Oil and Coal: trends in global energy substitution

Figure 1 shows annual percentage contributions to total global energy consumption by fuel type since 1965. I plotted it up to see if it contained any hidden messages about the world energy market that had escaped my attention (the data used to construct all the graphs presented in this post are from the BP 2014 Statistical Review):

Figure 1: Percent contribution to global energy consumption by fuel type

At first glance there was nothing new. Oil‘s contribution peaked in 1973 and has been trending down ever since. Coal’s contribution declined rapidly between 1965 and 1977, but then it flattened out and after 2002 began to stage a comeback. Gas slowly built market share before 2002 but has stabilized since then. Hydro’s contribution has hardly changed, nuclear forged ahead until 1985 but then flattened out before beginning to fall away, and other renewables (wind, solar etc.) have barely lifted off the zero line.

Then I noticed an interesting feature. The percentage contributions of oil and coal tend to move in opposite directions. Coal goes down, oil goes up. Oil goes down, coal goes up. Oil goes flat, so does coal. But neither oil nor coal show a clear overall relationship with the third major source of energy, natural gas. The suggestion is that oil and coal have been substituting for each other, with coal replacing oil or oil replacing coal depending presumably on market conditions at the time, but with gas remaining largely unaffected.

Let’s look at this a little more closely.

A very simple test for an oil-coal relationship is to add the two together. If one is periodically substituting for the other then the sum of the two should give us something approaching a smooth line. And it does, more or less.

Figure 2: Sum of contributions from oil and coal

Another approach is to compare the first vertical derivatives of the percentage contributions of coal and oil to the global energy mix, which although it sounds complicated is actually just a matter of calculating how much the oil and coal percentages changed in each year and plotting them up. Figure 3 shows the results:

Figure 3: Annual change in percent contribution, oil versus coal

Now we will turn coal upside down so it moves in the same sense as oil and add an XY scatterplot. The match isn’t perfect, but the annual percent changes are obviously related and the level of correlation is quite high (R=-0.84):

Figure 4: Annual change in percent contribution, oil versus coal (inverted)

Annual changes in the percentage contributions of gas and oil to the global energy mix, however, show no relationship whether the plot is turned upside down or not (R=0.00):

Figure 5: Annual change in percent contribution, oil versus gas

But there is a weak inverse correlation (R=-0.49) between gas and coal (note that coal is inverted again):

Figure 6: Annual change in percent contribution, gas versus coal (inverted)

And we get a particularly close match when we compare coal with the sum of oil and gas (R=-0.94, R squared = 0.88, coal inverted once more). The fact that the trend line has a gradient close to 1:1 further indicates that one MTOE of coal effectively offsets a million tonnes of oil, and vice versa:

Figure 7: Annual change in percent contribution, oil + gas versus coal (inverted)

What are the implications of these results?

Basically what they are showing us is a high level of flexibility in filling growing global energy demand, with wholesale substitution of oil for coal, or of coal for oil, as market conditions change. (The substitution has occurred dominantly in electricity generation; coal can’t substitute for oil in transportation unless we return to steam trains and coal fired steamships). Global energy consumption grew in all but four of the years since 1965 (1980, 1981, 1982 and 2009), and coal and oil have been used more or less interchangeably to fill increased energy demand, or at least that portion of it that wasn’t filled by something else, for the last 49 years. As illustrated in Figure 8 oil dominated before 1979, coal took over from 1979 to 1983, honors were roughly even between 1983 and 1988, oil regained the lead between 1988 and 2002, and since 2002, largely thanks to China, coal has prevailed:

Figure 8: “New” energy contributed by year, oil versus coal

The swings between coal and oil are also large. Between 1965 and 1979 oil contributed over 1,500 MTOE of “new” energy to the world and coal only about a quarter of that. Between 1979 and 1989 coal added about 400 MTOE and oil effectively none. Between 1989 and 2002 oil added about 600 MTOE, roughly six times as much as coal, and since 2002 coal has added some 1400 MTOE, roughly three times as much as oil.

As to what motivated the swings from oil to coal and back again, it was clearly oil price, at least before 2000 (Figure 9). When oil was cheap, new energy came dominantly from oil. When it wasn’t it came dominantly from coal. The situation after 2000 is complicated by the rapid expansion of coal-fired generation in China, but the trend is in the same sense.

Figure 9: Increase in coal & oil “new energy” by year versus oil price

Yet despite the large swings between coal and oil over the last 49 years the global economy has continued to grow and no one has run out of energy, suggesting a) that the world is largely indifferent to what form it gets its energy in as long as it gets it and b) that growth is not as sensitive to continuity of oil supply as is sometimes supposed. There are, however, reasons to believe that these conditions may no longer apply, because large-scale coal/oil substitution requires large amounts of oil-fired generating capacity and there is now very little left (oil now generates less than 5% of the world’s electricity compared to 25% in 1973). So as Euan Mearns observed in an editorial comment, “the next oil shock may be a biggie.”

The final question is why oil and coal consumption should be so closely related while gas consumption isn’t. Most probably this is a result of ease of transportation. Oil or coal can readily be shipped almost anywhere in large quantities and can easily be redirected somewhere else when market conditions change. Gas can’t, or at least not at the moment.

This entry was posted in Energy and tagged , , , . Bookmark the permalink.

27 Responses to Oil and Coal: trends in global energy substitution

  1. Hugh Sharman says:

    Most oil used in power generation tends to be distillate, used in diesels, overwhelmingly in Africa and the Indian sub-continent. At present distillate prices, the falling cost of PV and to a certain extent wind, I am seeing an increasing number of diesels giving way, when the sun shines and wind blows,, to renewables.

    This trend is set to continue.

    • Euan Mearns says:

      Hugh said:

      I am seeing an increasing number of diesels giving way, when the sun shines and wind blows, to renewables

      Examples please, preferably with data and links. Oil Drum friend Chris Vernon worked for a while on deploying solar PV powered cell phone networks in African which I gather was an enormous success. It is not difficult to imagine that solar ± some battery back up could revolutionise life in off grid African communities. But one needs to be very careful about cherry picking minuscule scale success stories in Africa and extrapolating that to Manhattan.

      I was surprised to see a repetition of the coal for oil trend post 2002 in Roger’s chart. I wish we had the data, but I suspect what we are seeing here is mainly influenced by China ± India where distributed diesel generation has been replaced by grid coal. Possibly some similar going on in Middle East (gas for oil) and Russia. I guess one could identify the countries if Roger repeated his analysis on a per country basis to see which ones experienced oil going down and something else going up.

  2. Aki Suokko says:

    Very nice observations again, Roger! What came to my mind is that is it fair to test correlation for variables that are not fully independent? I refer to right hand sides of figures 4..7. Since the changes of contribution are not fully independent (zero sum game if all the energy carriers included) this gives you a partly artificial correlation. I think that this does not mean that oil has not been substituted by coal and vice versa, but the statistical measures like R and R^2 for this are biased towards a higher numbers. You can test it by comparing the share of oil to the share of all others and you will get a perfect correlation? I am not a statistician and I am not sure about that but this came to my mind anyway. The effect of this might be small, but since the share of oil and coal are high ones, I think the bias in R and R^2 might be a big ones.

    • Graham Palmer says:

      Interesting post. Roger seems to be querying whether there is a substitution effect going on that maybe hasn’t been picked up. My thoughts are probably along similar lines to Aki – there may be correlation but perhaps the graphs are overstating the implied substitution. A quick couple of quick thoughts

      a) the relative decline in coal up to 1980 was direct substitution by nuclear and gas, and earlier also by oil
      b) the relative rise of oil up to 1975 was simply Marchetti’s Curve doing what it should until the oil crisis hit – shifting to higher quality energy. This was about the growth in motorisation before anyone worried about fuel economy
      c) the reversal of Marchetti’s Curve back to coal was due to nuclear getting too expensive and losing social support, then China and India doing what the developed states did over 100 years ago – use the cheapest, indigenous fuel to industrialise
      d) China also seems keen on coal-to-liquids, throwing Marchetti even more into reverse and (eventually) perhaps demonstrating Roger’s substitution
      e) in theory, high density will eventually win out and nuclear will eventually start to displace coal again but the social licence still seems some way off

    • Aki:

      That’s a tricky question you ask 😉

      You’re quite right in that if we match oil against non-oil we will get a perfect correlation because it is, as you say, a zero-sum game. But that doesn’t mean that we are going to get perfect matches when we correlate oil, coal and gas separately. And indeed we don’t. We get a fairly strong correlation between oil and coal, a weak one between gas and coal and none between oil and gas (which surprised me). If all the variables were dependent we wouldn’t see this. The whole point of the exercise was to determine the level of interdependence, and that’s what the correlation coefficients do. Hope I’m making myself clear.

      • Aki Suokko says:

        Yes I understand you, but just wanted to make clear that those R values are naturally high just because variables you studied are depending each others (x+y+z… = 1). In correlation studies the the variables under study should be independent from each others, so there is not allowed to be any mathematical relation such as x+y+z…=1. I was about to repeat your very nice calculations for a blog post, but decided not to do that because the “laws of correlation analysis” are broken with those variables whose sum is always below or equal to one. I liked very much from the other figures and still think that your arguments are most likely still valid, and very interesting, but not so certain than you think, because the R values calculated are at least partly artificial ones. Please take this as constructive critic, since I’m a big fan of your very original analysis. 🙂

        • In correlation studies the the variables under study should be independent from each others, so there is not allowed to be any mathematical relation such as x+y+z…=1.

          Coal + oil + gas = 1
          coal + oil does not equal 1
          coal + gas does not equal 1
          oil + gas does not equal 1

          the “laws of correlation analysis” are broken with those variables whose sum is always below or equal to one.

          Could you elaborate on this statement please?

          Constructive criticism always welcome 🙂

          • Aki Suokko says:

            What I was pointing out, Roger, is that the Pearson R correlation coefficient can be calculated only for random variables as far as I know. When there is a relation x+y+z… = 1, the variables x and y are not random but have a mathematical relation. I am not saying that the “non-randomness” is perfect here because the sum of x and y is always below one and not one, but anyhow this relation ensures that the R value will be artificially high just because the variables are mathematically relating. So I think that the R values you presented in fugures 4…7 are at least partly artificial and biased upwards taken that oil’s and coal’s shares altogether are actually quite high being 0.6 – 0.8. Can you follow me now? 🙂

          • What I was pointing out, Roger, is that the Pearson R correlation coefficient can be calculated only for random variables as far as I know.

            You can calculate R for any two data sets you like – dependent, independent, random, non-random, even downright fudged, and the answer will tell you how interdependent the data sets are, assuming a linear relationship between the two. That’s what Pearson is for. There’s nothing which says that Pearson only works for “random” variables. (How do you define “random” anyway?)

            … because the sum of x and y is always below one and not one … this relation ensures that the R value will be artificially high

            I can’t find anything which says that Pearson doesn’t work when the sum of the variables is less than one either.

            Aki, I think what you are really talking about is how to interpret the R values: “What does the R=-0.84 value for coal vs. oil in Figure 4 really mean?” Right? Well, what it means to me is that there is indeed a strong inverse relationship between the percentage of global energy demand filled by oil and the percentage filled by coal, thereby confirming what we see in Figure 1.

          • Aki Suokko says:

            Roger, I was wrong and you were right. I simulated the situation with a random number generator in Excel so that I checked the range for both oil and coal share for 1965-2013 and did let the shares to variate within these ranges randomly. Then I x-y scatter plotted them. There was no any correlation detectable in the data. So, you are right in this issue and we can move on. 🙂

          • Aki:

            No problem. You could have been right.

            Thanks for taking so much trouble.

  3. Ralph W says:

    Coal does substitute for oil in transportation more or less directly, by electrification, particularly of rail transport. This is a major factor in Europe, and some major urban areas in the developing world. Oil consumption has been in decline in OECD nations for a decade, but the railways are bursting at the seams, at least in the UK.

    However, correlation does not imply causation. The rise in coal use is largely the massive industrialisation of China and to a lesser extent India. Nuclear is in generational decline because of the fallout from Chernobyl and TMI, and the rise of cheap NG. Oil production per capita peaked 30 years ago, The fact is the human species is maximizing entropy in the name of growth ever more efficiently, but the second order effects from the pulse injection of stored sunlight are beginning to show. OECD nations have continued to grow by lending themselves money to buy cheap Chinese coal and labour, and pretended that their growth had de-linked from the price of oil, otherwise known as EROEI.

    Nuclear is two orders of magnitude too complex to survive the transition to a post fossil-fuel society and we will be hard pressed to prevent further core meltdowns, or waste fuel-rod fires when the lights go out. The situation in the US is particularly worrying.

    Oil at $100 is looking unsustainable as the bills become due, and the high depletion rate of shale wells means that we are likely to see a far stronger correlation between price and production going forward. However, neo-liberal economists will still fail to see the wood for the trees.

    • Ralph: I was thinking more in terms of short-term substitution, where coal-fired generation increases and oil-fired generation decreases at existing power plants, or vice versa, in response to short- or medium-term price changes. Electrification does ultimately substitute coal (or gas, or nuclear or whatever) for oil, but the process takes time.

      “The rise in coal use is largely the massive industrialisation of China and to a lesser extent India.”. See the graph in my reply to Willem Post below.

      • dennis coyne says:

        Hi Roger,

        I agree that the substitution of coal for oil may have reached a limit because there is not that much power generated by oil, maybe a little in remote locations. There may be some due to the electrification of rail, though in some cases the electricity may be produced by natural gas or nuclear (in Europe), along with some coal.

        Rutledge estimates around 700 Gt for a URR for coal. If correct, and current rates of coal consumption continue until 2020 we would be half way through by that point and the peak might follow within 5 years. A rise in prices might increase the URR, but at some point the price rises to a point where wind, nuclear and solar look better than coal.

        A suggestion for your coal oil substitution graph and including only oil price. Maybe the relative price of oil to coal on a $/BTU basis would make more sense?

        • Dennis: The problem with coal prices is that they vary so much by location. 2013 prices, for example, ranged from $70/tonne to $140/tonne depending on whether you were in the US, Europe or Japan. Natural gas is even worse.

  4. Willem Post says:

    Roger,
    The correlation might change somewhat going forward.

    China will increase its gas use more than increase its coal use (coal use may even decline due to more supercritical capacity and retiring obsolete capacity) for energy generation, as a result of contracts it signed with Russia to buy 100 bcm/yr of gas by 2020.

    Also interesting is the miserable RE graph.

    As a result of RE build-out investments of about $1,700 billion from 2002 to 2013 (excluding mostly “socialized” investments for grid adequacy, capacity adequacy, etc., of about $400 billion not mentioned in the report), worldwide RE generation increased from 1.6% to 5.3%, a 3.8% addition, of which:

    – Wind increased from 0.3% to 2.7%
    – Biomass from 0.9% to 1.8%

    – Solar (PV + CSP) from 0.0% to 0.5%

    – Geo from 0.3% to 0.3%

    – Marine from 0% to 0%

    • Willem:

      “The correlation might change somewhat going forward.”

      It is in fact already changing. The correlation between oil and coal began to fall apart as soon as China began its massive expansion of coal generation in the early 2000s:

      Not sure what this bodes for the future.

      • Willem Post says:

        Roger,

        I am amazed at you facility with analyses.

        The breakdown of the correlation is quite remarkable.

        Keep in mind, China (and India) is doing this with low-efficiency, dirty coal plants equipped with minimal scrubbing and particulate collection systems.

        My estimate is each ton of coal burned in China is at least 100 times worse regarding particulates and about 30% worse regarding CO2 emissions/kWh, than each ton burned in the much more efficient and cleaner plants US and Europe.

        I just had published an article on THE ENERGY COLLECTIVE, which gives a double-barreled hint regarding the future.

        If you like the article, please press the LIKE button in the upper left corner.

        http://theenergycollective.com/willem-post/2146376/renewable-energy-less-effective-energy-efficiency

        • Willem:

          Thank you for your kind words.

          I looked at your article. I liked the first bit but not the second. I agree that improvements in energy efficiency are important but I don’t think they are the solution to the problem.

  5. Luís says:

    Hi Roger, regarding this comment:

    and other renewables (wind, solar etc.) have barely lifted off the zero line.

    I would like to call your attention to a short discussion I had with Rune some days ago. Essentially, BP’s statistical review is not a reliable source on renewable energy; biomass is mixed with fossil fuels and the numbers reported for solar and wind are way too low.

    Cheers.

    • Luís says:

      Somehow the href tag was ignored in the comment above. Here’s the link in text mode:

      http://fractionalflow.com/2014/10/10/the-powers-of-fossil-fuels/#comment-215

    • Luis: Renewables don’t enter into the discussion here, but I would agree that BP’s estimates of biomass generation probably don’t include the biomass that’s burned along with coal (lignite) in German power plants.

      But as for BP not being a reliable source on renewable energy, my question would be, who is? Different sources conflict, sometimes alarmingly. The US Energy Information Agency, for example, reported 3,451 GWh of renewables generation in the US in 2012 while the National Renewable Energy Laboratory – a branch of the EIA – reported 12,557 GWh. Conflicts of this type are everywhere. I take everyone’s estimates with a grain of salt, but at least BP is a coherent source of data, and I don’t think BP has any particular incentive to manipulate the numbers either way.

    • Euan Mearns says:

      Luis, I’ve done a quick check on UK wind comparing the BM reports data as reported by Gridwatch with BP for 2012

      BP report UK wind consumption of 19.6 TWh
      Gridwatch report (I calculate) 12.6 TWh (from over 105,000 lines of data)

      In an earlier thread we found that BM reports are not metering all UK wind, I think they were out by about 30% which tallies with the numbers reported here. No evidence from this single example that BP are under-reporting.

      In going from TWh to Mtoe BP gross up primary electricity to account for thermal losses in thermal generation using:

      Converted on the basis of thermal equivalence assuming 38% conversion efficiency in a modern thermal power station.

      I seem to recall we thrashed this out in a bar in Paris many years ago 🙂

    • Willem Post says:

      Luis,
      This site shows 104.5 TWh (PV 100.4 + CSP 4.1) for 2012
      http://www.energies-renouvelables.org/observ-er/html/inventaire/pdf/15e-inventaire-Chap01-Eng.pdf

      Way too low?
      Do you have better numbers for 2012 and 2013? URLs please.

    • Willem Post says:

      http://www.ren21.net/Portals/0/documents/Resources/GSR/2014/GSR2014_full%20report_low%20res.pdf

      This site states 77.9% fossil, 22.1 RE, of which
      16.4% hydro,
      2.9% wind,
      1.8% bio,
      0.7% solar,
      0.4 % geo, CSP, ocean

      The 2013 BP report states 78.3% fossil, 21.7% RE, of which
      16.4% hydro
      2.7% wind
      1.8% bio
      0.5% PV + CSP
      0.3% geo
      0.0% ocean

      As a result of RE build-out investments of about $1,700 billion from 2002 to 2013 (excluding mostly “socialized” investments for grid adequacy, capacity adequacy, etc., of about $400 billion not mentioned in the report), worldwide RE generation increased from 1.6% to 5.3%, a 3.8% addition, of which:

      – Wind increased from 0.3% to 2.7%
      – Biomass from 0.9% to 1.8%

      – Solar (PV + CSP) from 0.0% to 0.5%

      – Geo from 0.3% to 0.3%

      – Marine from 0% to 0%

      http://theenergycollective.com/willem-post/2146376/renewable-energy-less-effective-energy-efficiency

      • Euan Mearns says:

        Thanks for the link and data Willem which I have added to my next post. The agreement is suspiciously good though suggesting they and BP are accessing the same sources. Over to you Luis!

        • dennis coyne says:

          Hi Euan,

          The solar numbers often leave out PV on homes and only collects data on large installations. I do not have data on how much this amounts to, perhaps very little, but it is quite difficult to track. I could not find World data. For the US the BP uses EIA data for solar output which includes only utility scale PV and CSP (concentrating solar power), the NREL data includes all solar, in 2012 the difference was 7 TWh in the US. It may be that Eurostat accounts for both Utility scale and other solar output, I could not find that information.

          The data is not very good, perhaps Luis knows more.

Comments are closed.