C H A P T E R S E V E N
Integrated Modelling Frameworks for
Environmental Assessment and Decision
Support
A.E. Rizzoli a, G. Leavesley b, J.C. Ascough II c, R.M. Argent d,
I.N. Athanasiadis a, V. Brilhante e, F.H.A. Claeys f, O. David b,
M. Donatelli g, P. Gijsbers h, D. Havlik i, A. Kassahun j, P. Krause k,
N.W.T. Quinn l, H. Scholten m, R.S. Sojda n, and F. Villa o
Contents
7.1. Introduction 102
7.1.1 A first definition 103
7.1.2 Why do we develop new frameworks? 103
7.1.3 A more insightful definition 104
7.2. A Generic Architecture for EIMFs 105
7.2.1 A vision 107
7.3. Knowledge Representation and Management 107
a IDSIA, Galleria 2, CH-6928 Manno, Switzerland
b US Geological Survey, PO Box 25046, MS 412, Denver Federal Center, Lakewood, CO 80225, USA
c USDA-ARS-NPA, Agricultural Systems Research Unit, 2150 Centre Avenue, Bldg. D, Suite 200, Fort Collins,
CO 80526, USA
d Bureau of Meteorology, Water Division, GIPO Box 1289, Melbourne 3001, Australia
e Federal University of Amazonas, Computing Science Department, Av. Gen. Rodrigo Octavio J. Ramos, 3000,
Aleixo, Manaus, AM 69077-000, Brazil
f Ghent University, Department of Applied Mathematics, Biometrics and Process Control, Coupure Links 653, B-
9000 Gent, Belgium
g IPSC, Agri4cast Action, Joint Research Centre, Via E. Fermi, 2749, I-21027 Ispra (VA), Italy
h WL Delft Hydraulics, Inland Water Systems, PO Box 177, 2600 MH, Delft, The Netherlands
i Smart Systems Division, Austrian Research Centers GmbH, ARC, 2444 Seibersdorf, Austria
j Department of Social Sciences, Information Technology Group, Hollandseweg 1, Building 201 (De Leeuwenborch),
6706 KN, Wageningen, The Netherlands
k Institut für Geographie, Lehrstuhl Geoinformatik, Friedrich-Schiller-Universität Jena, Loebdergraben 32, D-07743
Jena, Germany
l Berkeley National Laboratory, University of California, 1 Cyclotron Road, Bld. 70A-3317H, Berkeley, CA 94720,
USA
m Wageningen University, Social Sciences, Information Technology Group, Hollandseweg 1, 6706 EW, Wageningen,
The Netherlands
n Northern Rocky Mountain Science Center, USDI, Geological Survey, 212 AJM Johnson Hall, Montana State
University, Bozeman, MT 59717-3492, USA
o Ecoinformatics Collaboratory, Gund Institute for Ecological Economics, University of Vermont, 617 Main Street,
Burlington, VT 05405, USA
Environmental Modelling, Software and Decision Support © 2009 Elsevier B.V.
1574-101X. All rights reserved.
101

102 A.E. Rizzoli et al.
7.3.1 Challenges for knowledge-based environmental modelling 109
7.4. Model Engineering 110
7.4.1 Component-based modelling 111
7.4.2 Distributed modelling 113
7.5. Driving and Supporting the Modelling Process 114
7.5.1 The experimental frame 114
7.6. Conclusions 115
References 116
7.1. Introduction
As argued in Chapter 1, modern management of environmental resources de-
fines problems from a holistic and integrated perspective, thereby imposing strong
requirements on Environmental Decision Support Systems (EDSSs) and Integrated
Assessment Tools (IATs). These systems and tools tend to be increasingly com-
plex in terms of software architecture and computational power in order to cope
with the type of problems they must solve. For instance, the discipline of Inte-
grated Assessment (IA) needs tools that are able to span a wide range of disciplines,
from socio-economics to ecology to hydrology. Such tools must support a wide
range of methodologies and techniques like agent-based modelling, Bayesian deci-
sion networks, optimisation, multicriteria analyses and visualisation tools, to name
a few.
Sometimes EDSSs and IATs are built from scratch, often with limited resources,
by non-programmers. From a software point of view, these applications are custom-
made, by craftspeople rather than industrially developed by professionals. More
recently, the disadvantages of this approach, which can quickly become overly
expensive in terms of delivery time and resources required, have been addressed
by the development of suites of software engineering tools called Environmental
Integrated Modelling Frameworks (EIMFs). EIMFs have typically been designed
as a response to the increasing complexity of building and delivering EDSSs and
IATs.
Modelling frameworks are not a novelty per se, having made a first appearance in
the management science field towards the end of the 1980s (Dolk and Kottemann,
1993; Geoffrion, 1987). The framework concept later found its way into com-
mercial packages such as MATLAB for scientific computing, GAMS and AMPL
for management science and operations research applications. Moreover, modelling
and simulation tools and frameworks have been taken up on a large scale in other
disciplines, and standards for developing and expanding them have been adopted.
As a result, electrical circuit design toolkits and printed circuit board simulators have
contributed significantly to the advancement of electronics in science and industry.
The same holds for many other sectors, from the automotive industry to mechanical
systems design. In contrast, no modelling framework has been universally adopted
within the environmental modelling domain, and the number of environmental
modelling frameworks is still growing.

Integrated Modelling Frameworks for Environmental Assessment and Decision Support 103
A frequently asked question is: “why do we need yet another modelling frame-
work?” The reasons why MATLAB (http://www.mathworks.com), MathCAD
(http://www.mathsoft.com), Mathematica (http://www.wolfram.com) and simi-
lar software environments are not up to the task of deploying effective and usable
EDSSs are often unclear, and there is always the option of re-using an existing EIMF.
Yet, this option is often disregarded, again without clear reasoning behind it.
In this chapter, we strive to address the above issues and clearly identify the
essential characteristics of an EIMF. Moreover, we wish to: (1) point out the main
differences among the leading EIMFs present on the scientific market; and (2) assess
which characteristics justify the differences, and which characteristics are artificial
and should be ignored to better facilitate interchange of knowledge and experiences
in EIMF development. Finally, this chapter also advocates the development of open
standards for the exchange and re-use of modelling knowledge, including data sets,
models, and procedures in order to facilitate improved communication among the
leading EIMFs.
7.1.1 A first definition
Definitions are tricky in that just the simple act of defining something reduces and
limits its essence. Yet definitions are useful since they provide a common under-
standing of the fundamental nature of things. In this chapter, we attempt to identify
the essential characteristics of an EIMF while retaining the necessary flexibility to
allow for the different declinations of EIMFs in practice.
Thus, a first and very general definition of an EIMF is: “a set of software li-
braries, classes, components, which can be (re-)used to assemble and deliver an
environmental decision support system (EDSS) or an integrated assessment tool
(IAT).”
However, this definition is potentially too generic since it does not fully capture
the essence of an EIMF. It also depends on adequately defining the essential func-
tions provided by EDSSs and IATs. Moreover, if the EIMF itself is too generic, then
the programmer and the modeller will feel more comfortable using well-assessed
code development frameworks such as .NET and J2EE and mathematical mod-
elling tools such as MATLAB and Mathematica, rather than taking the trouble of
learning to use a new framework.
Yet, we still develop frameworks – examples are TIME (Rahman et al., 2004,
2003), OpenMI (Gregersen et al., 2007), Tornado (Claeys et al., 2006), OMS
(David et al., 2002; Ascough et al., 2005), JAMS (Kralisch and Krause, 2006), and
ModCom (Hillyer et al., 2003). Their proliferation leads us to ask the reason why.
7.1.2 Why do we develop new frameworks?
If we take a quick review of the most successful environmental models in different
domains, e.g. MODFLOW (http://water.usgs.gov/nrp/gwsoftware/modflow2000/
modflow2000.html) for groundwater modelling or the MIKE11 series of models
(MIKE11, MIKE21 and MIKE Basin) for hydrodynamic modelling (http://www.
dhisoftware.com/general/Product_Overview.htm), we notice that few of them

104 A.E. Rizzoli et al.
were developed using an EIMF. The Argus One Numerical Environment (http://
www.argusint.com) is as close to an EIMF that has been developed to support
popular groundwater models. This software provides graphical pre-processing and
post-processing tools available as a plug-in extension for several USGS groundwater
codes including MODFLOW as well as Arc/Info and Arcview. However, we can
also pose another interesting question: will the next generation of these types of
models be developed using more comprehensive EIMFs?
The answer is hopefully yes, but only if the EIMFs prove to be effective devel-
opment tools – otherwise they will not be used. This answer is trivial, but it also
means that we need to identify the most important features that make an EIMF a
powerful development tool.
The main reasons why we need an EIMF are: time, money, and quality.
• Time: we want to deliver a new application (for instance, a decision support
system for the management of water quality in a river stretch) in a reasonably
short period of time, e.g. within months rather than years.
• Money: we want to re-use what we have previously developed, and possibly re-use
and link to what others have developed.
• Quality: we need to deliver results of proven quality, and for this, we need a tool
that guides us through a proven development process.
The development of a new framework starts from these main drivers, usually
targeting a specific domain in order to solve a problem of moderate complexity.
Being a framework, it is therefore extended but evolution and growth bring in new
problems. For example the framework either becomes too difficult to maintain or
too complex to use, outlives the programming language in which it was written, or
becomes obsolete for another reason. Before this happens the framework may reach
the point where it takes less effort to build a newer, simpler to use EIMF (such
as one that is better suited to a well-defined class of problems) than to solve the
problem at hand using the existing framework. Therefore, a “generic” or “all-in-
one” EIMF seems to remain the holy grail for environmental simulation software
and this appears to be a driving force behind new EIMF development.
We believe that continually developing new EIMFs should not be the case, and
that now is the time to make better use of available resources in order to improve
existing EIMFs while controlling their growth, integrating the strengths of other
developers, and sharing common and re-usable knowledge in the form of data,
models and processes.
7.1.3 A more insightful definition
We can now assess that our previous definition of an EIMF is too generic and,
while in principle very powerful, could lead to the design and implementation of
frameworks of little re-usability because of the steepness of the learning curve of
such a framework. On the other hand, a definition that specifies, in the smallest
detail, the requirements of an EIMF would unavoidably tend to be too application-
specific, and we would end up with a one-to-one relationship between frameworks
and applications.

Integrated Modelling Frameworks for Environmental Assessment and Decision Support 105
We think that we need to shift the development process of EIMFs in order to
foster re-usability of knowledge, data and models across frameworks, thus minimis-
ing the re-coding and re-design of frameworks for fitting particular needs. In order
to achieve this aim, we need a new, more insightful definition that is able to cap-
ture the essence of an EIMF. We therefore start from the name itself, which is a
derivative of the keywords environment, integration, and models.
• Environment means that the framework must target the environmental domain
and, even more specifically, the particular environmental sector under inves-
tigation. It must therefore provide easy access to a domain-specific body of
knowledge. Yet the framework should be configurable in order to span differ-
ent domains, thereby allowing real integrated modelling.
• Such knowledge resides both in models and in data, which may pertain to different
sub-domains of the natural environment, and to socio-economic dimensions that
are essential in IA studies.
• Finally, the knowledge must be made operational by integration. This means es-
tablishing causal links across domains by means of modelling and simulation.
We can therefore extend the previous definition of an EIMF by specifying in
greater detail what we mean by “supporting the assembling and delivering” of envi-
ronmental applications. A more robust and insightful definition of an EIMF is: “a set
of software libraries, classes and components, which can be (re-)used to assemble
and deliver an environmental decision support system (EDSS) or an integrated as-
sessment tool (IAT), to support modelling and processing of environmental knowledge
and to enhance the re-usability and distribution of such knowledge.”
In the remainder of this chapter we focus on how EIMFs can support mod-
elling and the processing of models, and how environmental knowledge, models,
data and workflows can be efficiently stored, used, and exchanged across different
frameworks.
7.2. A Generic Architecture for EIMFs
Given that we have put forward sound arguments as to why an EIMF should
not be too generic, it remains rather difficult to come up with an encompass-
ing definition of the ideal architecture of such a framework. In the context of
the SEAMLESS project (http://www.seamless-ip.org), for example, an architecture
has been proposed that is rather generic and yet can accommodate essential EIMF
components as described above. In fact, the SEAMLESS project is quite unique in
trying not to develop a new EIMF from scratch, but to re-use existing ideas and
components from other EIMFs. An architecture such as this is shown in Figure 7.1.
This is a layered architecture. At the bottom we find the knowledge base which
is a semantically annotated collection of data structures, models, simulation tools,
optimisation algorithms, data analysis routines and workflows. The knowledge base
is populated by accessing large and heterogeneous databases. We will outline later
the role of ontologies in the mediation between the databases and the knowledge
base.

106 A.E. Rizzoli et al.
Figure 7.1 A generic architecture for an EIMF.
On top of the knowledge base, we find three “framelets” (Pasetti, 2002).
Framelets are lightweight and highly specialised frameworks. The knowledge manager
framelet provides software structures to access the knowledge base and allows the
other two framelets such access to the knowledge base. The model manager framelet
specifically targets modelling, while the experiment manager framelet allows the cre-
ation of workflows where tools are coupled with models to perform a number of
activities, such as simulation and calibration experiments, model sensitivity analy-
ses, output visualisation, simulation monitors, and so on. The separation between
the model manager and the experiment manager allows distinguishing between the
model and the operations that we perform on it. For example, in a calibration ex-
periment the calibration algorithm is a tool that operates on the model, varying its
parameters to find the best fit for instance to a given behaviour.
The framelets can then be combined into the modelling and experimentation en-
vironments. These software environments facilitate the development of end-user
applications. We can think of these environments as plug-ins to existing software
development frameworks such as Eclipse RCP (McAffer and Lemieux, 2005) and
the NetBeans platform (Keegan et al., 2005). Of course, applications are the final
product and they can be EDSSs, IATs or specific applications aimed at solving a
given problem.
It is important to note that this architecture provides a blueprint that not every
EIMF will follow, i.e. not all the components must be in place. The role of this

Integrated Modelling Frameworks for Environmental Assessment and Decision Support 107
architecture is to provide a common layout to compare different EIMFs and map
their architectures to a common reference.
7.2.1 A vision
We have clearly stated that the above architecture simply identifies some key el-
ements in the structural design of an EIMF. Yet we think that the generality of
the EIMF is guaranteed at the lowest level, the knowledge base, while it specialises
in the upper levels. We now introduce our “heretical” vision: the most important
parts of an EIMF are its components, described and annotated in the knowledge
base which, if properly designed (as specified later), can stand the test of time and
be used across multiple frameworks. The re-usability of knowledge base compo-
nents (data, models, ontologies, workflows) across frameworks is an achievable and
worthwhile goal that we should pursue.
In the next sections, we will focus on the features provided by the knowledge
manager, model manager and experiment manager. In particular, we will describe
software design and implementation features that will enhance the re-usability of
knowledge across frameworks.
7.3. Knowledge Representation and Management
Formal knowledge representation through ontologies has been suggested as a
viable solution for information and knowledge integration problems (Ludaescher et
al., 2001; Villa, 2007) on the grounds that they elicit the meaning of knowledge in
ways understandable by both computer systems and humans.
An ontology is a formalism for knowledge representation that comprises a
vocabulary of terms representing concepts, properties and relations, knowledge do-
main characterisation, and formal specifications of the intended meaning of such
terms (Uschold and Gruninger, 1996). As ontologies are founded on logical lan-
guages, automated reasoning can be employed in order to ensure model consistency
and ontology-compliance.
The integration of models and data is the principal problem faced when building
EDSSs and IATs. As we know, models and data are intrinsically related: “Science
consists of confronting different descriptions of how the world works with data,
using the data to arbitrate between the different descriptions, and using the best
description to make additional predictions or decisions. These descriptions of how
the world might work are hypotheses, and often they can be translated into quanti-
tative predictions via models” (Hilborn and Mangel, 1997). In modelling practice,
however, to consistently relate data to models is not an easy task because data, while
conforming to the same paradigms and world views that inspire model conceptu-
alisations, may not directly connect to the higher-level set of concepts necessary to
describe a model. This difficulty often leads to bias and mismatches between models
and supporting data sets.

108 A.E. Rizzoli et al.
Figure 7.2 Modelling through mapping between semantic data and semantic model compo-
nents.
Suppose a modelling framework exists which includes as resources (possibly dis-
tributed) ontologies, data sets, and model components. An ontology of domain
data can be used to semantically annotate a data set of interest, in this way elic-
iting data properties and relations. Similarly, model components can be annotated
through a model ontology interface, including variables, parameters and modes of
operation. As a result, we would have a description of data and model components
with a much narrower conceptual gap between them. Models can then be obtained
by mapping the semantically annotated domain data to the semantically annotated
model components. To give some examples:
• a data item annotated as being a measurement of an amount of some matter, and
hence expressed in a unit of the mass dimension contained in an entity of the
natural system, can be mapped to a stock concept;
• a data item annotated as being a rate of transfer of matter between entities over
time can be mapped to a flow concept;
• a data table whose cells contain measurements of a data item a in relation to
measurements of a data item b, annotated as a dependency relation of a with
respect to b, can be mapped into a causal relationship between a and b, expressed
as b being a variable in the equation that defines the variable a.
This approach can be extended to accommodate other characteristics of envi-
ronmental data, such as spatiotemporal relations, economic data value changes as
a result of inflation, management of complex data types, etc. Such mappings em-
bellish existing data sources and models with advanced semantics that may lead
to instantiation and linking of model components, and ultimately to declaratively
defined models (Villa et al., 2006). Figure 7.2 illustrates the approach.
Of course, the realisation of such a modelling approach requires efforts to de-
liver – building on existing means (techniques, tools, etc.) – not only the modelling
mechanisms capable of performing semantic annotation and mapping, but also the
formal knowledge itself. Note that the mapping can occur in both directions and
in combination: annotated data can be mapped to annotated model components
and vice-versa. The former may be advantageous in modelling exercises where
high-quality data are available, and the latter where suitable model components

Integrated Modelling Frameworks for Environmental Assessment and Decision Support 109
can be identified. Either way, the mapping between semantic data and semantic
model components would restrict the space of modelling solutions to plausible
knowledge-level matches between conceptual model structures and data properties
and relations. Modellers’ expertise can further arbitrate and refine the best models
out of this space of model solutions. Furthermore, the approach promotes the re-use
of both data sets and model components, that is the re-use of modelling knowledge
at large. Re-use of model components requires techniques to determine the right
scope of the components so as to make them small enough to maximise re-usability,
yet large enough to contain significant modelling knowledge.
7.3.1 Challenges for knowledge-based environmental modelling
Apart from facilitating smooth integration between data and models, the adoption
of ontologies as a mediation resource has other advantages that include:
1. The efficient definition of declarative models. So far, declarative modelling has been
concentrated on expressing model equations in a declarative way (e.g. as found
in environments such as Simile, http://www.simulistics.com, and STELLA,
http://www.isesystems.com, and in the Modelica framework, http://www.
modelica.org). While capturing the structure of causality contained in a model
remains difficult due to the simplified logics allowed by current ontology frame-
works, extending model components with rich semantics for variable and para-
meter definitions can lead to great improvements in performing logical, dimen-
sional and structural validation of models.
2. The automation of scaling and unit transformations. This is needed because data will
no longer be solely vectors of numbers but will be associated with units, di-
mensions and spatial and temporal references that explicitly give the appropriate
context to the numbers. For example, transforming an ozone concentration from
0.12 ppm into 235 µg/m3 could be done automatically through knowledge-
based tools, able to manipulate units and dimensions. Extending the coverage of
the knowledge base to the conceptual aspects of space and time also allows sys-
tems to perform automatic aggregation and propagation of values over extents
represented at different resolutions or with different paradigms, greatly facili-
tating the simulation of multiple-scale models and reducing data pre-processing
overhead.
3. Support for handling and communication of uncertainties. Semantic annotation can
greatly help the production of uncertainty records associated with measurements,
suggesting or mandating the definition of known factors associated with partic-
ular methodologies, and assisting the system in propagating the calculation of
uncertainties along the chain of computation when a model is executed.
A challenge that should not be underestimated is the recognition and accep-
tance of shared ontologies, even when the difficulties of developing, storing and
maintaining large ontologies are successfully addressed. A common resistance in the
scientific community to the use of ontologies is the fear of committing to a specific
conceptualisation that may not fully reflect one’s scientific view. Ontologies exist
for eminently practical reasons, and the level of conceptual sophistication reachable

110 A.E. Rizzoli et al.
by current, first-order approaches is low enough to make such concerns relatively
easy to dispel. Nevertheless, the time, training and discussion necessary to induce
acceptance of such approaches in the user community is easily underestimated.
Knowledge-based computing will put modelling back in the hands of mod-
ellers: environmental modelling may become a conceptual activity, focusing on
model design rather than on model implementation. Code generation and imple-
mentation of software components could be largely delegated to ontology-aware
tools. In this respect, we envision the whole model lifecycle to change drasti-
cally, becoming more of a theoretical activity and less of a coding-intensive, highly
engineering-oriented task. For now though, knowledge-based approaches remain
a little-understood black box in the minds of most environmental scientists and
engineers.
7.4. Model Engineering
While the creative activity of writing a model pertains to the area of knowl-
edge management (as described in the previous section), here we focus on the
support to model “engineering” provided by EIMFs and their model manager
framelets. By engineering, we mean the set of tools and machinery that enables
a modeller to transform a conceptual model, declaratively represented and semanti-
cally annotated, into computer executable code. We also include the infrastructure
and software solutions that enable this model to be linked with other models, inte-
grated in different modelling exercises, and distributed to a variety of end users and
platforms.
From this viewpoint, an EIMF should: (a) allow for rapid prototyping of mod-
elling exercises by accessing a library of models and solutions; (b) assure backward
compatibility with existing (legacy) models; (c) assure interoperability with other
software tools and protocols, e.g. accessing GIS tools for the purpose of building a
decision support system; and (d) allow the re-use of the models developed using the
EIMF within other frameworks. The ORCHESTRA network is an example of an
architecture for integration and interoperability (see http://www.eu-orchestra.org),
and OpenMI is another example of a framework focusing on the integration of
legacy models (http://www.openmi.org).
However, note that it is practically impossible to satisfy all these constraints at
the same time. Use of legacy models (point b) might impede some aspects of in-
teroperability (point c). Moreover, the rapid development of models (point a) often
assumes the use of EIMF specific libraries that imply dependencies which may be
difficult to solve when re-using the model elsewhere (point d). We need to ex-
plore alternative options to find a reasonable compromise to satisfy the conflicting
requirements. A solution is to approach the problem from the perspective of model
representation:
1. Run-time linkage of model engines (components) through well-defined interfaces. Each
model engine contains a model as well as its executor. Models can be imple-
mented using different formalisms. This approach is very well suited for re-use

Integrated Modelling Frameworks for Environmental Assessment and Decision Support 111
of legacy codes. However, it lacks maintainability and homogeneity. Linkage can
be based on “push/pull” mechanisms (e.g. OpenMI) or a “bus” approach (e.g.
TISC, http://www.tlk-thermo.com/tisc.html). Software agent and web-service
architectures can be utilised for implementing EIMFs as virtual enterprises.
Component-based modelling plays an important role in this approach.
2. Using a declarative modelling paradigm for the description of the overall model. Thanks
to the independence of the model, expressed in a declarative fashion from its
imperative implementation in code, the model is translated to an executable
model description using flattening and optimisation techniques and then run by
one single executor. This approach offers a high degree of homogeneity and
maintainability. Moreover, the declarative representation can be semantically an-
notated in an ontology, allowing automated processing of the model knowledge.
It is unrealistic, however, to assume that any one paradigm will ever be power-
ful enough to capture all modelling formalisms. In the area of physical system
modelling, Modelica is an example of a high-level unified modelling paradigm.
A good introduction to declarative modelling for environmental models is pro-
vided in Muetzelfeldt (2004).
3. Translation of models implemented in different formalisms to one single, low-level common
denominator. This is a third option, an example of which is the Discrete Event
System Specification (DEVS) that can be regarded as the assembly of modelling
languages (Zeigler, 1990). In the DEVS approach, a translator is needed for each
formalism and one executor is needed for the execution of the overall low-level
model. This approach offers high potential, but has not been fully explored yet
as it requires substantial development effort (de Lara and Vangheluwe, 2002).
The above approaches are not mutually exclusive and we should be able to
develop EIMFs which support a mix of them. More specifically:
• If the overall model to be built is entirely situated within one particular domain
(e.g. physical system modelling), one should be able to adopt Approach 2 above
since this offers the highest degree of clarity, maintainability and optimisation of
performance.
• If legacy models are to be used, one should be able to revert to Approach 1 above
and interact with these models through standardised interfaces at run-time.
We have seen how an appropriate model representation technique can support
model re-use within and across frameworks. In the next subsection we focus on the
role of component-based software engineering techniques for model linking and
re-use.
7.4.1 Component-based modelling
By committing to modelling frameworks as the major paradigm for model devel-
opment and application, a system concept can be adopted that allows the proper
assembly of models based on scientific building blocks. Such building blocks are
well-defined, documented, tested and packaged. We refer to them as components,
and thus advocate abandoning monolithic model development efforts in favour of

112 A.E. Rizzoli et al.
building models as a series of smaller, reusable parts. But how does general compo-
nent technology translate into model development and EIMFs?
The use of components in EIMFs is not a common practice to date. Scientific
building blocks in modelling frameworks do not always comply with the concept
of a component for various reasons. Programming languages, overall application ar-
chitectures, and legacy code requirements might constrain the design of scientific
modules from a technical perspective. Such frameworks provide at least a traditional
application programming interface (API) or communication protocol that can be
used to implement modules. Such modules typically stay within the realm of a spe-
cific framework and therefore have limited re-use. Hence the adaptation of general
component standards and their customisation for modelling has the potential for
interoperability and acceptance.
Component technology focuses on the component as the primary reusable piece
that allows tools to create, explore, and consume such components in a standard-
ised way. Components are typically objects that have a predefined, documented
and reusable behaviour. A scientific model component is an independent software
unit that is developed for a specific scientific purpose and not for a specific model
application. They are self-contained in terms of technical dependencies on other
components but may rely on software outside of the framework.
As another feature, components are linkable building blocks. They can be con-
nected at execution time using dynamic loading and linking techniques.
An EIMF that takes advantage of component technology should contain the
following features:
• The EIMF allows assembly of a model from various components that can be
classified into scientific, utility, control, input/output, analysis, or other types of
components. They share a common structure but typically have different seman-
tics.
• The EIMF provides flexible options to represent space and time for environmen-
tal modelling but does not constrain the user to a specific spatial or temporal
discretisation concept. The EIMF might offer control components that allow for
easy iteration across time and space.
• The EIMF is able to explore the component structure, a concept known as in-
trospection, achieved using software reflection as provided by modern languages.
Components might have metadata attached that specify application requirements
to help domain experts with model building. Component interfaces specify data
constraints or domain application requirements such as temporal and spatial con-
straints. Metadata structure and values could refer to a shared ontology to make
linking more effective, and allow the development of tools which will rely on the
type of information to implement in the EIMF, i.e. specific functionalities such
as optional data quality control at run time.
• The EIMF can use components that are developed from other institutions and
research groups assuming the components follow a common standard. In order
to promote framework adaptation to different component and module APIs, the
use of a common component standard that can be used in different EIMFs should
be strongly promoted.

Integrated Modelling Frameworks for Environmental Assessment and Decision Support 113
• Since the EIMF manages the connectivity of components, it also acts as a model
linker. The model execution environment locates components dynamically, and
loads and connects them for execution within the model.
Yet even if an EIMF provides all of the above features, the crucial issue of the
dependency of the component on the framework has to be dealt with. A compo-
nent is a piece of software that is always dependent on the specific platform and
in most cases on the framework itself. For instance, a watershed model developed
using the TIME framework depends on a number of TIME-specific libraries that
make it difficult to re-use it ‘as is’ in another framework. A possible solution is to
design the component as independently as possible from the framework, but we al-
ways hit the “data type” barrier. Whenever we want to exchange a data type more
complex than an integer or a float, we need to rely on complex data types which
have to be defined in the framework.
A possible solution is available if we allow model components to share a com-
mon interface that can be created from a public ontology, including the selection of
quantities and attributes to develop data structures. Data structures can be saved as
domain objects in a knowledge base, and can be further extracted in RTF or XML
formats to generate domain object code for specific implementations (Athanasiadis
et al., 2006). This approach goes hand-in-hand with declarative modelling where
the solution to a modelling problem in various implementations is via code gen-
eration, targeting, and optimisation with respect to a specific EIMF. In this case,
the constraint for re-usability is given by the appropriateness of model use in a
specific context whereby possible technical issues (e.g. language, platform, EIMF
functionalities) can be overcome.
Regardless of the choice of developing framework-specific or intrinsically re-
usable components, there is a basic choice that must be carefully evaluated be-
forehand. This choice is related, in general terms, to the framework as a flexible
modelling environment for building complex models (i.e. model linking), but also
to the framework as an efficient engine for calibration and simulation of model
components (i.e. model execution).
7.4.2 Distributed modelling
Taking the model component concept further, we envision model component in-
terfaces published on the Internet and accessible for re-use, either to a specific
community or the public in general. In a “marketplace-like” open source envi-
ronment, models of fine granularity could be registered and their services made
available in a collaborative fashion. In a service-oriented architecture, an EIMF
could eventually become a composite service and end-users will be able to build
and develop from the existing services available. Environmental data will operate as a
virtual resource shared among peers, instead of a scarce resource for which peers strive
(Athanasiadis, 2007). Employing software agents, web services or grid technology
for realising a service-oriented approach, the members of a “virtual modelling mar-
ketplace” will be able to construct scientific workflows for combining original data
sources with environmental models and reporting tools, all available as services. Al-
though such a vision seems very promising, a realistic implementation is hindered

114 A.E. Rizzoli et al.
by the lack of standards for exchanging environmental data. The development and
the wide adoption of community standards is a prerequisite for achieving smooth
information flow within a virtual modelling marketplace. Developments similar
to ebXML (http://www.ebxml.org), which is a widely adopted standard in the
electronic business sector, are required for maximising the interoperability and re-
usability potential of a web-based, open environment for modelling.
7.5. Driving and Supporting the Modelling Process
Modelling is an iterative process (see Chapter 2) during which several activities
need to be accomplished including pre-modelling tasks such as problem description
and requirements analysis. Common problems encountered during modelling in-
clude inadequate project setup, insufficient or inappropriate use of methods, and
lack of documentation and transparency. In multidisciplinary projects, modellers
from different disciplines often do not always understand one another because of
different methodological approaches and developments. To improve the quality and
credibility of modelling results and model-based decision support in general, an
EIMF needs to support conceptualisations that provide guidance on the modelling
workflow. Such guidance can be effectively delivered only if we acknowledge the
clear distinction between a model and its experimental frame.
7.5.1 The experimental frame
One essential requirement for re-usability and maintainability in complex EDSSs is
the strict separation between the model and its experimental frame. By model we
mean the (mathematical or other) representation of the physical system under study.
The experimental frame is the environment in which the model resides. It is the
experimental frame’s responsibility to provide to the model the input it needs, and to
accept and further process model output that is generated. Many well-known legacy
“models” are actually codes in which the representation of the physical system and
its experimental frame are fully intertwined. Options for re-use are therefore often
limited in these cases.
In order to tackle the complexity of models, techniques such as hierarchical
decomposition and object-orientation are commonplace. These techniques are es-
pecially relevant since in many cases a set of basic, atomic models (e.g. unit processes)
can be identified from which other more complex models can be derived through
coupling or inheritance.
Unfortunately it is not always understood that the experimental frame can also
be regarded as a model in its own right, to which hierarchical decomposition and
object-orientation can equally be applied. One can indeed identify a set of basic
operations that apply to nearly all models, and from which other more complex
types of model processing can be derived. A non-exhaustive list of the basic types
of model processing (sometimes also referred to as “virtual experimentation,” Claeys
et al., 2006) include:

Integrated Modelling Frameworks for Environmental Assessment and Decision Support 115
• dynamic simulation and steady-state simulation;
• optimisation with regard to various objectives (e.g. parameter estimation) in con-
junction with confidence information analysis;
• local and global sensitivity analysis;
• scenario evaluation;
• risk analysis; and
• optimal experiment design.
Only a few EIMFs exist in which a full set of basic model processing operations
is available. Even fewer allow for compound operations to be constructed from
these in a flexible manner. However, in order to facilitate the process of building
complex EDSSs, there is a need for tools that support hierarchical and object-
oriented modelling of experimental frames – this is the role of the experiment
manager framelet shown in Figure 7.2.
Various authors have aimed their work at providing tools and methodologies
to support the modelling process. Jakeman et al. (2006) provide a list of ten steps
to support a disciplined model development approach. Castelletti and Soncini-Sessa
(2006) show how a participatory and integrated planning procedure can be formally
specified in a sound methodological approach, also supported by a software tool.
Giupponi (2007) shows how the DPSIR approach can be used in the context of a
DSS for implementing the EU Water Framework Directive (WFD).
Business process modelling standards in general provide conceptualisations that
describe the type of information required by workflow systems, but these standards
lack the support of formal semantics and ontological structures to represent mod-
elling methods, activities and other related information. The HarmoniQuA quality
assurance framework tries to address some of these issues by proving quality assur-
ance guidelines, a knowledge base and an associated modelling support tool (MoST)
(Refsgaard et al., 2005).
In conclusion, an EIMF should support features that allow process management
tools to be integrated with the framework so that the modelling process can be
properly managed, transparent and quality-controlled.
7.6. Conclusions
In this chapter we have described the modern concept of the EIMF based on
past experience, current endeavours and future plans. The existence and adoption of
today’s EIMFs give us hope for future improvements as we continue to research and
develop, apply and experiment, and think and test. EIMFs such as JAMS, TIME,
OMS, OpenMI, Tornado and ModCom show how far we have come and some
of the problems we still face. Many of these frameworks share component-based
approaches, use object-oriented techniques, conform to sets of design patterns and
fit, in varying ways, the template offered by the generic EIMF architecture in Fig-
ure 7.1. However, they also differ fundamentally in their technological basis, such
as use of NetBeans or .NET, their structure of classes and methods, data handling
approaches, operational sequencing, and even in the way that they appear to a user.

116 A.E. Rizzoli et al.
These differences arise not through any particular disagreements or differences
of opinions amongst developers, but through the origins and evolution of the
frameworks, the institutional context of development, and the pressing needs in
application. Almost all existing EIMFs have previous lives, and have been built
and rebuilt to meet specific (and increasingly generic) needs. They often repre-
sent institutional knowledge that must be retained in the next generation, lest such
knowledge be lost and the new EIMF end in failure. Given this evolution, it is wise
to consider carefully new directions and the promise of converging on the some-
what “heretical” vision espoused in this paper. At the same time, user needs must be
met so we cannot forget the fundamental importance of providing helpful, familiar
and intuitive user interfaces, support for legacy systems, and seamless access to cur-
rent and expanding data sources that may be distributed, disconnected, incomplete,
incompatible and inconsistent.
In this chapter, we suggest that a promising trend for knowledge representation
is the use of ontologies that have the capacity to elicit the meaning of knowl-
edge in a manner that is logical, consistent and understandable by computers and
the knowledge worker community. This new path in knowledge-based computing
will support retention of institutional knowledge, while putting modelling back in
the hands of modellers. Environmental modelling will then become a conceptual
activity, focusing on model design rather than model implementation, with code
generation being delegated to some degree to ontology-aware tools. In this respect,
we envision the whole model lifecycle to change drastically, becoming more of a
theoretical activity and less of a coding-intensive, highly engineering-oriented task.
Environmental science and technology is multifaceted and individual disciplines
are unequal in their reliance upon models and computer-science-aided approaches.
Hence we need to retain flexibility and encourage creativity in the knowledge-
based approaches we develop. There has never been a better time to reflect on some
of these past successes to contemplate the next paradigm leap in computer-aided
decision support.
REFERENCES
Athanasiadis, I.N., Rizzoli, A.E., Donatelli, M., Carlini, L., 2006. Enriching software model interfaces
using ontology-based tools. In: Voinov, A., Jakeman, A.J., Rizzoli, A.E. (Eds.), Proceedings of the
iEMSs Third Biennial Meeting: “Summit on Environmental Modelling and Software”. Interna-
tional Environmental Modelling and Software Society, Burlington, USA, July 2006. ISBN 1-4243-
0852-6. ISBN 978-1-4243-0852-1. CD ROM. Internet: http://www.iemss.org/iemss2006/
sessions/all.html.
Athanasiadis, I.N., 2007. Towards a virtual enterprise architecture for the environmental sector. In:
Protogeros, N. (Ed.), Agent and Web Service Technologies in Virtual Enterprises. Idea Group Inc.,
pp. 256–266.
Ascough II, J.C., Flanagan, D.C., David, O., Ahuja, L.R., 2005. Assessing the potential of the Object
Modeling System (OMS) for erosion prediction modeling. In: Proceedings of the 2005 ASAE
Annual International Meeting, ASAE Paper 052011. Tampa, Florida, July 17–20.
Castelletti, A., Soncini-Sessa, R., 2006. A procedural approach to strengthening integration and par-
ticipation in water resource planning. Environmental Modelling and Software 21 (10), 1455–1470.

Integrated Modelling Frameworks for Environmental Assessment and Decision Support 117
Claeys, F., De Pauw, D., Benedetti, L., Nopens, I., Vanrolleghem, P.A., 2006. Tornado: A versatile
efficient modelling and virtual experimentation kernel for water quality systems. In: Voinov, A.,
Jakeman, A.J., Rizzoli, A.E. (Eds.), Proceedings of the iEMSs Third Biennial Meeting: “Summit
on Environmental Modelling and Software”. International Environmental Modelling and Soft-
ware Society, Burlington, USA, July 2006. ISBN 1-4243-0852-6, ISBN 978-1-4243-0852-1.
CD ROM. Internet: http://www.iemss.org/iemss2006/sessions/all.html.
David, O., Markstrom, S.L., Rojas, K.W., Ahuja, L.R., Schneider, I.W., 2002. The Object Modeling
System. In: Ahuja, L., Ma, L., Howell, T.A. (Eds.), Agricultural System Models in Field Research
and Technology Transfer. Lewis Publishers, CRC Press LLC, pp. 317–331.
de Lara, J., Vangheluwe, H., 2002. AToM3: A tool for multi-formalism and meta-modelling. In: Euro-
pean Joint Conference on Theory And Practice of Software (ETAPS), Fundamental Approaches to
Software Engineering (FASE). In: Lecture Notes in Computer Science, vol. 2306. Springer-Verlag,
Grenoble, France, pp. 174–188.
Dolk, D.R., Kottemann, J.E., 1993. Model integration and a theory of models. Decision Support
Systems 9 (1), 51–63.
Geoffrion, A.M., 1987. An introduction to structured modelling. Management Science 33 (5), 547–
588.
Giupponi, C., 2007. Decision Support Systems for implementing the European Water Framework
Directive: The MULINO approach. Environmental Modelling and Software 22 (2), 248–258.
Gregersen, J.B., Gijsbers, P., Westen, S.J.P., 2007. OpenMI: Open modelling interface. Journal of
Hydroinformatics 9, 175–191.
Hilborn, R., Mangel, M., 1997. The Ecological Detective. Princeton University Press.
Hillyer, C., Bolte, J., van Evert, F., Lamaker, A., 2003. The ModCom modular simulation system.
European Journal of Agronomy 18 (3–4), 333–343.
Jakeman, A.J., Letcher, R.A., Norton, J.P., 2006. Ten iterative steps in development and evaluation of
environmental models. Environmental Modelling and Software 21 (5), 602–614.
Keegan, P., Champenois, L., Crawley, G., Hunt, C., Webster, C., 2005. NetBeans 5.0 IDE Field
Guide: Developing Desktop, Web, Enterprise, and Mobile Applications. Prentice Hall.
Kralisch, S., Krause, P., 2006. JAMS—A Framework for Natural Resource Model Development and
Application. In: Voinov, A., Jakeman, A.J., Rizzoli, A.E. (Eds.), Proceedings of the iEMSs Third
Biennial Meeting: “Summit on Environmental Modelling and Software”. International Environ-
mental Modelling and Software Society, Burlington, USA, July 2006. ISBN 1-4243-0852-6, ISBN
978-1-4243-0852-1. CD ROM. Internet: http://www.iemss.org/iemss2006/sessions/all.html.
Ludaescher, B., Gupta, A., Martone, E.M., 2001. Model-based mediation with domain maps. In: 17th
Intl. Conference on Data Engineering (ICDE). Heidelberg, Germany.
McAffer, J., Lemieux, J., 2005. Eclipse Rich Client Platform: Designing, Coding, and Packaging Java
Applications. Addison-Wesley Professional.
Muetzelfeldt, R.I., 2004. Declarative Modelling in Ecological and Environmental Research. Euro-
pean Commission Directorate-General for Research, Position Paper No. EUR 20918. European
Commission, Brussels, Belgium.
Pasetti, A., 2002. Software Frameworks and Embedded Control Systems. Lecture Notes in Computer
Science, vol. 2231. Springer-Verlag, Berlin.
Rahman, J.M., Seaton, S.P., Perraud, J.-M., Hotham, H., Verrelli, D.I., Coleman, J.R., 2003. It’s
TIME for a new environmental modelling framework. In: Post, D.A. (Ed.), MODSIM 2003 In-
ternational Congress on Modelling and Simulation. Modelling and Simulation Society of Australia
and New Zealand Inc., Townsville, pp. 1727–1732.
Rahman, J.M., Seaton, S.P., Cuddy, S.M., 2004. Making frameworks more useable: Using model
introspection and metadata to develop model processing tools. Environmental Modelling and Soft-
ware 19 (3), 275–284.

118 A.E. Rizzoli et al.
Refsgaard, J.C., Henriksen, H.J., Harrar, B., Scholten, H., Kassahun, A., 2005. Quality assurance
in model based water management—Review of existing practice and outline of new approaches.
Environmental Modelling and Software 20 (10), 1201–1215.
Uschold, M., Gruninger, M., 1996. Ontologies: Principles, methods and applications. The Knowl-
edge Engineering Review 11 (2), 93–136.
Villa, F., 2007. A semantic framework and software design to enable the transparent integration,
reorganization and discovery of natural systems knowledge. Journal of Intelligent Information Sys-
tems 29, 79–96.
Villa, F., Donatelli, M., Rizzoli, A., Krause, P., Kralisch, S., van Evert, F.K., 2006. Declarative mod-
elling for architecture independence and data/model integration: A case study. In: Voinov, A.,
Jakeman, A.J., Rizzoli, A.E. (Eds.), Proceedings of the iEMSs Third Biennial Meeting: “Summit
on Environmental Modelling and Software”. International Environmental Modelling and Soft-
ware Society, Burlington, USA, July 2006. ISBN 1-4243-0852-6, ISBN 978-1-4243-0852-1.
CD ROM. Internet: http://www.iemss.org/iemss2006/sessions/all.html.
Zeigler, B.P., 1990. Object Oriented Simulation with Hierarchical, Modular Models. Academic Press.