Environmental Modelling

An Uncertain Future?

Keith Beven

• Editorial: CRC Press
• Año: 2018
• ISBN:
  9781498717977
• Edición: 1
• Categoría: Ciencias biológicas
• Categoría: Ciencia medioambiental

Uncertainty in the predictions of science when applied to the environment is an issue of great current relevance in relation to the impacts of climate change, protecting against natural and man-made disasters, pollutant transport and sustainable resource management. However, it is often ignored both by scientists and decision makers, or interpreted as a conflict or disagreement between scientists. This is not necessarily the case, the scientists might well agree, but their predictions would still be uncertain and knowledge of that uncertainty might be important in decision making.

Environmental Modelling: An Uncertain Future? introduces students, scientists and decision makers to:

• 
  
  

• the different concepts and techniques of uncertainty estimation in environmental prediction

• 
  
  

• the philosophical background to different concepts of uncertainty

• 
  
  

• the constraint of uncertainties by the collection of observations and data assimilation in real-time forecasting

• 
  
  

• techniques for decision making under uncertainty.

This book will be relevant to environmental modellers, practitioners and decision makers in hydrology, hydraulics, ecology, meteorology and oceanography, geomorphology, geochemistry, soil science, pollutant transport and climate change.

A companion website for the book can be found at www.uncertain-future.org.uk

Chapter 1

How to make predictions

As we know, there are known knowns, there are things that we know we know. We also know there are known unknowns, that is to say, we know there are some things we do not know. But there are also unknown unknowns, the ones we don’t know we don’t know.

Donald Rumsfeld, former US Secretary of Defense, February 12th 2002

What men really want is not knowledge, but certainty.

Bertrand Russell, 1964

1.1 The purpose of this book

This book is primarily intended as a discursive examination of the process of uncertainty estimation in environmental modelling as it is done now and how it might be done more “properly”, and perhaps more “realistically”, in the future. It is intended to be a book that might be useful to students, graduate students, and practitioners interested in increasing their understanding of different methods of uncertainty estimation and how that understanding might be used in decision-making. It is written with both users of environmental models and decision makers in mind; particularly those who have not much previous exposure to uncertainty concepts. It can be read without reference to all the detailed technical material in the Boxes that follow the different chapters. It cannot, of course, be a comprehensive account of uncertainty estimation in all the different disciplines that comprise environmental science. It is intended to be much more a first guide on how to think about the modelling process and choose uncertainty estimation and decision making techniques appropriate to a particular application. It cannot go into all the details and software available for each technique but references are provided to allow the reader to explore further.

It is based on a long experience of trying to cope with predictive uncertainties in one particular branch of environmental science, that of hydrology. As such, it is coloured by the particular problems of hydrology, although these are not so different from many other areas of environmental modelling. Modelling and prediction in hydrology are important in providing information for the practical management of natural resources and natural hazards. Hydrology, however, also has severe limitations as a science resulting primarily from limitations in measurement techniques at the scales at which we want to make predictions. This is particularly the case for the examination of flow processes underground which is where many of the interesting and active hydrological processes take place.

Similar limitations can be found in most areas of environmental science, whether physical, chemical or biological process in the earth, atmosphere or oceans. In all cases it will be difficult to make measurements at the scales at which we wish to make predictions. It will be difficult to define the boundary conditions for a system of interest and, for time-dependent processes, the initial conditions everywhere within the domain of that system. It will also be difficult to specify the physical, chemical and biological characteristics of the domain. Thus, even if we have some understanding of how the system is working, these factors will make prediction difficult. How, then, to make predictions in the face of such difficulties?

One, necessary, answer to this question is “approximately”. The very act of prediction involves simplification of the complexity that is the real domain of interest. The discussions that follow are primarily concerned with how to achieve that simplification in a scientifically rigorous way taking proper account of the uncertainties in the modelling process. In the remainder of this chapter we will consider how to approach the environmental modelling process and some of the difficulties involved in trying to model environmental systems “properly”.

The reader may also find that there will be some terms in this book that might be new. For clarity, a Glossary of Terms is provided, including a discussion of different types of usage for some of the entries. Words in the text in bold will be found explained in the Glossary. A brief revision of matrix algebra and a guide to sources of software for uncertainty estimation are also provided as Appendices at the end of the book.

1.2 The aims of environmental modelling

The way in which an environmental scientist might choose to make predictions will depend, in part, on the aims underlying the effort. One major aim for any scientist is to show that some of the understanding that has been gained about the controlling processes can be formalised into a system of mathematical relationships that result in verifiable predictions about that system. This is prediction as science to show “that we do, after all, understand our science and its complex interrelated phenomena” (W. M. Kohler, Head of Hydrology at the World Meteorological Organisation, 1969). For many scientists this is a sufficient aim in itself since, if the predictions do not prove to be correct, then it should force a revision of the science that underlies the formal statement of the model used. That this does not always happen will be discussed further later. For now, it is sufficient to note that a secondary aim of developing a scientific model that produces verifiable predictions is to use that model operationally for predictions that will be useful for management and decision-making purposes.

With the growth of computer power and computer modelling capabilities, the aim of producing operational predictions that will be useful for management has been increasingly driven by demand. Now that the results of complex computer simulations of weather systems are routinely shown on television it is perceived that computer predictions should now be possible in many other areas of environmental science, from the transport of toxic immiscible pollutants in groundwater to the impact of climate change on vegetation patterns and floods. This is despite the fact that our knowledge of the properties of specific groundwater aquifers is poor; despite the fact that our ideas about future climates rest on the results of global circulation models that are not yet very secure; and despite the fact that we often complain about television weather forecasts being wrong. In many areas of environmental science the demand for predictions has outstripped the scientific understanding on which predictions must be based. There are certainly some areas in which the answer to the question of how to make predictions should, as yet, be don’t (or at least don’t put too much trust in the model predictions when making decisions).

It is indicative that not many environmental modelling studies show true tests of predictions of the models in the form of post-prediction auditing. Many will show simulations that are compared with past data after some history matching or model calibration has taken place. Some will show similar predictions of periods not used in model calibration as a test of the capabilities of a model. Very few studies have made predictions that have then been verified (or not) by data collected later (something that we have all been taught should be part of the “scientific method”).

In fact, experience in this type of post-prediction audit has not been good, at least in the field of groundwater modelling (Konikow and Bredehoeft, 1992). Post-prediction audits made for a variety of different modelling studies showed that, in general, the results were generally poor (see also Anderson and Woessner, 1992). This was often for very understandable reasons, such as wrong assumptions about future boundary conditions, but this does not change the conclusion that the results were poor. What, then, should we conclude about the predictions of the much more complex coupled ocean–atmosphere global circulation models that are being used to predict the expected changes in climate as a result of changing concentrations of greenhouse gases into the future? That their predictions are wrong? Quite possibly, but not necessarily. The more common conclusion is that they are necessarily approximate at present but will be improved as computer power increases and as any mismatches between observed and predicted variables are evaluated and understood. The same would now be true in the case of the groundwater models. In most cases a post-prediction audit would lead to model improvements that would allow better predictions to be made with the benefit of hindsight about, for example, which boundary conditions actually occurred over the predicted period. This means that modellers are rarely forced to admit to false predictions since they can always revise their predictions with hindsight or with a new generation of models and auxiliary conditions. It is worth noting that, viewed in this way, model applications become part of a learning processes, not only about the models but also about the places they are applied to. The idea of modelling as a learning process will be a continuing theme in this book as it is essentially about reducing different forms of uncertainty in making predictions.

Modelling for understanding, modelling for prediction for practical applications and modelling as career are all part of the current practice of environmental modelling. Scientists and practitioners who model and make predictions tend, for the most part, to be pragmatic realists at heart. Their goal is to bring models based on the most comprehensive understanding to bear on prediction problems of operational or practical interest. This would combine the aims of prediction as science, of prediction as practical tool (and of prediction as career). Gradually, as the science progresses, the models used in prediction are expected to evolve to become a more and more realistic description of the real system. This pragmatic realism is one commonly held philosophy of environmental modelling. This is not, however, the only possible philosophical position to take and we will return to discuss this further in Chapter 2 after considering the nature of the modelling process and the different sources of uncertainty that arise in modelling environmental systems.

1.3 Seven reasons not to use uncertainty analysis

The issues that are raised by the uncertainty inherent in the application of environmental models have been discussed for two decades and more (e.g. notably Beck, 1987, in the field of water quality modelling). Pappenberger and Beven (2006) have considered why uncertainty estimation is still not yet standard practice in environmental modelling. It remains common to show results without uncertainty bounds to decision makers, at scientific conferences, in refereed publications or in consultancy reports. It seems that there is still significant resistance to the routine use of uncertainty analysis methods by environmental modellers, whether for reasons of expense, understanding of methods, or training in the requisite skills. Yet, the use of uncertainty estimation should be routine in environmental science. As yet, despite all of the research on methods of uncertainty estimation that is now available, it is not. Seven of the reasons why not are as follows:

1Uncertainty analysis is not necessary given physically realistic models.

2Uncertainty analysis is not useful in adding to process understanding.

3Uncertainty (probability) distributions cannot be understood by policy makers and the public.

4Uncertainty analysis cannot be incorporated into the decision-making process.

5Uncertainty analysis is too subjective.

6Uncertainty analysis is too difficult to perform.

7Uncertainty does not really matter in making the final decision.

The reader may well be able to add some other reasons to this list (for example that the whole idea of trying to assess the uncertainties makes his/her head hurt!). Pappenberger and Beven consider each of those seven reasons in turn and suggest that none of them is tenable in many applications, at least where uncertainty estimation is not limited by computational constraints. In particular, they discuss the interaction between scientists and policy and decision makers. The concepts of “uncertainty” and “ risk” are perceived and understood in a variety of different ways by different communities and different people. However, it can be shown that when both scientists and public work together this gap may be bridged. For example, several studies have shown that probabilistic weather forecasts can be understood by non-scientist users (e.g. Luseno et al., 2003). Moreover, policy makers derive decisions on a regular basis under severe uncertainties, because the scientific basis is not sufficient at the time. Studies suggest that decision makers actually want to get a feeling for the range of uncertainty and the risk of possible outcomes when it can be provided (e.g. McCarthy et al., 2007). This point is illustrated by the political demand of “handling uncertainty in scientific advice” to the UK Parliament (Ely, 2004). The response may, however, be subject to the type of decision under consideration. Tyszka and Zaleskiewicz (2006) report that people were much less interested in probabilistic information about scenarios when the decision to be made had an ethical dimension.

This is a very important point for the modeller since a misunderstanding of the certainty of modelling results can lead to a loss of credibility and trust in the model and the modelling process (Demeritt, 2001; Lemos et al., 2002). The communication of uncertainty to decision makers, the public and other stakeholders is all important in this process (e.g. Brashers, 2001; Fox and Irwin, 1998; Patt and Dessai, 2005; Faulkner et al., 2007; Stainforth et al., 2007b). Effectively, uncertainty estimation is embedded in the wider decision-making process (e.g. Refsgaard et al., 2005, 2006). The suggestion that scientific uncertainty cannot be understood by stakeholders and decision makers persists (on both sides). There would seem to be little reason why this argument should continue to be made in the future in terms of understanding. This book is, hopefully, a contribution towards easing the communication process and working towards the routine application of uncertainty estimation in environmental modelling. Other initiatives are also helping, such as the more widespread availability of software for uncertainty estimation (see Software Appendix at the end of this book) and the decision tree for uncertainty estimation methods described in Section 1.9 below.

An open scientific discourse on uncertainty would have important implications for the environmental decision process. Uncertainty clearly does matter in the current debate over the significance of future predictions of climate change and its implications for future global policies (and consequent impacts on future water resources management and capital investment). This is an area where the science has not yet matured to the point where an open discourse is possible and expressions of uncertainty are interpreted as simple disagreements amongst scientists. Some disagreements exist, of course, but neither side in the climate change deb...