The Challenges and Realities of Climate Modeling

Computer modeling is central to climate science. Climate models help us understand how the climate system works, why it has changed in the past, and most importantly how it might change in the future. But we have to keep in mind that usefully describing the Earth's climate, is one of the most challenging scientific simulation problem.

So, how good are our climate models? And how much confidence should we have in what they say about future climates? To answer that, lets talk about how the models work. All but the simplest climate models begin by covering the Earth's atmosphere with a three dimensional grid, typically ten to 20 layers of grid boxes stacked above a surface grid of 100 km by 100 km.

Grids covering the ocean usually have smaller grid squares closer to 10 km by 10 km. With the grids in place, the computer models use the fundamental laws of physics to calculate how the air, water and energy in each box at a given time move to neighboring grid boxes at a slightly later time.

This process is then repeated over and over. It all sounds straightforward, but its not at all easy, in fact, its excruciatingly difficult. And anyone who says that climate models are just physics, either doesn't understand them or is being deliberately misleading. One major challenge is that the models use only single values of temperature, humidity and other variables to describe conditions within the entire grid box.

For example, they ignore the temperature difference between Palo Alto and Stockton, which are 100 km apart. Another is that models have to be initialized. But even with our most sophisticated observation systems, getting the right starting points in each grid box is really difficult. Considerable judgment is required by the modelers, and since different modelers make different assumptions, results can vary widely.

Next comes tuning, which is the process of adjusting the model to deal with troublesome inconsistencies or irksome uncertainties. Sometimes modelers are tuning parameters in ways that aren't based on their knowledge of the parameter, but rather are aimed at producing a desired result. The documentation for one model tells of tuning a parameter related to convection in the atmosphere by a factor of ten, because the originally chosen value resulted in twice as much warming as had been observed.

You'd think, based on popular reports that all models arrive at nearly the same answer, but that's not true either. If you delve deeper into the IPCC models, you'll find that the results presented are averages of models that disagree, often wildly, with one another. One particularly jarring failure is that the simulated global average surface temperature, not its changes, varies among models by about three degrees Celsius, which is three times greater than the observed 20th century warming that they are attempting to describe and explain.

And incredibly, as the models have gotten more sophisticated with finer grids and more sub-grid parameterizations, the uncertainty has increased rather than decreased. Historical analysis of the same models also don't inspire confidence. An analysis of hundreds of simulations run by the latest models used by the IPCC, shows that they do a very poor job of describing the warming over the past century.

That models can't reproduce the past is a big red flag, it erodes confidence in their projections of future climates. Its uncommon for the popular media to discuss how problematic our climate models are. Those uncertainties should make us skeptical of their ability to predict human caused consequences in the climate.

I believe a fairer assessment would be something like, the uncertainties in modeling of both climate change and the consequences of future greenhouse gas emissions. Make it impossible today to provide reliable quantitative statements about relative risks, consequences, and benefits of rising greenhouse gases to the Earth's system as a whole, let alone to specific regions of the planet.

As the reports themselves often note, the models are the best we've got, and they are getting more sophisticated. But for now, they are beset with myriad problems, and I wish they were better.