9
Environmental modeling and simulation -
applications and future requirements
R. Griitzner
University of Rostock
Department of Computer Science,
Albert-Einstein-Str. 21, D-18059 Rostock, Germany
Tel.:+49(381)4983369, Fax: +49(381)4983426
e-mail: [email protected]
Abstract
This paper gives a short report on the state of the art in the method of problem solving by mod-
eling and simulation used in environmental protection. Application fields, problems, and future
developments will be discussed. Requirements and new approaches to the software tools for
model and experimental description, simulation system architectures for parallel and distri-
buted simulation experiments will be represented.
Keywords
modeling, simulation experiments, simulation system architecture, environmental applications
1 BASE DEFINITION
Up to now the state of the environment has been changed by human activities in a dramatic and
dangerous way. This situation must be stopped as soon as possible. Therefore, we need innova-
tive theoretical and practical scientific approaches. The urgency of the problem poses a chal-
lenge to computer science and particularly to environmentally oriented informatics. Important
methods and tools are modeling and simulation. Modeling and simulation is a problem solving
method where problems are solved not by experimentation with real world systems but by
experiments using models of real world objects. Using this method, the main task is to develop
adequate models of the real world systems which we want to examine. If a simulation model
adequately represents the important aspects of the real system, selected for investigations, a
R. Denzer et al. (eds.), Environmental Software Systems
© Springer Science+Business Media Dordrecht 1996
114 Part Three Modeling and Simulation
conclusion is then permitted from the simulation results to the behaviour of the real system.
Therefore, modeling and simulation is a very important method in practical and theoretical
environmental research and protection.
To solve a given problem every model pursues a special aim - the model or simulation aim -
which defines the scope of the model. Only these components of a real system are mapped in a
model which are significant for the defined model aim. That means modeling is a process of
abstraction and generalization neglecting all unimportant real world system components (with-
out any contribution to the model aim).
Simulation models are abstract mathematical descriptions of system relations, the relations
between the states of a system. For descriptions model definition languages (e.g. DYMOLA) or
simulation languages (e.g. SIMULA, ACSL) are used.
Depending on the value scope of the state variables given, we distinguish models or systems
as continuous, discrete or combined. Models are called continuous if the variables of the system
possess values from an intervall, discrete if system variables assume discrete values. In com-
bined models system variables assume discrete and continuous values.
To get simulation results, we have to experiment with models. In an experiment computation
methods are applied to models (models are executed by a method). Computation methods or
short methods are computation algorithms. A base method is the simulation method which con-
trols the time-dependent dynamical behaviour of the system state. Other examples of methods
at a higher level are optimization of parameters or system trajectories, sensitivity analysis, and
Monte-Carlo methods (stochastic changing of parameters or inputs).
Experiments require an appropriate experimental description to define the kind of problem
solving needed. If model and experimental descriptions are separated, we get an important base
for a multiple use of models.
Essential software systems for modeling and simulation activities are simulation systems. A
simulation system supports the user during the phases of model building and description, the
experiment definition process, the execution and supervision of simulation experiments, the
analysis and visualization of simulation results, the managment of models, data, and experi-
mentation methods.
2 MODELLING AND SIMULATION OF ENVIRONMENTAL SYSTEMS
Modeling and simulation in the environmental domain today is used to determine the dynamic
behaviour of systems in order to make decisions or to test the impact of human actitivities (e.g.
the use of solar energy for heating, construction, production processes, traffic and logistic plan-
ning and decisions) on the environment. Further questions will be added in the near future
because of the increasing complexity of the systems: the assessment of alternative decisions
and activities, the examination of sceneries, the analysis of eco-economical and of socio-eco-
nomical systems. Additional future assignments would be the evaluation and optimization of
the whole life-time cycle of products related to their impacts on the environment. The life-time
cycle of a product consists of its production, use, and finally, the recycling process.
To solve these assignments, new experimentation approaches, simulation methods, computa-
tion methods, and software tools are needed. Several of the application fields and some soft-
ware architectures for modeling and simulation will be represented in the following.
Environmental modeling and simulation 115
2.1 Problems of Modeling Environmental Systems
The modeling and simulation of environmental systems requires the consideration of a lot of
special system properties which complicate the analysis considerably.
We have to consider different groups of problems. One group results from the complexity of
the environmental systems (e.g. the biological, ecological, economical properties) and the other
group results from the degree of the user's qualification in managing these problems (using the
models, storing and preparation of data, interpretation of results and, the making correct deci-
sions).
Problems of modeling environmental systems:
. there is a lack of theoretical knowledge
causal relations between system components are frequently unknown
. most of the real systems can be observed by experiments, but frequently there isn't any pos-
sibility for controlled experiments
ecological systems very often possess long reaction times (up to years or centuries) and in
contrast, other components possess very short reaction times
system complexity is high because of the strong relations between the system components
nature is an open self-organizing system with adaptable components
inaccurate data and ill-defined system components exist
there is an increasing set of data: spatial data and spatial data with time dependence.
often it isn't possible to define exact and unique optimization aims for ecological and envi-
ronmental systems
the model description must be realized by analytical and non analytical (rule based) methods.
Another group of problems result from the different levels of knowledge by users in the field of
environmental protection. There are highly qualified specialists without any knowledge in the
field of simulation and in the use of complex software systems. This situation requires software
with excellent user interfaces and user support components for the whole model-life-time
cycle. Unfortunately, we can't very often find simulation software which possess only some of
these expected support components. Such systems are still in research.
2.2 Fields of Application
The most important tasks in the field of environmental modeling and simulation today are still
the determination and analysis of the behaviour of environmental systems, that means the com-
putation of the time dependent dynamic behaviour of state variables.
Some applications of modeling and simulation are given below. On the one hand, undesired
secondary effects of human activities (e.g. emissions of harmful substances of production proc-
esses) can appear. In such a situation, the secondary effects to the environment are to minimize.
On the other hand, actions of environmental protection are to optimize. These are investiga-
tions of the questions: how does a process work? or which effects are released in the environ-
mental system by a special input?
In the environmental domain, simulation models are especially used in four fields: emission
computation (air, water, ground pollution), process control, groundwater - economical and flow
investigations, and ecosystem research.
116 Part Three Modeling and Simulation
But further applications become more and more important: e.g., models on the use and balance
of resources (e.g. water, ground, materials, energy, food); models of the carrying capacitiy and
of loading limits of ecological systems with an input, which has been caused by human activi-
ties; quality models; product-life-time-cycle models and ecobalances; socio-economic models;
combined tasks: the use of simulation and optimization methods or experiments consisting of
coupled economical and ecological models.
Models of pollution extension
An important task in environmental protection is the computation of immission loads caused by
emission processes. Most environmental problems are caused by the emission of harmful and
dangerous substances which are spreaded in air, water, and on the ground. The computation of
the expansion of radiation, noise, vibrations, chemical reactions of substances and their effects
on human life are of high interest today.
The primary targets of extension models are:
computation of pollution values at geometrical points where we don't have any measurement
results
detection of pollution sources by simulation experiments
supporting position planning and permission processes for objects with emissions (e.g. of
factories, production processes, traffic projects, buildings)
determination of reactions to accidents (e.g. of chemical processes, of oil transports).
An overview of the state of the art is given by (Zannetti,I990), (Wrobel,1991), (Melli,1992).
Models are urgently needed in the future for the computation of pollution and distribution in the
micro scale range (near the ground). For example, there are problems regarding the distribution
of auto exhaust in street canyons depending on traffic density and kind of formation of build-
ings near the street.
Water models
The pollution and the increased use of water based on constant water resources has produced
evident environmental damage all over the world. This causes a high importance of water mod-
els. There are three main groups of models:
investigations of the influence of structures, buildings and other human activities on the
amount and flow of groundwater
investigations of the levels of groundwater depending on irrigation, draining, and soil ero-
sion.
waste water models and water cleaning processes.
An important future assignment will be the modeling and simulation of the relations between
vegetation and groundwater levels.
Models for process control simulation
Models are used to evaluate the consumption of recources of technical processes and the emis-
sion of these processes. The primary target is the minimizing of the environmental influence.
Models are used during the planning phase and the model supported process control to get an
optimal process behaviour. Examples of process models are given by (Keller, 1994), (Tuma,
1994), (Frtih, 1989).
Environmental modeling and simulation 117
Modeling of ecosystems
The two fields of environmental protection and ecosystems are strongly connected. But today a
broad use of simulation is still difficult. Simulation is used only in selected parts, e.g. commu-
nity ecology in water, ground, forest; nitrogen balance; steadiness and elasticity of ecosystems
- that means the capability of taking up harmful substances without any effects on the ecologi-
cal balance; effects of chemical substances in ecological systems.
More than in other applications, modeling and simulation is an interdisciplinary task. The fol-
lowing part of this paper will refer to the modeling and simulation of ecological systems con-
nected with environmental protection.
2.3 Problems and Expected Results
Modeling and simulation cause some problems when using existing simulation systems and
simulation software tools. The main problems are:
it is necessary to get a better understanding of the dynamic behaviour of the systems by using
structure edaquate models
missing system support for investigations of human processes in such a way, that damages to
the environment will be at a minimum
handling the increasing system complexity (based on extensions of system borders and/or
detailed system investigations) by modular-hierarchical modeling concepts
development of comfortable user interfaces of software tools for simulation experiments.
An improvement in the present situation can be achieved on the one hand by the building of
more precise models (e.g. of structure adequate models, (Bossel,l994)) and on the other hand
by the development of better simulation software. We will explain the last point below.
The complexity of environmental systems demands a model description so that the method of
description can be adapted to knowledge about the system. Models consisting of different
described modules may be called heterogenous. For instance one modul of the model may be
described by differential equations and other moduls by rules, difference equations, higher
Petri-nets (ecocontroling models), or other tools. The simulation systems must be able to simu-
late such heterogenous models. That means, that dependent on the description method of a
module appropriate simulation methods must be used. The structure of the simulation system
must be changed to realize simulation runs of heterogenous models.
An effective experimentation requires control of experiments either interactive or by an
experiment description. Experimental control means to fix the methods, sequence of use, and
data to solve a problem in a simulation experiment. Methods are used for optimization, sensi-
tivity analysis, investigation of transient states, steady-state detection, visualization, and so on.
3 MODERN STRUCTURES OF SIMULATION SYSTEMS
The following considerations are restricted to the modeling and simulation of ecological prob-
lems, that means the consideration of combined systems which are described by systems of
ordinary differential equations of first order (system state equations).
Computer science can support the modeling and simulation in three important tasks:
118 Part Three Modeling and Simulation
building and managment of models and experiments
supervision of simulation experiments by adequate simulation system architectures
simulation result analysis and visualization.
The first and second points are investigated in our projects SAME and SAMEC (advanced by
the Deutsche Forschungsgemeinschaft), (Griitzner,I993).
3.1 Models and Modeling
One possibilty in handling the system complexity and to guarantee the mUltiple use of models
is the separation of model and experiment description (Schwarze, 1990), (Zeigler, 1986), (Zei-
gler,1989). Up to now this has been realized in only a few systems.
Separation means that a model represents only the structural description of a class of systems.
All of the values of parameters, of initial state variables, and system input variables are defined
in the experimental description belonging to the model. Aside from the values an experimental
description also defines the operations (or methods) which are applied to the model (e.g. opti-
mization of parameters) during the experiment's execution. The experiments control the model
and the method execution during the simulation process.
But there is a disadvantage in the separation because we need two languages, which a user
must know; one language for the model and another one for the experimental description.
Present model description languages are simulation languages belonging to a special simula-
tion system. This is a disadvantage because we are unable to exchange models between differ-
ent simulation systems and the user has to learn another simulation language if he wants to use
a new simulation tool. For the considered field of application, it seems important for the future
to look for a general modeling language independent of any simulation system. Such a general
language will allow a better model exchange and system understanding. To do a simulation cur-
rently, the model description would have to transform into the language used by the system.
Such an approach has been used successfully by electrical and electronical model descriptions.
The proposed language VHDL-A will be standardized soon (Bausch-Gall, 1994).
CEDARBOG ENVIRONMENT
10 Const
K - P ORGANIC OCI
NATUR - H OS
OJ
C
SUN LAKE
P OJ II
S_ S
H 02 12
C 03 I3
pENVIRON OC
0
H ES
C
Figure 1 Graphical description of the modular-hierarchical model CEDARBOG.
Environmental modeling and simulation 119
Modell Description
The models in the environmental domain are modular-hierarchical (a second concept to handle
the system complexity) and of a variable structure. The systems are decomposed into base
objects. To find a solution to an existing problem, a further decomposition of a base object isn't
necessary. Base objects contain the equations for the dynamic behaviour of the system. Cou-
pled base objects are called submodels. These can be coupled again to get submodels at a
higher level. So a model can be used as a submodel in another model. This means object ori-
ented modeling, where a base object represents a real system object.
It is important for users to have a model description language and a graphical modeling tool
at the same time. A mix of both descriptions in one model should be allowed.
The graphical model description of a simple example CEDARBOG of a silting-up process in
a lake see Figur 1 and the description by the language SAME-MDL see Figur 2. The model
consists of two submodels, NATURE and ENVIRONMENT, which consist of two base
objects. SUN defines the energy input by the sun, base object LAKE defines the state of plants
(P), of herbivorous objects (H), and of carnivorous objects (C). The base object ORGANIC
Basicobjects Submodels
basicobj ect SUN (out: s) ; submodel NATUR(out: 01.02,03);
constants float PI::3 .14; components
var double 5=0.0; SUN(out: S);
equations LAKE(inp: S; out: P,H,C);
S == 95.9 * (1 + O.635*sin{2*PI*TIME» ; connections
end. SUN.S -> LAKE.S;
LAKE.P -> 01;
LAKE.H -> 02;
basicobject LAKE(inp:S; out:P,H,C); LAKE.C -> 03;
states float P=D.D, I*Plants*/ end.
H=O.D, /*Herbivorous objects*! submodel ENVIRONMENT(out: 01,02;
C=D.D; j*Carnivorolls objects*/ inp: IO,Il,I2,I3);
var double 8=95.9; components
equations ORGANIC (out: OS;inp: P,H,C,Const);
p'= S 4.03 * P; ENVIRON(out: ES; inp: P,H,C);
H':::: 0.48 * P -17.87 * Hi connections
C'= 4.85 * H - 4.65 * C; 10 -> ORGANIC.Const;
end. I1 (ORGANIC.P. ENVIRON. P) ;
I2 -> (ORGANIC,H,ENVIRON.H);
I3 -> (ORGANIC.C,ENVIRON.C):
basicobject ORGANIC (out:QS; inp:P,H,C,Const); ORGANIC.O -> 01;
states float 0=0.0: ENVIRON.E -> 02;
var float P,H,C; end.
equations sublllodel CEDARBOG(inp:K; out:OC1,OC2);
OS = 0 * 0.1; components
0'= Canst * P + 6.12*H + 1.95*C; NATUR(out: 01,02,03);
end. ENVIRONMENT (out:01,02;
inp: IO,Il,I2,I3);
connections
basicobject ENVIRON{out:ES; inp: p,H,e); K -> ENVIRONMENT.IO;
states float E=O. 0; NATUR.Ol ENVIRONMENT.Il;
var float P,H,C; NATUR.02 -> ENVIRONMENT.I2;
equations NATUR.03 -> ENVIRONMENT.I3;
ES :::: E * 100; ENVIRONMENT. 01 OC1;
E/= 1.00 * P + 6.90*H + 2.70*C; ENVIRONMENT.02 -> OC2;
end. end.
Figure 2 Model description of CEDARBOG using SAME-MDL, (Dimitrov,1994),
120 Part Three Modeling and Simulation
models the sediment process and ENVIRON the energy output to the environment.
Because of the above mentioned problems in environmental modeling, it's important to allow
the use of different mathematical descriptions of the dynamic behaviour of the base objects,
e.g. by differential and difference equations, rule based approaches, fuzzy rules, and neuronal
net model components. Up to now this hasn't been realized by SAMEC.
Experimental Description
The proposed procedural experiment description language (Dimitrov,1994) is based on the
frame concept. A frame is a structuring element for an experiment also to define subexperi-
ments and a parallel experiment execution. Subexperiments can be described depending on
additional conditions. One example of such an experiment description is given by Figure 3.
This is an experiment with the model CEDARBOG for two parallel experiments using this
modell. CEDARBOG may be used with two different parameter sets.
experiment EXAMPLE;
variables float konst;
timeseries OCll, OC21, OC12, OC22; /* result files */
frame
framel (model_name, inparam, resultl, result2, tstart, tend);
(simulate (model_name; RK4, tstart, tend; inparam,resultl,result2));
frame2 (model, inconst, resl, res2, tstart, tend);
{konst = inconst;
simulate (model; RK4,tstart,tend; inconst,resl,res2)};
control
parbegin
framel (CEDARBOG, 2.55, OCll, OC2l, 0, 250);
frame2 (CEDARBOG, 5.10, OC12, OC22, 0, 250);
parend
end.
Figure 3 Description of two parallel experiments with the model CEDARBOG.
3.2 Simulation Systems
Let us consider simulation systems for ecological systems. The architecture of simulation sys-
tems is affected by the following requests:
interactive use in all working phases and user friendly interfaces
support of the users in the building and managment processes of models and experiments
parallel and distributed execution of experiments
use of innovative methods: genetical algorithms, fuzzy set and fuzzy rule based concepts,
neuronal nets
coupling of simulation systems with information systems (geo-information systems).
Parallel simulation
Based on the modular-hierarchical model concept, we have realized a system architecture for
parallel and distributed simulations of base objects. The system of algebraic-differential equa-
tions is separated into subsystems, the equations of the base objects. For the parallel simulation
Environmental modeling and simulation 121
of the base objects special integration algorithms have been used (Schiifer,1994),
(Werner,1995). Another iterative algorithm for parallel simulation, which uses the classical
numerical integration methods for base objects, is proposed by (Singh,1986). But in contrast to
the first algorithms the iterative solution is inaccurate.
To control the couple relation between the base objects and the submodels special coordina-
tor routines are introduced. A coordinator realizes the data exchange between the model com-
ponents and some functions for the integration process. The external structure of the model is in
this architecture equivalent to the internal structure. Some results of measurements neglecting
the communication overhead indicate an evident speed up (Werner,1995).
Parallel experiments
An experiment description also allows a parallel experiment realization. Each experiment is
then controlled by its own experiment process. Parallel experiments are especially important
(high speed up) for parameter and trajectory optimization, sensitiveness investigations, and
other multiple simulation runs.
The communication between processes are based on the PVM concept (Parallel Virtual
Machine). A special communication protocol of simulation system processes (PSP) has been
defined for implementation (Nekien,I995). The implementation has been realized at an UNIX
workstation net.
3.3 Coupling of Simulation Systems with other Software Tools
Modeling and Simulation of environmental systems means working with a lot of data (e.g.
input, geographical, environmental, result data). For an effective simulation all data must be
provided automatically from different data bases (e.g. geographical information systems, envi-
ronmental information systems). The access to the data supposes a coupling of simulation sys-
tems and data bases. The data which are used in the simulation experiments must be defined in
the experiment description in such a way, that no manual actions by users are necessary. The
access description to data bases in experiments must be possible without any knowledge of
internal data base structures.
4 REFERENCES
Bausch-Gall,1. (1994) Modellaustausch zwischen Simulationsprogrammen. in Fortschritte der
Simulationstechnik, Bd. 9, ASIM 9. Symposium (ed. G. Kampe & M. Zeitz), Vieweg, Stuttgart.
Bossel, H. (1994) Understanding Dynamic Systems: Shifting the Focus from Data to Structure,
in 8. Symposium Informatikfor den Umweltschutz (ed. L. M. Hilty, A. Jaeschke, B. Page, A.
Schwabl .), Bd.l, Metropolis Verlag, Marburg, 63-75.
Dimitrov, E. (1994) Modell- und Experimentbeschreibungenfor Umweltproblemstellungen und
ihre Implementation. Preprint No. CS-01-94, University of Rostock, Dept. of Computer Sci-
ence, WG Modeling/Simulation.
Froh, K. F. (1989) Informationsorientierte ProzeBleittechnik flir die Chemische Industrie. Auto-
matisierungstechnische Praxis atp, 31(7), 301-305.
Griitzner, R. (1993) AbschluBbericht zum Projekt SAME. University of Rostock, Dept. of Com-
puter Science, Report of the WG Modeling/Simulation
122 Part Three Modeling and Simulation
Keller, H. B., Kugele, E., groBe Osterhues, B. (1994) Einsatz neuronaler Netze im UmweItbe-
reich am Beispiel der Miillverbrennung. in 8. Symposium lnformatikfiir den Umweltschutz
(ed. L. M. Hilty, A. Jaeschke, B. Page, A. Schwabl .), Bd.1, Metropolis Verlag, Marburg, 405-
414.
Melli, P., Zannetti, P. (1992) Environmental Modelling. Computational Mechanics Publications
Southampton.
Nekien, T. (1995) The communication protocol in the SAME-simulator. University of Rostock,
Dept. of Computer Science, WG Modeling/Simulation, in publishing.
Schafer, P. (1994) Parallele Verfahren zur Simulation modularer technischer Systeme. in ASIM
9. Symposium (ed. G. Kampe & M. Zeitz), Vieweg, Stuttgart.
Schwarze, G. (1990) Digitale Simulation. Akademie Verlag, Berlin.
Singh, M. G., Allidina, A. Y., Daniels, B. K. (1986) Parallel Processing Techniquesfor Simula-
tion. Plenum press, New York, London.
Tuma, A., Haasis, H. D. (1994) Real Time Production Scheduling of Environmental Integrated
Production Systems - A Comparision of Selected Knowledge-Based Methods and Maschine
Learning Algorithms. in 8. Symposium lnformatikfiir den Umweltschutz (ed. L. M. Hilty, A.
Jaeschke, B. Page, A. Schwabl), Bd.1, Metropolis Verlag,Marburg, 371-8
Wrobel, L. c., Brebbia, C. A. (1991) Water Pollution: Modelling; Measuring and Prediction.
Computational Mechanics Publication, Southampton.
Werner, G.(1995) Vergleichende Analyse von parallel en Simulationsmethoden zur Ausfiihrung
komplexer Modelle. University of Rostock, Dept. of Computer Science, Diplomarbeit.
Zannetti, P. (1990) Air Pollution Modelling. Computational Mechanics Publication, Southamp-
ton.
Zeigler, B. P., Oren, T. 1. (1986) Multifacetted Multiparadigm Modelling Perspectives: Tools for
the 90's. in Winter Simulation Conference. SCS Publications, Washington, 708-712.
Zeigler, B. P., Zhang, G. (1989) The Entity Structure: Knowledge Representation for Modeling
Design. in Artificial Intelligence, Simulation, and Modeling (ed. L. E. Widman, K. A. Loparo,
N. R. Nielsen), John Wiley&Sons, New York, 47 - 73.
5 BIOGRAPHY
After a study of applied aerodynamics and an activitiy of seven years in the industry follows an
activity as research worker at the Humboldt University of Berlin, fields of modeling and simu-
lation, computer performance analysis, and operating systems. Since 1989 professor for mode-
ling/simulation at the University of Rostock. Working fields: architecture of simulation
systems, parallel simulation, user interfaces, application of simulation in environmental
research.
