Sunteți pe pagina 1din 78

A.A.

van Dam, MSc, PhD

EE492/0x/x

ENVIRONMENTAL MODELLING

UNESCO-IHE Delft, The Netherlands December 2005

Contents
Unit 1 1-1 1-2 1-3 1-4 1-4-1 1-4-2 1-4-3 1-5 Unit 2 2-1 2-2 2-3 2-4 2-4-1 2-4-2 2-4-3 Unit 3 3-1 3-1-1 3-1-2 3-1-3 3-1-4 3-2 3-2-1 3-2-2 3-2-3 3-2-4 3-2-5 3-3 3-3-1 3-3-2 3-3-3 Unit 4 4-1 4-2 4-3 4-4 4-5 4-5-1 4-5-2 4-5-3 4-6 4-7 4-7-1 4-7-2 Unit 5 5-1 5-2 5-3 5-4 5-5 Title Introduction to modelling and the systems approach Environmental science Modelling as a scientific approach Systems, ecosystems and modelling The systems approach Reductionism and specialisation Holism and the systems approach Why modelling? The history of modelling Systems, integration levels and model types System classification Integration level and generalisation Model complexity versus model applicability Model classification Static versus dynamic models Descriptive versus explanatory models Deterministic versus stochastic models The modelling process Model formulation Problem definition and model objectives System boundaries Collection of available knowledge and data Formulation of conceptual model Model implementation Selection of modelling technique Formulation of the mathematical model Implementation of the model Calibration and sensitivity analysis Validation Model application Improvement of conceptual model and identification of knowledge gaps Prediction and scenario analysis Optimisation Dynamic simulation modelling Expressing biological processes in mathematical terms State-determined systems and feedback Dimensions and units System properties from differential equations Some simple models of biological processes First-order degradation Exponential growth Logistic growth Analytical and numerical integration Dynamic modelling: a few examples Pollution in a well-mixed lake Algal bloom Modelling in Stella Stella: the basic idea Stella components Defining the relations Making a graph Learning more about Stella Acknowledgements Selected Internet links Literature references Appendix 1: Exercises Appendix 2: Powerpoint slides Page no. 3 3 4 5 7 7 8 9 11 14 14 14 16 17 17 18 19 21 21 21 21 22 23 26 26 26 26 26 29 31 31 31 32 33 33 35 37 38 39 39 40 40 43 46 46 52 55 55 57 58 60 62 63 63 64

Environmental Modelling

UNIT 1. Introduction to modelling and the systems approach


1-1 Environmental science
Environmental science deals with problems arising from the impact of human activities on the environment and environmental resources. Human activities, such as agriculture, industry or tourism have impacts on the environment, e.g., through pollution or depletion of natural resources. The environment provides services for human activities, such as nutrients and energy. To reduce the negative impacts of human activities on the environment, a variety of action can be taken: households can change the way they deal with waste, farmers can change their ways of farming, industries can adopt cleaner production methods, etc. Governments can try to stimulate these actions by implementing environmental policies, such as taxes on the discharge of polluting substances or subsidies on clean production methods. Within this network of actors, activities and the environment, the environmental scientist can have different roles. He or she can collect information about human activities and on the environment, or be involved directly in influencing actions that have an impact on the environment. In other words, environmental scientists are dealing with various aspects of environmental problem solving (see Figure 1). Environmental modelling is one of the tools that can be used by environmental scientists.

Environmental science
Information

Human activities Agriculture Industry Tourism Etc.

Impact (e.g. pollution)

Environment and resources Water Land Air Fish Etc.

Information (e.g. through research)

Actors Households Business Government Etc.

Matter and energy

Action (e.g. policies, technology)

Action (e.g. policies, technology)

Figure 1 The roles of environmental scientists (adapted from the pressure-stateresponse model in Rockstrm 2002).

Environmental Modelling

This course is an introduction to ecological and environmental modelling. It will not turn the participants into modellers who are ready to apply modelling to environmental problems. Rather, it will give a first impression of what modelling is, why it can be used for solving scientific problems, and which modelling techniques are available. The principal learning goals for this introduction to environmental modelling are: to learn the basic concepts of modelling as a tool for problem-solving in environmental science research and decision-making to learn some important modelling approaches and methodologies to learn the relationship between an environmental science problem and the choice of modelling technique to learn the main features of dynamic systems and how they can be modelled After this course, the participants should be able to: distinguish between different environmental model types and judge the application of models in relation to the problem they were intended to solve translate some important environmental processes into mathematical equations build a simple dynamic simulation model using the Stella software As explained, after this introduction most participants will not be able to model environmental systems independently. To achieve that, a more advanced modelling course is needed. However, they should be able to critically assess the modelling efforts of others, e.g., as published in a scientific paper or written in a research proposal. They should have an idea of the tools available for modelling. They should also be able to judge the factors that determine to what extent a model can be relied upon for prediction of system behaviour and for management purposes.

1-2 Modelling as a scientific approach


In this environmental modelling course, modelling will be presented as a scientific approach to solving problems in environmental science and management. As such, modelling is one of several possible scientific appoaches to problem solving. Other options are empirical studies, comparative studies, and experimental studies (Jorgensen and Bendoricchio 2001). If we consider the example of pollution in a lake, these different approaches would result in different activities to find a solution, each with their own problems. In an empirical study, we could measure the concentrations of the polluting substances in the lake and relate them to other factors. We could look at variation in space and time. Based on the findings, we could try to formulate conclusions that lead to a solution of the problem. Such an approach would take time and would also involve costs for sampling and analysis.

Environmental Modelling

A comparative study would take a similar approach but also try to compare the results of the lake under study with those obtained in other lakes. This might lead to further insight into the causes of the pollution and/or the way to reduce or eliminate it. Again, this would take a considerable amount of time, and data on the other lakes would need to be available or collected. An experimental approach could involve creating pollution in the lake intentionally and studying the effects directly. Again, this would take time and money. Moreover, this approach might create ethical problems if the intentional pollution leads to mortality of wildlife or creates risks for human health. Experimental work with large ecosystems is thus not very realistic (although some experiments are known, e.g., applying fertilizers to lakes to investigate eutrophication and acid rain).

Modelling can be seen as another scientific approach to problem solving. In a modelling study, a model of the pollution in the lake with the factors affecting the pollution would be formulated. This model could then be used to predict the effect of different management scenarios that may lead to reduction of the pollution. Depending on the circumstances, modelling may be cheaper and faster than the other approaches. Obviously, the ethical problems arising from experimental manipulation of ecosystems can be avoided by manipulating the model. Modelling is often used in combination with the other approaches. Models can be supported by empirical data collected in the system itself. Such data can also be used to test the validity of the model. This is necessary if decision-makers want to use models for prediction and management.

1-3 Systems, ecosystems and modelling


First a few definitions:

A system is a limited part of reality that contains interrelated elements An ecosystem or environmental system is a system that is able to sustain life including all biological and non-biological variables A model is a simplified representation of a system based on a specific modelling objective

Environmental Modelling

These definitions have a few important consequences. First of all, they imply that a model itself is also a system. Furthermore, that an ecosystem is merely a special case of systems in general. Therefore, an environmental or ecological model (i.e., a model of an environmental system or ecosystem) is not essentially different from a model of any other system. The fact that a model is a simplified representation of a system implies that the model is always simpler than the system. If that were not the case, we might as well work with the system itself. A modeller therefore always has to decide which simplifications need to be made in the model. The simplifications are made in relation to the problem to be tackled and the modelling objectives. The model is made assuming that the simplifications will not affect the results of the model with regard to the problem to be solved. This means that a model is only suitable for answering questions related to the model objectives. E.g., a model that predicts the oxygen concentration in a river in relation to the stream velocity and the distance to a certain pollution source is not necessarily also a good model for predicting the ammonia concentration in that river. Another way of saying that models are a simplified representation of systems is: models are system analogues. Models share certain properties with the system, and this is the positive analogy of the model. But there are also system properties that are not present in the model. This is the negative analogy. As an example we can think of a computer war game (the war is the system, the game is the model): the decisions the players have to make about the fighting are a positive analogy with the system, but the fact that no real victims are made is a negative analogy. Similarly, a pilot in a flight simulator (the model) can use all the controls he will find in a real airplane, but he cannot crash (the positive and negative analogy, respectively). Obviously, one model can have several positive and negative system analogies. Figure 2 visualizes the relationship between model objectives and the structure and form of the model as a result of the simplifying assumptions. It is the objective of a doll to look real and be soft and cuddly, and therefore these are the main features of this model of the human body. The negative analogy of the doll is that it contains no real organs and tissue. Similarly, dummies in car accident tests, various forms of shopwindow dummies and minute quantities of blood are all models of the same human body. Each model has its own structure that is directly related to the objective of the model.

Environmental Modelling

Model objectives determine simplification, structure and form


System Objective
Look real and cuddly Looks real (but empty inside)

Model

No face Sensors inside Human body Test effect of car accident Shape important No head/hands/ feet/hair/legs/skin

Display clothes

Investigate disease

Only tissue involved

Figure 2 Relationship between model objectives and model structure/form.

1-4 The systems approach 1-4-1 Reductionism and specialisation


Modern science is based on the concept of atomism or reductionism, i.e., the idea that a whole system can be understood by its decomposition into basic units. These ideas started with early Greek thinkers like Leucippus and Democritus (fifth century B.C.) and was further developed from the 17th century by e.g., Descartes. This approach has been very successful for the development of modern science and its many applications. There are numerous examples: geologists determine the composition of a rock sample to say something about a whole mountain range. Biologists weigh a few birds and then say something about the whole bird population. The field of modern medicine is also known for this. Diseases are treated according to the part of the body where the problem is observed: think of the cardiologist, the neurologist, the gynaecologist, etc. In this way, science has resulted in specialisms and specialists: the bird ecologist, the

Environmental Modelling

entomologist, the soil scientist, all working on a part of the whole system. This specialisation is also visible in the way companies and government departments are organised. There are several reasons for the success of this specialisation approach. Practically, it is much easier to look at small parts of the system than at the whole system. Science needs quantitative data on the systems studied, and the only way to obtain such data is to look closely at parts of the system. It is possible to measure the size of an individual bird, but it is physically impossible to weigh a whole bird population. A geologist can determine the composition of one sample of rock, but not of the whole mountain range. Another reason for specialisation is that it is much easier to communicate with specialists in the same area. A biologist from The Netherlands who makes a study of large rivers may not have much in common with a colleague from China who studies aquaculture ponds, but if they would specialize in studying the common carp (Cyprinus carpio), a fish that may be found in both ecosystems, they would find it very easy to communicate about their research. Specialisation also makes it easier for scientists to distinguish themselves if they are specialists in their own field. If twenty researchers in a university department are all working on wastewater treatment, they are all competing with each other when it comes to getting credits for their results. However, if they specialise in the various aspects of wastewater treatment (bacteriology, mechanical treatment, denitrification, etc.), they can more easily distinguish themselves. There is also less need for specialists to work together with others. This makes it easier for people to work independently and little time is spent ("wasted", according to some ...) on discussing the work and organising the collaboration. Often, working together is made difficult by institutional and disciplinary barriers: specialists may be working in different departments and use different terminology. Their salaries and research projects are paid from different budgets. Thus, biologists and economists in a university will find it difficult to collaborate, e.g., on the same estuary or wetland. They will not meet easily (let alone work together) because of the way universities are organized, unless a special project is set up to make it happen.

1-4-2 Holism and the systems approach


What is the problem with reductionism and specialisation? The problem is that scientists often forget to put back together the information about parts of the system. They either fail to draw conclusions about the whole system, or they draw conclusions about the whole system based on their knowledge of only one part. There are many examples of problems resulting from this way of working. E.g., pesticides like DDT developed by chemical engineers initially seemed to be a perfect solution for

Environmental Modelling

agricultural pest problems and could increase crop yields, until ecologists found out that these compounds accumulate in foodchains and cause tremendous environmental problems, such as mortality among top predators and health problems in people working with DDT. Rachel Carson wrote about this in her famous book Silent Spring (Carson 1962). Another example are drugs that are extremely effective in treating a disease, but have unacceptable side-effects (like, e.g., being carcinogenous). The opposite of reductionism is called holism. Holism is "the theory that whole entities, as fundamental components of reality, have an existence other than the mere sum of their parts" (Pauly 1994). A holistic view of a system to be studied or managed means that all aspects of that system are considered. It recognizes that the behaviour of a system can only be predicted when the system as a whole is studied, and not just from the behaviour of its components. Another term for holism is systems approach. Modelling is very important in stimulating holistic approaches to research. It should be emphasized that science is not a contest between reductionism and holism. A holistic approach is needed to come up with results that are useful for the application of research results for the benefit of the ecosystem, including the human beings living in that ecosystem. However, rational analysis and reductionistic approaches by specialist scientists are necessary and valuable to obtain detailed knowledge about the functioning of parts of the system. A combination of reductionism and holism is likely to yield the best results.

1-4-3 Why modelling?


There are several reasons why models are extremely useful as a scientific approach, and particularly so in environmental science. Here are a few: 1. Models stimulate a multidisciplinary approach. Models provide conceptual frameworks in which knowledge about different parts of the system can be stored and analysed simultaneously. In this way, specialists can contribute their own knowledge to an analysis of the large system, and can begin to see how their work relates to the work done by other disciplines. E.g., bioeconomic models link ecological and economic processes in an ecosystem. This is important because the level of exploitation (e.g., fishing in a fishery system) depends on the income and profit that can be achieved, rather than on the ecological stability of the system. Bioeconomic models can help explain overexploitation by showing that fishing is still profitable when the fish population is moving towards extinction. Environmental problems are often multidisciplinary because they are caused by the interaction between humans and the environment. Environmental models therefore should incorporate both socioeconomic and biological components, and thus be multidisciplinary. The various disciplines often have their own methods and research tools, which makes it difficult to combine the results of different disciplines and analyse their

Environmental Modelling

interaction in a holistic way. Models allow for such integrated, multidisciplinary analysis. 2. Models can generate and consolidate knowledge. By combining knowledge about parts of the system, the behaviour of the whole system can be explained. This can lead to new knowledge about the system and how it can be managed. By constructing a model and documenting or publishing it, the knowledge about the system is automatically stored for use by others. Often, data are needed to run models and by constructing models, the underlying data are also stored and made available for other users. Models can also help to utilize historical information and "old" data. This is sometimes referred to as "data archaeology" or "data mining". In this way, data that cost a lot of time, money and effort to collect are not used just once, but can be used more often. 3. Models allow generalization and upscaling. Results from research done in one environmental system is often not directly applicable to other, similar systems in different places. This means that the research has to be done all over again. However, similar environmental systems have a lot of characteristics in common. If a model based on the underlying processes is constructed for one system, the same model may be applied to another system with minor modifications. The solutions for the first system may be applied relatively easily and rapidly to other locations as well. Models can support this process because they embody the generic structure and processes of an ecosystem. In this way, solutions designed for one particular situation or location may be applied on a much wider scale. This is called upscaling. Two types of upscaling can be distinguished. Horizontal upscaling means reaching large groups of end users with results of local initiatives. Research efforts are often limited in geographical scope and in the numbers of people that participate. There is a need to translate good results from local research into general concepts that can guide the process for larger numbers of people. Vertical upscaling is translating the site-specific results of community-level initiatives into more general concepts that can be applied by policy makers at higher aggregation levels. The limited geographical scope and a lack of quantitative analysis make it difficult for policy makers to benefit from the knowledge and information that surface from local research efforts. Models can support both types of upscaling. 4. Models can be used for training and education. Models can be used by others to learn about the functioning of systems and their management. Manipulation of the system can be trained using a model without the risk of damage to the real system. Different scenarios for ecosystem management can be simulated, and the results compared to gain insight. If models can visualise the results of such scenarios, they can also promote the participation of stakeholders in research and management.

Environmental Modelling

10

5. Models can be used for planning, management and decision-making. When properly validated, models can be used to evaluate various management options. See 3.3.2. 6. Models are easier to manipulate than natural/unique systems. See 1.2. 7. Models can be manipulated without safety or ethical limittions. See 1.2.

1-5 The history of modelling (see Figure 3)


Modelling or the systems approach as it exists today is the result of a number of developments in the 20th century. In the 1920s, the first models of the oxygen balance in a stream and of predator-prey relationships were developed. The Streeter-Phelps equation models the dynamics of oxygen in a river due to organic matter degradation and reaeration (Streeter and Phelps 1925). The Lotka-Volterra equations describe the competition between two populations of organisms (Lotka 1925; Volterra 1931). These equations are logistic equations (see 4.5.3.) that are linked through competition coefficients. Operations research now exists as a number of techniques for planning and policymaking. It has its origins in World War II, when the United States and Britain faced many (often logistical) problems that were too complicated to be solved by any single discipline on its own. Scientists from different disciplines were joined in multidisciplinary teams to solve problems related to, e.g., the development of radar technology, the size of fleet convoys and the detection of submarines. After the war, these scientists developed the new approaches further and provided them with a theoretical basis. Other related terms that are used are systems engineering (often used in relation with large, human-made systems), systems analysis and systems ecology. Systems ecology is also called ecological modelling and was described in a chapter in E.P. Odum's textbook 'Fundamentals of Ecology' by Carl Walters (Walters 1971). System dynamics is a term used by modellers at the Massachusets Institute of Technology (Forrester 1961). The Club of Rome is a global think tank that brings together scientists, economists, businessmen, international high civil servants, heads of state and former heads of state from all five continents. In 1972, the Club of Rome published a famous report entitled "Limits to growth" (Meadows et al. 1972). The report was based on a world model that was built specifically to investigate five major global trends: accelerating industrialization, rapid population growth, widespread malnutrition, depletion of nonrenewable resources, and a deteriorating environment.

Environmental Modelling

11

Several international and intergovernmental research programs have stimulated ecological and environmental modelling because they generated large amounts of data needed for calibration and validation of models. Examples are the International Biological Programme (IBP) between 1965 and 1975, the Man and the Biosphere Program (MAB), and the International Geosphere-Biosphere Programme (IGBP). Initially, modelers had to write their own computer programs using programming languages like Basic or Fortran. Later on, specialised software packages for statistical analysis and dynamic simulation facilitated the construction of models by many more scientists. During the 1970s, and especially after 1981 when IBM introduced the Personal Computer (PC), computers became smaller, cheaper and therefore available to individual researchers. Eutrophication models and complex river models were the first models to be used for formulation of management strategies for ecosystems. Using these models, it became clear that the availability of knowledge about the systems and data for calibration of models was a big constraint. Ecologists had to become more quantitative in the way they studied ecosystems, e.g. by looking for exact values for parameters.

Environmental Modelling

12

1920
Streeter-Phelps Lotka-Volterra

1950
Population dynamics River models

1970
Eutrophication models Complex river models

1975
Fixed modelling procedure Balanced complexity More ecology

1980
Ecotoxicological models More case studies Validation of prognoses

1990
Structurally dynamic models Ecological constraints New mathematical tools, including machine learning

2000

Figure 3 Development of ecological and environmental models (from Jorgensen and Bendoricchio 2001).

Environmental Modelling

13

UNIT 2. Systems, integration levels and model types


2-1 System classification
Environmental systems are often unique systems. There are no two Lakes Victoria or Tonle Sap. There are no two Amazon rivers or Sahara deserts. This means that a model of such a unique system cannot be applied directly to another system. Changes to the structure of the model will have to be made before it can be applied to another system. This means that it is difficult to verify the model. Verification is possible by comparing model results with data from a similar system, but for a unique ecosystem there exists no such system. Unique ecosystems often have a weak negative feedback (see 4.2.). This means that they can develop in any direction that may be widely different from other ecosystems. An exception to this are so-called recurring systems. These are natural ecosystems with a strong negative feedback. An example that is often mentioned are peat bogs, where the accumulation of organic matter and the resulting lack of oxygen gradually reduces the rates of transformation in the bog and leads to a similar situation in most bogs. This means that models of recurring systems can be applied to other, similar systems and are therefore verifiable. Another type of ecosystem are repeatable systems. Examples of repeatable ecosystems are agricultural or ecological experiments, where the conditions are controlled by the researcher and can be created again and again in subseqent experiments. Models of repeatable systems are thus verifiable. Note that according to the definitions of models and systems given earlier in these lecture notes, agricultural or ecological experiments are in fact themselves ecosystem models. Although many ecosystems are unique systems, models of such systems can be applied to other systems. The extent of modification to the structure of a model that is needed before it can be applied to another, similar system depends on the objectives of the model and its generality. As explained before (see 1.4.3.), one of the interesting uses of models is in upscaling and generalisation. If a model is sufficiently generic, it may be applicable to other unique systems.

2-2 Integration level and generalisation


Systems can be regarded at different integration levels. The lowest integration level is that of the atoms and molecules. In biological systems, these form the tissues and organs of living organisms, which then form communities and populations. At even higher integration levels we can discern e.g., a river, the watershed, the region and country and finally the whole world and even the universe(s?) (See Figure 4). A holistic

Environmental Modelling

14

scientist with a systems approach will study the system at the higher integration levels, leaving the lower integration levels to the more reductionist-minded colleagues. The integration level at which the system is studied has conseqences for the generality with which conlusions about the system can be drawn. The lower the integration level, the higher the generalisation level. E.g., if a chemist finds certain chemical properties for a molecule, it is generally assumed that these results will be valid for all molecules of this type wherever in the world (or indeed even the universe). If a new drug works for curing a disease in the liver of German patients, it is likely to work also in Chinese or American patients, although this somewhat higher integration level already is less generalisable (consider the possibility, e.g., that the German patient has a different set of enzymes than the Chinese patient). Working at even higher integration levels, the applicability is reduced because the systems become more and more unique. Examples of models at different integration levels are: plant or animal growth models at the level of individuals, plant growth models at the level of plant communities or agricultural crops, water quality models at the level of whole lakes or rivers, farm-level models for management or optimization, land-use models at the level of watersheds or river basins, and models of climate change or population growth at the world level.

Integration levels and generalisation


Integration level
HIGH . . . . . . . . . LOW

Knowledge level
World Country Region Watershed River Community Plant Organ Tissue Cell Molecule

Generalisation level
LOW . . . . . . . . . HIGH

Figure 4 Relationships between integration level, knowledge level and generalisation level.

Environmental Modelling

15

2-3 Model complexity versus model applicability


According to our definition, models are always a simplification of reality. Generally the degree of simplification is related to the applicability of the model. If a model is very simple, the potential use of the model will be limited. It may be very good in simulating a particular feature of a system, but as soon as conditions change, the model can no longer be used. On the other hand, if the model is very complicated, it means that relatively little simplification has occurred compared to the system that is modelled. This means that the model cannot be applied to other, similar systems. Again, the applicability and generality of the model are low. Models that are moderately complex are usually the most widely applicable. This notion is visualized in Figure 5.

Model complexity versus applicability


System definition / boundaries

Applicability/generality

reality

optimum range

Number of processes
Figure 5 Relationship between number of processes in a model (model complexity) and the applicability/generality of the model.

Environmental Modelling

16

2-4 Model classification (see Figure 6) 2-4-1 Static versus dynamic models
Static (or steady-state) models do not incorporate the effect of time. They represent the system at one moment in time, like a photo ("snapshot"), or represent the average system during a well-defined period of time. Ecopath models are typical examples of static models. An example is the Ecopath model of a mangrove system in Yucatn, Mxico (Figure 7). Although static models do not incorporate time, they can still be used for analysing the effect of time by comparing two or more static models representing different points in time or different periods (Figure 8). Dynamic models do incorporate time and analyze the changes of the system in relation to time. This implies that the model must describe the changes in the state of the system with respect to time. Because the rates of change of state variables are often related to the size of the state variable, this can lead to feedback (see also 4.2.).

Model classification
Static no time effect snapshot Dynamic time effect feedback

Descriptive / Empirical describes system does not generate new knowledge

Explanatory / Theoretical uses knowledge at lower integration level to generate new knowledge at higher level Stochastic variables have probability distribution same input can lead to different output

Deterministic fixed variables model results always the same with same input
Figure 6 Model classification.

Environmental Modelling

17

2-4-2 Descriptive versus explanatory models


Descriptive models (also called statistical or empirical) are constructed by fitting one of several possible forms of equations to experimental data using statistical techniques. Descriptive models merely describe the relationship between two or more variables as it was actually observed. In descriptive models, a distinction is made between the variables that need to be explained and the explanatory variables that are used for the explanation. Descriptive models are not based upon a theoretical understanding of the system and don't prove anything about the underlying reasons for the relationship. The system is treated as a "black box: inputs and outputs are known, but the underlying mechanisms are not. A particular dataset can be described by an infinite number of models. When no guidance is available from theory, the simplest equation which adequately fits the data should be chosen. Descriptive models contribute to biological theory by elucidating relationships between variables in complex systems. Some widely used techniques for constructing descriptive ecological models are: (multiple) regression analysis factor and principal components analysis cluster analysis canonical correlation analysis artificial neural networks Bayesian belief networks

Explanatory (also called mechanistic or theoretical) models are based on a theory or hypothesis about the nature of the system. Whereas an empirical model merely describes a particular dataset, a well-founded theoretical model may describe all such sets of data and in addition explain why the observed relationships exist. Failure of a theoretical model to fit a wide range of experimental values can be turned to advantage if it suggests how the underlying theory can be improved. Explanatory models usually require large amounts of input data that are often difficult to obtain. Calibration and validation of explanatory models are extremely valuable because possible areas for further study or more data collection are identified. Explanatory models are never 100% explanatory and always contain descriptive elements. E.g., many dynamic simulation models consist of process descriptions that are essentially explanatory, but contain empirical equations that link environmental variables to rates. The most interesting explanatory models are those that use knowledge at a lower integration level to explain the behaviour of systems at a higher integration level. Examples are crop and fish growth models based on the biochemistry of plant and fish growth (Penning de Vries et al. 1989; Machiels 1987).

Environmental Modelling

18

Figure 7 Example of an Ecopath model (source: Vega-Cendejas and Arregun-Snchez 2001).

2-4-3 Deterministic versus stochastic models


A deterministic model uses fixed values for the model parameters and produces the same results given the same set of input variables. In such models, the state of a given variable at any time is determined entirely by previous states of that variable and the other variable upon which it depends. A stochastic (also called probabilistic) model uses random values for model parameters, i.e., values drawn from a probability distribution. Repeated calculations with the same input variables will therefore yield different output variables with every calculation.

Environmental Modelling

19

Figure 8 Schematic representation of possible biomass trends in an ecosystem. (A) Strong, regular changes as, e.g., due to the succession of seasons, not well represented by an annual mean (Ba). (B) Rapid transition between two stable states, of which each is well represented by its own mean (B1, B2). (C) Example of a biomass that does not reach equilibrium. During a brief period (tc), this biomass can be represented by a single value (Bc) whose confidence interval will usually bracket the change of biomass during the interval tc . Source: Christensen and Pauly 1993b.

Environmental Modelling

20

UNIT 3. The modelling process


The whole modelling process consists of various steps that are followed more or less sequentially. After having completed a number of steps, the modeller can go back to one of the earlier steps and go through the process again, in an iterative fashion. In this way, the model can be improved or adapted (see Figure 9). Roughly, three main phases in model construction can be distinguished: formulation (4.1.), implementation (4.2.) and application (4.3.) of the model.

3-1 Model formulation


The formulation phase can be subdivided in four steps: definition of the problem and model objectives, definition of system boundaries, collection of available knowledge and data, and formulation of the conceptual model.

3-1-1 Problem definition and model objectives


The definition of the problem (including the questions to which answers are sought) often receives limited attention. In many cases, it is considered quite obvious what the problem is. Nevertheless, the definition of the problem is very important and should receive a lot of attention because the choice of modelling technique and the structure of the model will depend strongly on the formulation of the problem and the objectives. In fact, two models of the same system can look completely different when the model objectives are different (see 1.3. and Figure 2) The problem definition can be based on the research questions of the researcher, but this phase in the modelling process also offers opportunities for participation of stakeholders or clients (e.g., a government, a company, a community, etc.). The problem is normally stated in general terms and the modeller (possibly together with the client) should translate the problem into well-defined objectives for the model.

3-1-2 System boundaries


After defining the model objectives, the boundaries of the system and the model must be defined. This means that choices are made about which parts of the system under study will be part of the model and which parts will not. In some cases this is very clear, in other cases a difficult decision may have to be made. E.g., many water quality models include primary production by algae. Primary production affects many water quality parameters: dissolved nutrients, carbon dioxide and pH, oxygen. To model this, changes in irradiation should be taken into account. This can be included in the model based on cloud cover, season, etc. The modeler can also decide to include irradiation

Environmental Modelling

21

as a forcing function and not include these processes in the model. The decision about this will depend on the objectives of the model.

The modelling process

Define problem Define model objectives Define system boundaries Collect available knowledge / data Formulate conceptual model

Formulation

Select modelling technique

Implementation

Feedback / Iteration

Formulate mathematical model Implement mathematical model Parameterization / data collection Calibration / data collection Sensitivity analysis Validation / data collection

Model users
e.g., policy / decision making e.g., management Prediction / scenario analysis Optimisation Improve conceptual model Identify knowledge gaps

Application

Figure 9 The modelling process.

3-1-3 Collection of available knowledge and data


To make a good model, the modeller should know the components of the system, the dynamics of these components (in space and time) and their interrelationships. Before making the model, existing information on the system can be collected and summarized. Existing information can be in the form of scientific articles, student theses, reports, maps, local knowledge of community members, etc. After formulating and implementing the model, the need for more data collection (e.g. for calibration and/or validation of the model, see below) may arise. It is important to realize that modellers very often use data collected by other people. This is good, because in this way data that were expensive to collect initially can be used again and are utilized better. Traditionally, many (reductionist) scientists are educated to believe that they should always collect, analyze and write about their own data. However, models are perfect for utilizing and integrating data collected by many people. This enhances holism and a systems approach and prevents data from getting

Environmental Modelling

22

lost in dusty filing cabinets. Pauly (1994) listed some rules about using other people's data (see Table 1).

Table 1 Rules for using other people's data (based on Pauly 1994, page 148) Data from published material
Use any numerical data as long as you cite the source, including the page number For parts of a text, put thye material in quotation marks and cite the course, preferably with the page number When using a complete, published graph or table, cite the source and state that permission to reproduce the item has been obtained from the copyright holder, whol will ordinarily be the original publisher and/or author after you have obtained such permission in writing When using part of a published graph or table which is then incorporated into a new graph or table, or turning a table into a graph or vice versa, only a citation is needed

Data from unpublished material


Treat these the same way as the published items, except that materials from theses should be used only if the same material is not available in published version and the author can be contacted and her/his approval obtained, or if the thesis is more than 2-3 years old and can be expected to remain unpublished Written permission to cite certain sources is required (by some publishers) for preprints Use 'pers. comm.' with the name of the author who provided you personally with a communication and check your wording with the source person. She or he may also be mentioned in the acknowledgement section if your paper For any other unpublished data, it is most important to try hard to get explicit (written) permission to use the data

3-1-4 Formulation of conceptual model


Based on the system boundaries and knowledge about the system, a conceptual model can be formulated. A conceptual model is an image of the system based on available knowledge and experience. This image can have several forms: it can be a mental image in the head of a person, or a story in words about the system, or a picture or graph drawn by a person. It is important to realize that every person dealing with a system, whether a researcher, resource user, administrator, etc. uses conceptual models of that system. We all form mental models of the system in our minds and use these to make predictions about what will happen to the system when certain conditions change. E.g., many farmers know what will happen to their crops when the weather changes. They do not have formal models, but they do have (probably unconsciously) conceptual models about how their crops react under different climatic conditions.

Environmental Modelling

23

Word models and graphic models (pictures, diagrams) are other forms of conceptual models. Compared to mental models, they are easier to discuss with other people and can be adapted and improved based on inputs from others. Several systems exist for drawing schematized pictures of systems. A simple drawing of the system that includes the major components and processes of the system is a good start (Figure 10). Such drawings can be formalized using symbols such as those used by the simulation software Stella (Figure 11) or Forrester's symbols for relational diagrams (Figure 12).

Figure 10 Conceptual model of dissolved oxygen in a fishpond. Source: Meyer and Brune (1982).

Environmental Modelling

24

Relational diagrams Stella

Figure 11 Stella symbols for relational diagrams of dynamic models. Source: High Performance Systems, Inc., Hanover, USA. See 4.2. for an explanation of terms.

Relational diagrams Forresters symbols


Material flow Parameter or constant

State variable

Information flow

Auxiliary variable Rate variable Sink

Figure 12 Forrester's symbols for relational digrams. Source: Source: Forrester, J.W. (1961). See 4.2. for an explanation of terms.

Environmental Modelling

25

3-2 Model implementation 3-2-1 Selection of modelling technique


In this step, a choice of modelling technique is made based on the knowledge about the system and the objectives of the model. There is a wide range of techniques to choose from: statistical techniques, network analysis, dynamic modelling, etc. In this course, emphasis is on dynamic simulation modelling.

3-2-2 Formulation of the mathematical model


This is the core of the modelling where the conceptual model is translated into a number of mathematical equations that describe the system or the processes taking place in the system. Often, equations take the shape of difference or differential equations describing changes in certain system components in relation to time and/or to other system components. In statistical models, the form of the equation is given by the method and the modelling consists of finding parameter values for the equations to best describe the dataset at hand.

3-2-3 Implementation of the model


Implementation of the model consists of operationalizing the mathematical model using a computer application. Very often, specialized modelling software packages are used. Well-known examples of dynamic simulation modelling packages are VENSIM, STELLA, and EXTEND (see links to websites in 7.). For big models, special applications with user interfaces are usually written using programming languages like Delphi, VisualBasic or C++. For statistical models, there are numerous analysis packages such as SPSS, SAS, etc. There are also numerous specialized applications for Bayesian Belief Networks (e.g., Netica), artificial neural networks, etc. Because of the complexity of some modelling techniques, modellers tend to become specialists in one or a few approaches and the modelling tools that come with that approach. It is quite common to find students and lecturers who know how to use SPSS but not SAS (or the other way around), or modellers who always use Stella but have never used EXTEND.

3-2-4 Calibration and sensitivity analysis


Once the model is formalized and implemented, values for the model parameters must be selected. This is called calibration.

Environmental Modelling

26

Calibration is different for descriptive and theoretical models. For descriptive models, parameters form the link between explanatory variables and the variables that need to be explained or predicted. An example is the descriptive model in Figure 13, where suspended particulate matter (SPM) in lakes is predicted from a number of other variables: total phosphorous, pH and a morphometric parameter. The model consists of an equation with coefficients that are the parameters estimated in a multiple regression analysis. For such a statistical model, estimating the parameter values and their confidence limits based on the dataset is the main activity in the modelling process. Figure 14 shows an example of a theoretical model. The transformation of nitrogen in wastewater treatment ponds is modelled using a dynamic simulation model. In this case, the parameter values are derived from direct measurements in the ponds, or from values in the literature that were measured under similar conditions (temperature, oxygen concentration, etc.).

log(SPM) = 1.148 log(TP) + 0.137 pH + 0.286 log(DR) - 1.985 R2 = 0.87, n=26


Figure 13 Multiple regression analysis of particulate matter in Swedish lakes. Source: Lindstrom et al. 1999. The difference between the parameters in these two categories of models is that the descriptive model of SPM in lakes produces parameters that are not biologically meaningful, i.e., they cannot be measured in the real system. The parameters are derived theoretically from the relationship between the data. Their value certainly says something about the importance and the nature of the relationship, but the value in itself has no meaning. On the other hand, the parameter values in the theoretical model do have biological meaning. E.g., the reaction constant for the rate of nitrification can

Environmental Modelling

27

be measured in a real pond or even in the laboratory under different environmental conditions, and this same value can be used in the model. If the model gives acceptable results only with a value that is strongly different from a realistic value, the modeller knows that there is something wrong with the model. Sensitivity analysis is done to find out the response of the model to small variations in the values of the parameters. This gives insight into the importance of the parameter for the model results. If changing the value of a parameter has only a minor effect on the model result, the parameter is less important than a parameter that, when changed only slightly, creates a large effect on the model result.

Figure 14 Explanatory model of nitrogen transformation and removal in a primary facultative pond. Source: Senzia et al. 2002.

Environmental Modelling

28

Calibration and sensitivity analysis in dynamic simulation modelling are often done iteratively. Sometimes, the value of some key parameter is unknown and cannot be derived experimentally or from the literature. If the model is very sensitive to this parameter, we know that model results should be treated with care. More research should then be done to obtain more certainty about the value of that parameter and about the process which it helps describing. There are many different parameter sets within a chosen model model structure that may reproduce the observed system behaviour. This is called equifinality.

3-2-5 Validation
Before a model can be used for its intended purpose, it needs to be validated. Validation or verification of a model is essentially a test of whether the model does what it is supposed to do. The modeller can always learn something from validation: if the model does not pass the test, it means the model is not a good representation of the system and needs to be improved. If it does pass the test, it may be applied according to its objectives and within the framework of the assumptions made initially, in the situation/conditions for which it was developed. Validation can be split into a check of internal and external consistency. Internal consistency means that the model equations are correct, that the dimensions and units are consistent and correct, and that the model cannot create results that are unacceptable or unrealistic. E.g., a model that predicts fish to grow to a negative weight is not internally consistent. External consistency means that the model gives accurate predictions of whatever state variables it calculates. Obviously, this is difficult to test when we are dealing with predictions of things that will happen in the future. Nevertheless, there are ways of rigidly and quantitatively testing a model. Mostly this is done by comparing model output with historical or experimental data. E.g., if a water quality simulation model can mimick the variations in dissolved oxygen concentration over the period 1970-2000 well (taking into account the variation in climate and nutrient input during that period), it is reasonable to assume that predictions for a future period (under climate and nutrient input regimes in the same order of magnitude) will be accurate. It is extremely important for such a validation that the data used are completely independent of the data used for construction, parameterization and calibration of the model. Comparisons of simulation output with historical or experimental data can be done graphically (see Figure 15) but can also expressed quantitatively.

Environmental Modelling

29

Figure 15 Graphical comparison of model predictions with experimental measurements. Source: Senzia et al. 2002.

Environmental Modelling

30

3-3 Model application 3-3-1 Improvement of the conceptual model and identification of knowledge gaps
The modelling usually results in an improved understanding of the system. With this improved understanding, the conceptual model can be improved and the modelling process can be repeated. The modelling thus points the way to gaps in the knowledge about the system and provides ideas for collection of new data. The modelling process is often iterative and leads to gradual improvement of the knowledge of the system and of the model. It is important to note also that stakeholders or clients can be involved in the modelling process at various stages, e.g. when a first version of the model is completed and model results can be evaluated. The knowledge of stakeholders can be very valuable in improving the model and formulating the next version.

3-3-2 Prediction and scenario analysis


One major objective of modelling is predicting what will happen in the future. This is important particularly for managers and policymakers. They need to make decisions on the management of the system based on their assessment of what the consequences of these decisions will be. They have to consider the consequences of their decisions for the various dimensions involved: ecologically (the impact on the environment, the pollution effects, the effects on biodiversity, etc.), socially (what does it mean for the communities involved, how does it affect their lives) and economically (what is the effect on the incomes of households, but also of local and national governments). Effectively, this analysis of the impacts of decision-making is an analysis of the sustainability of the development that is foreseen. Models can be very useful for handling the complexity of this analysis. They can be used to answer 'what-if?'questions. One way of using a model for policy analysis is to formulate scenarios. A scenario is a hypothetical sequence of events constructed for the purpose of focusing attention on causal processes and decision points (Lein 2003). A model can be used to evaluate what happens to a system under different scenarios. Scenarios and models are used to explore alternative futures and to help ordering the ideas about what will happen in the future when certain decisions are made today. Often, a limited number (2-4) of scenarios are formulated that represent "plausible alternative futures" (Lein 2003). Scenarios should not represent the best, the most likely and the worst case because then the policy decisions have been made before the scenario analysis. Often a few key indicators are selected with which to compare the outcome of the different scenarios. Scenarios often resemble stories of what might happen with the system.

Environmental Modelling

31

The model is then used to quantify the story and produce measurable and comparable indicators upon which decisions can be based. Learning is therefore a main objective of scenario analysis.

3-3-3 Optimisation
Models can also be used for optimisation of systems. Certain types of models are built specifically for this purpose. E.g., linear programming is a technique that is used to optimize combinations of variables given certain objectives and constraints. But also other models can be used for optimisation. Sometimes a goal or objective function is defined. Such a function is a weighted sum of several indicators (e.g., ecological, economic and social). The value of the objective function can be calculated for different management scenarios and the alternatives can thus be evaluated quantitatively. Ofcourse, the assignment of weights to the different components of the goal function has a strong influence on the results.

Environmental Modelling

32

UNIT 4. Dynamic simulation modelling


4-1 Expressing biological processes in mathematical terms
Dynamic simulation models belong to the class of explanatory models. They try to explain the functioning of the system on the basis of a description of the underlying procesess and mechanisms. Table 2 gives an overview of the processes that can occur in ecological systems. A detailed explanation of the modelling of all these processes is beyond the scope of this course. In this chapter, a small selection of processes is discussed as an example (4.5.). The selection is based on the processes that occur in the exercises belonging to this course. Together with these processes, some essential elements of dynamic modelling are presented. Dynamic simulation modelling consists of translating the processes in the system under study into mathematical equations, finding values for the parameter values and then "running" the model, starting with the initial state of the system (at time = 0) to calculate the future state of the system.

Table 2 Processes occurring in ecological systems (based on Jorgensen and Bendoricchio 2001).
Physical processes mass transport advection diffusion turbulent diffusion dispersion mass transfer at a two-phase interphase mass balance well-mixed systems non well-mixed systems energy solar radiation light extinction temperature settling and resuspension Chemical processes chemical reactions reaction types reaction kinetics temperature effects enzymatic reactions chemical equilibrium hydrolysis redox acid-base Chemical processes (ctd.) adsorption and ion exchange equilibrium of adsorption partition of ionic organic compounds dynamics of adsorption ion exchange volatilization Biological processes biogeochemical cycles in aquatic environments nitrogen cycle phosphorous cycle prediction of the phosphorous concentration in a lake oxygen cycle oxygen dynamics in a river (Streeter Phelps) photosynthesis photosynthetic rate algal growth nutrient limitation zooplankton growth fish growth single population growth ecotoxicological processes biodegradation equilibrium between spheres bioaccumulation

Environmental Modelling

33

To illustrate this process of making mathematical descriptions of processes, consider the following example: A stagnant pond is polluted with a constant inflow of organic pollutant. The pond has no outflow, and the only way the pollutant disappears from the pond is by microbial degradation inside the pond. This description of the situtation is a word model. A word model describes the system to be studied and the processes that cause changes in this system. Another way of expressing our word model is: the change in pollutant inside the pond = inflow of pollutant - degradation of pollutant This is still a word model but it is beginning to resemble an equation. From this equation in words, the step to a mathematical equation is small:

dP = Pin k P dt
in which

(Eqn. 1)

dP = rate of change in P dt P = amount of pollutant in the pond Pin = constant inflow rate of pollutant k = degradation constant
Equation 1 is called a differential equation. It describes the process of change in the concentration of pollutant instantaneously, relating the rate of change in the amount of pollutant (dP/dt) to the amount of pollutant (P). (Note that a difference equation is not the same as a differential equation. A difference equation describes the process by expressing the next state of the system in terms of the present state. The development of the system is described as a sequence of the system's state. E.g., if a population grows by 5% each year, the size of the population can be described by the difference equation yn+1 = 1.05 yn; Doucet and Sloep 1992) To obtain a global process description of the value of P at any point in time, the differential equation needs to be solved to obtain an explicit or general solution. This

Environmental Modelling

34

general solution relates the amount of pollutant to time. There are several methods for deriving analytical solutions for differential equations (beyond the scope of these modelling lectures, not treated here). By integration, a solution for Equation 1 can be obtained:

Pt =

Pin P + P0 in e kt k k

(Eqn. 2)

with

Pt = amount of pollutant in the pond at time t Pin = constant inflow rate k = degradation constant P0 = P at time 0

Equation 2 describes the relationship between the amount of pollutant (P) and time. If the values of the constants in the equation (Pin, P0 and k) are known, the value of P can be calculated at any value of t. Figure 16 compares this process of translating the word/conceptual model into a mathematical equation for two systems: (1) a one-time lake pollution with degradation; and (2) a continuous inflow of pollutant and degradation.

4-2 State-determined systems and feedback


When making dynamic explanatory models of environmental systems, it is assumed that the state of a system can be described and quantified at any point in time. In dynamic models, several types of variables are used to describe the system: State variables (or stocks) describe amounts or sizes or numbers, e.g., biomass, number of animals, amount of nitrogen, etc. Rate variables (or flows) describe the rate at which state variables change. Usually, a rate equation describes one of the processes in Table 2, and can include state variables and driving variables. Driving variables (or forcing functions) describe the effect of external factors on the system and are not affected by the system itself. Typical examples are climate variables such as temperature, wind speed, etc. They can also be variables that are beyond the boundaries of the system as defined by the modeler (e.g., the inflow of nutrients with wastewater into a wetland).

Environmental Modelling

35

system:

One-time lake pollution, only degradation (1)

Continuous inflow of pollution and degradation (2)

conceptual model:

differential equation:
integration

dP dt

= - kP

dP dt

= Pin - kP

solution:

P(t) =

P0e-kt

P(t) =

Pin k

+ P0 -

Pin k

e-kt

Figure 16 Conceptual model, differential equation and analytical solution for two simple models of lake pollution.

These are the main model components but more types of variables can be distinguished. Auxiliary variables are variables that are computed in a model but are not directly part of the model structure. Sometimes, the values of state variables need to be converted into other units or expressed as a density (no. per unit of surface area). For this, an auxiliary variable can be introduced. In Figures 11 and 12, it was shown how these different variable types can be represented graphically in a conceptual model. An important concept in dynamic models is feedback. Feedback exists when a rate variable depends on a state variable. This is the case e.g., in Equation 1 where the rate of change in P depends on P itself. In Equation 1, there is negative feedback because the rate decreases when the state increases. With time, as the value of the state variable increases, the value of the rate of change gradually decreases. Systems with negative feedback eventually reach a state of equilibrium when the rate has decreased to the point of becoming zero. Then, the state does not change anymore and is constant. In Equation 1, this happens when the rate of inflow of the pollutant equals the rate of breakdown within the pond (by the way: note that Pin is a driving variable). Positive feedback exists when the rate variable has a positive relationship with the

Environmental Modelling

36

state variable. In that case, both increase continuously. Exponential growth is an example of positive feedback (see 4.5.2.).

4-3 Dimensions and units


When constructing a dynamic simulation model, it is very important to keep a close check on the dimensions and units of the equations. The very word "equation" means that the parts to the left and to the right of the equality sign should be equal, also in terms of dimensions and units. As an example, Equation 1 can be examined more closely:

dP = Pin k P dt

(Eqn. 1)

in which

dP is the rate of change in P, P is the amount of pollutant in the pond, Pin is dt the constant inflow rate of pollutant, and k is a degradation constant. The rate of

change is always expressed as a quantity per unit of time, e.g., g d-1. The right side of the equation should then also be expressed in g d-1. This means that Pin should have the unit of g d-1 and k of d-1. A more general term is dimension, which is the kind of unit in which a quantity is expressed. The dimension of the rate of change is: 'mass per time', and the dimension of k is 'per time'. The dimension of a quantity is denoted by dim X, dim(X), or [X]. For Equation 1:

dP [mass] ]= dt [time]

[ P ] = [ mass ]

[ Pin ] =

[mass] [time]

If a quantity has no dimension, this is indicated with [*]. There are some basic rules for dimensions (from Doucet and Sloep 1992): 1. Quantities may be equated, added, or subtracted only if they have the same dimensions. 2. Dimensions of products and quotients can be multiplied and divided in a straightforward way: [XY] = [X] [Y] and [X/Y] = [X] / [Y]. 3. Operations that cannot be expressed in terms of Rule 1 and Rule 2 are only allowed if the quantities concerned are dimensionless. E.g., ex is a legitimate expression only if [x] = [*].

Environmental Modelling

37

Equations should always be checked for dimensional correctness. It is also useful to check the dimensions and units of parameters, e.g. to find out if they need to be converted into other units.

4-4 System properties from differential equations


It is not always necessary to obtain the analytical solution of a differential equation to say something about the characteristics of the system. E.g., it is possible to calculate the amount of the pollutant when the pond reaches an equilibrium state. Equilibrium means that there is no change in the concentration of the pollutant, or in other words:

dP =0 dt
or

(Eqn. 3)

dP = Pin k P = 0 dt
Re-arranging Equation 4 gives:

(Eqn. 4)

Pin = k P
or

(Eqn. 5)

P=

Pin k

(Eqn. 6)

This is the equilibrium amount of the pollutant. It can be easily seen that when e.g., k increases (more rapid degradation), the equilibrium amount of P decreases. Or when the inflow of pollutant (Pin) increases, the equilibrium amount of P increases. Figure 17 compares the equilibrium amount of pollutant for a one-time lake pollution with degradation and a continuous inflow of pollutant and degradation (see Figure 17). Some more examples of system characteristics derived from differential equations are given in section 4.7.

Environmental Modelling

38

Solving for equilibrium: no change in amount of pollutant input degradation = 0 Only degradation dP dt = - kP = 0 P =0
Inflow and degradation

Inflow and degradation dP dt = Pin - kP = 0 Pin = kP


Only degradation

Pin k
0
Time

Figure 17 Equilibrium amount of pollutant in a lake with and without continuous inflow of pollutant.

4-5 Some simple models of biological processes 4-5-1 First-order degradation


First-order degradation is often used to model the decomposition of organic matter in ecological systems and was already shown in the example in 4.1. The rate of decay is related to the amount of organic matter as follows:

Pollutant

P =

Pin k

dC = k C dt

(Eqn. 7)

in which C is the amount of organic matter and k is the first-order reaction constant. The analytical solution of this differential equation is:

Environmental Modelling

39

C t = C 0 e kt

(Eqn. 8)

in which Ct is the amount of organic matter at time t, C0 is the initial amount of organic matter, and k is the first-order degradation constant.

4-5-2 Exponential growth


A quantity exhibits exponential growth when it increases by a constant percentage of the whole in a constant time period. Exponential growth occurs when an individual or population can grow without limitations. The growth rate of, e.g., a population is then proportional to its size:

dN = rN dt

(Eqn. 9)

in which N is the number of individuals and r (in d-1) is the instantaneous net growth rate of the population. Figure 18 (top) shows the relationship between dN/dt and N: a straight line. Analytical integration of Equation 9 (with constant r) gives the following solution, relating the number of individuals to time:

N t = N 0 e rt

(Eqn. 10)

where Nt is the number of individuals at time t, N0 is the initial number of individuals, and r is the instantaneous net growth rate. Figure 18 (bottom) shows the relationship between time and N: as time passes, N keeps increasing.

4-5-3 Logistic growth


Exponential growth usually occurs only during a limited time period. At some point, a population or organism encounters limiting conditions that cause the growth rate to become reduced: food, oxygen or space become limiting, or the concentration of a toxic substance becomes too high. This effect can be modelled with the equation:

dN N = r N 1 dt K

(Eqn. 11)

in which r is the instantaneous growth rate (in d-1), N is the number of individuals, and K is the carrying capacity of the system.

Environmental Modelling

40

Exponential growth
dN dt

dN =rN dt integration

N N

Nt = C ert t
Figure 18 Exponential growth: differential equation and analytical solution. Figure 19 shows the relationship between dN/dt and N. When population size N = 0, the growth rate is also 0. When N = K (i.e., the population size equals the carrying capacity), N/K = 1 and dN/dt becomes zero. When N>K, the growth rate becomes negative (<0). When N = K/2, the growth rate is maximum. An analytical solution of Equation 11 is:

Nt =

K 1 + c e rt

(Eqn. 12)

in which Nt is the number of animals at time t, K is the carrying capacity of the system, r is the instantaneous growth rate, t is time and c is a constant. Equation 12 is called the logistic equation. Figure 19 shows the relationship between Nt and time. In fact, Equation 12 is a special solution of the differential equation for those cases when N0 < K. The general equation is:

Nt =

K K rt 1 1 N e 0

(Eqn. 13)

Figure 20 shows the relationship between Nt and time for N0 < K, N0 = K and N0 > K.

Environmental Modelling

41

Logistic equation
dN dt K N N

dN N = r N (1 ) dt K integration K Nt = K 1 + c e-rt logistic equation

N0

Figure 19 Logistic growth: differential equation and (an) analytical solution.

Logistic equation
N Nt = K 1 + c e-rt logistic equation t Nt = K t 0 N0 = Nt = K 1- (1- K ) e-rt N0 K (1c) general solution K N0 < K t N0 > K

Figure 20 General solution of the logistic equation.

Environmental Modelling

42

4-6 Analytical and numerical integration


It is not always easy or even possible to find analytical solutions for differential equations. The equations can be complex, and one model can contain many equations. Without an analytical solution, it becomes difficult to achieve the objective of a simulation model: to calculate the value of the state variable after a certain period of time. However, there is an alternative approach to analytical solution techniques: numerical integration. Numerical integration is based on the principle that when the state of a system at a certain time t is known, and the rate of change of the system is known, the state of the system at time = t + t (in which t is a relatively small period of time during which the rate of change is assumed not to change) can be calculated as follows: State at time (t+t) = state at time (t) + change during t or statet+t = statet + t ratet statet+2t = statet+t + t ratet+t statet+3t = statet+2t + t ratet+2t (Eqn. 14)

etc.

Figure 21 shows how this works for a simple first-order degradation reaction (dP/dt = -kP with k = 0.05, P0 = 100 and t = 1). By adding the change in the state variable achieved during each time step to the value of the state variable at the beginning of that time step, the change in the state can be simulated. Because in numerical integration it is assumed that the rate of change during one timestep is constant, an error in the calculation of the value of the state is introduced. When the rate in reality is constantly increasing, the assumption that the rate is constant introduces an under-estimation of the state variable. This is illustrated in Figure 22 for a model of first-order degradation. The size of the time step is very important for the size of the error. To reduce the error, the time-step should be sufficiently small. However, a smaller time-step increases the calculation time of the model because the computer has to perform more calculations. There are various methods for numerical integration. The most simple method is the rectangular or Euler integration which assumes a constant rate during the whole timestep of integration. Other method are the trapezoidal and the Runge-Kutta methods of integration. Figure 23 shows how the rectangular and trapezoidal methods work and how the rectangular integration method can lead to overestimation or underestimation of the state variable.

Environmental Modelling

43

Numerical integration
dP dt t 0 1 2 3 4 5 6 = - kP Pt P0 = 100 95.00 90.25 85.74 81.45 77.38 etc. with k = 0.05 dP/dt -0.05 * 100 = -5.00 -0.05 * 95.00 = -4.75 -0.05 * 90.25 = -4.51 -0.05 * 85.74 = -4.29 -0.05 * 81.45 = -4.07 -0.05 * 77.38 = -3.87 At t = 0, P = 100 (P0) t = 1 : Pt+t 100 + 1*-5.00 = 95.00 95.00 + 1 * -4.75 = 90.25 90.25 + 1 * -4.51 = 85.74 85.74 + 1 * -4.29 = 81.45 81.45+ 1 * -4.07 = 77.38 77.38 + 1 * -3.87 = 73.51

Figure 21 Calculation scheme of numerical integration for first-order degradation reaction.

Numerical integration: error


Numerical t 0 1 2 3 4 5 6 Pt 100 95.00 90.25 85.74 81.45 77.38 etc. dP/dt -5.00 -4.75 -4.51 -4.29 -4.07 -3.87 Pt+t 95.00 90.25 85.74 81.45 77.38 73.51 Analytical Pt = P0 * e-kt 100 95.12 90.48 86.07 81.87 77.88 0 0.12 0.23 0.33 0.42 0.50 Difference (A-N)

Figure 22 Difference between results from numerical and analytical integration.

Environmental Modelling

44

f(b) f(a)

f(b) f(a)

Euler (=rectangular)

Trapezoidal

f(b) f(a)

f(b) f(a)

increasing rate amount underestimated

decreasing rate amount overestimated

Figure 23 Rectangular and trapezoidal integration methods (A) and the errors of rectangular integration in relation to the rate equation (B).

Environmental Modelling

45

4-7 Dynamic modelling: a few examples 4-7-1 Pollution in a well-mixed lake


Consider a small lake which is fed by a small stream. From the stream, an organic pollutant is discharged into the lake. Let's assume that the local government responsible for the water quality in the lake wants to know how the pollution in the lake can be managed. More specifically, they want to know: Which processes play a role in determining the concentration of pollutant in the lake? What is the relative importance of the processes? What is the time-scale of changes in pollutant concenctration?

To address these questions, a model will be constructed. The first step in the modelling process is to create a conceptual model. There are three ways for the pollutant to disappear: it can settle to the bottom; it can be degraded by microbes; or it can be washed out of the lake through the outflow. Figure 24 shows a graphical representation of the lake. This description in words and the picture of the lake form our first conceptual model of this environmental system. To keep the model simple, the following simplifying assumptions are made: (1) the lake is completely mixed; (2) the particles in the lake are homogeneously distributed; (3) the temperature and other environmental parameters are constant; (4) the water volume is constant. Dynamic modelling is chosen as the modelling technique. The next step then is to formulate a mass balance for the pollutant in the lake. This means that the input of pollutant into the lake should be balanced with the different pathways of disappearance. In words, this is as follows: Loading = Settling + Degradation + Outflow or: Rate of change = Loading - Settling - Degradation - Outflow (Eqn. 16) (Eqn. 15)

Note that the equations are formulated in terms of the processes that occur in the system, i.e., in terms of rates.

Environmental Modelling

46

Figure 24 Graphical representation of pollution in a lake.

Model of a well-mixed lake


Pollutant outflow g d-1

Pollutant loading

Amount of pollutant M g

Pollutant degradation

g d-1

g d-1 Pollutant settling g d-1

Figure 25 Relational diagram of the lake model using Stella symbols.

Environmental Modelling

47

Figure 25 shows a relational diagram of the model using the symbols of the Stella software. In the diagram, the units of the rate and state variables (or flows and stocks) are also indicated. The dimension of the stock 'amount of pollutant' is chosen as [mass], and therefore the dimensions of the flows into and out of that stock should be [mass]/[time]. The units were chosen as gram (for [mass]) and day (for [time]). The next step is to translate each part of Equation 16 (loading, settling, degradation and outflow) into a differential equation. For each process, the dimensions and units will be checked. According to rule 1 for the dimensions (see 4.3.), the processes need to have the same dimensions if they are to be added. They all need to be expressed in g d-1. Loading The loading of the pollutant into the lake depends on the volume of water flowing into the lake per unit of time (Q, in m3 d-1) and the concentration of pollutant in that inflow (Cin, in g m-3). We assume that Cin is constant. We can then formulate the loading rate as:

Loading = Q C in
A check of the units shows:

(Eqn. 17)

m3 g g 3 = , which is the right unit for the loading rate. d m d

Settling The settling of particles in a water column is affected by a number of factors, which makes it a rather complex process to model. The size and density of the particle, turbulence, physiological state of the cell (in the case of phytoplankton), viscosity of the water and other factors play a role. To model this process in detail, various intricate models were developed (Jorgensen and Bendoricchio 2001). For our purpose of modelling the removal of the organic matter in a small well-mixed lake, a simple approach is taken. It is assumed that the time needed for an individual pollution particle to settle is proportional to the depth of the water column. We assume a constant 'apparent settling velocity' for the particles (vs, in m d-1), which is independent of particle size. This means that for a given depth H (in m), the time needed to settle to the bottom can be calculated as:

settling time =

H vs

(Eqn. 18)

The settling rate in the model then becomes:

Environmental Modelling

48

Settling =

M H vs

(Eqn. 19)

or

Settling = M

vs H

(Eqn. 20)

Units check: g Degradation

d =g m d

(OK)

Degradation of an organic pollutant is a biological process performed by bacteria and other microbial organisms. The rate of degradation is therefore controlled by a range of factors such as the density and activity of the bacteria, temperature, and the composition of the pollutant. In a simple model like this, it is assumed that the breakdown of the organic matter is a first-order process (see 4.5.1.). The equation is:

Degradatio n = k M

(Eqn. 21)

in which k is the degradation constant (in d-1). It can be easily seen that the units are in order. Outflow Modelling outflow is very similar to modelling inflow. The only complication here is that the concentration in the outflow needs to be known. Since the lake is assumed to be completely mixed, the concentration in the outflow is the same as the concentration in the lake (C). C can be calculated from the amount of pollutant M and the lake volume (V, in m3). The complete rate equation (mass balance) for the change in the amount of pollutant in the lake (dM/dt) can now be assembled:

v dM = Q Cin M s k M Q C dt H
Figure 26 shows a more elaborated conceptual model.

(Eqn. 22)

Environmental Modelling

49

Q V

Outflow k M

Loading Cin

Degradation

Settling

vs

Figure 26 Elaborated conceptual model of the well-mixed lake model

Using the mass-balance equation, a number of system properties can be derived: the equilibrium concentration, the transfer function and the residence time. The equilibrium concentration is reached when there is no more change in the concentration of pollutant in the lake, i.e., dM/dt = 0, or:

Q C in M

vs k M Q C = 0 H

(Eqn. 23)

The mass of pollutant (g) is equal to its concentration (g m-3) multiplied by the lake volume (m3), so

Q Cin C V
or

vs k C V Q C = 0 H

(Eqn. 24)

v Q C in = C V s + k V + Q H

(Eqn. 25)

Environmental Modelling

50

The surface area of the lake A (in m2) equals V/H, so the equilibrium concentration can be expressed as:

C eq =

Q C in A vs + k V + Q

(Eqn. 26)

The transfer function () of the lake is expressed as:

C C in

(Eqn. 27)

or, in the case of our model:

Q Q + k V + vs A

(Eqn. 28)

The transfer function indicates how the input into the system is transformed or transferred to an output. If is much smaller than 1, the concentration in the lake is much smaller than the concentration in the inflow and the lake's removal mechanisms are active to reduce the level of pollutant in the water. If 1, then the processes in the lake are not very successful in reducing the level of pollutant. The residence time () of a substance represents the average time that a particle resides in the sysem. It is defined as:

mass loading or removal rate

(Eqn. 29)

The equation is valid for a lake that is in steady state (i.e., dM/dt = 0). From the equation, it can be seen that when the mass in the lake is high and the loading (or removal) is low, the residence time is high. On the other hand, if there is a lot of flow through the system and the mass is low, the residence time is low. For water molecules, the residence time (w) can be expressed as:

w =

V Q

(Eqn. 30)

The inverse of w is called the flushing rate of the lake.

Environmental Modelling

51

For the pollutant, the residence time can be calculated as:

p =

M C A vs + k C V + Q C

(Eqn. 31)

With M = C V, this becomes

p =

V A vs + k V + Q

(Eqn. 32)

Using this model, some of the questions about the pollution in the lake can be answered. During the computer exercises, you will work out a Stella version of this model and explore the dynamics of the system.

4-7-2 Algal bloom


A typical phenomenon that results from eutrophication is an algal bloom. When nutrient input in aquatic ecosystems is high, the algal population may suddenly explode, and significantly affect the quality of the water. Especially blue-green algae may become quite a nuisance, as they can excrete toxic substances. Normally, algal population densities are kept low by competition by water plants or by predators such as water flees (Daphnia spp.). This section introduces a simple model of the interaction between Daphnia and a microalgae species, and the dynamics of this interaction under changing nutrient inputs in the water. Blooms are often the consequence of increased availability of phosphorous. Therefore, the dynamics of the system are explored under a phosphorous load that increases in time. The system consists of a lake with a constant inflow and outflow of water. The inflow water contains a constant concentration of phosphorous. As in the previous examples, the lake is considered fully mixed so the concentration of phosphorous in the outflow is equal to the concentration in the lake. Phosphorous is taken up by algae. The mechanism by which this happens is modelled using so-called Michaelis Menten or Holling Type II kinetics. The equation is:

Uptake = U max

P A P + kA

(Eqn. 33 )

Environmental Modelling

52

where P is the phosphorous concentration in the water (g l-1); A is the algal density in the water (number of algal cells, in no. per litre); Umax is the maximal P uptake rate per alga (g P cell-1 d-1); kA is the half saturation constant of algae for phosphorous (i.e., the Pconcentration at which algal uptake is half maximal in g P l-1) A constant proportion of the algal population is also flushed out of the lake in the outflow (just like the phosphorous). The algae are consumed by the Daphnia, also according to a Holling Type II equation:

Consumptio n = C max

A D A + kD

(Eqn. 34)

where A is the algal density in the water (number of algal cells, in 103 per litre); D is the density of Daphnia's (individuals l-1); Cmax is the maximal uptake rate of algae per daphnia (cells individual-1 wk-1); kD is the half saturation constant of daphnia for algae (i.e., the algal density in the water at which consumption by Daphnia is half maximal, in cells l-1); You will build this model in Stella during the computer exercises.

Environmental Modelling

53

Daphnia density

maximum consumption rate consumption rate half saturation constant

Algae Algal outwash

Natural mortality

maximum P uptake rate

P uptake

half saturation contstant P uptake

P input

P outwash Phosphorous

~ P release P loss rate

Figure 27 Stella diagram of the algal bloom model.

Environmental Modelling

54

UNIT 5. Modelling in Stella


(based on original text by Dr. Johan van de Koppel) In Chapter 4, we discussed how to translate a real-world ecosystem into a schematic system description and then develop a dynamic mathematical model based on this description. This process is the basis of a simulation software called "Stella". In Stella, you design a model by building a conceptual model of boxes and arrows. After that, you describe the flows and relations in the model by attaching the appropriate equations or values to the symbols in the scheme. Stella then simulates the model dynamics based on numerical integration of the differential equations, and writes the results in graphs and tables. You do not have to do any programming yourself nor write the differential equations of your model. Stella does this for you automatically. This chapter describes how Stella works, and how to make a model in Stella.

5-1 Stella: the basic idea


Modelling in Stella is structured around making a conceptual scheme of the system that you want to study. This scheme is central, but can become very big and difficult to understand. For this reason, a Stella model consists of a number of levels (Figure 28). The first or top level, called Interface level, gives an overview of the main components of the model, which are called "blocks". These blocks can harbour a number of state variables, rate variables or parameters. This layer is used to present the model structure, without much detail. This level can also be used to present model results. The second or middle level, called Map/Model level, is used to draw the model diagram. This diagram consists of boxes (the state variables), double-lined arrows (the rate variables), curved, single-lined red arrows (relations), and circles (parameters). Using these four building blocks, a Stella model can be made. The third or bottom level, called the Equations level, contains the model equations. These equations are generated by the program as you draw the model in the Map/Model level. When you start the Stella program, you will see the Map/Model level (middle level). If you want to move between the levels, press the two little arrows, and that you can see on the left side of the screen (Figure 28). They move up to the Interface view, or down to the Equations view.

Environmental Modelling

55

Figure 28 An overview of the Ithink/Stella environment

Environmental Modelling

56

5-2 Stella components


A model is constructed in Stella in the Model Construction layer using the four building blocks presented as the first four buttons. These are: Stocks ( ), that represent the state variables Flows ( ), that represent rate variables, Converters ( ), that represent the parameters, Connectors ( ), that represent the relations between model components.

Pollutant inflow

Pollutant degradation

Pollutant

Pollutant release

Degradation constant

Figure 29 Graphical representation of a polluted pond. The pollutant is released at a constant rate, while a constant fraction of the pollutant in the pond degrades.
To build the model presented in Figure 29, the following steps can be made: In the Map/Model level, click on the Stock button (see Figure 30) and click somewhere on the white sheet in the middle of the screen. Now a box appears, which represents the "pollutant" state variable. Click on the name ("Noname"), and type "Pollutant". You have now created the pollutant state variable. The next thing to do is to create the rate variables, also called flows in Stella. There is one flow of pollutant entering the state variable box, and one is leaving. Click on the "Flow" button (the second from the left; see Figure 30) and now click a few centimetres to the left of the pollutant box. Keep the mouse button pressed! Drag the cursor over to the pollutant box, and then release. You will see the pollutant box become dotted when you move over it, indicating a possible link. Repeat this procedure for the outgoing flow, releasing the mouse button to the right of the pollutant box. Type in the names of the fluxes, as are given in Figure 29, by clicking on the flow names under the flow symbols.

Environmental Modelling

57

Figure 30 The button bar on the Map/Model level.

Now we have to bring the parameters called "Pollutant release" and "degradation constant" on the sheet. Click on the "Converter" button (third from the left), and insert a converter underneath each flow symbol, as in Figure 29. You will have to press the button on the button bar again for the second converter. Give them the appropriate name. Finally, we have to link those components that have a relation to each other, as is indicated by the dashed arrows in Figure 29. Click on the "Connector" button (fourth from the left) on the button bar, and click on the "Pollutant release" converter. Keep the mouse button pressed, and drag it to the "Pollutant inflow" tap. Then release the button. Do the same for the other two relation arrows as given in Figure 29. Whenever you make an error, you can delete components by using the "Dynamite" button ( ). This is the second-to-last button on the button bar (see Figure 30). Click first on the dynamite button, and then press on the object that you want to delete. If it becomes dashed, it is ok to release the mouse button, and the object is deleted. We have now finished entering the model structure in Stella. It is possible to add descriptions to the boxes and flows in the model by double clicking on them. This has no effect on the model but can help you remember what the box or flow represents later on. Especially in bigger models, this is very helpful.

5-3 Defining the relations


Before we can enter the mathematical relations between the components, we first have to put the programme in "equation" mode. This you do by clicking on the little globe that

Environmental Modelling

58

you can find on the left side of the screen. If you click it, it changes to a 2. A number of question marks pop up in the model. These will disappear as the model is completed. We will have to: 1. define the relations; 2. enter the parameter values; 3. give the initial values for the state variable(s). Let us start with entering the relations that define the flows in the model. Click the globe button if you haven't done that already. Double-click on the "Pollutant release" tap, and you will see a pop-up window popping up. In the lower part of this pop-up window, you see a white area, containing in blue: "{ Place right hand side of equation here... }" (Figure 31). Here you can formulate the equation that defines the inflow. In this case, the inflow is equal to "Pollutant release". Enter the variable by clicking on "Pollutant release" in the "Required inputs table", situated in the upper part of the window. Now close this window by pressing "OK". You will see that the question mark has disappeared. Now open the "Pollution degradation" flow window by clicking on the tap. The equation that we have to enter here is more complex: "Pollutant * Degradation constant". Enter this by clicking on "Pollutant" first, then press the multiply button in the calculator to the right, and then clicking "Degradation constant". Close the window by clicking "OK". We now have to define the parameter values. First double-click on "Pollutant release", and enter the value of 10. Close the window. Then enter 0.1 as the degradation constant. Finally, enter 0 as the starting value in the "Pollutant" box. We have now defined the model. In Figure 29, the system is bordered by a so-called system-boundary. Stella also allows you to define the boundaries of your system, or sub-system if there are more then one. We can put a system boundary around the model by using the "Sector frame" button ( ), which is the first one in the second block. Click on this button and then just slightly above your model to position the frame. Use the red square tags at the corners to drag the frame to the right position. Finally, enter a sector name by clicking on the temporary "Sector 1" indication at the top of the frame. The sector frame tool can be used to compartmentise your model.

Environmental Modelling

59

5-4 Making a graph


What good is a model if you cannot see the results? So, we should make a graph or a table in which we can see the model results. First move to the Map Layer by clicking the "up" arrow in the upper left corner of the screen. You will now see the sector frame we just produced on the Model Construction layer. Click on the graph button in the button bar. Then click on the screen. You will now see a graph window appearing. This graph is empty, because we first have to define what the graph should show. Double click on the graph window to open a second window that allows you to define what is presented. In the window that has popped up you can indicate the variables that you want to show. The model we made can be used to model the pollutant concentration in a pond. Therefore we will have to select that one from the list. Click on the "Pollutant" box in the "Allowable" list, and push the upper double arrow (>>) to move it to the "Selected" list. Than close the window by pushing the "OK" button. Finally, push the needle button in the upper left part of the graph window to fix the graph to the screen. If the graph, for some reason, disappears, you can open it by double clicking the "graph" icon that sits on your Map layer (not on the button bar). Now still your graph is empty. You need to run the model to produce results first! To do this, open the "Run controller window". Click on the running man button in the lower left corner of the screen. This opens the run controller window. This window contains a "play" button to start the model, as in a CD player. First, however, we first have to define some simulation specifications. Press the "Specs" button, and select "Time Specs". A window pops up which defines, amongst others, the length and time step of the simulation. Set the simulation time to 100 by entering 100 in the "To" box. Now press "OK". You can now run the model by pressing the play button.

Environmental Modelling

60

Figure 31 The flow dialog window

Environmental Modelling

61

The unit of time in Stella depends on the magnitude of the data you have entered. E.g., if you enter 4 as the inflow, Stella does not care if it is 4 litre/minute or 4 litre/decade. It is important to make sure that you use the same unit of time everywhere in your model. So, as soon as you start building a model, decide upon the time unit and type in the unit as comments with your flows. If you need the actual simulation time in an equation or a graph, you can use the built-in function TIME. If you want Stella to display the time unit you have chosen in the output graphs, you have to tell Stella what your unit of time is. This is done in "Time Specs". Note that this has no effect whatsoever on the simulation.

5-5 Learning more about Stella


This chapter is just a brief tutorial on how to use Stella. Stella can do much more than what is explained here, and not all the details of the dialogs and windows that we used were discussed. The best way of learning more about Stella is by building more models yourself and discovering how the software works (see the Exercises in Appendix 1). There is a Help function in the programme that explains the functioning of Stella. During the course, more exercises with Stella will be done. Stella has a detailed manual and there are also resources on the Internet that can explain how to work with Stella. See, for example, http://www2.chemeng.lth.se/Edu/stella/

Environmental Modelling

62

Acknowledgements
These lecture notes are partly based on lecture notes written by Dr. Johan van de Koppel for the Environmental Modelling course in the former EST and WERM programmes at UNESCO-IHE.

Selected Internet links


Subject
Agricultural Systems (scientific journal) Agriculture, Ecosystems and Environment (scientific journal) CAMASE Methods for research on agricultural systems and the environment Club of Rome Delft 3-D (integrated ecological modelling software) Ecological Engineering (scientific journal) Ecological Management and Restoration (scientific journal) Ecological Modelling (scientific journal) Ecopath with Ecosim (trophic modelling) Ecotools (Tools for wetland ecosystem resource management in Eastern Africa) Extend (simulation software) International Institute for Applied Systems Analysis(IIASA) International Society for Ecological Modelling MIKE-SHE (hydrological modelling software) and other software Stella (simulation software)

Link
http://www.elsevier com http://www.elsevier com http://www.bib.wau.nl/camase/ http://www.clubofrome.org/ http://www.wldelft.nl/soft/d3d/intro/index.html http://www.elsevier com http://www.blackwellpublishing.com/listofj.asp http://www.elsevier com http://www.ecopath.org http://www.unisi.it/ecotools http://www.imaginethatinc.com http://www.iiasa.ac.at http://dino.wiz.uni-kassel.de/isem-eu/index.html http://www.dhisoftware.com/ http://www.hps-inc.com http://www2.chemeng.lth.se/Edu/stella/

US Environmental Protection Agency (EPA) http://www.epa.gov/epahome/models.htm models page Vensim (simulation software) WWW-server for ecological modelling (Register of ecological models) (add your own favourites) http://www.vensim.com http://eco.wiz.uni-kassel.de/ecobas.html

Environmental Modelling

63

Literature references
Carson, R. 1962. Silent spring. First Mariner Book Edition 2002. Christensen,V. and D. Pauly (eds.) (1993a) Trophic models of aquatic ecosystems. ICLARM Conf. Proc. 26, 390 p. Christensen, V. and D. Pauly (1993b) On steady-state modelling of ecosystems, p. 1419, In: V. Christensen and D. Pauly (eds.) Trophic models of aquatic ecosystems. ICLARM Conf. Proc. 26, 390 p. Cuenco, M.L. (1989) Aquaculture systems modeling: an introduction with emphasis on warmwater aquaculture. ICLARM Studies and Reviews 19, 46 p. de Wit, C.T. and J. Goudriaan (1978) Simulation of ecological processes. PUDOC, Wageningen. Doucet, P. and P.B. Sloep (1992) Mathematical modeling in the life sciences. Ellis Horwood, New York. 490 p. Forrester, J.W. (1961) Industrial dynamics. MIT Press, Massachusetts. Gold, A.J. and D.Q. Kellogg (1997) Modelling internal processes of riparian buffer zones. p. 192-207, In: Buffer zones: their processes and potential in water protection. Ed. by N.E. Haycock, T.P. Burt, K.W.T. Goulding and G. Pinay. Quest Environmental, Harpenden, UK. Jrgensen, S. and G. Bendoricchio (2001) Fundamentals of ecological modelling, 3d Ed. Developments in Environmental Modelling 21, 530p. Elsevier, Amsterdam. Kelly, L.A., A. Bergheim and M.M. Hennesy (1994) Predicting output of ammonium from fish farms. Water Research 28, 1403-1405. LeCren, E.D. and R.H. Lowe-McConnell (eds.) (1980) The functioning of freshwater ecosystems (International Biological Programme 22). Cambridge University Press, Cambridge, 588 pp. Leffelaar, P.A. (ed.) (1999) On systems analysis and simulation of ecological processes, with examples in CSMP, FST and Fortran, 2nd Ed. Current Issues in Production Ecology, Vol. 4. Kluwer Academic Publishers, Dordrecht. Lein, J.K. (2003) Integrated environmental planning. Blackwell Science, Oxford. 228 p. Lindstrom, M., L. Hakanson, O. Abrahamsson and H. Johansson (1999) An empirical model for prediction of lake water suspended particulate matter. Ecological Modelling 121, 185-198. Lotka, A.J. (1925) Elements of physical biology. Williams and Wilkins, Baltimore. Lovelock, J.E. (1979) Gaia, a new look at life on Earth. Oxford University Press, Oxford UK. Lovelock, J.E. (1988) The ages of Gaia, a biography of our living earth. Oxford University Press, Oxford UK. Machiels, M.A.M. (1987) A dynamic simulation model for growth of the African catfish, Clarias gariepinus (Burchell 1822). PhD-thesis, Wageningen University, 110 p. Meadows, D.H., D.I. Meadows, J. Randers and W.W. Behrens (1972) The Limits to Growth. Universe Books, New York. Meyer, D.I. and D.E. Brune (1982) Computer modeling of the diurnal oxygen levels in a stillwater aquaculture pond. Aquacultural Engineering 1, 245-262. Milstein, A., M.A. Wahab and M.M. Rahman (2002) Environmental effects of common carp Cyprinus carpio (L.) and mrigal Cirrhinus mrigala (Hamilton) as bottom feeders in major Indian carp polycultures. Aquaculture Research 33, 1103-1117. Odum, E.P. (ed.) (1971) Fundamentals of ecology, Third edition. W.B. Saunders Co., Philadelphia. Pauly, D. (1994) On the sex of fish and the gender of scientists. Fish and Fisheries Series 14. Chapman and Hall, London.

Environmental Modelling

64

Penning de Vries, F.W.T., D.M. Jansen, H.F.M. ten Berge and A. Bakema (1989) Simulation of ecophysiological processes of growth in several annual crops. Simulation Monographs 29, PUDOC, Wageningen. Rockstrm, J.F. (2002) Environmental Policy Making. Lecture notes, Environmental Planning module, WERM 025/02. UNESCO-IHE, Delft. Senzia, M.A., A.W. Mayo, T.S.A. Mbwette, J.H.Y. Katima and S.E. Jorgensen (2002) Modelling nitrogen transformation and removal in primary facultative ponds. Ecological Modelling 154, 207-215. Streeter, H.W. and E.N. Phelps (1925) A study of the pollution and the natural purification of the Ohio River. Public Health Bulletin No. 146. U.S. Public Health Service. Vega-Cendejas, M.E. and F. Arregun-Snchez (2001) Energy fluxes in a mangrove ecosystem from a coastal lagoon in Yacatan Peninsula, Mexico. Ecological Modelling 137, 119-133. Volterra, V. (1931) Variations and fluctuations of the number of individuals in animal species living together. In: R.N. Chapman, 'Animal Ecology', McGraw-Hill, New York. Walters, C.J. (1971) Systems ecology: the systems approach and mathematical models in ecology, p. 276-292. In: E.P. Odum (ed.) Fundamentals of ecology, Third edition. W.B. Saunders Co., Philadelphia.

Environmental Modelling

65

Exercise 1

Appendix 1. Exercises ENVIRONMENTAL MODELLING

EXERCISE 1. Pollution in a stagnant pond (introduction to Stella) See Figure 1, a simple model of pollutant inflow and degradation. Stella is used to implement this model. The following steps in the process are followed:
Pollutant release Degradation constant

Pollutant

Pollutant inflow

Pollutant degradation

Figure 1: A graphical representation of a polluted pond. The pollutant is released at a constant rate, while a constant fraction of the pollutant in the pond degrades.

A. CREATE THE MODEL 1. Start Stella and look at the three layers of the program: Interface, Map/model and Equations layer. Toggle between the layers by clicking on the arrows in the top left of the screen. 2. In the Map/model layer, build the model by placing stocks, flows, convertors and connectors on the sheet. Type the names for each variable. Note that the position of the names can be changed by dragging with the mouse. 3. Click on the Map/Model toggle (little globe) and note that question marks appear in the diagram. Double-click on each variable to open the dialog box belonging to that variable. Click on the "Document" button and enter information about the variable (like what it represents, what its units are, etc.). 4. Enter the equations, values of constants and initial values of stocks. Use the following values: Pollutant release = 10 Degradation constant = 0.1 Initial(Pollutant) = 0 5. Click on the small arrows in the top left corner of your screen and go the Equations layer. Look at the equations and compare them to what you typed in the dialog boxes. When you've finished, return to the Map/Model layer.

Environmental Modelling

66

Exercise 1

B. MAKE A GRAPH 6. Place a graph on the sheet by selecting the graph pad from the menu and then clicking on the screen. Double-click on the graph to open its dialog box. Select the variable to be plotted. For this graph, select "Pollutant". Click "OK" and fix the graph on the sheet by clicking on the pin in the top left corner. 7. Specify the simulation conditions by clicking on the running man in the bottom left corner. In the Run Specs, set the unit of time as "day", simulation length at 100 days and DT at 0.25 day. Run the model and look at the graph output. What is the equilibrium value of "Pollutant"? 8. Click on the variable "Pollutant release" and change its value to 5. Close the dialog box and run the model again. Look at the difference with the first simulation. Explain the difference. 9. Now double-click on the graph again and select the "Comparative" option. Click "OK" and run the model several times, each time changing the value of "Pollutant release". See the effect in the graph. 10. Now double-click on the graph again and unselect the "Comparative" option (this automatically unselects all previously selected variables). Add the variables "Pollutant inflow" and "Pollutant degradation" to the list of variables to be plotted. Run the model and look at the results. Note that Stella automatically scales the graph. To put variables on the same scale, open the graph and click on the arrows to the right of the selected variables. Put the two rate variables on the same fixed scale of 0-10 by filling in the minimum and maximum variables. Click on "Set" to confirm the values. Run the model again and look at the result. What is the advantage of plotting the rates on the same scale? There are other features for graphs, like adding more pages to one graph pad, and using scatter or bar graphs. Try these features yourself. You can always make use of the Help function of Stella, which contains all the information you need to find out how the software works.

Environmental Modelling

67

Exercise 1

C. CREATE TABLE OUTPUT (and understand rectangular integration) 11. Complete the table below without using Stella by using numerical integration (rectangular). You can use a pocket calculator if you wish (k=0.1, t=0.25, Pin=10, P0=0).
t 0 0.25 0.5 0.75 1 1.25 Pt 0 2.5 Pin 10 10 kP 0 0.25 dP/dt = Pin - kP 10 9.75 Pt+t = Pt + t(dP/dt) 2.5 4.94

12. In your model, place a table on the sheet (use the Table pad from the menu) and double-click on it. Select the same three variables you had in the graph. Select an output interval of DT (=0.25 day) and fix the table on the sheet using the pin. Run the model (use Pollutant release = 10) and see what happens. Compare the results in the top of the Stella table to your own calculations. 13. Change the report interval of your table and look at the results after running the model again. D. CREATE A USER INTERFACE We will now create an interface layer that makes it easier for a user to work with this model. 14. Unpin the table and close it. Do the same for the graph. They will remain on your Map/Model sheet, so you can always use them again later on. If the dialog box for running the model is still on your screen, close that too. Now place a sector frame around your model by selecting the tool from the menu and placing it on the screen. Expand it so that it includes all model components. Give the sector a name if you like. 15. First, go to the File-Default Settings menu and make sure the Link high-level map to model option is selected. Now click on the small arrows in the top left corner of the screen and move to the Interface level. You will notice that a model "map" has appeared on the interface screen. In this way, sector frames on the Map/Model level can be used to group variables or parts of the model that belong together. On the Interface level, the overall structure of the model can thus be represented in a clear way. For this simple model, this function is not

Environmental Modelling

68

Exercise 1

very useful but for bigger models it can help the user to understand the structure of the model. In the sector maps, pictures can be placed to visualise the function of that sector. You can click on the arrow in the Interface level map to go directly to the right part of the model on the Map/Model level. 16. Place a graph that shows model output on the Interface level. Use the same variables that you created before and fix the scales of the state and rate variables at 0-300 (Pollutant) and 0-25 (Pollutant inflow and Pollutant degradation). Run the model and check that the graph shows the correct results. 17. Next, place a "slider input device" (select from the menu on top) on the screen. Double-click on the device and select "Pollutant release". Set the minimum and maximum values to 1 and 20, respectively. Press "OK" to finish. Note that you now have a sliding device on your screen which allows you to easily change the value of the pollutant inflow between 1 and 20. Also note that on the Map/Model level, the appearance of the "Pollutant release" converter has changed slightly. Run the model and note the differences in the results. Note that you can pause the simulation, change the value of "Pollutant release" using the input device, and resume the simulation to see the effect. To make this easy, change the simulation speed in the Run Specs menu (e.g., by selecting 0.2 seconds real time for each time unit in the model). 18. Try out other interface elements yourself, e.g. special buttons to start and stop the simulation or other types of input devices. E. QUESTIONS Please answer the following questions (this is easy once you have made the model correctly). 19. When "Pollutant release" has a constant value of 13.92, what is the equilibrium level of Pollutant? How much time does it take for the pollutant to reach 99% of the equilibrium value (other values the same: initial Pollutant = 0, degradation constant = 0.1). time = Pollutant = _____________ _____________

20. Set Pollutant Release at 10 use a comparative graph. Find the value of k at which 99% of the equilibrium value is achieved after 76 days. What is the equilibrium value? Pollutant release = _____________ Pollutant = _____________

Environmental Modelling

69

Exercise 2

EXERCISE 2. The well-mixed lake Figure 2 shows the simplified relational diagram of the simple model of a well-mixed lake. This is the same model that is presented in Section 4.7.1 of the lecture notes Environmental Modelling.

Water volume Pollutant outflow

Pollutant concentration Amount of pollutant M

Pollutant loading

Pollutant degradation

Note: Figure 2 does not show the complete model. You still need to add a number of variables! Include the variables needed according to the model equations specified in section 4.7.1.
Pollutant settling

Figure 2. Simplified relational diagram of pollution in a well-mixed lake 1. Construct this model in the Map/Model layer of Stella. First construct the whole model using the stocks, flows, convertors and connectors without entering any numbers or equations in the Map model. Note that you can create bended flow symbols by pressing the SHIFT-key when you are drawing the flow. Then switch to Model model (by clicking on the globe) and click on each variable to enter the values of constants, initial values of stocks, etc. Use the values in Table 1 to parameterise the model. Enter information about the variables and their units in the document field of each variable. Check the units and dimensions and make sure you convert the values to the right units before entering them into the computer. Save your model in a file.

Environmental Modelling

70

Exercise 2

2. In the model layer, create a graph showing the changes in concentration of the pollutant in the lake with time. Create another graph which compares the rates affecting the concentration of the pollutant. Create a table showing the value of all the variables every 5 days. Run the model with the values given and determine which process affects the concentration of the pollutant most. Use the graph scaling options to facilitate comparison. Use a time step of 0.25 day and run the program for 25 days. Table 1. Parameter values for well-mixed lake model
Variable Value Unit

Initial concentration of pollutant in lake Concentration of pollutant in inflow Lake water volume Mean depth Inflow = outflow First-order degradation rate constant at 25 C Apparent settling velocity

0 19 50,000 2 5.2 0.25 0.01

mg L-1 mg L-1 m3 m m min-1 d-1 m d-1


3

3. Use the model to find the equilibrium concentration of the pollutant in the lake and compare this value with the analytical solution for the equilibrium concentration (see lecture notes, equation 26, page 51). You should get the same value. If not, find out where the error is. Ceq = ____ 4. Because of global warming, river flow will increase to 10,000 m3 d-1 and the degradation constant is expected to increase to 0.35 d-1. What happens to the equilibrium concentration? Check with the analytical solution. Discuss the result (is this what could be expected?). Ceq = ____ 5. Table 2 gives the river flows and concentrations of pollutant in the inflow and in the lake during the months of January to December, 1999 (as measured by a researcher). We assume temperature to be constant at about 25 C. Assuming that an equilibrium situation existed in 1999, estimate the value of the degradation constant by calibrating the model to the equilibrium situation.

kD = ____

Environmental Modelling

71

Exercise 2

6. Assuming that the source of the pollution is subsequently removed, estimate how much time it will take for the concentration in the lake to approach zero when the river flow rate remains constant (use the parameter values from Table 1 again). teq = ____

Table 3 gives the water temperature in the lake and the river in relation to the time of year (mean monthly temperature). The value of the degradation constant is assumed to double with an increase of 10 degrees in water temperature (Q10 = 2). Use the following formula to relate the value of the degradation constant to the water temperature: k = k25 * Q10(TEMP-25)/10 with k = degradation constant, k25 = degradation constant at 25 C, TEMP = actual temperature, and Q10 = Q10-value. Table 2. Monthly river flow (m3 min-1), and pollutant concentrations (mg L-1) in river and lake in 1999.
Month
January February March April May June July August September October November December

Flow (m min )
0.375 0.370 0.360 0.345 0.340 0.335 0.315 0.310 0.315 0.310 0.355 0.360
3 -1

Concentration river (mg L )


18.31 18.42 18.34 19.53 19.65 19.52 20.58 20.43 20.94 19.60 19.45 18.75
-1

Concentration lake (mg L-1)


6.25 6.27 5.80 6.05 5.91 6.25 6.45 5.95 6.33 6.25 6.30 5.99

Environmental Modelling

72

Exercise 2

7. Make the necessary modifications to your model to relate the degradation constant to temperature (using the built-in TIME function of Stella or by creating a MONTH-variable). Run the simulation model for a period of one year and show the effect of temperature on the concentration of the pollutant in a graph. How can you run this temperature-dependent model if you don't want to simulate the effect of temperature? Table 3. Mean monthly temperatures, 1975-2000.
Month January February March April May June July August September October November December Mean temperature ( C) 27.2 29.3 30.1 30.4 30.5 31.4 32.3 34.1 35.4 33.1 30.8 28.4

Environmental Modelling

73

Exercise 3

EXERCISE 3. Algal bloom Section 4.7.2. of the lecture notes gives an introduction of a simple model of algal growth in relation to phosphorous uptake and grazing by zooplankton.

Daphnia density

maximum consumption rate consumption rate half saturation constant

Algae Algal outwash

Natural mortality

maximum P uptake rate

P uptake

half saturation contstant P uptake

P input

Phosphorous

P outwash

~ P release P loss rate

Exercises 1. Build the model in Stella. Construct the model on the Model/Map layer and make a graph of phosphorous and algae concentrations on the interface layer. Daphnia density (in individuals l-1) is assumed to be constant, so is not modelled as a stock but as a convertor. Use the following values:

Environmental Modelling

74

Exercise 3

Variable
Initial algae concentration Half saturation constant P uptake Maximum P uptake rate Natural mortality of algae Daphnia density Half saturation constant algal uptake by Daphnia Maximum consumption rate Daphnia P loss rate Initial phosphorous concentration

Value
10 60 0.02 0.1 1 50 10 0.1 10
3

Unit
10 cells L-1 g L-1 g P cell-1 d-1 d-1 ind. L-1 cell ind-1 cells ind-1 d-1 d-1 g L-1

Relate the the value of P_release to TIME. The initial input should be 10 g d-1, and it should increase by 5 g d-1 every 100 days. Simulate the model for 1,000 days, with a DT of 1 day. Note: pay attention to the dimensions of the variables. Algal density is expressed as number of cells (thousands) per litre while phosphorous is expressed as g L-1. Therefore it is necessary to use unit conversion in the flow that connects the phosphorous and algae stocks. The conversion efficiency of phosphorous equals 20 103 algal cells per g of phosphorous. If you have successfully constructed this model, you will see that the algae suddenly bloom after about 300 days. 2. Create sectors for Daphnia, algae and phosphorous and create slider input devices on the interface layer for the following parameters: initial P concentration (do this by creating a separate converter and using this variable as initial value for the stock no need to connect the initial value convertor and the stock, just enter the converer name in the appropriate field) initial algal concentration half saturation constants maximum uptakes

Change the values of each parameter (one at a time) by plus and minus 10% and look at the effect on the state variables. Create a table with all state and rate variables and look at the detailed simulation results. Can you say something about the sensitivity of the model for the various parameters?

Environmental Modelling

75

Exercise 4

EXERCISE 4. Fishing

Step 1. Consider the logistic equation :


dN N = r N 1 dt K

Construct a Stella model for the growth of a fish population (N), using one state variable (stock) and one rate variable (flow), as well as the necessary auxiliary variables (converters) and connectors. Use the following conditions: simulation time = 40 years time step = 1 year r = 0.5 /year K = 10,000 tonnes of fish initial value of N = 500 tonnes Make two graphs: 1. a time series graph of N with time 2. a scatter plot of dN/dt with N Explain in words what the graphs mean.

Step 2. Add another rate variable to the model: fishing (C). Assume that fishing is constant and reduces the fish population N. Assume that C = 1000 tonnes per year. Questions: 1. Write the differential equation for change in the fish population, including the effect of fishing. Draw the model in a graph. 2. Run the model and observe what happens. Describe the situation. 3. Now increase the initial size of the fish population and look for the size at which model behaviour changes fundamentally. This is the first equilibrium point. Make a note of this population size. 4. Now increase the initial size of the population further and look for the second equilibrium point. Note down this population size too. What is the difference between these two equilibria?

Environmental Modelling

76

Exercise 4

Step 3. Climatic conditions (such as e.g. El Nio events) can induce temporary changes in water temperature and therefore in growth rate. Introduce a temporary growth reduction by relating "r" to time, converting it to a graph and reducing its value to 0.05 when time is about 15 years (use the model from Step 2). Look at the effect on population size N. How long does it take for the population to recover from this change in growth rate?

Step 4. Add to the model an increasing fishing pressure by relating "C" to time and having it increase from 500 at time = 0 to 3000 at time = 40 years. What happens to the fish population?

Environmental Modelling

77

Powerpoint slides

Appendix 2: Powerpoint slides

Environmental Modelling

78

S-ar putea să vă placă și