Climate Change Models

All models are wrong. Some models are useful. George Box

The quote above appears to be a reasonable guideline on which to discuss the different interpretations of the climate change suggested by the two graphs. We will start by acknowledging that much scientific development, both theoretical and applied, has been predicated on the insights provided by models of all manner of description. However, it also needs to be accepted that every model is invariably designed to reduce the complexity of the real-world to some manageable subset of parameters, which can be increasingly problematic given the ambition of many climate models. So, while recognising their usefulness, we also need to understand their limitations, as history is littered with models that have been proved wrong, either in their assumptions and/or conclusions.

Note: Today, it seems that even those with the ‘weight of authority’ to simply question the accuracy of climate change models may be dismissed by branding them as a ‘sceptic’ or ‘denier’ in the fashion of religious heresy. This seem unworthy of a scientific debate and while some may hold to the idea that the science of climate change is already proven, such that there is no debate, this seems unworthy of the scientific method. So, while this discussion may seem to challenge the possible accuracy of climate models, it is not intended as a sceptical viewpoint or a denial of the importance of the on-going development of these models. Rather it is simply a discussion that attempts to understand the complexity and limitations of the climate models on which so much rests.

Irrespective of the truth of the claim, there is an assumption that 95% of climate scientists agree with the climate change paradigm, which we might assume is often predicated on the predictions of various climate change models. However, climate models are a complex subject in their own right, such that most of us have to defer to the weight of authority in order to understand some of this complexity. Unfortunately, authority exists on both sides of the debate, such that we will first consider the concerns of those who question the accuracy of the current climate models.

Note: By way of a quick introduction, the reader might wish to review a 5-minute YouTube video by Will Happer, who is the Emeritus Professor of Physics at Princeton University. Alternatively, there is a more detail paper entitled ‘Climate Model for the Layman’ published in 2017 by Professor Judith Curry, who is the author of over 180 scientific papers on weather and climate and is currently President of Climate Forecast Applications Network.

While the reader is urged to review the Judith Curry paper for themselves, this discussion will provide a brief overview of the workings of climate models. So, by way of an initial summary, it might be stated that climate models continue to be a relatively coarse-grained simplification of the complexity of Earth’s multiple climate sub-systems. On the macroscopic scale, these climate sub-systems have to consider the atmosphere, ocean and land surface, inclusive of sea ice, glaciers plus other possible factors, e.g. space weather. Within the confines of the atmospheric sub-system, multiple factors have to be taken into consideration, e.g. energy input, altitude variance, cloud types and formations, winds, temperature, humidity and atmospheric pressure etc. Naturally enough, such complexity extends into the other major climate sub-systems, i.e. oceans and land, where the climate model attempts to approximate all the interactions using mathematical equations.

Note: Many processes in the atmosphere and oceans are known to be non-linear, such that there may be no simple relationship between cause and effect. The nonlinear dynamics of the atmosphere and oceans are often linked to the Navier–Stokes equations, but where the Navier–Stokes existence and smoothness problem is described in terms of fluid mechanics subject to turbulence. However, theoretical understanding of these equations is still incomplete, such that accurate modelling might also be questioned.

So, while some of the equations in the climate models can be based on the classical laws of physics, e.g. Newton’s laws of motion and the laws of thermodynamics, there other important processes in the model that can only be approximated. As suggested in the note above, climate models often to have attempt to resolve the interaction between two essentially chaotic fluids, i.e. the atmosphere and the ocean, which can be highly dependent on any initial assumptions, which are then amplified by the near-chaotic non-linearity associated with the Navier–Stokes equations.

What other complexities need to be considered?

 In addition to the macroscopic complexity, there is also a complexity associated with the microscopic scale of interaction, such that the models are often limited to the processing capacity available.

For example, many models divide the atmosphere, oceans, and land into 3D cells, where latitude and longitude define a surface on the scale of 100-200 kilometres, while the height of the atmosphere and depth of oceans may be restricted to about one kilometre. Interactive events between these cells is then time-stepped with a typical granularity in the order of 30 minutes. While higher resolution may become increasingly possible, given Moore’s law, a doubling of the resolution outlined may require a 10-fold increase in processing capacity. So, given the implied coarseness of the resolution, important processes associated with clouds and rainfall have to be aggregated using simplified approximations, although they may be derived from observations or from more specific and detailed process models. However, today, the detailed modelling of clouds and precipitation still appear to be a problem, which may be responsible for the wide predictive variance of so many of the climate models.

So what about the model results?

Clearly, there may be an interpretive bias depending on which side of the climate debate is preferred. However, some have argued that because of the intrinsic limitations of the climate models, as outlined, there is evidence that these models may be overstating both the temperature increases and the sensitivity to CO2, as reflected in the opposing graphs at the beginning of this discussion. However, these two graphs also highlight another problem in that there is now disagreement of the accuracy of the climate models when judged against different interpretations of measured temperature data over the last 100 years or so, although this issue will be deferred to the next discussion. Of course, the controversy also extends to those who question the accuracy of the climate models because it is believed that they do not give sufficient emphasis to other factors, such as cooling associated with the various solar cycle mechanisms, such that the IPCC predictions for the 21st century may need further scrutiny, research and debate.

Note: While there may be legitimate reservations about climate models, their supporters may rightly highlight that they do provide the best insight into what is happening to the global climate, past, present and future. For they claim that these models provide an understanding of how the climate system works, reproduces past climate states, predicts future global climate change, explains extreme weather, provides the information for policy change and helps quantify the social cost of carbon.

However, the lay-person might reasonably assume that the problem of retro-fitting the climate models to match past data and projecting current assumptions into the future to be very different propositions. For we might reasonably assume that the past is known, such that the climate models can ‘simply’ be run backwards in time from current data, such that they can be proved to match historical climate records, although it is unclear to what extent this reversal has been proved. Of course, future predictions will always be more speculative, especially given the complexity of so many variables and the fact that there is no definite reference against which to judge these predictions. Therefore, we might simply try to summarise this situation as follows:

Models tend to converge on what we know,
but not necessarily on what we do not.