Uncertainty in warming since pre-industrial times

The consequences of the Paris Agreement’s choice of the pre-industrial as its baseline have been discussed previously on this blog. This choice makes sense from a climate forcing perspective (as radiative forcings are measured with respect to a quasi-equilibrated state, and the well-observed recent past is not close to have finished responding to anthropogenic drivers). Looking back into the pre-industrial period, there are fewer instrumental observations of the temperature across the globe. So naturally our knowledge of the pre-industrial baseline temperature is uncertain.

Recent work, such as Hawkins et al. (2017) and Schurer et al. (2017), have looked to assess and quantify this uncertainty in light of future targets. The magnitude of this uncertainty, although small, becomes important when you consider the amount of warming left between today and the 1.5°C target.

Guest post by Chris Brierley (UCL)

What I’d like to discuss in this post is how we should be visualizing global temperature changes to represent the uncertainty. For example, if you look at the recent WMO press release (below), the uncertainty is only implied by the inclusion of multiple instrumental datasets. The 1°C above pre-industrial is included as single horizontal line. The only nod to there being uncertainty in this perfectly horizontal value is a tilde (~) in its label.

So how might you go about considering the pre-industrial uncertainty? Firstly, we need to use a dataset with uncertainty. I’m going to use the new Ilyas et al. (2017) version of HadCRUT4 (which uses a multi-resolution lattice-kriging ensemble to estimate coverage uncertainty). All of the error bars here are 5-95% confidence intervals.

We compute the pre-industrial offset as combining the uncertainty in the 1850-1900 observational record with an additional component from the warming derived from climate model simulations. This component is roughly 0.1 +/- 0.1K, based on Schurer et al., 2017 which used model simulations to estimate the expected warming from 1400-1800 to 1850-1900.

In its conventional presentation (above), we see the observations becoming less uncertain with time – as coverage increases and our observational errors reduce. We now dress the 1.5°C marker with an error bar relating to the uncertain pre-industrial offset. Personally, I don’t feel that the inclusion of pre-industrial uncertainty in the 1.5°C line works well. Instead, it looks as if there is some flexibility in the target itself. The spirit and ambition of the Paris Agreement’s text does not imply flexibility in the target – at least in my reading.

Alternatively, you could change the vertical axis of the figure to be “temperature above pre-industrial” (as below). This means that the uncertainty in the pre-industrial temperature must be folded into the temperature record, along with the coverage and instrumental uncertainties. This is a bit more challenging to compute, but it means that the 1.5°C target is now robustly represented as a single line. You get a rather different interpretation about our knowledge of the climate system from this figure.

What is immediately obvious from the new figure is that the uncertainty range stays pretty much the same throughout the past century – demonstrating the importance of our definition of the pre-industrial.

About Ed Hawkins

Climate scientist in the National Centre for Atmospheric Science (NCAS) at the University of Reading. IPCC AR5 Contributing Author. Can be found on twitter too: @ed_hawkins

3 thoughts on “Uncertainty in warming since pre-industrial times

    1. The Schurer paper advocates defining pre-industrial as the average between 1400-1800 CE.

      The way they suggested doing this was to use the observed record to calculate the early instrumental value (so 1850-1900). And then use an ensemble of last millennium model runs to estimate the offset between this early instrumental value and the full 1400-1800 temperature. My statistics colleagues didn’t like just taking the 22-member last millennium distribution directly, so we resampled the offset with a kernal density estimator.

Leave a Reply

Your email address will not be published. Required fields are marked *