Why Climate Simulations Fail

As an engineer, I've written several computer simulations for electrical circuits and the behavior of networks in FORTRAN (the language used by climate scientists at the CRU), in BASIC (a FORTRAN-like language that is more interactive) and to a lesser extent in C (a more modular, reusable language than the other two). I don't have to know every detail of climate science to know there is something rotten in the simulation models that predict temperature rise a century from now.

My simulations taught me several things:

1. There is the feeling of exhilaration with getting the computer to actually do work for you and produce a result that you can use. It's an amazing feat.
2. Some simulations exhibit divergent behavior, that is, as you add more precision, they do not converge toward the same result. This is an important warning to those that would attach too much importance to today's early results.
3. Even a simple computer model involving less than ten variables, all of which are well understood, can quickly balloon into an vastly complicated and slow simulation that would take years to complete without shortcuts. By contrast, earth's climate involves millions of variables, some with small impact, others large, all of which are not precisely understood.
4. The computer is equally loyal about producing bad results as good (remember early programmable calculators and the popularity of biorhythms?). Unless the result is outrageous (such as a climate model generating millions of degrees or below absolute zero, both being impossibilities), it's impossible to discern good data from bad without further investigation. For example, the climate simulation will never tell you if you forgot to take into account the shapes of continents or clouds or mountains when formulating your conclusions. Independent methods of verification are needed, the best being a reality check against the climate itself.
5. A simple nonlinear system is hundreds of times harder to simulate than a linear one. An example of a linear system is a guitar amp where if you pluck the string twice as hard, the speaker plays it twice as loud. An example of a nonlinear system is the same amplifier driven to distortion. When you pluck louder, the amplifier puts out a fuzzier sound at almost the same volume, the exact details depending on how the system clips at the power limits, what kind of transistors or tubes are used, how the circuit is biased, the power supply and so forth. Climate simulation is a non-linear system on many levels, just one example being the idea of "tipping points", which by their hidden nature are obscure and not well understood. It is like predicting how pleasingly an amplifier distorts and exactly at what volume, without ever having played it so. Simulating small amounts of climate change is beyond our ability today. Simulating tipping points is hundreds of times beyond that.
6. Once the first simulation is done, a second one, known as sensitivity analysis is needed to avoid another important gotcha. Let's say you simulate a circuit, double-check it, build it and prove it works as expected. So you launch your new product, only to find the factory yield is dropping as none of the circuits seem to be working. On further examination, it turns out you needed capacitors with a precision of one tenth of a percent, which are simply not available. A sensitivity analysis proves to be an important practical part of any simulation. Sensitivity analysis also provides the best hope for climate change: A slight nudge here has a larger effect there. Expensive brute force efforts to control climate won't work, not only because we can't afford them (or stay sustainably focused on them), but it goes against nature itself. Nature gives us the secrets of the best ways to solve problems, we just have to find them, that is, understand the relationships between the variables.

Climate simulations lack the robustness that comes from being testable. In electronics, many a simulation has fallen by the wayside after failing the simple reality check. The only tests so far on climate have been the results of the last fifteen years, which already contradict the simulations that predicted significant warming. But notice that my article did not even depend on this fact. My point was that the climate simulations were incapable of generating worthwhile results at the current state of knowledge. I would have discredited the models even if the last fifteen years have been correctly predicted, since it would have been a roll of the dice.

In electronics, simulations of well-defined linear circuits of ten variables (components) took years to evolve and improve. Climate simulations have just begun. They have millions of variables, the variables are not well understood, are nonlinear and the testibility of the simulations is virtually nonexistent. In short, we have decades to centuries to go before we can say the simulations predict anything of relevence.