Wednesday 5 August 2020

The Very Model of Climate Change Concern


A friend of mine, JP, started all of this when writing:

If you were to ask me 2 years ago what my key understandings were about climate change, I would have said the following:

Sea ice is rapidly shrinking (summer arctic sea ice to be gone by 2015)
Sea levels are rising and accelerating
Polar bear populations are under stress (have increased in the last 20 years)
The levels of glacial retreat around the world are unprecedented (similar retreats have been seen in the last century)
97% of scientists agree that global warming is real and an urgent problem
Any scientist who is skeptical about the claims made about climate change is a "denier" and is funded by oil/resource companies
We are seeing an increase in extreme weather events (they are actually getting less common)
Climate models are accurate in their predictions 

Every one of those things is either totally false, or a largely exaggerated claim.

This is the eighth in a series based on my response, which itself was split over a few emails.  The first was Ice Extent Challenge (in which I provided a little more context about JP) and was followed by Sea Levels Rising, Polar Bears and Climate Change, Glacial Retreat, A Worry of Climate Change Scientists, Denying Denialism and Weathering a Storm of Climate Denial.  Some of the issues may also be touched on in a series of articles on the nature of climate denialism.  Please also note the caveat.

---


Note that JP is most familiar with statistical modelling.  JP has modelled demand on a supply chain and it’s amazingly accurate.  Based on statistical data from previous years, and possibly a whole suite of data that I am not aware of, JP can get very close to current demand.  It’s brilliant stuff.  However, it should also be noted that climate models are not, repeat NOT, and – just to make this perfectly clear – climate models are NOT statistical models.  They are models of physical processes.

I’ve long been of the opinion that models are never entirely accurate, no matter what sort of model they are, and that the only completely accurate model of the universe is … the universe.  This applies also to systems that are smaller than the universe, even though processing power necessary to do the modelling decreases significantly as you approach more human-sized scales.

You don’t need to take my word for it, you can just consider the three body problem.  Start with two bodies in otherwise empty space that are acting on each other gravitationally, say a star and a planet.  They will orbit the centre of their combined mass with a speed and separation that is inversely proportional to their relative mass (the heavier body will have a smaller separation from the centre of the combined mass than the lighter one, they will share the orbital period and thus the lighter one will have a higher speed having a greater distance to cover – with something as massive as a star compared to a planet, ).

Then consider three bodies, specifically two stars and a planet – like Tatooine:


Note that the planet orbits the pair of stars, since it’s the light one of the three-some.  This is not solvable analytically, it can only be solved numerically, meaning that you have to plug the figures in an run the problem (like at the link).  Now in the second program window, if you fiddle with the code to make one of the stars more than twice as heavy as the other, something strange happens (at a factor of 2.2):


At higher values, the binary stars lose the planet entirely (pretty much immediately at a factor of 3).

Now imagine trying to predict where the planet in the image above will be after 50 years.  It’s easy enough to run the simple program and find out, but you have some issues that are going to affect your results.  How accurate is your value of G?  How accurate are your masses?  How accurate were your initial conditions (speed and direction)?

Your inaccuracy in any of these figures are going to affect your results and that’s when you are assuming a spherical cow.  The bodies in question are not perfect point masses, they are two balls of gas and … a planet, which might be a third, smaller ball of gas or a conglomeration of rocks – and we are assuming only one single rock in the system.

The upshot of this is that we can predict with great accuracy where the planets in our solar system will be in the short term, but have no idea where they will be in 100 million years – and it’s not that we just haven’t worked it out yet but rather that, at that longer scale, the solar system is chaotic.

Now consider a model involving 1044 gas molecules (the atmosphere), interacting with 10 million cubic kilometres of water (admittedly only the top layer of it, so more than 350 million square kilometres times the average significant depth, whatever that is) and just under 150 million square kilometres of land surface which is covered variously by deserts, forests, ice, mountains, lakes, grasslands and, relatively recently, cities and towns.  Then factor in the vagaries of the sun, which has its own cycles which do affect us (for example with solar minima, grand or otherwise) and all the different products that end up in the atmosphere (CO2, methane, aerosols, water, etc).

Climate models don’t even bother trying to be accurate.  Instead, they do what is call parametrisation, breaking the planet up into gridboxes – the size of which vary depending on the precise application (and the computing power available).  The smallest gridboxes are usually in the range of about 5km, with clouds sometimes being modelled at 1km – and this is way too coarse for most clouds.

Even at the finest parametrisation used today, there will still be inaccuracies inherent in the model.  These are expected and discussed with an eye to minimising them (see “Are Climate Change models getting better?” which starts on page 824 at the linked document).  In part because the models are never perfect, the IPCC and climate scientists in general talk about trends and projections rather than predictions.  The trend observable in all the models across the standard scenarios (RCPs 2.6, 4.5, 6.0 and 8.5) is that the more you increase the amount of CO2 in the system, the more the temperature rises.

Again, you don’t need to take my word for this.  The outputs of the models are freely available (admittedly they take some time to figure out).

I compared the results against the temperature measurements available from NASA and NOAA which, without a model overlay, looks something like this:


Once I converted to absolute (rather than anomaly) values and corrected for an oddity in starting temperatures, the HADGEM2-ES model results look like this:


Be very careful about just accepting this though.  I don’t understand the starting temperature oddity, but basically not all models had the same temperatures at the beginning of the run, or the same date for the beginning of the run.  I fixed for that and used this model as an example because, without that correction, the results were offset by about 0.2 degree (noting that I used an RCP8.5 run).  The corrections I made did not change the trend, they just shifted the entire yellow line up a bit.  Additionally, I don’t know precisely what was fed into the model.  They seem to have included some volcanoes (which pump aerosols into the atmosphere and push down the temperature for a short period) and anthropogenic aerosol production seems to be accounted for, but this is just guesswork on my part.

With those caveats in mind, this is what the results look like from a raft of models (note that HADGEM is not in this one, so the yellow line represents output from a different model):


There’s a range of about a degree between the models, and about four of them are consistently higher than the measurement (although, prior to my adjustment, they were all below so perhaps I’ve messed it up somehow – and that messing up could be contributing to the spread as well) … but note that the overall trend is precisely the same.  Over the period 1880-2020, there is a rise in temperature of about a degree – which is consistent with the measurements.

Therefore, noting that the models are not expected to be wholly accurate, indicating temperature rise in pretty much precisely what we have experienced, and that they provide projections rather than predictions, JP is simply wrong when claiming the climate change models predictions projections are inaccurate.  They are a lot more accurate than one might have predicted.