Models Blur Science and Advocacy: A Note from Ian Read
MANY mainstream media science, economic and environmental journalists are not sufficiently trained to be aware of the limitations of models when they present climate-modelled output computated projections not only as data but also advocate this output as supposed proof of the threat posed by anthropogenic global warming, particularly with regard to runaway or catastrophic climate change. This disjunct between the scientific and media presentation when contained within the paradigm of advocacy represents a threat to the integrity and falsifiability of science.
Science seeks the truth in knowledge; (some) media advocacy seeks to propagandise this knowledge. The impact is reinforced if a climate scientist/modeller is directly quoted as an expert, further blurring the line between science and advocacy. This has societal repercussions as the science of anthropogenic global warming (AGW) and the perceived impact of runaway or catastrophic climate change is so model-dependent that the citizenry is not always able to differentiate between the science and advocacy – the implications of which, as regards policy development in term of climate change mitigation, are likely to have a profound effect on society.
Climate models are used, in part, to determine future climate change scenarios related to anthropogenic global warming (AGW) and are described by the Intergovernmental Panel on Climate Change (IPCC) as “mathematical representations of the climate system, expressed as computer codes and run on powerful computers.”
Furthermore, the IPCC states that climate models:
“Are derived from fundamental physical laws (such as Newton’s laws of motion), which are then subjected to physical approximations appropriate for the large-scale climate system, and then further approximated through mathematical discretization. Computational constraints restrict the resolution that is possible in the discretized equations, and some representation of the large-scale impacts of unresolved processes is required (the parametrization problem).”
In other words a climate model is a numerical model or simplified mathematical representation of the Earth’s climate system, or parts thereof. It includes data from real world observations and creates parameters or variables for the unresolved or unknown processes.
The ability of a model to simulate interactions within the climate system depends on not only the level of understanding of the physical, geophysical, chemical and biological processes that govern the climate system but on how accurately these processes are expressed as algorithms within the model, and how closely they represent real-world data. These models do contain some well-established science but they also contain implicit and explicit assumptions, guesses, and gross approximations, referred to as parameters (the parametrization problem mentioned above), mistakes in any of which can invalidate the model outputs when compared to real world observations. In other words computer models are just concatenations of theoretical calculations; as such they do not constitute evidence.
Climate models are data and parameters dependent. Data is based on direct or indirect observations from the environment; parameters (or parametrizations) are defined by the IPCC as:
“Typically based in part on simplified physical models of the unresolved processes . . . Some of these parameters can be measured, at least in principle, while others cannot. It is therefore common to adjust parameter values (possibly chosen from some prior distribution) in order to optimise [author’s italics] model simulation of particular variables or to improve [author’s italics] global heat balance. This process is often known as ‘tuning’.”
Tuning is considered justifiable if two conditions are met: that parameter ranges do not exceed observational ranges where applicable (though this does not necessarily constrain parameter values, which could lead to model output problems); that adjusted (or tuneable) parameters are allotted less degrees of freedom than in the observational constraints used in the model’s evaluation. The IPCC states that,
“If the model has been tuned to give a good representation of a particular observed quantity, then agreement with that observation cannot be used to build confidence in that model. However, a model that has been tuned to give a good representation of certain key observations may [author’s italics] have a greater likelihood of giving a good prediction than a similar model . . . that is less closely tuned.”
Herein lies a problem with modeling: that last sentence implies a subjective judgment on the part of the modeler regarding the greater likelihood of the model providing a good prediction than a less closely tuned similar model. In other words there is the possibility that tuneable parameters can be used as a ‘fudge’ factor, in either model prediction or hindcasting (making the model fit already observed data).
Prominent climatologist Richard Lindzen, writing in “Climate Science: Is it currently designed to answer questions?” a paper he presented at the “Creativity and Creative Inspiration in Mathematics, Science, and Engineering: Developing a Vision for the Future” conference held in San Marino, August 2008, summarises the problem thusly:
“Data that challenges the [AGW] hypothesis are simply changed. In some instances, data that was thought to support the hypothesis is found not to, and is then changed . . . Bias can be introduced by simply considering only those errors that change answers in the desired direction. The desired direction in the case of climate is to bring the data into agreement with models, even though the models have displayed minimal skill in explaining or predicting climate. Model projections, it should be recalled, are the basis for our greenhouse concerns. That corrections to climate data should be called for is not at all surprising, but that such corrections should always be in the ‘needed’ direction is exceedingly unlikely. Although the situation suggests overt dishonesty, it is entirely possible, in today’s scientific environment, that many scientists feel that it is the role of science to vindicate the greenhouse paradigm for climate change as well as the credibility of models. Comparisons of models with data are, for example, referred to as model validation studies [author’s italics] rather than model tests.”
It needs to be kept in mind that computer climate models do not output data: their results are simply computations of the input data. Obviously then, the accuracy or otherwise of the computated output is dependent upon the accuracy of the input data. Furthermore, a climate model’s output is only reliable to the degree that the model’s performance can be validated, not necessarily by comparisons with other models but from raw data recorded or observed from the real world. Of course, tuned parameter corrections may be legitimate but only if they include both those corrections that bring observations into agreement with the model, and those that do not – to exclude the latter is to obfuscate the model’s outcome through omission.
In climate science the most notorious example of obfuscation through omission is what has become known as Mann’s Hockey Stick. Lindzen again:
“In the first IPCC assessment (IPCC, 1990), the traditional picture of the climate of the past 100 years was presented. In this picture, there was a medieval warm period that was somewhat warmer than the present as well as the little ice age that was cooler. The presence of a period warmer than the present in the absence of any anthropogenic greenhouse gases was deemed an embarrassment for those holding that present warming could only be accounted for by the activities of man. Not surprisingly, efforts were made to get rid of the medieval warm period . . . The most infamous effort was that due to Mann et al . . . which used primarily a few handfuls of tree ring records to obtain a reconstruction of Northern Hemisphere temperature going back eventually a thousand years that no longer showed a medieval warm period. Indeed, it showed a slight cooling for almost a thousand years culminating in a sharp warming beginning in the nineteenth century. The curve came to be known as the hockey stick, and featured prominently in the next IPCC report, where it was then suggested that the present warming was unprecedented in the past 1000 years. The study immediately encountered severe questions concerning both the proxy data and its statistical analysis.”
The Mann Hockey Stick has since been discredited by two independent assessments, both statistically and by reference to historical and archeological records, though his initial claim that the current (late 20th century) warming is unprecedented remains within the lexicon of adherents to the AGW hypothesis.
There is a problem here for the reliability of science when models fail, either through prediction or hindcasting, but are still given the same validity as observed or model input data. One could suspect that advocacy is overriding science in this instance. While advocates and politicians might think that the science of AGW is settled scientists and climate modelers need to be able to, and be seen to, separate clearly what is science and what is advocacy otherwise their research may be subjected to political manipulation.
The computated output of climate models, often used in conjuction with models from outside the field of climate science, have been used to construct climate change scenarios, often abbreviated as SRES, an acronym for Special Report on Emission Scenarios. SRES was developed by the IPCC to develop scenarios with which to analyze, according to SRES:
“How driving forces may influence future [greenhouse gas] emission outcomes and to assess the associated uncertainties. They assist in climate change analysis, including climate modeling and the assessment of impacts, adaptation, and mitigation. The possibility that any single emissions path will occur as described in scenarios is highly uncertain . . . Any scenario necessarily includes subjective elements and is open to various interpretations.”
The output of SRES models, alternative views of how the future may unfold, are termed projections. Projections are often stated or implied erroneously, particularly in the media in connection with runaway climate change, as forecasts. This creates the impression that the SRES model output is new data, even proof, as opposed to being simply a projection of computated input data and parameters from a number of sources within and beyond the field of climate science.
Multi-model SRES climate change scenarios are said to create an ensemble of climate change projections. Modellers then consider the spread of these SRES projections, upon which has been built the notion that if the spread is close together then they can have confidence in the projections while if the spread significantly differs then there is uncertainty about the projections even though they may offer a range of possibilities.
The SRES approach is problematic: it is assumptive; prone to exaggerated errors; unscientific. Firstly, assumptions are made about unresolved processes by using tuned parameters while, secondly, errors may be exaggerated by (i) using the computational outputs as new ‘data’ by reintroducing that ‘data’ as inputs into a new model, upon which new projections are predicted and (ii) it assumes not only that a close spread within an ensemble raises the confidence of the prediction but also that a broad spread, rather than disproving the accuracy or otherwise of the models, indicates a range of possibilities, though with a lower (assumed) confidence. Thirdly, multi-model SRES outputs are based on so many assumptions that its use is inherently unscientific because many of the model elements are not falsifiable. It is, nonetheless, a good tool for advocacy though, especially when presented in the guise of science.
This SRES approach has no place in the scientific processes as its outputs can not be verified with real world data: projections are not records and models are not data generators. As yet there is no scientific principle that says that one can derive valid estimates from model outputs until the model output resembles the observed non-modelled data. The uncertainty of an ensemble of climate change projections will always depend on the accuracy of the raw data input irrespective of the spread of projections.
This is not to say that there is no place for models in climate science: even though they are a tool and not data generators there are there are many examples of statistical climate forecasting models providing good projection examples over short time frames. It is important that models, in this context, remain a tool of climate science and not a tool of advocacy.
A problem with climate modelling is that of replication and validation. Because models are tools used in order to calculate, usually complicated, data inputs it is important that the computated outputs can be tested (the models are validated, or not) and repeated (the models’ results are replicated, or not) – by doing so helps remove bias (the models’ outputs are easily influenced by data inputs, flux adjustments and parameterisation) thus increasing the confidence that the models do in some way represent what they are seeking to show. Not to do so means that the models’ computated outputs could be used for non-scientific purposes, such as advocating a predetermined position, without the ability for others, perhaps affected by this advocacy, having the ability to ascertain that the models’ computated outputs actually represent what they seek to show.
This is especially true for climate science, and the repercussions that the computated outputs have on public policy as regards AGW, climate change, tipping points, emissions trading schemes, etc. Furthermore, as much climate science is publicly funded through government grants, etc then it is even more imperative that the funders, ie the public, receive information that can be trusted. It is unfortunate, then, that there is a reticence for some climate scientists and modellers not to share data, especially all the codes or algorithms used that would allow the models to be fully replicated and, if necessary, challenge the validity of the models. This reticence goes to the core of scientific thought and process.
Notes and Links
Ian Read is a researcher, author and geographer with a special interest in climatology and vegetation. He has written over twelve books including The Bush: A Guide to the Vegetated Landscapes of Australia and Australia: The Continent of Extremes – Our Geographical Records.
This article was previously published at On Line Opinion on 23 June 2009 and is reproduced here with permission from the author.