|
Supplementary material to “Modeling Earth Dynamics: Complexity, Uncertainty, and Validation”
21 December 2010
A. Neri, Istituto Nazionale di Geofisica e Vulcanologia, Pisa, Italy
Citation:
Neri, A. (2010), Modeling Earth dynamics: Complexity, uncertainty, and validation, Eos Trans. AGU, 91(51), 506, doi:10.1029/2010EO510012. [Full Article (pdf)]
A special session with the title "Can our models only predict the irrelevant?” was also held during the conference. The session originated from an idea of Prof. Albert Tarantola, former President of CMG, who unfortunately passed away in December 2009. "The Committee on Mathematical Geophysics could perhaps take seriously the issue of the very worth of mathematical models in the geophysical and environmental sciences (…). Are we honest when building and "selling" mathematical models that are dangerous, because they give to society a false sense of security? Can our models only predict the irrelevant?". This was the stone that Albert threw in the pond during the early stages of the conference organization. The theme was introduced by five leading scientists who discussed the worth and role of mathematical models in terms of predictability and decision making for different geophysical and environmental applications. Sceptical and supporting voices were both present, and talks were followed by a wide-ranging panel and floor discussion around this theme.
The session started with a thought-provoking talk by Orrin Pilkey (Pilkey and Pilkey-Jarvis, 2010) who argued that mathematical models cannot predict the evolution of processes on the Earth surface. The author claimed that quantitative modelling of the Earth surface processes, at accuracies useful for engineering purposes, is not possible. The reasons for that included the observation that many models (including the climate models) involve human behaviour (considered unpredictable) and often over-simplified or meaningless parameters. Moreover the critic claimed that quantitative models are often verified by hindcasting but this will not guarantee that the model will successfully predict the next event. Examples were discussed with specific reference to the modelling of shore retreat and parameters affecting the long shore sand transport (e.g. Bruun model). According to Pilkey, for such systems, because of ordering complexity, the predictive capabilities of simple correlations are very limited. However, it was agreed that such criticisms particularly apply to correlation-based models that do not aim to describe the dynamics of the system by solving the full set of transport equations but simply provide a rough estimate of the process to be used for engineering purposes.
The session continued with the talk by Warner Marzocchi (Marzocchi, 2010) about the present capabilities to forecast large earthquakes and volcanic eruptions. The author discussed the challenges of this task and particularly the difficulties produced by the lack of data for large events that prevents an accurate testing of the predictions. Two possible approaches were outlined: a) the extrapolation of statistical distribution of small-to-moderate events to the largest ones, and b) the use of subjective probability to define ad hoc statistical distributions based on all available strands of information available. However, in both cases the problem of testing remains given the scarcity of data. The main conclusion of the talk and following discussion was that there is no best strategy for all problems but this depends on the problem type, end-user, and final purpose of the prediction.
Uncertainty and limitations of models of terrestrial ecosystem processes were then introduced and discussed by Victor Brovkin (Brovkin, 2010). Such processes are in fact largely controlled by changes in climate and carbon dioxide concentration that in turn control heat and momentum fluxes between land surface and atmosphere. Numerical experiments with coupled atmosphere-ecosystem models highlight a strong coupling between the sub-domains as well as the existence of multiple steady-states of the system. Moreover, the neglect in the models of the ecosystem response to increase of carbon dioxide concentration and temperature as well as of nutrient limitation of carbon uptake introduces further large uncertainties in the predictions. On the other side, an adequate representation of the land surface heterogeneity would require very high spatial resolution beyond the present computational capabilities. However, it was recommended that new accurate ecosystem processes need to be added to high-resolution Earth system models to reproduce palaeodata and improve the accuracy of predictions.
The discussion of the worth of climate models was extended further by introducing the risk management perspective of climate change. Roger Cooke (Cooke, 2010) presented an original mathematical look at economics of climate change. In particular Cooke highlighted how current economic models of social costs of carbon are based just on expected values without proper account of the uncertainty affecting the system evolution. This is very different from what others risk takers do by guaranteeing solvency in a 1-in-500 year loss event. Based on the current estimates of the International Panel for Climate Change, expected increase of temperature from doubling CO2 concentration would largely overcome any available resource. The author concluded that quantifying the uncertainty of the system in a risk management perspective would greatly contribute to create an effective policy and plan to face a global temperature increase. But at the same time, the author highlighted the sensitivity of the cost/benefit analysis on the way damages and capital terms are expressed and quantified.
The session was closed by a comprehensive talk by Lenny Smith (Smith, 2010). In this case the question was: can the diversity of our models inform our uncertainty of the future? Smith introduced the topic by clarifying that the real question is not whether models are perfect but whether they are useful. Starting from the well-known Tukey's maxim "Far better an approximate answer to the right question, which is often vague, than an exact answer to the wrong question which can always be made precise", he pointed out that, given all our models are empirically inadequate, precise probability forecasts based on model-output should not be interpreted naively as decision-relevant probabilities. So, how our simulations can provide a useful contribution? Examples of several "vague answers" were illustrated highlighting the crucial contribution given by model results in terms of insight. Therefore, even if one cannot often take model outputs literally given the range of systematic errors, many of today's models of the various geophysical processes play a critical, positive and informative role. For instance this appears the case for climate global models, that all confirm how increasing greenhouse gases will inevitably result in a warmer planet.
A wide discussion followed the five introductory talks, bringing into the debate also recent examples of policy-relevant modelling applications, such as the ash dispersal modelling forecasts produced during the recent Eyjafjallajökull volcano eruption and the short-term earthquake forecasts proposed for risk mitigation purposes. It was summarized that, although it is not possible to draw the same conclusions for all applications, model outcomes can actually provide fundamental insights for our understanding of many geophysical systems as well as, quite often, relevant key information for decision making. Defining the "right questions" and quantifying the approximation of our "vague answers" appear to remain the key future challenges.
References
Brovkin, V., (2010), Limitations of global terrestrial biosphere models used for future climate projections, Book of abstracts of the 28th International Conference of the Committee of Mathematical Geophysics, Tipografia Editrice Pisana, May 2010, 200 pp.
Cooke, R., (2010), The risk management perspective in climate change, Book of abstracts of the 28th International Conference of the Committee of Mathematical Geophysics, Tipografia Editrice Pisana, May 2010, 200 pp.
Marzocchi, W. (2010), Forecasting large earthquakes and eruptions: is it a scientific issue?, Book of abstracts of the 28th International Conference of the Committee of Mathematical Geophysics, Tipografia Editrice Pisana, May 2010, 200 pp.
Pilkey, O., and Pilkey-Jarvis, L., (2010), Predictive modeling of processes on the surface of Earth doesn't work, Book of abstracts of the 28th International Conference of the Committee of Mathematical Geophysics, Tipografia Editrice Pisana, May 2010, 200 pp.
Smith, L., (2010), Extracting insight from predictions of the irrelevant: can the diversity in our models inform our uncertainty of the future?, Book of abstracts of the 28th International Conference of the Committee of Mathematical Geophysics, Tipografia Editrice Pisana, May 2010, 200 pp.
Acknowledgments
The author is grateful to all conference participants and organizers that contributed to the success of the meeting. A special thanks goes to Daniel Rothman and Alik Ismail-Zadeh for their help in the organization of the programme, to Enzo Boschi for his support to the initiative, and to Sara Barsotti, Mattia de'Michilei Vitturi, and Tomaso Esposti Ongaro for their daily enthusiastic contributions to this event.