actual human use of the area.

Next, they would figure out NPP_h which would be basically the carbon in the harvested corn based on national agricultural statistics (and in agricultural residues if those were harvested and statistically tracked also, but not likely in the case of corn). So NPP_h is the part that we humans really use (either by eating or feeding to our animals).

Given the actual NPP_act, and the NPP_h they would then compute the difference, NPP_t – basically the carbon in the corn stover which gets returned to the ground, eaten by mice, or whatever happens to it.

So then the human appropriation of net primary productivity (HANPP) is defined as 1 – NPP_t/NPP₀. That is to say, if you look at the carbon that the prairie would have fixed, and then the carbon in the corn-stover, the difference is what is considered to be human appropriated. And that’s the thing in the map that’s 60-100% in the midwest (and other heavily utilized major cropland areas). However, this is not the same as the theoretical yield. In particular, a lot of the appropriated carbon comes about due to the difference between NPP₀ and NPP_act – the corn field doesn’t fix as much carbon as the prairie, probably mainly because it starts the season out as bare soil and has to grow an annual crop from seed, instead of being a set of perennial grasses that can sprout from last year’s roots and cover the available area in chlorophyll much faster.

Let’s look at their Table 2 to make this clearer. This table shows the global breakdown of HANPP by food class. If we look at the “Cropping” category, we can see the different figures.

Summary of human appropriation of net primary productivity. NPP₀ is modeled carbon fixation in wild condition. NPP_act is carbon fixation in actual human usage. NPP_h is carbon harvested or unfixed by harvest. NPP_t is residual carbon flowing into ecosystem. Source: Haberl et al: Quantifying and mapping the human appropriation of net primary production in earth’s terrestrial ecosystems

As you can see, the average m² of cropfield (worldwide) would fix 0.6kg of carbon if it wasn’t actually a field, but instead was covered in whatever the climactic climax vegatation is in that location. As a square meter of a field instead, it fixed 0.4kg of carbon, and of that humans got, on average 0.3kg as food and straw etc, leaving 0.1kg to go to the ground. So the HANPP is considered to be 5/6 (1 – 0.1/0.6). (The authors insist on three significant figures (83.5%), but I’m skeptical that the calculations are really that accurate). However, hopefully it should be clear by now that that doesn’t mean there’s a theoretical limit of only increasing yield by a further 1/5. Instead, there are multiple targets for the agronomists and geneticists to go after. The gap between the 0.4kg of NPP_act and the 0.6kg NPP₀ could be addressed with plants that had a longer growing season, covered the ground earlier, etc. To the extent some cropland is water-limited, irrigation could potentially increase the total NPP feasible. To the extent the 0.3kg of NPP_h is showing up as straw rather than food, then potentially that could be increased further.

A few decades down the road, one imagines heat-loving genetic mutant corn plants that pop up in the spring from perennial roots, promptly cover the ground with leaves that flatten themselves to the soil, and then start spitting out corn kernels, which can be harvested several times a year. It might not look much like a corn plant, but made into Doritos, people would probably still eat it (well, Americans would, anyway).

In short, another factor two of global cropland yields seems not to be ruled out on theoretical grounds. However, much more than that would appear to require the geneticists to come up with better photosynthesis (black plants basically – on which there has been no progress, as far as I understand).

Finally, it’s worth mentioning that the FAO thinks there is considerable potential to use more land for agriculture:

There is still potential agricultural land that is as yet unused. At present some 1.5 billion ha of land is used for arable and permanent crops, around 11 percent of the world’s surface area. A new assessment by FAO and the International Institute for Applied Systems Analysis (IIASA) of soils, terrains and climates compared with the needs of and for major crops suggests that a further 2.8 billion ha are to some degree suitable for rainfed production. This is almost twice as much as is currently farmed.

Here’s the breakdown for where the alleged potential cropland is:

Regional breakdown of land considered available for cropping, compared to land in present use for that purpose. Source: FAO report World agriculture: towards 2015/2030

However, “much of the land reserve may have characteristics that make agriculture difficult, such as low soil fertility, high soil toxicity, high incidence of human and animal diseases, poor infrastructure, and hilly or otherwise difficult terrain.” Caveat emptor!

If you look carefully at this figure – with the available land mainly in South America and Sub-saharan Africa, and the HANPP map above, you’ll realize that much of what the FAO is talking about is cutting down the remaining tropical rainforests and using them for agriculture. I don’t think that’s a very good idea for a host of different reasons – better that we eat mutant corn, I think. The great bulk of the best land is almost certainly in production already.

Soil Loss

It appears to me that until recently, there has been a good deal of scientific confusion on the seriousness of soil erosion, estimates of the rate of erosion vary by more than an order of magnitude, and the overall data situation make global oil reserves look like a model of precision. As such, I don’t think it’s possible to make a clear evaluation of how near term the threat is globally. My best impression is that it’s regionally quite severe, especially on fragile and marginal lands (dry, steep, or thin-soiled), but is probably not a near-term (next few decades) threat on the core agricultural regions from which most food comes (which tend to be flatter places with deep soils that don’t erode quickly). It is certainly a major concern on the century timescale. However, there are many cultural practices that can help while still allowing good yields and, if I’m reading the literature correctly, erosion appears to be controllable, even within the context of fairly industrial styles of agriculture. Let me quickly sketch some of the debate.

The last global evaluation appears to have been GLASOD done by Oldeman et al and published in 1990. They produced a map which looks like this:

Global map of soil degradation. Source: GLASOD map, as shown in FAO report World agriculture: towards 2015/2030

This looks really bad – everywhere humans are, the soil is degraded, and much of the world’s core crop land is in the “severely degraded” category. However, that did not yet have much noticeable effect on global yields, which have continued to increase by leaps and bounds since then. Moreover, this map was produced by what amounts to a survey of soil scientists, who used their subjective judgement. The instructions for filling out the questionaire describe how to set up the map cells, and then say:

The next step involves evaluation of the degree, relative extent, recent past rate and causative factors for each type of human-induced soil degradation, as it may occur in the delineated physiographic unit. This evaluation process should be carried out in close cooperation with national and/or international experts with local knowledge of the region. The evaluation process results in a list of of human-induced soil degradation types per physiographic unit, ranking them in order of importance.

So this doesn’t sound like a precise, quantitative sort of estimate. And more quantitative estimates are dogged with problems. A central issue is that most soil eroded from place A (let’s say a steep field on the side of a valley) isn’t necessarily lost to cultivation. Instead, it may end up in place B (let’s say the flood plain of the river in the bottom of the valley) where it may still be of use in cultivation.

The US is the best measured place, in that we at least have a national agency charged with regular quantitative assessments of soil erosion (a legacy of the dustbowl years). The last assessment was the 2003 National Resources Inventory.

NRCS maps of US soil erosion in 2003. Source: US National Resources Conservation Service 2003 National Resources Inventory

These estimates are made by applying models (the Universal Soil Loss Equation and the Wind Erosion Equation) to topographical and climate data. The model inputs are things like the rainfall data, the slope of the field, the erodibility of the particular soil, etc. The overall amounts of erosion are decreasing, and the amount is not imminently scary. The current national average of 4.7 tons/acre/year corresponds to a little more than 1 kg/m²/yr, which in turn is about 1mm/year, or an inch in twenty five years. That’s not good, but doesn’t sound like a likely disaster before 2050, particularly given that the rate of erosion is dropping quite rapidly.

However, these estimates in one way overstate the problem because the USLE and WEE are designed to assess how much soil is removed from its original location, but not where that soil goes. Most of it is unlikely to make it all the way out to the ocean, but instead end up somewhere else where it may be put to use. An extraordinary paper by Trimble in 1999 assessed the details of where soil went in a single valley in Wisconsin by doing detailed samples and cross sections of the alluvial plains. His estimates of the trends and disposition of soil is as follows:

Disposition of soil erosion in Coon Creek watershed, Wisconsin. Source: S. Trimble Decreased Rates of Alluvial Sediment Storage in the Coon Creek Basin, Wisconsin, 1975-93

Clearly, the soil erosion is decreasing, but also, most of it hasn’t gone that far, and, therefore, could potentially be put back at some point in the future if that becomes economically desirable.

Still, in the long term, it seems that eroding an inch every few decades from upland areas is certainly not sustainable, though it’s not an imminent crisis either. In an important meta-analysis last year, D. Montgomery compiled erosion rates for a wide variety of situations and plotted the following cumulative density function for the probability of different erosion rates:

Cumulative distribution function of soil erosion and formation rates from numerous studies around the world. Hollow circles represent rates of soil formation, solid line is geological erosion rates, triangles are soil erosion rates under native vegetation, while diamonds are soil erosion rates under various conservation tillage methods (terracing or no-till agriculture). Solid circles represent plough-based agriculture. Source: D. Montgomery, Soil erosion and agricultural sustainability

The key things to note are these:

• Rates of soil production and erosion under native vegetation are roughly similar, suggesting soil depths are naturally in equilibrium.
• Rates of “agricultural” erosion are a couple of orders of magnitude higher, suggesting that ploughing is not a long-term proposition.
• Rates of “Conservation” erosion are roughly comparable to to natural erosion rates under native vegetation. This covers more sustainable management regimes such as terracing and no-till agriculture.

This suggests that the long-term sustainability of industrial agriculture requires the use of no-till farming systems in which ploughing is not done, crop residues are left on the field, and weeds are managed another way (primarily via herbicides today).

Fertilizer

The three major fertilizer nutrients applied in industrial agriculture are Nitrogen (N), Phosphorus (P), and Potassium (K). None appear to be a critical constraint on agriculture to the 2050 timeframe, though there are significant issues with nitrogen in the short term.

Nitrogen fertilizer is manufactured via the Haber-Bosch process in which nitrogen gas (which forms almost 80% of the atmosphere) is heated with hydrogen over an iron catalyst at high temperatures and pressures to form ammonia (NH₃) which is subsequently reacted with other compounds to form urea, ammonium sulphate, and other compounds used as fertilizer. Presently, almost all the hydrogen input to this process is produced by steam reformation of natural gas, and this is the cause of the short term problem since natural gas supplies are problematic, and likely to worsen with both Europe and North America probably at or past peak natural gas. Fertilizer manufacture is exiting these regions and moving to the Middle East, Trinidad, and other places with more natural gas.

However, in the long term, there’s no reason nitrogen fertilizer has to be made from natural gas. In my scenario in which energy production is dominated by renewable/nuclear electricity by 2050, the natural source of hydrogen for Haber-Bosch is by electrolyzing water. Producing nitrogen fertilizer is unproblematic as long as society has ample energy.

The reserves and reserve-base for phosphorus are enormous. According to the USGS, 2006 global production of phosphate rock was 145 million tons, while reserves were 18 billion tons, and the reserve base was 50 billion tons. For the 2050 timeframe, I consider reserve base to be the more appropriate number for the same reasons discussed under lithium. The reserve base for phosphate rock is 350 times larger than 2006 production, so there is no evidence of a problem at present.

Some bloggers are concerned that the Hubbert linearization suggests peak phosphorus has already past. However, Hubbert linearization is not very reliable if there is no independent evidence to suggest peak is at hand, due to the problem of dual peak structures giving rise to misleading linear regions (eg see the UK oil linearization). In this case, with enormous reserves, and stable phosphorus prices (they haven’t varied outside the range of $27-$28/ton from 2002-2006), it seems very unlikely that phosphorus is in trouble. JD has made a similar point (snark warning).

Potassium comes from the mining of potash. The USGS estimates the global reserve base to be 550 times larger than current usage. So potassium is unlikely to limit civilization any time soon.

Fuel use in Farming and Food Transport

I don’t have global statistics, but at least in the US, agriculture is a minor user of oil. In total, it only used 2.2% of oil in 2000. This contrasts with cars and light trucks, which used 40%, heavy trucks which used 12.7%, air travel at 6.7% etc. Since agriculture is such a critical industry, we can ensure it is preferred for oil usage.

Furthermore, all shipping trade only uses 2.5% of US oil use. Most of that is shipping things other than food, but the bulk of food transportation is in there too. Amongst critics of globalism, the image of strawberries being flown from Chile is a popular thing to pick on. However, things like strawberries form a miniscule fraction of our diet. A more representative image of global food trade would be a grain ship like this one:

Grain ship docked in Australia.

Shipping is extremely energy efficient – two orders of magnitude better per ton-mile than air freight. Thus, long-haul shipping of food will be cost effective long after oil has peaked. Ships can also be run on nuclear power, as the US navy has been demonstrating for decades.

In Conclusion

There seems to be reason for cautious optimism that if other global problems can be solved, food production will not be a critical constraint on civilization to 2050. If industrial agricultural yields maintain their historical trajectory, there will be enough food without needing much more land. In case yields fail to continue increasing, more land is potentially available globally, though likely of poor quality. Soil erosion is an important problem, but not a critical emergency, and can seemingly be solved permanently with no-till farming methods. Fertilizer does not appear to be seriously constrained in the long-term, though nitrogen fertilizer needs to be transitioned away from reliance on natural gas. Agriculture only needs a tiny fraction of global liquid fuel use to operate, and this can be maintained for a long time, since food production is a critical infrastructure.

However, if we were to keep growing the conversion of food into biofuels, all bets would be off.

Other sources

In addition to the sources linked directly above, I consulted the following references

• Smil, V. The Earth’s Biosphere: Evolution, Dynamics, and Change (book).
• Wild, A. Soils, Land and Food: Managing the Land during the Twenty-First Century (book)
• Brown, L. Plan B 2.0: Rescuing a Planet Under Stress and a Civilization in Trouble (book)
• McKibben, B. Deep Economy: The Wealth of Communities and the Durable Future (book)
• T. Dyson: World food trends and prospects to 2025
• Millenium Ecosystem Assessment, Ecosystems And Human
  Well-Being: Desertification Synthesis

• P. Waggoner and J. Ausubel: How Much Will Feeding More and Wealthier People Encroach on Forests?
• Great Transition: The Promise and Lure of The Times Ahead.
• IPCC Special Report on Emissions Scenarios, Chapter 4 (An Overview of Scenarios).
• Climate Change 2007: Impacts, Adaptation and Vulnerability, Chapter 5 (Food, fibre and forest products)
• J. Pretty et al. Resource-Conserving Agriculture Increases Yields in Developing Countries
• J. Pretty and R. Hine Reducing Food Poverty with Sustainable Agriculture: A Summary of New Evidence
• D. Pimentel et al Environmental and Economic Costs of Soil Erosion and Conservation Benefits

This is a guest post by Yair Wallach. Originally from Jerusalem, he is completing his PhD in Cultural History in Birkbeck College, the University of London (writing about Palestine/Israel between 1858 and 1948). During his five years of study in London he has lived in precarious conditions, spending many months without electricity or hot water. These experiences have made him aware of issues of environmental sustainability, especially relating to energy, water, waste and the global food market. He currently makes his living by writing articles of economic analysis on the Middle East.

Abstract

The use of food crops for biofuels is one of the key factors driving a dramatic increase in the global price of cereals. As Stuart Staniford demonstrated here in the past few weeks, this trend is set to intensify. This article will look at the potential implications of rising wheat prices for countries in the Middle East, taking Egypt and Morocco as examples. Government food subsidies in both countries have so far protected the poor urban population from much of the global hike in cereal prices. However, as food prices continue to spiral, subsidies will demand a growing share of national budgets. Subsidies cuts seem inevitable, leading to riots and political instability.

The further development of biofuels could make food too costly for millions of poor in the Middle East, and destabilise the region which supplies most of the world’s oil exports.

[break]

Introduction

Stuart Staniford’s article Fermenting the Food Supply exposed the dangerously rapid manner in which food crops have been diverted to biofuels in the USA, and the likelihood that this pattern will be copied elsewhere. Staniford attempted to gauge the impact of price rises on the global poor. Looking at the elasticity of food expenditure, he suggested a grim possibility of 60% of the globe’s population priced out of the food market within the next five years. In a later article, Death Rates and Food Prices he considered the mitigating effect of subsistence farming, which could support a considerable part of the global poor.

Staniford established convincingly that the impact of biofuels on food crops will be almost immediate – that is, within the next decade or even five years. However, within such a short time span, assessment based on universal parameters will give a very limited picture. I believe that a more detailed attention needs to be given to specific regions and countries. Which ones are most at risk?

The Middle East is my home region, with which I am familiar personally and professionally. It is natural for me to be interested in the dangers for the region’s population. But furthermore, a food crisis in the Middle East may have far reaching consequences, due to the importance of the region for oil and natural gas exports.

My starting assumption is that countries that import a large percentage of their cereal utilisation will be more exposed to the rising prices. Where hard currency has to be paid for cereal, the increase in price will be most visible. By this criterion, the Middle East is especially vulnerable. As the chart below shows, out of 20 countries that import 10% or more of their cereals, 7 countries are found in the Middle East: Saudi Arabia, Algeria, Morocco, Egypt, Iran, Iraq and Sudan.

Imported Cereals as share of utilized cereals in selected countries. Source: FAO, Food outlook. Data for 2006-2007 is estimated; data for 2007-2008 is projected. The data is selective and probably includes only countries with substantial population.

The vulnerability of the region also lies in the fact that wheat-based bread is the main staple. Without bread there is no life – indeed, in Egypt the same word is used for both (‘aish). The global commodity price of wheat has gone up most drastically, tripling between 2000 and 2007. Maize and rice prices have doubled during this time. Countries in which wheat is the main cereal are likely to be more severely affected.

Outside the rich pockets of wealth in the Gulf, poverty is widespread in the Middle East. In Egypt, 45% of the population are estimated to live on US$2 per day or less (2007). The population in the region spends on average a third to half of its income on food. Poor urban households are in a precarious position to begin with, and they will be affected badly by any prices increases. However, the price of bread is not dictated directly by global cereals prices, because of generous government subsidies. Before examining the possible implications of the crisis by looking at the specific cases of Egypt and Morocco, a few words on the economics and politics behind food subsidies in the Middle East.

Oil and Food subsidies

Government intervention in the food market is a crucial mitigating factor that has to be taken into account when trying to assess the impact of the current price hike. In virtually all countries in the Middle East and North Africa, governments offer generous subsidies for food and, in most cases, for fuel. There is an unwritten pact between governments and peoples in the region that guaranties that the price of bread and fuel remains affordable, and any cut in subsidies is seen as a direct attack on people’s most basic rights.

The IMF and various other global consulting bodies have persistently preached against subsidies, arguing that they are not an effective means to alleviate poverty. The argument has merit: the subsidies benefit poor and rich alike; they encourage corruption and waste. This is especially true with fuel subsidies, of which the middle classes take full advantage. The IMF has consistently called for replacing the subsidies with other mechanisms that would support directly the population in need, such as cash transfers. However, the population in the region has real concerns about such suggestions: Will cash-grants be sufficient? Will they rise with inflation? Will they reach everyone in need? Will governments be competent enough to administer the scheme? The general sentiment is that the answer to all of these is ‘no’; a recent survey showed that 88% of Egyptians are opposed to any subsidy reform, fearing that ‘reform’ would mean in effect elimination.

The subsidies form a considerable part of all national budgets in the region, but for some countries they are a bigger strain than others, especially as the bill is getting higher. The rich oil and gas producing countries – Saudi Arabia, UAE, Algeria and others – are able to pay the rising price with high revenues from hydrocarbon exports. Other countries are in a far more precarious situation: these include not only resource-poor countries like Jordan, Tunisia and Morocco, but also oil producers such as Egypt, Iraq and Iran, which, for various reasons (resource depletion, internal strife or failing infrastructure) are fiscally vulnerable. Egypt, which has a substantial fiscal deficit, is expected to spend 30% of its budget for 2007/08 on subsidies.

Middle Eastern governments have been wary of eliminating food subsidies or replacing them, as it is clear that the issue is politically explosive. Subsidy cuts lead to riots. This has been the case in Egypt (1977), Sudan (1979), Morocco (1981, 1984, 2007) Jordan (1989, 1996), and Tunisia (1984). The riots are perceived as serious challenge for the regimes. In some cases (Morocco 1981) hundreds of demonstrators were killed. After clampdown of arrests and emergency measures, governments usually back down from the subsidy cuts. We have seen this happen in the last bread riots in Morocco (September 2007). This scenario will become increasingly unlikely as the subsidy bill becomes much more costly. As prices of oil and food go up, removing subsidies will become politically impossible, but sustaining them could become economically unviable.
Whatever happens, subsidies are unlikely to be eliminated completely, and global price rises will be mitigated and not hit the population in their full toll. Famines are therefore not to be expected in the immediate future. Yet political unrest is unavoidable. Even if governments succeed in repressing food riots, popular disapproval will remain and the political situation will be much more volatile.

Egypt

Egypt has the biggest population in the region – 77 million people, and a high growth rate. The country is also one of the biggest wheat importers in the world, importing about 38% of its cereals in 2006-7. The price of bread is very low – less than one cent in 2007, and subsidised bread is available mainly for the urban population, which made 42% of the total population in 2007.
In 2007 rising wheat prices cost the Egyptian government an additional US$ 2.5 billion in subsidies. The government could afford this because of windfall oil and gas revenues, and strong economic growth since 2004 in non-oil sectors. In 2007 Egypt had a US$ 5 billion trade surplus. In the recent Davos conference, Egypt was hailed as a success story for liberalisation reforms, and as one of the next emerging economies.

But in 2008 things are set to change. Egyptian oil production peaked in the mid 1990s. Oil consumption is growing strongly, due to economic growth. In 2008, Egypt is set to become a net importer of oil for the first time. From a dwindling source of income, oil will become a substantial fiscal burden. The government would have to import oil and sell it at a subsidised price – which would be a heavy burden, since fuel subsidies already made 20% of the government budget in 2005/6 (source: IMF).

Will the Egyptian government sustain bread prices at their current levels? After announcements of possible changes to the subsidy system, the Government recently announced that no major reform will take place. The current system will continue and will be extended. But can the government afford it to sustain bread prices at their current levels? Natural gas exports will continue to bring hard currency, but subsidies cuts seem inevitable. In 2007 the price of fuel went up by 30%. Further rises are no doubt on the way.

Egypt’s production and consumption of crude oil, in million tons, between 1973-2006. Source: BP

Morocco

Morocco has a large agricultural sector and therefore is in a better position to fall back onto subsistence farming. However, in recent decades Morocco has been plagued by recurrent droughts, in what is widely seen as the effect of climate change. The frequency of droughts has increased from once every five years to every other year; the length of the growing season has shortened considerably. (Source: Karrou). Yields vary considerably between years, and in 2007 they were especially low. As a result, Morocco is forced to import a growing share of its cereals: about a third of its cereals in 2006/7, and in 2007/8 it is expected to import about 56%.

Both fuel and food subsidies in Morocco are much lower than in Egypt. To give some indication, in 2004 the retail price of a litre of gasoline was US$ 1.10, compared with 28 cent in Egypt. Diesel was 70 cent compared with 10 cent. (source). Bread is sold at 1.20 Dirham or US 15 cents. Yet oil and food subsidies still made up about 10% of the government budget in 2007; if they were to double, this would create a considerable fiscal strain.

There are some early signs of crisis. In September 2007, just before the month of Ramadan (in which bread consumption rises) the government raised the price of bread by 30%. Bread riots followed, and after clashes between police and demonstrators, the government backed down and restored the lower price. The decisions on subsidies cuts, interestingly, was taken by the Ministry of Interior, in charge of internal security. (source: ecomaroc.blogspot.com, French).

Also there are indications of falling demand for oil. The volume of crude oil imports in 2007 was about 2% lower than in 2006. However, when November 2007 is compared to November 2006, we find an alarming drop of 43% in the volume of oil imports. (source: Moroccan Statistics). With no substantial hydrocarbon industry, a more urbanised society (60% urban compared with 42% in Egypt), and greater dependency on wheat imports, Morocco seems more vulnerable to the impending crisis than Egypt.

Conclusion

Cereal prices in the Middle East are mediated through state subsidies. So far, the urban poor have not been exposed directly to the rise in prices. It seems inevitable, however, that at some point the price rises will be passed on to the public through subsidy cuts, either in 2008 or in 2009, in countries such as Egypt, Morocco, Tunisia, Iraq, and Jordan.
Subsidy cuts will, without doubt, result in immediate riots. The urban poor will not wait until they reach a starving point: they will act immediately, as they have done before, against what they will see as the government betraying its fundamental duty to provide affordable food prices.

Egypt and Morocco are among the US’s closest allies in the region. Belonging to the so-called “moderate Arab/Muslim countries”, they have been the most accommodating in terms of supplying the US with intelligence and military cooperation against Islamist groups. In return the US has supported these regimes militarily and economically, through direct support (Egypt) or Free Trade Agreements. Political instability in these countries will put in serious risk the position of the US in the Middle East. The notion that food prices have gone up because of American (and other developed countries’) use of biofuels will not make the US more popular among people in the region.

The American policy on biofuels is repeatedly presented as a means to improve US national security, by reducing dependency on imported oil from the Middle East. Articles on Ethanol production here in the Oil Drum (by Robert Rapier and others) have shown this to be a fiction at best, because of ethanol’s poor EROI. Now it becomes clear that the subsidising of biofuels will make the world less safe for the US, by destabilising “friendly regimes” in the Middle East and beyond.

A few more words. Egypt, Morocco and other Middle East countries are regularly covered by Western Media, because of their economic and geo-political importance, as well as their proximity to Europe. Other countries – for example in sub-Saharan Africa – may be even more vulnerable, as many of them depend on cereal imports (although perhaps not to the same extent). It would seem likely that governments in sub-Saharan Africa have less power to mitigate price rises through generous subsidies. However, many such countries are off the radar for Western media, and the developed world will learn about the problems only through news of famines or refugee crises.

To forecast the impact of cereal price rises, one should take into account food subsidies (where they exist) and the ability of governments to sustain them. In the Middle East, it seems, the political consequences will be almost immediate, and will come before actual food shortages. In other regions it may take a different course. In Mexico, for example, subsidies have been eliminated long ago. But as I am no expert on Mexico, I will leave this for others.

If this short article dealt with the problem in strategic terms, in grand summaries of numbers (population, oil, food), it is important to remember that behind all these are people, real people, and many of them. Poor families in Egypt and Morocco, for whom life is already very difficult, and who survive on the bare minimum, are going to be badly hit in the next two years, when even a pita bread will become too expensive. The important issue here is not the survival of certain political regimes, but rather the survival of these families.

Sources:

• IMF latest report on Egypt
• IMF latest report on Morocco
• Akhter U.Ahmed, Howarth E. Bouis, Tamar Gutner, and Hans Lofgren (2001). The Egyptian Food Subsidy System: Structure, Performance, and Options for Reforms. International Food Policy Research Institute.
• Bread, the (subsidized) stuff of life in Egypt, International Herald Tribune.

Average daily total liquid production, by month, from EIA (green), IEA (plum), and OPEC (indigo) plus daily crude+condensate production from EIA (teal), and Oil and Gas Journal crude oil production estimate (dark red). Each series has the 13 month centered moving averages of each line, recursed once. Click to enlarge. Graphs are not zero-scaled. See below for sources.

[break]

Back when the October IEA total liquids number came in, Jim Hamilton at Econbrowser cited it (and other things) as evidence that there were “signs of gains in global oil production”. JD at Peak Oil Debunked had a similar reaction. I felt that was premature since that one high datapoint was not a statistically significant departure from the overall flat plateau that oil supply then appeared to have been on for the last 1-3 years (depending on your choice of data).

However, a few months have passed, and the evidence for a change in trend in oil supply now seems somewhat stronger, albeit there is still a lot of uncertainty and conflicting data. I won’t draw firm conclusions in this piece, but I’m starting to lean towards a bump up in 2008, rather than a bump down.

Let’s start with the numbers for total liquids from the EIA, the IEA, and (for the first time) the OPEC monthly oil market report (MOMR). I should say a few words about the last: I was able to extract monthly figures from graph 19 in the OPEC MOMR from January 2005 on. Before that, I could not find global monthly figures, and instead I constructed quarterly figures by adding together the numbers for OPEC crude, OPEC NGLs, and Non-OPEC oil production. I placed the quarterly number in the middle month of the quarter (Feb for Q1, May for Q2, etc) and then linearly interpolated to get the other months. So that’s the situation in this graph from Jan 2002 – Dec 2005:

Average daily total liquid production, by month, from EIA (green), IEA (plum), and OPEC (indigo) together with 13 month centered moving averages of each line, recursed once (LHS). WTI spot price (blue – RHS). Click to enlarge. Graphs are not zero-scaled. See below for sources.

Both OPEC and the IEA have initial estimates through January, but the EIA only goes through November. As you can see, both OPEC and the IEA show Dec and Jan stronger even than Oct/Nov. Since there’s a lot of correlation between the fluctuations in these series, it’s a decent bet that the EIA will also show a strong Dec/Jan, which will likely be enough to start to pull the EIA moving average line upwards again, after a couple of years of being consistently very flat.

For many people of course, the total liquids number is not really oil since it includes coal-to-liquids and biofuels in addition to things that are somewhat more colorably oil, such as natural gas liquids and refinery gains. I track two series that are more just “oil” as it comes out of the ground: the EIA’s Crude + Condensate series (in Table 1.1 of the International Petroleum Monthly), and the Oil and Gas Journal’s estimates of global crude oil production. Those two series (which only go through November at present) are as follows:

Average daily crude+condensate production, by month, from EIA (teal), and Oil and Gas Journal crude oil production estimate (dark red), together with 13 month centered moving averages of each line, recursed once (LHS). WTI spot price (blue – RHS). Click to enlarge. Graphs are not zero-scaled. See below for sources.

To get a better sense of the trends of all these series I have replotted them on the same graph. I have started each moving average at an arbitrary offset (2mbd, 4mbd, 6mbd, etc), but there is no rescaling of the data – just a fixed vertical offset. This allows us to compare the shape of the various series:

Average daily total liquid production, by month, from EIA (green), IEA (plum), and OPEC (indigo) plus daily crude+condensate production from EIA (teal), and Oil and Gas Journal crude oil production estimate (dark red). Each series has the 13 month centered moving averages of each line, recursed once. Click to enlarge. Graphs are not zero-scaled. See below for sources.

As you can see, the last four strong months have caused the IEA line to rise significantly, but also started the OPEC line (which was flat) to begin heading up. The others are not heading upwards, but it seems likely the next two months will be strong, and then they likely will begin to head up. No certainties, of course, till we actually see the data. But historically, the correlation in these month to month fluctuations is quite good.

What is not so good is the agreement on the underlying trend. If I replot just the moving averages as anomalies from their Jan 2002 level, you can see how large the divergence is getting:

Moving averages of daily total liquid production, by month, from EIA (green), IEA (plum), and OPEC (indigo) plus daily crude+condensate production from EIA (teal), and Oil and Gas Journal crude oil production estimate (dark red),. All curves are expressed as the anomaly from the Jan 2002 value. Click to enlarge. See below for sources.

I am still mystified by the large and growing discrepancy between the EIA and IEA total liquids numbers. It does not seem to come from one or a small number of countries, and it does not seem to arise out of any particular component of oil supply (at least in so far as it’s possible to tell).

However, all of the curves share a common feature when looking at the year on year growth rates (here in the moving averages):

Caption

As the growth decelerated in 2004/2005, it was possible to think that it would continue through zero and the world would go into decline. That hasn’t happened, and instead the growth rates have been fluctuating at or near zero – we have been on the bumpy plateau, with the forces trying to increase production and those trying to decrease production in approximate balance. And so the endless debate on whether the plateau would eventually break upwards, or downwards, or continue more-or-less flat for a long time.

The reason I had a bias towards the bump down before was based on a country by country analysis. The growth in global oil supply from 2001 to 2004 came primarily from two sources Saudi Arabia and Russia. In both cases, I believe this growth was primarily a result of putting in service spare capacity – explicitly spare in the Saudi case, and in the Russian case disabled following the Soviet collapse.

The Saudi’s stopped increasing production in 2005, and then declined in 2006, before making up a small fraction of that decline in 2007. My interpretation of this sequence of events was that they didn’t have any deployable spare capacity left (they may well have a final reserve that will be used only in the event of a real geopolitical emergency interrupting the oil markets, but not just to moderate prices). And it seems unlikely that they would produce large increases in production soon, since the new projects coming on line have to balance considerable depletion of key old fields (North Ghawar being the case we understand best).

Meantime, the Russian’s have been working harder and harder for smaller and smaller increases in production, suggesting that declines (albeit likely very slow ones) cannot be too far off.

In the meantime, the world has had flat production with new capacity just offsetting declines (as well as the depredations of various terrorists). However, there is some preliminary evidence that 2008 new capacity is much larger than recent new capacity:

Gross new capacity from Megaprojects by year. Source: Wikipedia.

Of course, the Wikipedia oil megaprojects tabulation is still not finished and the totals may change. Still, my sense is it’s not likely that the large step up in 2008 will entirely disappear. What is certainly the case however is that not all the gross additions will actually come on in 2008. Ace has estimated that only about 2/3 of the new capacity will actually arrive in 2008. If that was typical of prior years, then we’d have a hangover of new capacity coming from them too, but about 2/3 of the step up (from about 3-4mbd of new capacity in recent years to 7mbd in 2008) would lead to rise of about 2/3 x 3-4, or a couple of mbd rise.

I’m a bit sceptical that we can really expect to see a doubling of the industry’s ability to bring on new capacity from one year to the next, particularly given that global rig counts are not increasing nearly that rapidly:

Global monthly rig counts from Baker Hughes. Excludes US and FSU.

Furthermore, a couple of mbd is small enough that a revolution, war, or hurricane in any one of a large number of locations on the planet could easily offset it. Still and all, the balance of the incoming information seems to be shifting towards a modest increase in 2008 oil supply, rather than a small decrease.

Sources for Oil Series

Monthly data are from:

The Tata Nano will sell for about $2500 (US) in the base model, and get about 51 mpg (US). Source: Wikipedia.

[break]

After taking a hiatus from my regular Monday blog spot for a couple of weeks (to focus on another obligation) I want to pick up where I left off – exploring the quantitative barriers to getting most of the way to a sustainable planetary civilization by 2050. Last time, I laid out what I was trying to do:

This post is the start of an attempt to sketch out what an integrated solution to the world’s food, energy, land, climate, and economy problems might look like. My basic goal is to get to a somewhat defensible story of how civilization could get to 2050 in reasonable shape, despite the problems of climate change, peak oil, global population growth, etc.

Since it’s not possible for me to entirely solve this problem in a week of part-time work, I put this out as a hasty straw-man. Feel free to point out the parts of this that don’t work, or where my ignorance of some of the relevant issues shows particularly badly. Of course, I don’t make the claim that I can predict what will happen forty years ahead. Nor do I expect the global population to pay much attention to what I think they should do. Instead, the value of a scenario is to try to think through the general issues that society faces, and the value of an integrated scenario is that we can think about how all the parts fit together holistically, whereas usually they get projected separately by specialists, and even the obvious interconnections get missed by decision-makers (if we try to solve our fuel problems by converting food to fuel, perhaps the price of food might go up).

With that said, for the remainder of the piece I’m arrogating to myself sole authorship of all relevant international treaties and implementing legislation at the national level. Here’s how I’d go about it. In this first piece, I’ve analyzed the overall requirements for the problem, but only fleshed out any detail on the population, economy, and energy sectors; I did not have time to write up my analysis of transportation and agriculture/land issues. I will do so in a future piece.

and some of the requirements I saw as necessary in order to consider a solution viable:

• Population: The global population is able to grow and go through its demographic transition with death rates continuing to go down. No die-offs.
• Economy: The world economy is able to grow on average over the period – modestly in developed countries, faster in developing countries.
• Carbon emissions: The global energy infrastructure will be mainly replaced with non-carbon-emitting energy sources by the end of the period, and residual emissions will be rapidly diminishing.
• Fossil fuels: I assume that peak oil is here about now but that declines will be governed by the Hubbert model (and thus will be gradual). I assume natural gas and coal are globally plentiful enough that climate policy is required to prevent their full use.
• Technology: I do not assume any massive breakthroughs – no technological miracles that solve problems in ways completely unknown or untested today. However, where technological sectors have long established rates of progress in key metrics, I extrapolate the metric to continue improving at the historic rate (eg the economics of solar power, or the yields/acre of agriculture are assumed to keep improving on the historical trajectory).

Then I looked at the energy sector and we saw that there are several potentially feasible ways to power civilization. They aren’t cheap or easy, but solar, wind, and nuclear all have good to excellent EROEI and a fairly large resource that could be exploited. Solar in particular has the best learning curve historically (the rate at which price falls for each doubling of the installed base of the technology) and the highest growth curve and fewest barriers to early adoption (except cost). However, the renewable options are growing from a tiny base, and nuclear faces ongoing political resistance, so in the short term conservation and efficiency are critical requirements if we are to make it to the long term.

Still, if as a society we were serious and determined about solving our energy/climate problems, and we made the right investments, there seems to me little doubt that there are a number of feasible technical paths to a non-fossil-fuel energy infrastructure for civilization. Indeed, I argued that energy would likely become cheap again after a couple of decades of being expensive, once a renewable civilization was over the hump (the hump having been caused in part by failing to make more progress in the 1980s and 1990s).

However, there are many other resource constraints that we might hit along the way. So I want to continue surveying the terrain at a very high level and look at the automobile sector under the rough assumptions I outlined in Powering Civilization to 2050. In particular, how many cars might we expect by 2050, and how can we possibly power them, given that there will be less oil, not more, by that time. I think most readers would intuit that if society was wealthier in 2050, as I postulated, then if they possibly could, the planet’s citizens would tend to drive more, not less.

But how much more? In my earlier scenario, I postulated a world GDP of about $350 billion in 2006 dollars by 2050 (on a purchasing power parity – PPP – basis) which arose by assuming reduced economic growth in the near term due to problems of recession and energy constraints, but then renewed growth as those ultimately lift. Given the UN’s medium population projection of 9 billion people, that gives a world GDP/capita in 2050 of about $28,000 (versus about $11,000 recently).

Now, global figures on auto ownership have proven hard to find. The best I’ve been able to get is some data from the EIA for a small selection of countries, and the world as a whole, for 1990 and 1999. However, it’s probably enough. I combined that data with GDP data from the IMF, and UN population data to come up with the following graph:

Cars per thousand population versus GDP/capita for selected countries in 1999 (using 1999 PPP dollars). Bubble area is proportional to population. Sources: Auto ownership from the EIA, GDP data from the IMF, and population data from the UN.

As you can see, at least for the available data, there is a pretty strong linear relationship between income and car ownership. 93% of the variance in the latter is explained by the former. For each additional $1000/year in average GDP per person, you get about another 25 cars per 1000 people. This strong relationship between income and car ownership seems to match the views of transportation economists, who believe that people everywhere are willing to expend a roughly constant fraction of their time and money on getting around:

Both travel budgets are of very rough nature only. However, since they apply to virtually all people, independent of income, space, and time, strong regularities in aggregate travel patterns are observed when we compare cross-sectional and longitudinal data of all travel surveys, including those from the developing world. The travel money budget along with country-specific characteristics of the transportation system (land-use, prices, etc.) translates disposable income into daily distance traveled. All other patterns can be largely explained by the travel time budget. Using this approach, travel patterns of countries with very different characteristics at first glance evolve on nearly uniform trajectories. Thus, despite their only rough stability, the travel budgets offer a simple, elegant framework on the basis of which average travel behavior characteristics can be approximated on aggregate levels.

While there isn’t enough data above to prove this statistically, it rather looks like the major secondary variable controlling car ownership would be population density. The country with the lowest car ownership for their income level is the Netherlands, which is one of the most densely populated countries in the world, and the region with highest car ownership for income level is Australia/New Zealand, which is one of the most sparsely populated parts of the world. This matches the intuition one would get from US data, where public transportation ride share is highly correlated with population density, and auto usage inversely so:

Transit and private vehicle share as a function of census tract density. Source: Commuting in America III.

There’s a similar pattern in vehicle ownership – households in areas with the highest population density (over 10,000 persons/mile) are much more likely to have no car, and much less likely to have lots of cars:

Number of private vehicles per United States household as a function of population density in 2001. Source: 2001 NHTS Summary of Travel Trends.

This is somewhat encouraging for keeping the car count down, since the average world citizen in 2050 is likely to live in a very dense city in what we today call the developing countries (a lot of them will be pretty developed by 2050 under my assumptions). However, the discouraging thing is that those countries are growing the fastest economically, and that means rapid growth in car ownership also:

Annual growth rate in cars per thousand population versus annual growth rate in GDP/capita between 1990 and 1999 (using current PPP dollars). Bubble area is proportional to population. Sources: Auto ownership from the EIA, GDP data from the IMF, and population data from the UN.

Whereas high car ownership is entirely a function of being a wealthy developed country, high growth in car ownership comes from being a low income but fast growing country. Particularly striking is Korea, with an extraordinary 14% growth rate average in car ownership over the decade of the 1990s. Korea was a middle-income rapidly-industrializing country, which obviously leads to a lot of car buying. These days, China and India, with a third of the world’s population, are working their way into that status.

So let’s try to roughly guesstimate the number of cars people might buy if they weren’t resource constrained under this scenario, and then look at the resource constraints that might prevent them from having that many cars. I’ll do the guesstimation three different ways which should give us a rough sense of the ballpark.

The first method is to note that the $28k/person/yr GDP in 2050 (expressed in 2006 dollars) would be about $23k in 1999 dollars. On the straight line in the ownership versus GDP graph above, that would place us at around the 500 cars/person mark. However, if we figure the average citizen is at Netherlands densities, we might knock 200 cars/person off that total to come out around 300 (give or take). With 9 billion people, that’s about 3 billion cars in round numbers.

Another way to get to it is to notice that in the growth rate comparison, the average elasticity is a little more than 1 (ie 1% growth in income/capita leads to 1% growth in car ownership/capita). The EIA says that global car ownership in 1999 was 122 per 1000 people, which was 730 million cars. This source says there were a little less than 1 billion cars in 2006, so let’s figure 1 billion in 2007. So car ownership grew 4%/year, and the global economy averaged 4.3% growth over the same period. Close enough to an elasticity of 1. So if we extrapolate that out to 2050, we go from 150 cars/person and $75 trillion today to $350 trillion and thus about 700 cars/person in 2050. If we again knock a couple of hundred off for high density city effects, we would get down to around 500 cars/1000 people, or about 4.5 billion cars. This is effectively to say that another 40 years of economic growth at something like current rates would place the world average roughly at current European levels of car ownership, which sounds reasonable if only we can find some way to power that many cars.

The third method is to use some data from here which show production of autos (rather than ownership). Production from 1997 to 2005 grew at an average rate of 2.43%. If we apply that growth rate to ownership and extrapolate to 2050, we get 2.8 billion vehicles on the road at that time.

So all three methods come out somewhere in the range of a few billion vehicles on the road in 2050. Whether it’s 3, 4, or 5 we can’t know, but clearly it would take something on the order of a major economic collapse somewhere along the way for there to still only be 1 billion cars on the road then. (For example, Soviet car ownership declined from 357 per thousand people in 1990 to 134 per thousand people in 1999, so that’s what a major economic collapse can do). Let’s take 4 billion as a reasonable working number, with the understanding that this is &plusmn 25% (at least). Is there any way that many cars could be built and powered? Let’s first look at powering them, and then building them.

Running Four Billion Cars

The Tata Nano will sell for about $2500 (US) in the base model, and get about 51 mpg (US). Source: Wikipedia.

Ok, so I don’t think we need to spend very long on the idea that that many cars could be run primarily on oil. Let’s try to roughly figure what kind of oil that might need. I don’t have global mileage numbers, but in the US, vehicles do a fairly stable 10-12000 miles/year:

US annual vehicle mileage for vehicles of varying age grades. Source: 2001 NHTS Summary of Travel Trends .

Older vehicles do less, newer vehicles do more. Part of this might be income related, and since the average world citizen in 2050 will still not quite have achieved today’s US income level, and will live in a denser city than today’s average US resident, let’s figure 10,000 miles/vehicle. So we’re looking at somewhere in the neighborhood of 40 trillion vehicle miles per year. If we take the Toyota Prius/Tata Nano as exemplifying 2050 average fuel economy – about 50mpg – then we’d need a little over 50mbd of gasoline/diesel to run the 4 billion cars, even under the assumption they were pretty efficient. This won’t work at all. In my scenario, we’d be down to about 35mbd of total oil production by 2050, and we’d want most of that for other things (aviation, heavy machinery, petrochemicals etc).

I don’t really see doing too much supplementation of this with biofuels. Even the 1mbd of biofuels the world is already producing is causing a lot of problems, and it has the potential to get much worse quickly. Although cellulosic ethanol in theory could help, in reality most of the good agricultural land on the planet is already in use, and expansions onto the remaining land will tend to create far more carbon emissions than they save. (See two recent Science papers by Fargione et al, and Searchinger et al which are pretty convincing on this point).

So to run 4 billion cars, we should be looking at more like an average fleet economy of 200mpg to 250mpg, to keep the fuel bill down around 10mbd. That makes it likely that most of the energy would have to be coming from something other than liquid fuel.

There are two basic possibilities. The first is the hydrogen economy, in which renewable/nuclear power is used to produce hydrogen via electrolysis. The hydrogen is then used to power vehicles (and other things). I’m deeply sceptical about this whole idea. My objections are not primarily technical (though there are technical concerns) but rather based on the market diffusion problems.

Generally, diffusion of a new technology requires that there be early adopters who see value in it, then a larger group of less early adopters who are willing to do it once the worst bugs have been worked out, then the bulk of customers who only convert once the technology is really well established and their friends are starting to do it, and then finally the holdouts who cling to the old way of doing things until it becomes really not viable. This is the mult-stage diffusion process that has to occur.

Hydrogen has the huge problem of requiring a new infrastructure. So there need to be both early adopters on the infrastructure side (investors willing to fund hydrogen pipelines, gas stations owners willing to put in a hydrogen pump, etc), and early adopters on the consumer side (people willing to buy hydrogen cars). And in the early stages, both of these kinds of early adopters are going to have a miserable time because there won’t be enough of the other kind close by. (I buy a hydrogen car, but have to drive 100 miles to buy hydrogen, or I open a hydrogen station and I lose my shirt because there are only three hydrogen car owners in my city).

Now, if hydrogen cars were the only way to get around at a decent speed, people would find a way to get over these hurdles (after all, cars succeeded in displacing horses). But hydrogen cars will have the problem that there are already lots of gasoline cars on the road. Gasoline of course is expensive and likely to get more so over time. But hydrogen is even more expensive and, even in my scenario, is not likely to get cheap for decades. In the meantime, a hydrogen car is at a serious disadvantage to a gasoline car. Therefore, they won’t get adopted any time soon.

The other story, which I think is a lot more appealing, is that the present trend to hybrid gasoline/electric cars moves onto a plugin-hybrid stage in which the car has a larger battery and motor and gets plugged in to the electric grid at night or during the day at work. This has a far less serious adoption barrier. We already have distribution infrastructures for electricity and liquid fuel, so the only early adopter needed is the buyer of the plug-in hybrid. To the extent the grid needs to get expanded over time due to increased electricity usage by plug-ins, this will be done on the basis of clearly proven demand trends and can be a relatively low-risk decision. The speed of adoption of plugins will essentially be controlled by the relative prices of electricity and liquid fuels (including any carbon emission surcharges and governmental incentives).

In this scenario, power for cars will be predominantly coming from electricity by 2050, which I have already argued could be plentiful if we make the necessary infrastructure investments. So then the issues become whether it might be feasible to build that many plug-in hybrids.

Building Four Billion Plugin Hybrids

I stress of course that I’m not proposing that we make any crash program to build plugins. I’m simply proposing that as the economy grows and people, particularly in developing countries, get wealthier and want more cars, we create incentives to shift the car population gradually to hybrids and then plugin hybrids. Such incentives are already in place in a number of countries (eg the hybrid tax credit in the US). If this is done sufficiently, we would end up with a few billion plugin-ins by 2050. Market forces will do a lot of the work, since electricity is already cheaper per unit energy than liquid fuel, and the gap is likely to widen over time.

So then the question is what other resource constraints might we run into along the way, given that energy is not one in my scenario (at least not in the long term). Some things are fairly clearly not problems. The bulk of the car is made from steel, perhaps aluminum in future (lighter), and plastics. Iron and aluminum are the two most common metals in the earth’s crust and are unlikely to be serious resource constraints this century. Plastics will be available from oil as long as we can mostly stop burning the oil in automobiles. The necessary roads can be made from concrete, and we are very unlikely to run out of limestone to make the cement, or sand and gravel.

Two issues seem to me to be potentially pressing – lithium for batteries and copper for motor wiring. I will examine lithium in detail now, but defer copper, which is a more general concern, to a future piece. (Roughly speaking, those familiar with the copper issue can imagine that it comes down to an argument about how much copper usage can be substituted by much more plentiful aluminum).

Lithium

The best battery chemistries known for future automobiles all appear to involve lithium. Lithium-based rechargeable batteries have more energy density per unit weight, as well as carrying more energy per unit volume:

Energy density (weight and volume) for various battery technologies. Source: Wikipedia.

For these reasons, lithium is coming to dominate a number of rechargeable battery markets (laptops, cordless power tools) and it seems reasonably forseeable that it would also dominate the plug-in hybrid market. So in this section, I want to take up the question of whether we would run out of Lithium before 2050. An important analysis for considering this issue is The Trouble with Lithium which seemed quite alarming when I first read it, though now I have run the numbers myself, I am more sanguine. It’s also useful to read the most recent USGS commodity summary, as well as the longer 2006 Minerals Yearbook entry on Lithium. Also, Costs of Lithium-Ion Batteries for Vehicles has much useful data. I also recommend this nice overview presentation on mineral resource limitations on Earth in general (though some of the graphs are dated).

The concerns raised in Trouble With Lithium are two-fold. One is that the best and cheapest sources of lithium are limited to a few geographic regions (principally certain high desert regions of South America and Tibet) and that therefore the world would be as vulnerable to political problems with these regions as it is now with oil and the Middle East. The second is that the expansion of lithium mining required to support a plug-in hybrid world would be enormous relative to present day production. There is some validity to both these concerns, but let’s first look at the total amount of lithium available, see if that’s enough, and then come back to the potential bumps along the way.

The best estimates available are these (expressed as thousands of tonnes of elemental lithium).

As you can see, there is not an urgent lithium problem – reserves/production is currently about 250. (As far as I know, no-one has raised a serious question about the validity of lithium reserve numbers). However, this is with hardly any cars containing lithium batteries. How does four billion of them change the picture?

Let’s take a moment to look at the definition of reserves and reserve base:

Schematic of various categories of reserves and resources. Source: Resource Limitations on Earth.

Basically, reserves are the material that we know where it is today, and we believe with high confidence that we know how to get it out of the ground profitably at today’s prices with today’s technology. The reserve base additionally includes known or reasonably inferred deposits of material that are technically recoverable but marginally economic. After that, we get into resources that either haven’t been discovered, or are sufficiently dilute or inaccessible that no practicable way is known to extract them at near current prices.

Thinking about 2050, it seems to me that the reserve-base is the best guesstimate of how much lithium might be available if really needed (ie if we haven’t figured out a better idea in the meantime). This allows for improvements in extraction technology and/or higher prices (manageable by a wealthier society), but we aren’t getting out into “lithium from seawater” territory – the “reserve base” is lithium in known deposits.

Furthermore, since cars are 95% recycled even now, it seem reasonable to assume that long before 2050 we can pretty much be recycling all or nearly all of the lithium. So the calculation I’m going to assume (and I freely admit this is just a rough back-of-the-envelope sort of exercise) is to just divide the 13.4 MT of lithium reserve base by the 4 billion cars, which gets us an average of 3.3kg of lithium per car. Now, currently, it takes 0.3kg of lithium to get 1 kWhr of battery capacity, so that 3.3kg of lithium represents about 11 KWhr of electricity storage.

A reasonable assumption is that a plugin would require about 0.2 kWhr/mile. Thus the 11 KwHr average battery is 55 miles of all-electric range, or about 90 km. Now if we assume that they all get charged at night (figuring that the extra charging during the day for some cars is cancelled by others that can’t or don’t charge at night) then we can basically treat this as the maximum amount of daily mileage that can be covered by electricity, rather than liquid fuel.

Which allows us to make use of this data on the cumulative distribution of daily miles:

Cumulative distribution of personal daily travel distance in the US, UK, and an estimate for developing countries. Source: The Geography of Transport Systems, summarizing data from Regularities in Travel Demand: An International Perspective.

The average world citizen in 2050 is probably going to do a little less mileage than the US/UK numbers (given very dense cities). Thus more than 95% of vehicle days would be covered on electricity alone, and therefore it’s pretty comfortable that the 50mbd of oil requirement (assuming 50mpg) would fall below 10mbd.

(Obviously, powering even more cars after 2050 would require at some point that we find more lithium, figure out how to extract it from seawater profitably, discover some better battery technology, or somehow otherwise work around the problem. I’m willing to live with this – who knows what 40 years of innovation will come up with. We wouldn’t be worried about peak oil if it was 40 years off, and so I’m not going to worry about running out of lithium then — we have to leave our children something to do…).

Now, let me go back to the other concerns raised in Trouble with Lithium. The author, William Tahil, spends a lot of time concerned about the disconnect between the current production of lithium and the amount required to turn out all of today’s car production with lithium batteries, or convert all of today’s cars on the road to lithium. But these are not reasonable models for the time path of lithium/plugin adoption occurring. Instead, the right way to frame the question is to assume that the market for lithium-ion batteries in plugins grows gradually over time, and then look at the required growth rate in lithium production and see if it looks outrageous. In my case, for a first quick calculation, I’m just going to look at what constant growth rate is required to produce a cumulative 13.4MT of lithium by 2050, and then compare that to recent history. Here is the recent history:

Recent history of global lithium production (exclusive of the US). Source: USGS. Note that the USGS doesn’t publish US production because there is only a single US producer. The US will not be a major factor in lithium production going forward.

The darker green curve is actual lithium production. The compound annual growth rate from 1994 to 2007 is 11.46%, and the implied smooth curve is shown. Now, it turns out that if we were to assume that growth rate going forward, then we would mine almost 26 million tonnes by 2050. To get only the reserve-base number of 13.4 million tonnes, the average growth rate in lithium production between now and 2050 should be 9.35%. This doesn’t seem an obviously outrageous growth rate.

The second major objection Tahil raises is the geographic concentration of lithium that will thus give some countries a lot of leverage over global supplies. He is undoubtedly right, but this will be only one of many such world problems, and far from the most serious. The world is going to be ever more dependent on the Middle East for oil in coming decades. The world will be critically dependent on Asia for a lot of manufactured goods. Asia and the Middle East will be critically dependent on big food exporters like the US and Brazll to eat. If we don’t trade, we are all going to be in a world of hurt. In this context, concentration of lithium exports doesn’t seem like the worst problem (at least if lithium exports stop, it only hurts the ability to build new cars, not run the existing ones).

In conclusion, these stylized facts seem to be roughly true:

• With the existing known reserve base of 13.4 million tonnes of lithium and less than 10 mbd of oil, we could run 4 billion cars in 2050.
• If we assume most residents of the planet are living in dense cities in the third world with degrees of public transportation comparable to dense western cities today, then 3-5 billion cars should be enough to satisfy people’s aspiration for automobile transport by that time.

Oh, and I think we could reasonably hope that the darn things will drive themselves by 2050!