Global warming is projected to force climatic variables in some places beyond the range of historical experience, perhaps permanently. A reassessment shows that this could begin sooner or much later than recently estimated.
Earth will continue to warm under plausible 'business-as-usual' scenarios for future greenhouse-gas emissions1, which assume that little, if anything, is done to abate emissions. This warming, in turn, will cause further widespread changes in the climate system1. Future climatic conditions will also depend on naturally occurring climatic variability1,2, a large component of which arises spontaneously within the climate system2,3; the El Niño–Southern Oscillation is one example of such internal climate variability2,4. In a paper published in Nature last year, Mora et al.5 estimated when permanent departures — or expulsions — from the historical range of variability will commence. Hawkins and colleagues6 now dispute these estimates in a Brief Communications Arising published on Nature's website, and question the magnitude of the associated uncertainty provided by Mora and co-workers.
The timing of an expulsion (tE) of a particular climate variable at a given location depends on the magnitude of the underlying warming, and on both the level and the timing of future natural variability (Fig. 1). However, the ability of climate scientists to predict internal variability beyond the next year or two is limited2, and so the sequence of variability that will unfold during the remainder of the twenty-first century is largely unknown. This uncertain future sequence will either tend to bring expulsions forward in time, or delay their commencement.
Mora et al. and Hawkins et al. estimated tE using information from many of the world's latest generation of climate models1. The models generate a wide range of tE values because each model has a different sequence of future natural variability, and differ in their sensitivity to the imposed increases in greenhouse-gas concentrations. Hawkins and colleagues contend that Mora and co-workers' estimates of tE are, in some cases, too low (that is, some expulsions were estimated to occur too soon), and that the associated ranges of possible values are too narrow for many locations and climate variables.
Hawkins et al. argue that these problems arise for two reasons. First, Mora and colleagues analysed model results only up to the year 2100. In some places, expulsion seemed to occur before then, but in the remaining regions, Mora and colleagues set tE to 2100. This approach underestimates tE in regions where expulsion occurs after 2100, and misrepresents the situation in regions where expulsion might never happen. Second, Mora and co-workers based their discussion of the range of possible future values of tE on a statistical measure of the uncertainty in the model-average value of tE. Hawkins et al. regard this as an inappropriate measure of future uncertainty in tE.
Mora et al. concede the first point in their response7 to Hawkins and colleagues' arguments (also published on Nature's website), but not the second. The source of this disagreement can be clarified with a simple analogy. Imagine that a doctor receives a text message from Harry, who is doing a school project on life in the nineteenth century: “What is the average age at death of people born in our city in 1860?” Having just read a report on the topic, the doctor promptly replies “61 ± 1.2 years.” This is the estimated average, together with an indication of the accuracy with which the average value is known.
A few days later, Harry sends a second message: “How old were they when they died?” The doctor, recalling her previous text, is about to re-type “61 ± 1.2 years,” but then realizes that this is a different question. The range of possibilities is substantially broader than ± 1.2 years. Some people died soon after birth, others lived beyond 80 years of age. She then replies: “0 to 80+ years.”
In their original paper5, Mora et al. gave standard errors that provide (under certain assumptions8) confidence intervals associated with the model-average values of tE. Although these are valid statistical measures, they are more closely related to the uncertainty in the doctor's first response than to that in the second. This is because the real world does not behave like the model-average — it will be much more like a single model realization with a particular sequence of future internal climate variability.
The problem is that we do not know which model sequence to choose, or even if any of the model sequences coincides with the one that will actually unfold. The best that we can do is to indicate the range of possible future sequences. This is what Hawkins and colleagues do. Their ranges6 are much broader than those provided by Mora and co-workers5, and are analogous to the uncertainty in the doctor's second answer. Hawkins et al. conclude, for example, that there is more than an 85% chance that tE at any location will fall outside the range provided by Mora and colleagues. In my opinion, the approach taken by Hawkins et al. provides a more appropriate estimate of the range of tE that could occur in the real world over the coming century and beyond.
Despite these issues, important conclusions of Mora and co-workers' original paper remain valid. Expulsions are indeed expected to occur under business-as-usual scenarios over wide areas before 2100. These will tend to occur sooner under scenarios involving higher emissions, and are more likely to happen soonest in regions that include biodiversity hot-spots and many low-income countries3,9.
Highlighting the genuine risk of expulsions5 and providing better estimates of when they can occur6 are both valuable. It is crucial to realize, however, that more-modest changes occurring well before an expulsion can also be a major concern. This is because such changes can still be large and rapid enough10 to have severe impacts on humans and ecosystems11.
About this article
International Journal of Climatology (2016)