Saturday, June 23, 2018

Hansen's 1988 predictions - 30 year anniversary.

It is thirty years ago since James Hansen's famous Senate testimony on global warming. This has been marked by posts in Real Climate and WUWT (also here), and also ATTP, Stoat, Tamino. As you might expect, I have been arguing at WUWT.

The more substantive discussion is on the accompanying 1988 prediction paper. This was a remarkable achievement, which used runs of an early GISS GCM to forecast temperatures for the next thirty years. These forecasts are now often checked against observations. I wrote about them here, here and here. Each post had an active plotter which allowed you to superimpose various observation data on Hansen's original model results. It's annual data, to match Hansen's prediction. Since sientists can't know how much carbon society will choose to burn, Hansen analysed three scenarios (A, B and C) which covered the range between no restraint and successful limiting. Arguments then ensue as to which scenario actually happened. At least, that is what people should argue about, although they have a tendency to drift off into "Hansen said..." or even orse "we were told...".

Anyway, I'll leave discussion of that for the moment, and show the interactive plotter. The diagram in the background is Hansen's original plot, which is of anomalies relative to base years 1951-80, and uses GISS Ts as the observed data set (this had recently been published (Hansen and Lebedeff)). I have used that base where possible, else I match the dataset to GISS Ts over 1981-2010 (satellite data). Data is annual to end 2017. Sources are linked here.

To operate, just choose datasets to plot using the radio buttons, and Clear All if you want to start again. You can't erase curves without restart.

In interpreting these, I think weight should be given to GISS Ts, since it is what Hansen had available and used. Later indices incorporating SST rise more slowly. And I have reluctantly included troposphere data, which is definitely not what Hansen was predicting. Properly interpreted, I think the predictions are excellent. But that comes back to deciding which scenario is appropriate. I discussed this extensively here. We have detailed versions of the sequences of gas concentrations that quantified the scenarios, and while CO2 followed scenario B, others were much lower. CH4 and CFCs were below scenario C, so overall a result between B and C is to be expected. And that is what is mostly observed, though GISS Ts is higher.

Update. I have a zipfile online here which has numerical data for both scenario gases and temperature prediction; details here. I used it to calculate trends, in °C/Century, for the 30 years 1988-2017: (Further update - I fixed an error in scenario rates - now trend for B is larger)

Scenario AScenario BScenario C

In that analysis of scenarios, I showed some old plots. Gavin Schmidt, at Real Climate, has shown some updated values, and I'll show his plots. I mentioned that there are two sets of scenario data. One is IMO the original, as I discuss there, but Gavin uses a slightly different set, which I think was digitised from graphs. Anyway, here is the RC plot:

For the CFC plots; scenario C assumed that the Montreal agreements on curbing them, still being negotiated, would be approved and would work. A and B were more sceptical, but C was right. For methane, the concentration not only rose rather slowly, but was revised downward even before 1988.

Overall, in placing the outcome between scenarios B and C, Gavin gives this plot of combined forcings:

What the showing of combined temperature records shows is that Hansen's 1988 prediction is about as good as it could be, because it sits within the scatter of modern records. The difference between GISS Ts and GISS land/ocean is comparable to the difference between GISSlo and scenario B.

As a check on my active plot above, here is RealClimate's rendition of the case for GISS land/ocean with the same scenarios:


  1. Hi Nick, great post. Thank you.

    On my laptop screen I cannot see the radio buttons. They are cut off the screen. If anyone else has this problem, the way around it is to reduce the zoom, to 90% in my case. They do show up straight away at full zoom when I attach a larger monitor.

    1. Thanks Sou. I put the interactive in an iframe hoping that would generate a horizontal scroll bar in those circumstances, but alas, no.

  2. Nick, Gavin seems to disagree that weight should be given to Ts; see

    1. Yes. I think Jim Hansen also said at one stage that the appropriate measure was probably between Ts and Ts+SST. I agree that the GISS model doesn't actually predict an index, but rather it's best estimate of the true global temperature. But in this case, that is air temperature, and in a sense GISS Ts is the adjustment of GISS LO SAT. As to being small, yes, but the discrepancy of the forecast is small.

      I think that the introduction of indices with SST has slowed down measured warming a little, and if JH had foreseen that that would be the measure, he might have made a slightly different prediction. Anyway, I don't think it is a big deal; Jim H's summary is probably best.

  3. Nice work Nick. I would expect that ideally for validating GCM output we should be looking at 2m temperature above both land and water integrated globally. Consequently the commonly used LO+SST integrations are not quite apples to apples, but still should be close as you have pointed out.

    I'm a bit puzzled though. How is the "GISS Ts" derived? It obviously produces a much higher result in recent years than "GISSlo" on your graph. Why is that? I previously thought that global land only temperature assessments were the highest assessments.

    Also, in working with the CFSR/CDAS data I have been using the 2m temperature output which I assume is 2m above both land and water and would thus be more appropriate for comparison with GCM 2m temperature output. The NCEP/NCAR Reanalysis 1 has been run back to 1948 and could potentially be used to cover this period for comparison, since the CFSR only goes back to 1979.

    1. Bryan,
      "How is the "GISS Ts" derived?"
      It was developed in Hansen and Lebedeff, 1987. It takes fairly large grid cells and seeks to estimate each with land data - so islands play a big role. There wasn't really usable SST data at the time. There were still ocean areas that it couldn't cover. You can get a picture of what it covers by going to the GISS maps page and asking for a map with no ocean cover. It shows you Tsurf, the GISS Ts calc.

      People have worried about the discrepancy between GCM global averages (lowest air level) and global indices (air on land +SST). Cowtan et al wrote about it here, and yes, using reanalysis for the observation part does seem to be the best answer.

    2. Thanks for the info Nick. So I assume the reason GISS Ts is higher in recent years than GISSlo is that recent higher anomalies over land must be having a greater influence on the Ts grid cells over water thus causing the global integration to be higher than not using any grid cells over water as with GISSlo?

      Below are NCEP/NCAR R1 global 2m surface temperature annual anomalies referenced to 1951-1980, as calculated from monthly global temperature output provided online by UM CCI.
      1960 -0.01
      1961 0.01
      1962 -0.05
      1963 -0.03
      1964 -0.24
      1965 -0.17
      1966 -0.08
      1967 -0.05
      1968 -0.07
      1969 0.08
      1970 0.04
      1971 -0.17
      1972 0.00
      1973 0.11
      1974 -0.18
      1975 -0.15
      1976 -0.22
      1977 0.12
      1978 0.04
      1979 0.16
      1980 0.33
      1981 0.30
      1982 0.03
      1983 0.24
      1984 0.08
      1985 0.06
      1986 0.15
      1987 0.26
      1988 0.30
      1989 0.15
      1990 0.36
      1991 0.33
      1992 0.07
      1993 0.01
      1994 0.10
      1995 0.28
      1996 0.15
      1997 0.34
      1998 0.55
      1999 0.27
      2000 0.27
      2001 0.45
      2002 0.54
      2003 0.54
      2004 0.46
      2005 0.65
      2006 0.57
      2007 0.56
      2008 0.43
      2009 0.57
      2010 0.63
      2011 0.49
      2012 0.54
      2013 0.58
      2014 0.62
      2015 0.76
      2016 0.96
      2017 0.82


  4. Tomas and others at Curry’s are trying to marginalize your scientific credentials.
    I think the main problem is that those guys don’t seem to understand that chaos theory is an abstract mathematical notion that is not in fact “recognized as a specific field of physics” as Tomas claims.

    He also defines chaos only by whether a time-series has a certain range of Lyapunov exponent. This is questionable because one can find a sufficiently complex yet obviously non-chaotic time series that also has a chaotic exponent.

    1. Indeed, Web. It's interesting debating scientific credentials with pseudonymous characters.

      I've been mulling a new post on the Lorenz butterfly, basically developing the Lyapunov exponents, which I think of as the eigenvalues of the linearised coefficient matrix. It's an interesting story.

    2. If there are eigenvalues that are both damped and undamped I can see how this may be connected.

  5. Tomas is further confused by the use of a Lyapunov exponent to characterize a chaotic process, since one can not do that via measurement data alone. The only way to determine the exponent is by invoking a model and evaluating the divergence by changing initial conditions, but time-series data (such as ENSO) has no intrinsic model associated with it. So only by fitting the data to a model, and then evaluating the model can one determine if it is a chaotic process. But to do this one has to rule out all possible non-chaotic models that match the time series. There are a lot of hoops that one must go through and even then there is no guarantee that you can characterize a time-series as chaotic.