A Case Study in Application of Vee Model of
Systems Engineering to System Dynamics Modelling of
Dryland Salinity in Australia
Naeem U Khan and Alan C McLucas
School of Information Technology and Electrical Engineering
UNSW @ADFA, Northcott Drive
Campbell ACT 2600
e-mail: naeemk@ tpg.com.au, a.mclucas@ adfa.edu.au
ABSTRACT
This paper describes an application of the Vee Model of Systems Engineering in
developing a System Dynamics model of dryland salinity in the Murray Darling Basin. A
modular approach was adopted. Simple modules of salt affected land were developed
using Powersim Studio following decomposition, definition, integration and verification
processes. Individual modules were verified, integrated and provided with options for
policy testing. The use of Vee Model provides a structured way for developing computer
simulation model with a top-down approach for requirements elicitation and ensures
that the computer model meets requirements and limitations elicited through qualitative
System Dynamics and defined at the start of the modelling exercise. It also provides
rigorous verification processes. The results of this research provide an avenue for
further exploration of the synergistic use of the two approaches to improve model
quality.
Key Words: System Dynamics, Systems Engineering, Vee Model, Dryland Salinity,
Australia.
1 INTRODUCTION
The purpose of a System Dynamics inquiry is to improve our understanding in and about
complex systems. The goal of modelling is to improve understanding of the relationships
between feedback structure and dynamic behaviour of a system, so that policies for
improving the problematic behaviour may be developed (Richardson 1981). This purpose is
achieved through development of qualitative as well as computer simulation models.
Development of computer simulation model poses a special challenge to a modeller as the
qualitative models like causal loop diagrams, concept maps, systems diagrams carry a rich
picture of the problem to be modelled. The computer simulation, however, may not include
all those variables due to what Coyle (1999; 2000) calls limits to quantification.
Version 1.1 1
Richardson (2001) and Homer and Oliva (2001) suggest that simulation nearly always adds
value to policy analysis, even in the face of significant uncertainties and soft variables.
Sterman (2000) highlights the need for quantitative models as our mental models are
dynamically deficient, that is, they omit feedbacks and time delays, accumulations and non-
linearity with the consequence that simulation is the only practical way of testing our
mental models, noting that the complexity of our mental models vastly exceeds our
capacity to understand their implications.
Systems Engineering has a strong tradition in complex project management and provides an
opportunity for development of simulation models through a structured process. McLucas
and Ryan (2005) highlight the strengths of Systems Engineering in design, building and
testing quantitative System Dynamics models including the detailed transition from
conceptual representation to quantified model. They examined the System Dynamics
modelling process and suggested the following benefits by using the Systems Engineering
process:
e deliberate and careful management of the complexity introduced at each stage of the
model building process;
e discipline and rigour associated with requirements engineering;
e aid in managing and coping with complexity through a top down approach; and
e rigour in validation and verification.
This paper describes founding research on application of Vee Model of Systems
Engineering to development of a simple System Dynamics model of dryland salinity in the
Murray Darling Basin, Australia. First, Vee Model is described that is followed by the
description of the activities undertaken at each step of model development and resultant
models.
2 VEE MODEL OF SYSTEMS ENGINEERING
Sage (2005) defined Systems Engineering as a process that is comprised of a number of
activities that will assist in definition of requirements for a system, transform this set of
requirements into a system through development efforts and provide for deployment of the
system in an operational environment.
System Engineering uses a distinct system development process. This process coincides
with a systems lifecycle and lists key steps in the system development from concept to
development, integration, testing and deployment of the system. This process is usually
represented in a diagram that is in the shape of English letter “V” and is also called as a
“Vee Model”. A traditional Vee is based on project cycle and represents a progressive
product development process.
Version 1.1 2
Forsberg and Mooz et al. (2005) presented a Vee Model addressing architecture
decomposition, integration and verification. The model presents both system development
planning and system realization activities. Forsberg and Mooz et al. (2005) also presented
detailed V ee models for decomposition, integration and verification. Figure 1 shows a high
level Vee Model that presents decomposition and definition on the first leg and integration
and verification on the second leg of Vee Model. Thickness of the Vee increases near
bottom of the V ee and indicates the increasing number of activities at that level.
Solution /
System
Realization
System
Development Integration, Verification & Validation Planning
Subsystem
Development | Integration, Verification
& Validation Planning
1, V &V planning
Subsystem
realization
Decompositio} .
and definition Integration and
LCI- Lowest LCI - Lowest Verification
configuration configuration
item item
development realization
T T
Figure 1 Vee Model of Systems Engineering re-drawn from Forsberg, Mooz et
al. (2005)
Lead authors in Systems Engineering have used multiple variants of the Vee Model.
Despite variations, Vee Models present similar activities though with different terminology
and the use of various levels of decomposition and integration. The first leg of the Vee
Mode shows the activities involved in planning for the development of a system, e.g.,
concept, requirements or expectations form the system and design. A system is first
Version 1.1 3
decomposed to identify requirements. Components are designed and then integrated to
progressively prove performance and compatibility of all components of the system.
In Systems Engineering process, verification is used to ensure a product or system is built
according to its specifications while validation is the user satisfaction regardless of
specifications. Forsberg and Mooz et al. (2005)’s Vee Model shows that integration,
verification and validation activities are planned at the system decomposition level when
the subsystems, modules and lowest configuration items (LCIs) are developed.
Neudorff, Randall et al. (2003) presented a different Vee Model with multiple levels in
decomposition and integration, for example, they further divided requirements and design
into high level and detailed requirements and high level and detailed design. The first leg
his Vee Model addresses concept of operation, requirements, design, implementation,
integration and testing, subsystem verification, system acceptance and operation and
maintenance.
For this paper, a general approach to Vee Model presented by Forsberg and Mooz et al.
(2005) is adopted. Key activities of the System Dynamics modelling process are grouped
and represented according to this Vee model.
For this study, a System Dynamics model for the dryland salinity was considered as a
system to be developed. The knowledge gained through qualitative modelling process, e.g.,
learning cycles, reference modes and concept mapping constitutes the baseline
requirements for model development and informs the decomposition and definition for
identification of modules. Table 1 shows the Systems Dynamics modelling activities used
for this study against Vee Model.
Version 1.1 4
Tablel: System Dynamics Modelling activities organized acoording to Vee Model of Systems Engineering.
‘Vee Model Components System Dynamics Modelling activities
Architecture - System level decomposition- Qualitative System Dynamics:
Decomposition and - Leaming cycles
Definition Concept of operation. - Reference modes
Requirements Elicitation -Concept maps
-Systems arch-types analysis
- Causal loop diagrams
Subsystem level decomposition. - Identification of stocks and flows.
Development of specifications. Identification of the individual modules needed to represent the problem.
- Stock and Flow Diagrams.
Planning for integration,
verification and validation of
subsystems.
LCI- Lowest configurationitems | Identification of Auxiliaries, constants.
development:
Architecture - integration, | System realization. Integration of individual modules into a whole model including interfaces,
verification and validation policy levers, input controls, output objects, data transfer facilities etc.
Subsystem realization. - Built-up of individual modules with stocks, flows, auxiliaries, constants
using a system dynamics modelling software, e.g., Powersim Studio
- Development of interfaces and model input controls, e.g, slider bars,
- Development of the output objects, e.g., graphs, tables, gauges,
- Development of individual modules
LCI Solution system realization.
Development of individual rate models with auxiliaries and constants.
Compliance to baseline - Examination of each equation to verify that it represents real world.
verification & validation. counterparts and follows the model logic.
- Model validation tests.
Version 1.1 5
3 DRYLAND SALINITY IN AUSTRALIA: A SNAPSHOT OF THE
PROBLEM
A detailed description of the dryland salinity was presented in System Dynamics Society
Conference 2006 (Khan, McLucas et al. 2004; Khan and McLucas 2006). As the focus of
this paper is on demonstrating Systems Engineering approach, therefore, detailed
description of dryland salinity is not given to accommodate the detailed description of the
ways in which Systems Engineering was applied. Here brief conclusions about the problem
from earlier papers (Khan, McLucas et al. 2004; Khan and McLucas 2006) are reproduced.
Australia is facing a serious environmental problem in the form of salinity. Approximately
5.7 million hectares are reported to be either affected by or at risk of dryland salinity. The
problem has developed over a long time as a result of the feedback interactions among
various climatic, geographic, environmental and human factors.
The landscape in the Murray Darling Basin has changed overtime. The major human
induced impacts had been settlement and land clearing for agricultural urban and industrial
uses. Agriculture is one of the major sectors for land use change (Crabb 1997). A major
expansion in agricultural development during 1950s to 1980 was due to extensive clearing
and increase in cultivated area.
Land clearing started in the Murray Darling Basin many years ago and it is still continuing.
The term land clearing refers to removal of the natural cover (e.g. forest) from the land for
altemative uses. The current motivators for land clearing include land availability, clearing
controls, environmental and social influences, financial and Institutional incentives,
agricultural research and development, and market forces (AGO 2000). One of the causes
of land clearing was conditional purchases. For example from 1860's to 1960’s leases and
conditional purchases were issued on the proviso that a certain percentage of tree cover was
to be removed each year (BRS 2000). Graetz, Wilson et al. (1995) assessed that, at national
level, 1,029,640 sq km have been thinned and cleared within intensive landuse zones and
most of this is in the Murray Darling Basin.
4 DECOMPOSITION AND DEFINITION (THE FIRST LEG OF
VEE MODEL)
Decomposition refers to further break down of a perceived system into manageable
components. During decomposition and definition, a system is partitioned into hardware,
software components and operator activities and each component is then designed, built
and coded according to its functional and physical content (Forsberg, Mooz et al.
(2005:110). Requirements of a system (system dynamics computer simulation model in this
Version 1.1 6
case) are elicited through multiple methods (mentioned in the column three of the Table 1),
analyzed and subsequently converted into specifications of the system to be developed.
4.1 REQUIREMENTS ELICITATION
A qualitative analysis of the problem was undertaken and reference modes of the problem,
concept maps and causal loop diagrams were prepared. The Reference modes and causal
loop diagrams were presented at the System Dynamics Society Conferences in 2004 and
2006 (Khan, McLucas et al. 2004; Khan and McLucas 2006) respectively. Detail about
qualitative analysis is not presented here due to limitations on scope of the paper. This
qualitative analysis provided requirements for the model in terms of model purpose,
performance expectations and module specifications that are describes in the following
paragraphs.
4.2 MODEL PURPOSE, SCOPE AND INTENDED USES
The main purpose was to develop a simple model that helps to understand the impacts of
land clearing and time delays on different land categories undergoing dryland salinity in
Australia. The model does not intend to predict the quantity of actual salt affected lands or
the quantity of salt at a certain geographical location. The model should be concise and
simple enough to be used for communication purposes and it should provide a user
interface to allow users to change the inputs. The model should exhibit past behaviour close
to the one identified in the reference modes. As the reference modes were prepared using
descriptive data, the model output is not expected to provide statistical correlation.
The model provides an opportunity for understanding causal mechanisms underlying the
perceived system, and it should not be used as a framework for statistically based
inferences.
The qualitative analysis in the form of causal-loop diagrams, concept maps and the
reference modes presented the qualitative analysis of the dryland salinity problem. The
qualitative analysis presented a large number of variables that are important in
understanding this problem. The simulation model takes few key variables related to the
land cover as related to the dryland salinity.
The model must conform to the following expectation:
e The model should address land clearing issue as related to dryland salinity and help
develop a strategic view of the problem.
Version 1.1 7
Model should help in understanding the impacts of time delays on different land
progressing through different stages.
The model should aid learning about the impacts of land component of the dryland
salinity problem.
The model should provide policy levers for experimentation.
The model is to be used as a research tool to investigate synergies between System
Dynamics and Systems Engineering for model development.
The modelling should be started simple. However, the model should have flexibility
to allow additional detail/modules to be added, as needed to aid analysis and
subsequent learning.
Model should use the important variables that can directly influence dryland
salinity.
The model should address the timeframe and model boundary requirements elicited
through qualitative System Dynamics.
4.3 MODEL SPECIFICATIONS
The simulation model consists of land cover sector. However, the model has flexibility for
other sectors for example population and socio-economic sectors to be added in future
research. The model provides a framework for learning about the impacts of time delays,
different land clearing rates, and application area and effectiveness of control treatments
over dryland salinity in the Murray Darling Basin.
A simple model depicts these interactions by three stocks:
e
Land under natural vegetation.
Cleared land neither salt affected nor at risk of becoming salt affected.
Cleared land either salt affected or at the risk of becoming salt affected: A piece of
land is considered at risk of becoming salt affected if it has the watertable within 2
meters beneath the surface consistent with the NLWRA (2001).
The initial values of the stocks are user controlled and provide a room for experimentation.
These stocks are linked by four flows:
e
e
e
Rate of land clearing.
Rate of land becoming salt affected.
Rate of land reclamation.
Rate of land either salt affected or at the risk of becoming salt affected returning to
natural vegetation.
Version 1.1 8
5 MODEL REALIZATION: DEVELOPMENT, INTEGRATION
AND VERIFICATION (THE SECOND LEG OF VEE MODEL)
5.1 DEVELOPMENT OF INDIVIDUAL MODULES.
The individual module structure was developed using a generic module specified by
McLucas (2003) and defined by its boundary and functionality. According to McLucas
(2003) the functionality of a module means the operations it performs on the inputs, e.g.,
accumulating, draining, etc. The outputs from a module are either lost across a boundary or
made available to another module.
The generic model consists of one stock and two flows. One flow is into the stock and
accumulates stock and the other is out of the stock and drains it. Flows and their
determinants are within a module boundary. Across the boundary, there are physical and
information flows as well as datasets that provide it connectivity to other modules and its
environment. The generic module is shown in the Figure 2.
Physical Outflow
Definition to include:
flow type discrete or YW
continous; flow N,
direcion;max flow ‘
rate; dt;simulation
time step;simulation
Physical Inflow Export to Dataset
Definition to include: Definition to include:® “\
flow type discrete or write format; write
continous; flow direction;conversion
/° direcion;max flow factors;units of
; rate; dt;simulation measurement. \
Hy time step;simulation { time horzon;units;
H time horizon; units; | idimentions: Hi
\ — dimentions /
Constant_2 /
Rate_1 Level Rate_2
Constant_1
Import from Dataset 4 1
Definition to include: i
read format;read
direction; conversion
actors; units of
measurement,
Information Outflow
Definition to include:
sampling rate;dt;
simulation timestep;
simulation time
horizon; calendar.
Information Inflow
Definition to include:
sampling rate;dt;
simulation timestep;
simulation time
horizon; calendar.
Figure 2 Structure of a generic module. Redrawn from McLucas (2005:179)
Version 1.1
5.2 STOCKS AND FLOWS
The model presents a simple three land stocks, i.e., land under natural vegetation, cleared
land neither salt affected nor at the risk of becoming salt affected, and land either salt
affected or at the risk of becoming salt affected. These stocks are linked by four flows, ie,
‘rate of land clearing’, rate of land becoming salt affected’, ‘rate of land reclamation and
the rate of land at risk of becoming salt affected that is returning to a natural vegetation
cover. For developing modules, a modular approach was adopted that encourages starting
simple and then adding details as necessary.
5.2.1 Land Stock 1: Land under natural vegetation.
The stock ‘Land under natural vegetation’ represents land either bush or forest that has not
been cleared for agricultural purposes under land clearing operations. This stock is
represented in the model diagram by a rectangle. Land clearing rate drains it while rate of
land becoming salt affected returning to natural vegetation adds to this stock. This simple
formulation is represented in the Figure 3. A negative feedback loop manages the level
stock. As the rate increases of land clearing increases, it decreases the stock of land under
natural vegetation.
LAND UNDER NATURAL VEGETATION
a 0
Rate of land at risk Rate offLand
of becoming salt ? ’ Clegting
affected returning im sv
to natural Ss--e
vegetation
Initial land under
natural vegetation
Figure 3 ‘Sub-model land under natural vegetation
5.2.2 Stock 2: Cleared land neither salt affected nor at risk of becoming salt affected.
Cleared land represents a piece of land that was previously under natural vegetation either
bush or forest and is cleared for the purposes of bringing it under agricultural production.
Two rates ‘Rate of land clearing’ and cleared land neither salt-affected nor at risk of
becoming salt-affected’ compound this stock while the rate ‘Rate of land becoming salt
affected drains it. There is a negative feedback loop that manages this stock. The rate
reduces the stock (shown in dashed lines) and low stock causes a reduced rate of land
Version 1.1 10
becoming salt affected’ as rate equation is formulated as a fraction of the stock. Rate
formulations are discussed in the following sections. The inflows and outflows of this stock
are presented in the 4.
Rate of land
reclamation
CLEARED LAND NEITHER SALT AFFECTED NOR AT RISK OF
BECOMING
Rate of Land
Clearing
Initial Cleared land
not at risk of
becoming salt
affected
Figure 4 Sub-model cleared land neither salt affected nor at the risk of becoming
salt affected.
5.2.3 Land stock 3: Cleared land either salt affected or at the risk of becoming salt
affected.
As it is clear from the stock name, this stock represents pieces of land that is already salt
affected or is at the risk of becoming salt affected, i.e., watertable is at or within 2 meters
from the ground surface. This stock has one inflow that compound the stock and two
outflow that drain this stock. A simple representation of the stock is presented in the
Figure 5. The inflow is ‘Rate of land becoming salt affected’.
There are two outflows, ie., ‘Rate of land at risk of becoming salt affected returning to
natural vegetation’ and the rate of land reclamation. There are two negative feedback loops
that manage this stock (shown with broken lines in the Figure 5).
Version 1.1 11
Rate of land
reclamation
+
CLEARED LAND EITHER SALT AFFECTHD OR AT RISK DF BECOMING SALT
AFFECFED
Rate Land becoming
Rate of dnd at risk
Salt Affected
of becgming salt
affected returning
---7 ‘to natural
vegetation
Initial land at risk of
becoming salt affected
Figure 5 Sub-model land either salt affected or at the risk of becoming salt affected.
5.2.4 Rate of Land Clearing
A sub-model providing land clearing rate is shown in the Figure 6. For this sub-model, rate
of and clearing is defined as a fraction of the land under natural vegetation.
Rate of land clearing = Land under natural vegetation * Fraction of land under natural
vegetation./Time delay in land clearing
Rate of Land
Clearing
Fraction of land
under natural
vegation being
Cleared
LAND UNDER
NATURAL VEGETATION
Random Time Delay
in Land Clearing
Time Delay in Land
Randoms: Clearing
Figure 6 Rate of land clearing
Version 1.1 12
Fraction of land under natural vegetation being cleared is modelled on the basis of the
historical data of land clearing developed from different references. The fraction of land
under natural vegetation that is being cleared is considered to be varying overtime. The
input data is given through a graph. The maximum rate is considered between 30-35%
during the middle of the last century. Under the current environmental pressures and data
provided by the Australia Greenhouse Office (AGO 2000), it was considered that during
later part of the last century, land clearing rates were started to decline.
The time delay in land clearing is a user defined variable and includes the time that is
consumed in planning, land acquisition, getting permissions for land clearing, arrangements
for the machinery, acquisition and movement of machinery and felling and export of logs
from the area. As there may be varying time for different areas, land clearing operations,
communities. To check sensitivities, a random variable is used.
Time delay in land clearing provides for the time spent in planning for land clearing,
getting approvals/permissions, accessibility to the area and finally clearing the land of its
natural vegetation either bush or forest. Time delay is a user controlled parameter. Default
value is 10 years. As the actual time delay will vary over simulation period, therefore, a
random number has been used that fluctuates between 6 and 10 years.
Random 1 = Random (0.5, 1.0, 0.9)
5.2.5 Rate of land becoming salt-affected
Rate of land becoming salt affected is depicted in the Figure 7 and is defied as:
Rate of land becoming salt affected = Cleared land neither salt affected nor at the risk of
becoming salt affected* Fraction of land not at the risk of becoming salt affected
Rate Land becoming
Salt Affected
CLEARED LAND
NEITHER SALT a1
AFFECTED NOR AT.
RISK OF BECOMING
SALT AFFECTED
Time delay in land Fraction of land not
becoming salt at risk of becoming
affected salt affected that
becomes at risk
Random 2
Figure 7 Rate of land becoming salt affected
Version 1.1 13
Fraction of cleared land that becomes salt affected is provided by historical evidence
developed in the process of developing reference modes. The fraction near negligible at the
start of the simulation, reaches a peak around 0.3 around 2000 and reduces to 0.1 near the
end of simulation period. As accurate estimates of the rates could not be obtained, a random
number given below fluctuates the fraction around the points in the table function.
Random 2 = Random (0.5,1.0, 0.9)
Time delay in a land becoming salt affected or at the risk of becoming salt affected is not
actually known. It would vary according land policies, specific geo-physical and social set-
up and market forces. Time delay in land becoming salt affected is a user controlled
parameter. The default value is rough estimate of 30 to 40 years. A random variable
(Random (0, 1, 0.5)) fluctuates this time delay between 0 and 40 years over the simulation
period.
5.2.6 Rate of land reclamation
In this model, the rate of reclamation has been defined as function of the fraction of land
either salt affected or at the risk of becoming salt affected on which a control treatment is
applied, time delay and effectiveness of the control treatments. The model that provides the
rate of land becoming salt affected is shown in the Figure 8.
Rate of land
reclamation
Effectiveness of
Control treatments
& =
CLEARED LAND
EITHER SALT
AFFECTED OR AT RISK
OF BECOMING SALT
AFFECTED
Fraction of land at
risk of becoming
Time delay between salt affected on
land at risk of which control
becoming salt treatment is applied
affected and land
not at risk of
becoming salt
affected under a
certain control
treatment
Figure 8 Rate of land reclamation
Version 1.1 14
For simplicity, a single category of salt affected land or land at risk of becoming salt
affected. However it is acknowledged that the process of land becoming salt affected is
gradual. The actual statistics about hectares of salt affected land in the Murray Darling
Basin is not available. The National Land and Water Resources Audit (NLWRA 2001) used
a parameter ‘salt affected land or land at risk of becoming salt affected that means a land
that as a water table within 2 meters of the ground surface. As stated before, this model
uses the parameter that is consistent with the one used by the National Land and Water
Resources Audit (NLWRA 2001).
Fraction of land either salt affected or at risk of becoming salt affected is a user controlled
parameter. The default value is 0.1 that means a land control treatment is applied at 10% of
the salt affected or at risk of becoming salt affected land.
Effectiveness of a control treatment is a treatment specific parameter. It is also a user
controlled parameter. The default value is 0.5 (50% effective). A 50 % effective control
treatment means that if applied it can bring about the half of the impact of a hypothetical
fully effective control.
Time delay in a piece of land going out of risk of becoming salt affected is also user
controlled. The default value is kept at 30 years.
Rate of land reclamation=cleared land either salt affected or at the risk of becoming salt
affected* friction of land at risk of becoming salt affected on which control treatment is
applied* effectiveness of control treatments./time delay between land at risk of becoming
salt affected and land not at risk of becoming salt affected under a certain control treatment.
5.2.7 Rate of salt affected land returning to natural vegetation cover.
The model accommodates another pathway for the salt affected land, i e., the land is left out
of agricultural operations. No further land reclamation control treatment is applied.
Overtime, the unattended land starts to return to the bush/natural vegetation cover. The
formulation that provides this rate is graphically shown as in the Figure 9
Version 1.1 15
Rate of land at risk
of becoming salt
affected returning
to natural
vegetation
Time delay between
land at risk of
becoming salt
affected and
returning to natural
vegetation
CLEARED LAND
EITHER SALT ie
AFFECTED OR AT RISK
OF BECOMING SALT ©
AFFECTED
Random Fraction of
land at risk of
becoming salt
affected that is
returning to natural
vegetation
O Fraction of land at
Random 4 risk of becoming
salt affected that is
returning to natural
vegetation
Figure 9 Rate of land return to natural vegetation cover
The rate of land returning to natural vegetation is defined as:
Rate of land returning to natural vegetation cover= Cleared land either salt affected or at the
risk of becoming salt affected* Fraction of land at risk of becoming/time delay between land
either salt affected or at the risk of becoming salt affected and returning to natural
vegetation.
Both the fraction and the time delay are user controlled parameters. The default value is
0.5, ie., 5%. The actual fraction may vary over the period of simulation. A random number
generator (random(0, 1, 0.5)) fluctuates this fraction.
Time delay may vary based on a number of factors, e.g., location of a piece of land, type of
vegetation and other geo-physical conditions. It is a user controlled parameter. The default
value, i.e., the maximum time a piece of land takes in returning to its natural vegetation
cover is kept at 50 years.
5.2.8 Delays
All time delays in the model are considered as material delays. Although the symptoms of
land becoming salt affected, e.g., reduction in crop yields, salt crust, surface appearance
and change in vegetation cover may appear after a certain time, the processes involved in a
piece of land becoming salt affected start in an early phase. The simple formulation is
shown below:
Time Delay = material/time taken
Version 1.1 16
5.3 BUILDING CONFIDENCE IN MODEL: INTEGRATION,
VERIFICATION AND VALIDATION (THE SECOND LEG OF VEE
MODEL)
5.3.1 Integration
Forsberg, Mooz et al. (2005) defined integration as the successive combining and testing of
system hardware assemblies, software components, and operator tasks to progressively pro
eth performance and capability of all entities of the system. In terms of a System Dynamics
model, it would mean the integration of modules to each other. Modules were
incrementally integrated one by one with each other. A top down integration approach was
adopted. Model behaviour was checked after each step in integration.
5.3.2 Verification and Validation
In Systems Engineering, confidence is progressively gained in a model or system through
verification and validation. Verification ensures the model is built rightly, i.e, it does not
has internal inconsistency or flaws in its equation formulations and is built conforming to
the baseline/requirements ether identified at the start of modelling or evolved during the
process. In development of a System Dynamics model, verification would mean that the
governing business rules have been correctly identified and coded and the structure in
which those rules operate results in correct replication of the reference modes of behaviour
identified in earlier stages (McLucas 2005).
A mix of System Dynamics and Systems Engineering model verification and validation
approaches were applied. Authors consider that a complete description of the verification
and validation undertaken for this model is beyond the scope and available length of this
paper. The issue of model verification and validation will be comprehensively addressed in
the next paper in this series of research papers. A brief overview of the flow sequence,
dimensional consistency and the mass balance test is below.
A flow sequence test was conducted to ensure the integrity of stocks and that the outflows
do not precede inflows. Dimensional consistency was assured by:
a) using software that does not allow to simulate if there is dimensional inconsistency.
Powersim Studio has been used to confirm dimensional consistency.
b) each equation was individually analysed to ensure dimensional consistency
c) a simple model was built without using complex technicalities including multi-
dimensional arrays.
A mass balance test ensures that the functions performed by algebraic operators do not
result in the inadvertent creation or destruction of flows (McLucas 2005). A mass balance
test was performed to ensure that the algebraic operators do not result in inadvertent
Version 1.1 17
creation or destruction of flows. The method described by McLucas (2005) was used. A
new variable called Delta was created for each module. Delta represented sum of all flows
into a stock. The structure of the Delta check model is shown in the Figure 10. The delta
check indicated zero mass balance error.
) Count cleared la
Rate of Land
Clearing
__|
|
LAND UNDER
NATURAL VEGETATION
955,322.10 sq km
Delta check - land under natural TAA
vegetation module
Wa
Initial land under
w Count land returied to natural
natural vegetation
Rate of land at risk vegetatlgn
‘of becoming salt 5,307,345.00 ha
—
affected returning
=
CLEARED LAND
to natural
vegetation
NEITHER SALT
AFFECTED NOR AT
&) ( y RISK OF BECOMING
ci
SALT AFFECTED
wm, ‘ount reclaimed land
Rate of land
reclamation 5,307,345.00 ha
TAT
Wa
Initial Cleared land
not at risk of
becoming salt
affected
0.00 ha
o) Count land at risk of becoming
Rate Land becoming salt affected
Salt Affected
Delta check - land either salt
affected or at risk of becoming salt
affected module
TAG
L : a
Initial land at risk of
becoming salt
affected
-_
CLEARED LAND
EITHER SALT
AFFECTED OR AT RISK
OF BECOMING SALT
AFFECTED
Figure 10 Delta Check: Mass Balance Test of Model Structure and Behaviour
Version 1.1 18
6 SUMMARY AND CONCLUSION
This paper demonstrated the process in which Systems Engineering Vee Model was applied
for developing a System Dynamics-based dryland salinity model. Basic requirements for
the model were identified and listed at the start of the model building process. Each module
was described with its stocks, flows and auxiliaries. Module verification tests like
dimensional consistency, flow sequence and mass balance test were described.
System Dynamics model is a simpler system than the broad category of engineering
systems for which Systems Engineering process is applied. However, there are certain
model building aspects, like requirements identification, subsystem planning and model
verification, in which the application of Vee Model can improve the quality of a System
Dynamics model. It can reduce some of the inconsistencies in the model development
process and can help in development of robust and ‘responsive to purpose’ models.
The use of the Vee Model can also provide a mechanism for model validation as it helps to
emphasise the importance of planing and evaluation in almost all stages of model
development. This application highlights the synergies between system engineering and
System Dynamics and provides an avenue for further exploration of such synergies
between the two methods to improve model quality.
7 REFERENCES
AGO (2000). Land Clearing: A social History, The Australian Greehouse Office.
BRS (2000). Landcover Change in Australia. Canberra, Bureau of Resource Science.
Coyle, R. G. (1999). Qualitative Modelling in System Dynamics or What are the Wise
Limits of Quantification? The Proceedings of the 17th Intemational Conference of the
System Dynamics Society and 5th Annual Australian & New Zealand Systems Conference,
Wellington, New Zealand, System Dynamics Society.
Coyle, R. G. (2000). "Qualitative and Quantitative Modelling in System Dynamics: Some
Research Questions." System Dynamics Review 16(3): 225-144.
Crabb, P. (1997). Murray Darling Basin Resources. Canberra, Murray Darling Basin
Commission.
Forrester, J. W. (1994). "System Dynamics, Systems Thinking, and Soft OR." System
Dynamics Review 10(2/3): 245-256.
Version 1.1 19
Forsberg, K., H. Mooz, et al. (2005). Visualizing project management : models and
frameworks for mastering complex systems. Hoboken, Wiley.
Graetz, R. D., M. A. Wilson, et al. (1995). Landcover disturbance over the Australian
continent : a contemporary assessment. Canberra, ACT, Dept. of the Environment, Sport
and Territories,.
Khan, N. U., A. C. McLucas, et al. (2004). Development of a Reference Mode for
Characterization of the Salinity Problem in the Murray Darling Basin. 22nd International
System Dynamics Society Conference, Oxford, England, UK.
Khan, N. U. and A. C. McLucas (2006). Development of A Strategic System Dynamics
Model of Dryland Salinity. International System Dynamics Society Conference, Nijmegen,
The Netherlands, System Dynamics Society.
McLucas, A. C. (2005). System Dynamics Applications: A Modular A pproach to
Modelling Complex World Behaviour. Canberra, Australia, Argos Press.
McLucas, A. and M. Ryan (2005). Meeting Critical Real-World Challenges in Modelling
Complexity: What System Dynamics Modelling Might Learn From Systems Enginee. The
23rd International Conference of the System Dynamics Society, Boston, USA.
McLucas, A. C. (2003). Decision making : risk management, systems thinking and
situation awareness. Canberra, Argos Press.
Neudorff, L. J., J. E. Randall, et al. (2003). Freeway Management and Operations
Handbook. Washington DC, Office of transportation management, Federal Highway
Administration.
NLWRA (2001). Australian Dryland Salinity Assessment 2000. Canberra, ACT, Australia,
National Land and Water Resources A udit.
Homer J and R Oliva (2001) Maps and Models in System Dynamics: A Response to Coyle
System Dynamics Review, 17, 347-355.
Richardson, G. P. (1981). Feedback thoughts in Social Science and Systems Theory.
Portland, OR 97213-0390, Productivity Press.
Richardson, G. P. (2001). Mapping versus Modeling : The Answer to the Debate. The 19th
Intemational Conference of the System Dynamics Society, Atlanta, Georgia, System
Dynamics Society.
Sage, A. P. (2005). Systems of Systems: Architecture Based Systems Design and
Integration, Department of Systems Engineering and Operations Research, George Mason
University, Fairfax, VA 22030-4444.
Version 1.1 20
