We will produce a so-called “return time plot” of total accumulated precipitation for DJF2013/2014 (ie. for winter that year) for South England/Wales for the two ensembles: the world as observed and the world that might have been. We will start plotting these as soon as the first results are returned from our participants and will update them every day, adding the new ensemble members that we received during the previous day. Return time plots are not straight forward to interpret but by showing how they evolve when we add more ensemble members each day from the two different experiments, we hope that people will understand better what these plots tell us.
The plots show the chance of an event (on the vertical axis, for example a total rainfall amount of more than 450 mm during the whole winter) occurring in a given year. In the next Figures, each individual ensemble member is represented by a dot. In blue, these are ensemble members from the “current winter as observed” experiment and in green, the ensemble members of the “current winter in the world that might have been” experiments.
If the blue and green dots lie on top of each other, then this means that the chance of a given rainfall event to occur is the same in the observed world and in the world that might have been, and in that case, we would conclude that human-induced climate change did not have an influence on the rainfall amount for that particular winter.
If the blue and green dots lie clearly on different “lines” (as for example in Figure 4 in Otto et al.,2012), then for a given extreme event, for example a rainfall total of more than 450 mm for the winter 2014, we can read different chances of occurrence on the horizontal axis. This means human-induced climate change increased the chances of that extreme event occurring from 1/500 years to 1/200 years, and therefore climate change has more than doubled the chances of this type of extreme event occurring compared to the world that might have been.
The next Figures illustrate how an increasing number of model simulations changes the shape of the return time curves using data obtained from previous experiments, for the winter 2011/2012 (which was not an extreme winter).
In the 2014 experiment we have set up, we will have 12 different possible “worlds that might have been” (as described above, we used 12 different SST patterns from the CMIP5 models), so this will give us the information about the uncertainty. The uncertainty in simulating “the world that might have been” is larger than in our simulations of the “world as observed” because we do not have observations from the “world that might have been” and thus can only estimate the Sea Surface Temperatures (SSTs) associated with a climate without human greenhouse gas emissions. To account for this larger uncertainty we do not use only one possible pattern of associated SSTs but 12 different ones.
We will plot big green dots for the results from all models taken together and, we will show each of the ensemble members from these 12 different “worlds that might have been” as little light green dots. The green dots then (like on the specimen below) show the average over the 12 individual experiments.
The little green dots form a sort of cloud representing the uncertainty: if the blue dots lie outside of this green cloud, we can conclude that climate change played a significant role in the wet winter. It is key for the results to have this indication about the uncertainty, because we can say that climate change played a role (making wet winters more or less likely) only if the two experiments are statistically different from each other (in other words, when the blue dots lie outside the green cloud).
The first Figure below has been produced with only ten ensemble members per experiment. By definition, the lowest chance of an event to occur is 1/10 years, as we only have 10 realisations per experiment. These 10 realisations can simply be interpreted as what we would randomly obtained from 10 members of the public who ran the experiment a couple of days after the launch of the experiment. At this stage, some blue dots lie above green dots, some below.
The second Figure shows the full ensemble (around 2000 members, the one we plan for 2014 will be much larger, around 30,000). The rare events (1/100 years and rarer) appear indistinguishable between both experiments, but for less rare events, there does seem to be a tendency for risks in the “world that might have been” to be somewhat lower than risks in the “world that is”.
These examples illustrate that very large ensembles are needed to capture rare events and that with too small ensembles, no thorough conclusions can be drawn. We can also see how the result will emerge as runs are contributed by the general public: each point on these plots is someone’s simulation.