A little while back, in a discussion with Sparko that I can't find, there was a consideration of the performance of climate models. Specifically, he wondered whether the climate models we ran a few decades ago were any good at projecting the sorts of temperatures we're currently seeing. We ended up talking about a graph produced by Gavin Schmidt, who runs NASA's GISS institute. In a tweet today, Gavin gave an update of the graph based on 10 months of 2019, so i thought it was a good follow up.
What are you looking at? These are the results of a panel of climate models run in about 2001. Colored lines are actual temperature data, analyzed by 4 different groups. The black line is the average of multiple climate models (the averaging is why it's smoother than temperatures), and the grey region is the 95% confidence interval of the model output.
Back when the models were run, we had data on actual temperatures through 2000; those are shown in the "Hindcast" part of the graph, which confirms that the models can accurately reproduce the behavior of the climate when given actual data on things like El Niño status and volcanic eruptions.
From 2000 onwards, however, the models are just being run with "typical" behavior - average levels of volcanic activity, generating their own El Niños when the conditions they simulate call for one, using the expected increase in greenhouse gasses etc. So, what you're looking at is what climate models do when left to simply use the physics of the atmosphere as we understand it.
While we don't know precisely where 2019 is going to end up (those vertical bars at the far right of the colored graphs are the expected ranges given the first 10 months), it's pretty clear that it's going to be one of the top-5 warmest years on the instrument record. And will be right in around where the models projected it might be.
Over time, things like carbon emissions will probably change in a way that differs from the expectations used in this model run, as solar and wind have gotten cheap much faster than anyone expected. So, we'd expect this particular run of the models to diverge with actual temperatures over time. But we've probably got at least another decade before that becomes an issue, so this will be an ongoing test. And if i recall correctly, Sparko asked for models run "20 years ago", so we've only got another year or so until we get into the range he was looking for.
What are you looking at? These are the results of a panel of climate models run in about 2001. Colored lines are actual temperature data, analyzed by 4 different groups. The black line is the average of multiple climate models (the averaging is why it's smoother than temperatures), and the grey region is the 95% confidence interval of the model output.
Back when the models were run, we had data on actual temperatures through 2000; those are shown in the "Hindcast" part of the graph, which confirms that the models can accurately reproduce the behavior of the climate when given actual data on things like El Niño status and volcanic eruptions.
From 2000 onwards, however, the models are just being run with "typical" behavior - average levels of volcanic activity, generating their own El Niños when the conditions they simulate call for one, using the expected increase in greenhouse gasses etc. So, what you're looking at is what climate models do when left to simply use the physics of the atmosphere as we understand it.
While we don't know precisely where 2019 is going to end up (those vertical bars at the far right of the colored graphs are the expected ranges given the first 10 months), it's pretty clear that it's going to be one of the top-5 warmest years on the instrument record. And will be right in around where the models projected it might be.
Over time, things like carbon emissions will probably change in a way that differs from the expectations used in this model run, as solar and wind have gotten cheap much faster than anyone expected. So, we'd expect this particular run of the models to diverge with actual temperatures over time. But we've probably got at least another decade before that becomes an issue, so this will be an ongoing test. And if i recall correctly, Sparko asked for models run "20 years ago", so we've only got another year or so until we get into the range he was looking for.
Comment