Chaos and Complexity Theory For Management Nonline... - (Section 4)

Section 4
Nonlinearity in Water and Land

Cover Management
Copyright © 2012. IGI Global. All rights reserved.
Chaos and Complexity Theory for Management : Nonlinear Dynamics, edited by Santo Banerjee, IGI Global, 2012. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/acu/detail.action?docID=3312270.
Created from acu on 2018-11-08 20:41:47.
270
Chapter 13
Managing Fisheries in Light of
Complexity and Chaos Theories
Manuel Alberto Martins Ferreira José António Candeias Bonito Filipe
Instituto Universitário de Lisboa (ISCTE-IUL), Instituto Universitário de Lisboa (ISCTE-IUL),
BRU-UNIDE, Portugal BRU-UNIDE, Portugal
Manuel Francisco Pacheco Coelho Maria Isabel Pedro

SOCIUS/ISEG-UTL, Portugal CEGIST/IST, Portugal
ABSTRACT
Chaos theory - and models related to non-linear dynamic systems - has increased in importance in recent
decades. In fact, chaos is one of the concepts that has most rapidly expanded in research topics. Chaos
is ordinarily disorder or confusion; scientifically, it represents a disarray connection, but basically, it
involves much more than that. Change and time are closely linked, and they are essential when consid-
ered together as chaos theory foundations are intended to be understood. Given the large number of
applications in several areas, the goal of this work is to present chaos theory - and dynamical systems
such as the theories of complexity - in terms of the interpretation of ecological phenomena. The theory
of chaos applied in the context of ecological systems, especially in the context of fisheries, has allowed
the recognition of the relevance of this kind of theories to explain fishing events. It raised new advances
in the study of marine systems, contributing to the preservation of fish stocks.
INTRODUCTION rule was a life with the strictly necessary for the
populations’ day-to-day.
Mankind’s oldest concern inhabits necessarily on The development is an exception in the history
its survival. For thousands of years the Humanity of the Humanity that exists just for a period not
always lived with an almost steady level in the longer than 250 years. The Humanity had never
way of living. Production followed the growth had a so significant, consistent and intense trans-
of the population. To the generality of the times formation in its History. In this period, a complete
and to the generality of the people, the historical revolution in the life style of the world-wide
DOI: 10.4018/978-1-4666-2509-9.ch013
Copyright © 2013, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Managing Fisheries in Light of Complexity and Chaos Theories
population happened to the countries currently It is on the basis of the existence of this type
appointed as developed countries. However, its of position divergences that we intend to analyze
effects were not bordered to these zones. Its pres- the framings and this idea relative to the Earth
ence has been extensive to the whole world. In resources. Our study intends to conjugate the need
particular, with what it is assigned as external aid, of resources preservation - specially the common
the rich countries changed the stages of develop- resources - with the perception of the way the
ment of a great part of the globe, shortening the involved agents face the subjects studied.
development stages of many countries and pro- We intend to answer to the necessity of find-
voking several kinds of unbalances and disturbs ing solutions that aim the preservation and the
- see, for example, what happened with vaccines improvement of the conditions of life in society
on the field of health, with the pollution, with all compatible with the environment.
the problems emerged on the feeding area and The existing risks to the Humanity and to all
with other many forms of intrusion, with visible the live resources require that some theoretical
effects and heavy direct impacts. contexts are posed in a way that allow investiga-
This development represents a large step for tors to develop possible solutions and to find out
the Humanity. The comfort, the best levels of the best ways to achieve them.
health and the best standards of life correspond Since its beginning, chaos has become rapidly
to a set of advantages which medal reverse rep- a developing field and many of the progresses in
resents so big costs that their effects we are not this area were revealed just since the 1970s. This
able to study yet. means that many facets of chaos are distant from
It is not the progress and the development being understood or determined yet, making of it
that may be contested but the way the societies a field with great potential. It is important to note
do not find ways to get a supported and balanced that nowadays chaos is extremely difficult to be
development, respecting the principles of equity identified in real world information in order to be
that must exist among our specie and the others. workable. It is possible to find it in mathematical
Throughout the 20th century the extreme computer problems to be solved and laboratory
exploitation of resources brought out very big research. As soon as the idea of nonlinearity1 is
concerns about their preservation. Along the time introduced into theoretical models, chaos gets
the immoderation has shown the necessity of mea- obvious. A very complex structure is observed
sures to avoid the depletion of many resources. in the field data. Simple patterns can be found
That is why common resources become studied and approximated; complex patterns are another
worldwide. Nowadays, they are an important matter. In any event, we can’t just grab a nice
subject to analyze resources problems. In the last little set of data, apply a simple test or two, and
few decades, international organizations have declare “chaos” or “no chaos.” (Williams, 1997).
successively looked for balances to the problem The chaos theory involves multiple interac-
of fragile resources preservation and environ- tions and supposes the existence of an enormous
mental questions, searching for solutions. The number of interrelations, with direct developments
international agents and the private interests are in vast fields of study. It got an important role in
numerous and the convergences seem to be tradi- the context of recent theoretical developments
tionally very difficult. National and international of the non-equilibrium theories. The word chaos
organizations and States have seen the balances of assumes the idea of the existence of turbulence
life and resources on Earth to be frequently broken. and disorder; an unwanted chance or even the
271
idea of an “abyss”. The predisposition to a pro- main agents in this process may be questioned?
found change in the direction of a phenomenon The supranational institutions just emit simple
generates an own force, understood as a depth indicative rules?
change that results from small changes in their
initial conditions. The chaos is, from this point of
view, something extremely sensitive to the initial CHAOS IN DYNAMICAL
conditions. It is interesting to note, however, that SYSTEMS’ THEORIES
the chaotic system normally seems to develop
itself in a very smooth and orderly way, although A dynamical system represents moving, chang-
inside changes may be complex and paradoxical. ing or evolving in time. For this reason, chaos
Recent developments in the dynamical systems deals with dynamical systems theory - the study
theories, which require the existence of an inher- of phenomena that vary with time - or nonlinear
ent complexity of the systems themselves that dynamics - the study of nonlinear movement or
are based on a set of large inside interactions, led evolution.
in some cases to understand and highlight self- Dynamical systems fall into one of two cat-
organizing systems, revealing strong strengths egories, depending on whether the system loses
and reinforcing their internal cohesion factors. energy. A conservative dynamical system has
Given the non-linearity conditions of the no friction and it doesn’t lose energy over time.
nature phenomena, theories which are based on In contrast, a dissipative dynamical system has
the dynamics of non-equilibrium seem to explain friction; it loses energy over time and therefore
quite well the spatial and temporal heterogeneity always approaches some asymptotic or limiting
observed in ecological systems. Chaos occurs often condition. That asymptotic or limiting state, under
in deterministic, nonlinear, dynamical systems. certain conditions, is where chaos occurs (see
The disturbances and heterogeneity are inter- Williams, 1997).
dependent factors that create opportunities for The dynamical systems theories have been
re-colonization and determine the structure of applied to numerous areas of knowledge. In the
communities. It is important to note, for reflec- 80’s, several exact sciences - physics, chemistry
tion about the effects of the disturbances in the or biology, for example - and some social sciences
systems, that the ability to recover the ecological - economics or management or even the sociol-
systems depends, to some extent, on the existence ogy - still had their own objects of study and their
of refuge areas both for flora and fauna, acting as own methods of analysis and each one of them
reservoirs of re-colonizers, after the disturbances was different from the others. The science has
occurred in the ecosystem. been branched and specialized, so that each one
Many live resources, particularly many ma- uses to have its own world. Recently new forms
rine resources have suffered drastic reductions of analysis, looking for an integrated study, have
motivated by their overexploitation. Populations emerged (Filipe, 2006; Filipe et al, 2007).
of many species have been led to the rupture and The theory of chaos and complexity theory
close to the extinction. How can Humanity modify itself reflect the phenomena that in many activi-
this state of things? Is there a line of evolution ties, such as fisheries, are translated into dynamic
that helps to explain this kind of events? How forms of analysis and reflect a very complex and
can we shape these facts in this perspective? Is widespread reality, specific of complex systems.
there any ways to invert these trends? Or simply That reality falls within a range of situations in-
to find the principles that underline the facts? The tegrated in a broader context, which is expressed
Chaos and Complexity 272

Theory for Management : Nonlinear Dynamics, edited by Santo Banerjee, IGI Global, 2012. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/acu/detail.action?docID=3312270.
in the theory itself but also in terms of their own In contrast, difference equations usually can be
realities - fisheries, for example - dynamic, com- solved right away. Furthermore, they are often
plex and often chaotic features in their essence. acceptable approximations of differential equa-
The theory of chaos stresses that the world does tions. Olsen and Degn (1985) say that difference
not necessarily work as a linear relationship with equations are the most powerful vehicle to the
perfectly defined or with direct relations in terms of understanding of chaos.
expected proportions between causes and effects. Many scientists see, with particular interest, the
The chaos occurs when a system is very sensitive theory of chaos as a way to explain the environ-
to the initial conditions. These initial conditions ment. Therefore, the theory of chaos stresses the
are the measured values for a given initial time. fundamental laws of nature and natural processes
The presence of chaotic systems in nature seems and requires a course for a constant evolution and
to place a limit on our ability to apply physical recreation of nature. The theory of chaos allows
deterministic laws to predict movements with to realize the endless alternative ways leading to
any degree of certainty. Indeed, one of the most a new form or new ways that will be disclosed
interesting subjects in the study of chaotic systems and that eventually emerge from the chaos as a
is the question of whether the presence of chaos new structure. The reality is a process in which
may or may not produce ordered structures and structure and chaos rotate between form and
patterns on a wider scale. In the past, the dynamic deformation in an eternal cycle of death and
systems showed up completely unpredictable and renewal. Conditions of instability seem to be the
the only ones who could aspire to be understood rule and, in fact, a small inaccuracy in the condi-
were those that were represented by linear relations of departure tends to grow to a huge scale.
tionships, which are not the rule. On the contrary, Basically, two insignificant changes in the initial
there are some situations clearly isolated. conditions for the same system tend to end in two
Today, with the help of computers, it is possible situations completely different. This situation is
to make extremely complex calculations and to known, poetically, as the “butterfly wing effect”.
understand better the occurrence of chaos. A small movement of the wings of a butterfly can
As Williams (1997) says, phenomena happen have huge consequences. It is the microscopic
over time either as at discrete, separate or distinct, turbulence having effects in a macroscopic scale
intervals2 or as continuously3. Discrete intervals - an effect called by Grabinski (2004) as hydro-
can be spaced evenly in time or irregularly in dynamics. Mathematically, the “butterfly wing
time. Continuous phenomena might be measured effect” corresponds to the effect of chaos, which
continuously. However, we can measure them at can be expressed as follows.
discrete intervals4. Special types of equations apply Given the initial conditions
to each of those two ways in which phenomena
happen over time. Equations for discrete time x 1 , x 2 , x 3 , ..., x n

changes are difference equations and are solved
either analytically or by iteration methods. In it is possible to calculate the final condition
contrast, equations based on a continuous change given by
- continuous measurements - are differential
equations. The term “flow” often is associated FinalResult = f (x 1 , x 2 , x 3 , ..., x n )
to differential equations5.
Differential equations are often the most ac-
curate mathematical way to describe a smooth If the initial conditions x i have a margin of
continuous evolution. However, some of these error (variation), the final result will be influenced
equations are difficult or impossible to solve. by the existence of this margin. If these margins
273
in x i are small as the margin of error in the final and there is chaos when λ > 0.
result, we have a situation non-chaotic. Otherwise Through this function the chaos exist when
if the margins of error in x i are small but the final arbitrary small variations in initial conditions grow
result has a big variation, there is a chaotic situ- exponentially with a positive exponent.
ation. Therefore, small variations in initial condi- Grabinski also says that the nonlinearity is
tions can lead to a major effect in the final outcome. the main characteristic of a chaotic situation.
Sometimes small changes in x i have exponential Mathematically, the nonlinear functions to be
effects on the final result due to the passage of considered chaotic should be based on variables
time. with some resistance. The author also argues
This effect can be demonstrated mathemati- that is not enough to describe the chaotic situ-
cally6 using the Lyapunov Exponent7 (see Gra- ations, such as turbulence, but it is necessary to
binski, 2008). Given the initial value x 0 and ε find ways to better cope with the nonlinearity. A
smooth flow of a river, non-chaotic, that can be
being its - arbitrarily small - variation, we are
described in quantities like the flow velocity can
conducted to an initial value between x 0 and
reach a chaotic behavior with variations of many
x 0 + ε.The general form of Lyapunov indicator
situations. The best example is a waterfall where
is presented by the speed of the flow reaches a certain point. In
a smoothly flowing river it is easy to calculate or
x n +1 = f (x n )
predict the flow velocity of the river at any point.
However, to calculate it in a river with a waterfall,
that after N iterations leads to a value for x n it is necessary to introduce chaos. In an attempt
between to make this calculation, man has focused on the
construction of super computers that have shown
f N (x 0 ) to be useless due the infinity of factors that may
cause turbulence in the flow of the river. Thus,
the analysis of frequency on the change of flow’s
and
velocity is much more promising than the analysis
of velocities themselves.
f N (x 0+ε )
Moreover, Grabinski, shows the situation in
which there is chaos on a microscopic scale but
being the difference between these two values not on a macroscopic scale - the hydrodynamics.
An example is a glass of water resting on a table,
f N (x 0+ε ) − f N (x 0 )  εe N λ(x 0 ) a not chaotic event. A slight disturbance on the
table causes a small flow on a macroscopic level
in the water. However, a microscopic observation

where λ is a parameter.
reveals a great agitation of millions of molecules,
Dividing both sides by the variation ε and
a chaotic event. This is a situation where there is
assuming the limit ε → 0, we have a differential
chaos on the microscopic scale but a smooth flow
quotient. Making its logarithm and assuming the
on the macroscopic scale. Mathematically Gra-
limit N → ∞, we get the final definition of the
binski presents hydrodynamics equations which
Lyapunov Exponent
combine the chaos theory with business situations.
For that, he presents the function about the value
1
λ (x 0 ) = lim log | df N (x 0 ) / dx 0 | of a company (v ) that depends on two variables,
N →∞ N
the revenue (r ) and the number of employees (n ).
Its general form is

v (r, n) = v 0 + a10 r + a 01n + a11rn + a 20 r2 + a 02 n 2 + … The ecology where many things are random
and uncertain, in which everything interacts with
being aij general parameters. For n = 0 (no em- everything at the same time is, itself, a fertile
ployees) or r = 0 (no revenue) the company area for a cross search to the world explanations
doesn’t exist because the function value is equal (Filipe et al, 2005).
to 0. Lansing (2003) states that the initial phase of
So some terms of the function must be the research of nonlinear systems was based on
removed the deterministic chaos, and it was later redirected
to new outbreaks of research focusing on the
(v 0 = a10 = a 01 = a 20 = a 02 = … = 0). systems properties, which are self-organizing.
What is called anti-chaos. It also says that the
The general form comes as study of complex adaptive systems, discussed in
the context of non-linear dynamic systems, has
v (r, n) = a11rn + a 21r2 n + a12 r n2 + a 22 r2 n2 + … become a major focus of interest resulting from
the interdisciplinary research in the social sciences
Now to the case of symmetry, r and n could and the natural sciences.
be negative. A negative employee means that the The theory of systems in general represents
employee is paying to work and negative revenue the natural world as a series of reservoirs and
means that the company is paying the costumer streams governed by various feedback processes.
to consume. So the previous formula can lead to However, the mathematical representations were
negative results if r and n change signs simulta- ignoring the role of these adjustment processes.
neously. Only these terms are allowed for which The theory of complex adaptive systems part
the sums of the powers of r and v are even num- of the theory of systems, although it has in specific
bers. Thus the general expression of the equation account the diversity and heterogeneity of systems
is rather than representing them only by reservoirs.
It explicitly considers the role of adaptation on the
v (r, n) = a11rn + a 22 r2 n2 + … control of the dynamics and of the responses of
these heterogeneous reservoirs. This theory allows
ecologists to analyze the reasons inherent to the
CHAOS, DYNAMICAL SYSTEMS’ process at the lower levels of the organization that
THEORY, AND ECOLOGY lead to patterns at higher levels of organization
and ecosystems. The adaptive systems represent
Chaos, that is, deterministic chaos, deals with one of the means to understand how the organi-
long-term evolution - how something changes over zation is produced to a large scale and how it is
a long time. A chaotic time series looks irregular. controlled by processes that operate at lower levels
Two of chaos’s important practical implications of organization. According to Lansing (2003),
are that long-term predictions under chaotic con- came to be a general idea involving physical and
ditions are worthless and complex behaviour can mathematical complexity that is hidden behind
have simple causes. Chaos is difficult to identify systems very simple.
in real world data because the available tools Considering a system composed by many in-
generally were developed for idealistic condi- teractive parts, if it is sufficiently complex, it may
tions that are difficult to fulfil in practice (see not be practical or even not be possible to know
Williams, 1997). the details of each interaction place. Moreover, the
275
interactions can generate local non-linear effects to which the components of the system fit, creat-
that often it becomes impossible to find a solution ing a meta-stability that depends not only on the
even for simple systems. However, diverting us internal interactions within the system but also on
from causal forces that move the individual ele- external forces that can regulate and strengthen the
ments, if we focus on the system behavior as a internal factors of cohesion (see Lévêque, 2002).
whole we can highlight certain global behavior Scoones (1999) argues that should be con-
standards. However, these behavior standards may cluded a new commitment in research on the
hide an associated cost: it cannot be expected to ecological new thinking and he develops its
understand the causes at the level of individual search precisely in the area of ecology around
behavior. the concepts of chaotic dynamics and systems
Indeed, the systems do not match the simple of non-equilibrium. In turn, Levin (2003) shows
decomposition of the whole into parts and there- that in the study of complex adaptive systems
fore do not correspond to the mere sum of the anti-chaos involves the understanding of how the
parts, as living systems are not the juxtaposition cooperation, alliances and networks of interac-
of molecules and atoms. Since the molecule to the tions emerge from individual behaviors and how
biosphere, the whole is organized and each level it generates a feed-back effect to influence these
of integration leads to properties that cannot be behaviors within the spontaneous order and self-
analyzed only from mechanisms that have explana- organization of ecosystems.
tory value in the lower levels of integration. This
corresponds to the appearance of new features to
the level of the set that does not exist at the level DYNAMICAL SYSTEMS AND
of the constituent elements. Lansing (2003) be- CHAOS THEORY AND FISHERIES
lieves that the adoption in the social sciences of
the idea that complex global patterns can emerge In order to frame some methodological devel-
with new properties from local interactions had a opments, it must be mentioned, first of all, that
huge impact here. some characteristics associated with some spe-
The ecological systems are comparable to cies support strategic survival features that are
systems self-organized as they are open systems exploited by the present theory. Its aim is to find
which arise far from thermodynamic equilibrium. the reasons and the way in which these strategies
On self-organized and self-regulated systems, the are developed and the resulting consequences.
reciprocal interactions within the system between The species use their biological characteristics
the structures and the processes contribute to the resulting from evolutionary ancient processes to
regulation of its dynamics and the maintenance establish defense strategies.
of its organization; partly due to the phenomena However, given the emergence of new forms of
of feedback (see Lévêque, 2002). These systems predation, species got weaker because they are not
seem to develop themselves in accordance with prepared with mechanisms for effective protection
the properties referred to the anti-chaotic systems. for such situations. In fisheries there is a preda-
Indeed, we have auto-regulated systems that chan- tor, man, with new fishing technologies who can
nel different initial conditions for the same stage, completely destabilize the ecosystem. By using
instead of what is happening with chaotic systems, certain fisheries technologies, such as networks
which are very sensitive to initial conditions (see of siege, allowing the capture of all individuals
Kauffman, 1993). These systems would be rela- of the population who are in a particular area of
tively robust for a particular type of disturbance, fishing, the fishers cause the breakdown of certain

species, particularly the pelagic ones, normally inevitable. Since the efficiency of the school is
designated by schooling species. proportional to its size, the losses due to the effects
To that extent, with small changes in ecosys- of predation are relatively high for low levels of
tems, this may cause the complete deterioration stocks. This implies non-feedback in the relation
of stocks and the final collapse of ecosystems, stock-recruitment, which causes a break in the
which in extreme cases can lead to extinction. curves of income-effort, so that an infinitesimal
These species are concentrated in high density increase on fishing effort leads to an unstable
areas in small space. These are species that tend condition that can lead to its extinction.
to live in large schools. Considering however the fishing as a broader
Usually, large schools allow the protection issue, we may consider the modeling of the stocks
against large predators. The mathematical theory, of fish on the basis of an approach associated with
which examines the relationship between schools the theory of chaos instead considering the usual
and predators, due to Brock and Riffenburgh (see prospect based on classical models. Indeed, the
Clark, 1974), indicates that the effectiveness of issue can be placed within this framework from
predators is a reverse function of the size of the two different prisms: the traditional vision and the
school. Since the amount of fish that a predator vision resulting from theories of non-equilibrium.
can consume has a maximum average value, over- Around the traditional Newtonian view, the facts
coming this limit, the growth of school means a can be modeled in terms of linear relationships:
reduction in the rate of consumption by the preda- involving the definition of parameters, identifying
tor. Other aspects defensive for the school such relevant variables and using differential equations
as intimidation or confusing predators are also to describe the processes that change slowly over
an evidence of greater effectiveness of schools. time. For a given system, it should then carry out
However this type of behavior has allowed the measurements in a context that remains stable
development of very effective fishing techniques. during various periods.
With modern equipment for detecting schools - Moreover, we may have models based on
sonar, satellites, etc. - and with modern artificial the theory of chaos. These models are based on
fibers’ networks - strong, easy to handle and quick non-linear relationships and are very close to
placement - fishing can keep up advantageous for several disciplines, particularly in the branch of
small stocks (Bjorndal, 1987; Mangel and Clark, mathematics that study the invariant processes of
1983). As soon as schools become scarce, stocks scale, the fractals, and in a huge range of other
become less protected. Moreover, the existence subjects in the area of self spontaneous creation of
of these modern techniques prevents an effect of order: the theory of disasters or complex systems,
stock in the costs of businesses, as opposed to for example.
the so-called search fisheries, for which a fishery The first way is largely used by the majority
involves an action of demand and slow detection. of biologists, economists and environmentalists,
Therefore, the existence of larger populations is scientists and technical experts that conduct studies
essential for fishermen because it reduces the in marine search and senior technicians from state
cost of their detection (Neher, 1990). But, the and transnational agencies in the area of fisheries.
easy detection by new technologies means that It treats nature as a system, which has a regular
the costs are no longer sensitive to the size of the order. But today there are many responsible for
stock (Bjorndal and Conrad, 1987). fisheries management who also base their deci-
This can be extremely dangerous due to poor sions on models of chaos.
biotic potential of the species subject to this kind The classical models center on a particular sys-
of pressure. The reproductive capacity requires tem and depend on a local analysis, studying sev-
a minimum value below which the extinction is eral species, age, class, sub-regions of the marine
277
eco-niche, the various ports and their discharges, MODELLING FISHERIES’

depending on the account of an even wider range DISSIPATIVE EFFECT
of other factors. Probably, the classic expression
of linearity on the dynamics of the population - In order to frame some additional methodological
the principle that nature is orderly, balanced and developments, it must be mentioned that some
that has a dynamic balance - is due to Maynard characteristics associated with some species sup-
Smith (1968), which argues that the populations port strategic survival features that are exploited
either remain relatively constant or regularly vary by these species. Considering that, it is important
around an alleged point of balance. In the specific to find the reasons and the way these strategies
case of commercial fisheries, biologists believe are developed and the resulting consequences.
that the fishing effort is often relevant to explain The species use their biological characteristics
the deviations of actual populations’ values for the resulting from evolutionary ancient processes to
model. They say that, specially based on studies establish defense strategies.
made in the last decade, fish stocks sustainability However, given the emergence of new forms of
should be ensured by the control made through predation, species got weaker because they are not
fisheries regulation. prepared with mechanisms for effective protection
Moreover, some people see nature as not ca- for such situations. In fisheries there is a preda-
sual and unpredictable. The natural processes are tor, man, with new fishing technologies who can
complex and dynamic, and the causal relations completely destabilize the ecosystem. By using
and sequential patterns may extend so much in certain fisheries technologies, such as networks
time that may seem to be non-periodical. The of siege, allowing the capture of all individuals
data appear as selected random works, disorderly, of the population who are in a particular area of
not causal in their connections and chaotic. The fishing, the fishers cause the breakdown of certain
vision provided by nature leads to consider the species, particularly the pelagic ones, normally
fish stocks, time, the market and the various designated by schooling species.
processes of fisheries management as likely to To that extent, with small changes in ecosys-
be continuously in imbalance rather than behave tems, this may cause the complete deterioration
in a linear fashion and in a constant search for of stocks and the final collapse of ecosystems,
internal balance. It is this perspective that opens which in extreme cases can lead to extinction.
the way for the adoption of the theory of chaos These species are concentrated in high density
in fisheries. However, the models of chaos do areas in small space. These are species that tend
not deny, for themselves, some of the linearity to live in large schools.
resulting from the application of usual bionomic For the application of a mathematical model
models. What is considered is that there are no to fisheries, let’s inspire in the model presented
conditions to implement all significant variables in Berliner (1992) related to dissipative systems
in a predictive model. Moreover, in finding that in the presence of chaos.
a slight change in initial conditions caused by a In Berliner (1992) it is referred that nonin-
component of the system may cause major changes vertibility is required to observe chaos for one-
and deep consequences in the system itself. So, dimensional dynamic systems. Additionally it is
the application of the theory of chaos to fishing said that “everywhere invertible maps in two or
is considered essential, by many researchers. The more dimensions can exhibit chaotic behavior”.
theory of chaos depends on a multitude of factors, The study of strange attractors shows that in the
all major, and in the prospect of this theory all very long term, as time proceeds, the trajectories of
important at the outset, on the basis of the wide systems may become trapped in certain bounded
range of unpredictable effects that they can cause. regions of the state space of the system.

The model presented in Berliner (1992) is an Considering the particular case below:
example in two dimensions of the Hénon map
(displaying the property of having a strange at- x t +1 = x t − yt ,
tractor).
The Hénon map appears represented by the and
equations:
yt +1 = bx t .
x t +1 = 1 + yt − ax t2
So,
and
x t +2 = x t +1 − yt +1
yt +1 = bx t ,
and
for fixed values of a and b and t = 0, 1, ...
This invertible map possesses strange attrac- x t +2 − x t +1 + bx t = 0.
tors and simultaneously has strong sensitivity to
initial conditions.
The Hénon map, representing a transformation Now, after solving the characteristic equation
from the Cartesian plane to itself, has Jacobian associated to the difference equation (see Ferreira
equal to −b. and Menezes, 1992) it is obtained:
If 0 < b < 1, the Hénon map contracts the
domains to which it is applied. These maps are 1 + 1 − 4b
k=
said to be dissipative (on the contrary, maps that 2
maintain the application domain are said to be
conservative). or
Presented the model, it is possible now to sug-
gest a model on this basis for fisheries. 1 − 1 − 4b
So, if a general situation is considered, the k= ;
2
following equations may represent a system in
which fish stocks, at time t, are given by x t and
calling ∆ = 1 − 4b and being 0 < b < 1, comes
catches by yt . The model is as follows:
that −3 < ∆ < 1.
So,
x t +1 = F (x t ) − yt
0 <∆ <1
and
1 1
yt +1 = bx t . if 0 < b < and −3 < ∆ < 0 if < b < 1, being
4 4
1
∆ = 0 when b = .
It is a generalization of Hénon model. The 4
1
Jacobian is equal to b. As yt +1 is a portion of x t , Consequently for 0 < b < ,
4
0 < b < 1. So, it is a dissipative model and the
values of x t are restricted to a bounded domain.
279
t t
1 + 1 − 4b    that study the invariant processes of scale or the
x t = A1   + A2 1 − 1 − 4b 
 2   2  fractals, for example.

Chaos theory got its own space among sciences
and has become itself an outstanding science.
and for However there is much left to be discovered. Any-
way, many scientists consider that chaos theory
1 is one of the most important developed sciences
b= ,
4 on the twentieth century.
Aspects of chaos are shown up everywhere
1
t
around the world and chaos theory has changed
x t = (A1 + A2t )   the direction of science, studying chaotic systems
 2 
and the way they work.
We cannot say yet if chaos theory may give us
Finally, for solutions to problems that are posed by complex
systems. Nevertheless, understanding the way
1 chaos discusses the characteristics of complexity
<b <1
4 and analyzes open and closed systems and struc-
tures is an important matter of present discussion.
       Considering the fisheries in a broad context,
( b ) A cos arccos 2 1b  t  + A sen arccos 2 1b  t 
t
xt = 1 2
  the modeling of stocks of fish may be considered
on the basis of an approach associated with the
In these solutions, A1 and A2 are real constants. theory of chaos instead considering the usual
Note that the bases of t powers are always prospect based on classical models.
between 0 and 1. So, lim x t = 0 and whatever the On the fisheries analysis, it is interesting to see
t →∞ that overfishing may cause a problem of irrevers-
value of b, the dissipative effect is real, even ibility in the recovering of several species, after
leading to the extinction of the specie. Of course, certain stages for the stocks. Anyway, to analyze
this is evident according to the hypotheses of this the specific situation of each case, it is necessary to
particular situation of the model. obtain enough data to analyze the kind of function
Concluding this approach, the general model which is specific for that particular case, and it
does not allow to obtain in general such explicit must be analyzed the situation for certain phases
solutions by analytical means. But, of course, with of fishing and it must be seen the consequences
simple computational tools it is possible to obtain for these species.
recursively concrete time series solutions after In this paper, a general model that evidences
establishing the initial value x 0 and to check the the dissipative effect of catches on fisheries was
dissipative effect. proposed. Moreover, a particular case, with a
system of two difference equations analytically
treatable, shows how fishing stocks are dissipated
CONCLUSION and may tend to the extinction.
On the fisheries analysis it is interesting to see
The models of chaos are based on non-linear that overfishing may cause a problem of irrevers-
relationships and are very close to several disci- ibility in the recovering of several species, after
plines, particularly in the branch of mathematics certain stages for the stocks. Anyway, to analyze

the specific situation of each case it is necessary Ferreira, M. A. M., & Menezes, R. (1992). Equa-
to obtain enough data to analyze the kind of func- ções com Diferenças – Aplicações em problemas
tion which is specific for that particular case and it de Finanças, Economia, Sociologia e Antropolo-
must be analyzed the situation for certain phases gia. Lisboa, Portugal: Sílabo.
of fishing and it must be seen the consequences
Filipe, J. A. (2006). O Drama dos Recursos
for these species.
Comuns. Um caso de aplicação da Teoria dos
Jogos aos comuns da pesca. PhD thesis. Lisboa,
Portugal: ISCTE.
REFERENCES
Filipe, J. A., Ferreira, Coelho, M., & Pedro, M. I.
Bergé, P. Y., & Pomeau, C. V. (1984). Order within C. (2010). Chaos, anti-chaos and resources: Deal-
chaos. New York, NY: John Wiley. ing with complexity. Aplimat-Journal of Applied
Berliner, L. M. (1992). Statistics, probability and Mathematics, 3(2), 84–90.
chaos. Statistical Science, 7(1), 69–122. Filipe, J. A., Coelho, M., & Ferreira, M. A. M.
Bjorndal, T. (1987). Production economics and (2005). Sistemas Dinâmicos, Caos e os Comuns
optimal stock size in a North Atlantic fishery. da Pesca. Revista de Economia Global e Gestão,
The Scandinavian Journal of Economics, 89(2). 2. Lisboa, Portugal: ISCTE.
doi:10.2307/3440061 Filipe, J. A., Ferreira, M. A. M., & Coelho, M.
Bjorndal, T., & Conrad, J. (1987). The dynamics (2007). O Drama dos Recursos Comuns nas
of an open access fishery. The Canadian Journal Sociedades Actuais: à procura de soluções para
of Economics. Revue Canadienne d’Economique, os Ecossistemas em perigo. Lisboa, Portugal:
20(1). doi:10.2307/135232 Edições Sílabo.
Clark, C. W. (1974). Possible effects of school- Filipe, J. A., Ferreira, M. A. M., & Coelho, M.
ing on the dynamics of exploited fish popula- (2008). The relevance of chaos theory to explain
tions. Journal du Conseil Internatinal pour problems of overexploitation in fisheries. Work-
L’Exploration de la Mer, 36(1). ing Paper, WP/24/2008/DE/SOCIUS. Lisboa,
Portugal: ISEG.
Ferreira, M. A. M., Filipe, J. A., Coelho, M.,
& Pedro, M. I. (2010). Fishing policies and the Filipe, J. A., Ferreira, M. A. M., Coelho, M., &
contribution of chaos theory for fisheries manage- Pedro, M. I. (2010). Managing complexity: A
ment. Proceedings International Conference on problem of chaos in fisheries policy. China-USA
Technology and Business Management. Business Review, 9(3), 15–24.
Ferreira, M. A. M., Filipe, J. A., Coelho, M., & Filipe, J. A., Ferreira, M. A. M., Coelho, M., Pedro,
Pedro, M. I. C. (2011). Chaos effect in fisheries M. I., & Andrade, M. (2010). Analysing fisher-
management. Journal of Economics and Engi- ies management through complexity and chaos
neering, 2(1), 4–13. theories framework. Journal of Mathematics and
Technology, 1(2), 5–12.
Ferreira, M. A. M., Filipe, J. A., Coelho, M., &
Pedro, M. I. C. (2011). Modelling the dissipative Filipe, J. A., Ferreira, M. A. M., Coelho, M., &
effect of fisheries. China-USA Business Review, Pedro, M. I. C. (2009). Complexity, theory of
10(11), 1110–1114. chaos and fishing. In Porath, D., & Bayer, A.
(Eds.), International supplement special update.
FH Mainz. Mainz, Germany: University of Ap-
plied Sciences.
281
Grabinski, M. (2004). Is there chaos in manage- Rasband, N. S. (1990). Chaotic dynamics of

ment or just chaotic management? International nonlinear systems. New York, NY: John Wiley.
Conference on Complex Systems, Intelligence and
Scones, I. (1999). New ecology and the social sci-
Modern Technology Applications, Paris.
ences: What prospects for a fruitful engagement?
Grabinski, M. (2008). Chaos – Limitation or even Annual Review of Anthropology, 28.
end of supply chain management. Istanbul: High
Williams, G. P. (1997). Chaos theory tamed.
Speed Flow of Material, Information and Capital.
Washington, DC: Joseph Henry Press.
Hastings, A., Hom, C. L., Ellner, S., Turchin, P.,
& Godfray, H. C. J. (1993). Chaos in ecology: Is
mother nature a strange attractor? Annual Review
ENDNOTES
of Ecology and Systematics, 24, 1–33.
Kauffman, S. (1993). The origins of order: Self-
1
Nonlinear means that output isn’t directly
organization and selection in evolution. New York, proportional to input, or that a change in
NY: Oxford University Press. one variable doesn’t produce a proportional
change or reaction in the related variable(s).
Lansing, J. S. (2003). Complex adaptive systems. 2
Examples are the of occurrence earthquakes,
Annual Review of Anthropology, 32, 183–204. rainstorms or volcanic eruptions.
Retrieved from http://www.ic.arizona.edu/ 3
Examples are air temperature and humidity
lansing/GompAdSys.pdf doi:10.1146/annurev. or the flow of water in perennial rivers.
anthro.32.061002.093440 4
For example, we may measure air tempera-
Lévêque, G. (2002). Ecologia: Do ecossistema à ture only once per hour, over many days or
biosfera. Lisboa, Portugal: Instituto Piaget. years.
5
To some authors (see Bergé et al., 1984), a
Levin, S. (2003). Complex adaptive systems: flow is a system of differential equations.
Exploring the known, the unknown and the un- To others (see Rasband, 1990), a flow is the
knowable. Bulletin of the American Mathematical solution of differential equations.
Society, 40. 6
Several statistics may indicate chaos and
Mangel, M., & Clark, G. (1983). Uncertainty, can express how chaotic a system is. One
search and information in fisheries. Journal du of the most important statistics to measure
Conseil International pour L’Exploration de la magnitude of chaos is at present Lyapunov
Mer, 41. exponents. Other statistics could be pre-
sented as the Kolmogorov-Sinai entropy or
Maynard Smith, J. (1968). Mathematical ideas in
the mutual information or redundancy.

biology. Cambridge, UK: Cambridge University 7
A Lyapunov Exponent is a number that
Press. doi:10.1017/CBO9780511565144 reflects the rate of divergence or conver-
Neher, P. (1990). Natural resource economics: gence, averaged over the entire attractor, of
Conservation and exploitation. Cambridge, UK: two neighbouring phase space trajectories.
Cambridge University Press. Trajectory divergence or convergence has to
follow an exponential law, for the exponent
Olsen, L. F., & Degn, H. (1985). Chaos in biological to be definable.
systems. Quarterly Reviews of Biophysics, 18(2),
165–225. doi:10.1017/S0033583500005175

283
Chapter 14
A Comparative Study of Four
Different Satellite Image
Classification Techniques for
Geospatial Management
Devanjan Bhattacharya
Indian Institute of Science Bangalore, India & University of Pardubice, Czech Republic
Santo Banerjee
Institute for Mathematical Research, University Putra Malaysia, Malaysia
& International Science Association, Turkey
ABSTRACT
Satellite imagery interpretation has become the technology of choice for a host of developmental, scien-
tific, and administrative management work. The huge repository of geospatial data and information that
are available as satellite imageries datasets from platforms such as Google Earth need to be classified
and understood for natural resources management, urban planning, and sustainable development. The
classification and analysis procedures involve algorithms like maximum likelihood classifier, isodata,
fuzzy-logic classifier, and artificial neural network based classifier. Amongst these classifiers the optimum
has to be selected for classifications which involve multiple features and classes. Herein lies the moti-
vation for the present research, which can facilitate the selection of one amongst the many algorithms
available to a decision maker/manager. The aforementioned techniques are applied for classification,
and the respective accuracies in the classes of forestry, rock, water, built-up area, and dry river bed have
been tabulated and verified from ground truth. The comparison is based on time and space complexity of
the algorithms considering also the accuracy. It is found that traditional methods like MLC and Isodata
offer good time and space consumption performance over the recent more adaptable algorithms as fuzzy
and ANN. But the latter group excels in accuracy of assessment. The study suggests points and cases
for ranking the techniques as best, 2nd best, and so on, where each technique could be optimally utilised
for a given geospatial dataset based on its contents.
DOI: 10.4018/978-1-4666-2509-9.ch014
Copyright © 2013, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
A Comparative Study of Four Different Satellite Image Classification Techniques
INTRODUCTION the techniques wherein the time taken by each

algorithm is noted as well as the memory space
Classification is the process of partitioning a digital occupied inside the computing system. Based on
image into multiple classes (sets of pixels). The the collective results, suggestions are made on the
goal of classification is to simplify and/or change selection of the optimum classification techniques
the representation of an image into something for better land use and land cover management.
that is more meaningful and easier to analyze
(Ballard & Brown, 1982). More precisely, image
classification is the process of assigning a label BACKGROUND
to every pixel in an image such that pixels with
the same label share certain visual characteristics The algorithms considered for the study are all
(Hord, 1982). pixel based techniques that work on raster data
The result of image classification is a set of sets in the domain of digital image processing and
segments that collectively cover the entire image. classification. Raster data is a two dimensional
Each of the pixels in a region is similar with respect array of digital numbers (DN) where each cell of
to some characteristic or computed property, such the array corresponds to a pixel of the image. The
as colour, intensity, or texture. Adjacent regions image could be a gray-scale (b/w) or a coloured
are significantly different with respect to the image. Accordingly the array values would vary.
same characteristics. Several general-purpose A gray-scale would have a single layer of digital
algorithms and techniques have been developed values and a colour image would consist of hues
for image classification. Since there is no general of RGB. Each DN value represents a unit intensity
solution to the image classification problem, these captured in the image (Lillesand & Kiefer, 1994).
techniques often have to be combined with domain The algorithms try to best utilize the categori-
knowledge in order to effectively solve an image zation of these DN values. Maximum likelihood,
classification problem for a problem domain. isodata, fuzzy techniques and ANN have been
The objective of the present study is to apply used for numerous satellite image analysis studies
and compare the results of the four different im- but have not been compared amongst themselves
age classification techniques namely maximum on a single platform anywhere. The measures of
likelihood classifier (MLC), Isodata, fuzzy-logic comparison that effect the performance and utility
classifier (FLC), and artificial neural network most are space occupied in memory, time taken
(ANN) based classifier, all on the same satellite for processing, accuracy delivered, and future
image of a region acquired from Google Earth upgradability. To understand the study better, it
for which actual ground truth data has already is imperative to get a look into the backgrounds
been collected for later verification of results. related to the study. In this section some common
The comparative study details out the procedures terminologies are discussed to lay the foundation
undertaken, the parameters considered for evalu- for their applications later.
ations, and the standing of each technique, put in The simplest method of image classification
a tabular manner. The accuracy assessment has is called the thresholding method. The key of
been conducted as per the variations of classi- this method is to select the threshold value (or
fied pixels with respect to actual ground truth. values when multiple-levels are selected) for each
The evaluation parameters of time complexity, category that exists in a digital image. Several
space complexity, accuracy and upgradability popular methods are used in this field and in in-
have been considered for performance testing of dustry including the maximum entropy method,

Otsu’s method (maximum variance), and et al. Mc = the mean vector of the sample of class c
Fuzzy “k-means” or simply “k-means” clustering
can also be used. The present study analyses and ac = percent probability that any candidate pixel
compares MLC, Isodata, Fuzzy classification and is a member of class c (defaults to 1.0, or is
ANN based classification. entered from a priori knowledge)
Maximum Likelihood Classifier Covc = the covariance matrix of the pixels in

the sample of class c
One of the most powerful classifiers in common
|Covc| = determinant of Covc (matrix algebra)
use, Maximum Likelihood Classifier is based
on statistics (mean, variance, covariance), and
Covc-1 = inverse of Covc (matrix algebra)
a (Bayesian) Probability Function which is cal-
culated from the inputs for classes established
ln = natural logarithm function
from training sites. Each pixel in the image to
be classified is judged by MLC as to the class to
T = transposition function (matrix algebra)
which it most probably belongs (Barbieri, Arruda,
Rodrigues, Brunoa & Costa, 2011).
The pixel is assigned to the class, c, for which
The maximum likelihood decision rule is based
D is the lowest.
on the probability that a pixel belongs to a particular
Among the advantages of MLC is the observed
class. The basic equation assumes that these prob-
fact that if the input samples have a normal
abilities are equal for all classes, and that the input
distribution, then the classifier is very accurate,
bands have normal distributions. The inclusion of
because it takes the most variables into consider-
a-priori knowledge about the probabilities being
ation. Also it takes the variability of classes into
not equal for all classes may be specified as weight
account by using the covariance matrix. But these
factors for particular classes. This variation of the
are computation intensive processes involving
maximum likelihood decision rule is known as the
an extensive equation that takes a long time to
Bayesian decision rule (Hord, 1982). If a priori
compute (Unnikrishnan, Panpofaru & Herbert,
knowledge of the probabilities is unavailable, it
2005). Also the computation time increases with
is better that they not be specified. In this case,
the number of input bands. Further, maximum
these weights default to 1.0 in the equation. The
likelihood is parametric, meaning that it relies
equation for the maximum likelihood (Bayesian)
heavily on a normal distribution of the data in
classifier is as follows:
each input band. Also, it tends to overclassify
signatures with relatively large values in the
D = ln(ac) - [0.5 ln(|Covc|)] - [0.5 (X-Mc)T
covariance matrix (Bishop, 2006).

(Covc-1) (X-Mc)]
Isodata
where
ISODATA stands for Iterative Self-Organizing
D = weighted distance (likelihood)
Data Analysis Technique. It is iterative in the
c = a particular class sense that it repeatedly performs an entire clas-
sification, outputting a thematic raster layer, and
X = the measurement vector of the candidate recalculates statistics. Self-Organizing refers to
pixel
285
the way in which it locates the clusters that are The Fuzzy Convolution operation creates a
inherent in the data (Meyer & Beucher). single classification layer by calculating the total
The ISODATA clustering method uses the weighted inverse distance of all the classes in a
minimum DN value distance formula to form window of pixels. Then, it assigns the center pixel
clusters. It begins with either arbitrary cluster in the class with the largest total inverse distance
mean values or mean values of an existing signa- summed over the entire set of fuzzy classification
ture set, and each time the clustering repeats, the layers.
mean values of these clusters are shifted. The new This has the effect of creating a context-based
cluster mean values are used for the next iteration. classification where classes with a very small
The ISODATA utility repeats the clustering of distance value remain unchanged while classes
the image until either: with higher distance values may change to a
neighboring value if there is a sufficient number
• A maximum number of iterations has been of neighboring pixels with class values and small
performed, or corresponding distance values. The following
• A maximum percentage of unchanged pix- equation is used in the calculation:
els has been reached between two itera-
tions (Comaniciu & Meer, 2002). s s n wij
T [k ] = ∑ ∑ ∑ D [k ]
i =0 j =01 =0 ij 1
Fuzzy Logic Based Classifier
(Fuzzy K-Means)
where
The Fuzzy Classification method takes into ac-
i = row index of window
count that there are pixels of mixed make-up,
that is, a pixel cannot be definitively assigned to
j = column index of window
one category. Fuzzy classification is designed to
help with data that may not fall into exactly one
s = size of window (3, 5, or 7)
category or another. Fuzzy classification works
using a membership function, wherein a pixel’s
l = layer index of fuzzy set
value is determined by whether it is closer to one
class than another. A fuzzy classification does n = number of fuzzy layers used
not have definite boundaries, and each pixel can
belong to several different classes (Zhang, 1996). W = weight table for window
Fuzzy classification also uses training, but
through the training it is also possible to obtain
k = class value
information on the various constituent classes
found in a pixel which may be of a mixed nature. D[k] = distance file value for class k
In a fuzzy classification done in ERDAS Imagine
software environment, the fuzzy convolution util- T[k] = total weighted distance of window for
ity performs a moving window convolution on a class k
fuzzy classification with multiple output class as-
signments. Using the multilayer classification and The center pixel is assigned the class with the
distance file, the convolution creates a new single maximum T[k].
class output file by computing a total weighted
distance for all classes in the window.

The K-means algorithm is an iterative tech- to the expectation-maximization algorithm for

nique that is used to partition an image into K mixtures of Gaussians in that they both attempt
clusters. The basic algorithm is: to find the centers of natural clusters in the data.
The model requires that the object attributes
1. Pick K cluster centers, either randomly or correspond to elements of a vector space. The
based on some heuristic. objective it tries to achieve is to minimize total
2. Assign each pixel in the image to the cluster intra-cluster variance, or, the squared error func-
that minimizes the distance between the pixel tion (Ge, Wang & Liu, 2006).
and the cluster center. Since the algorithm is extremely fast, a com-
3. Re-compute the cluster centers by averaging mon method is to run the algorithm several times
all of the pixels in the cluster. and return the best clustering found. A drawback
4. Repeat steps 2 and 3 until convergence is of the k-means algorithm is that the number of
attained (e.g. no pixels change clusters). clusters k is an input parameter. An inappropriate
choice of k may yield poor results. The algorithm
In this case, distance is the squared or absolute also assumes that the variance is an appropriate
difference between a pixel and a cluster center. measure of cluster scatter (Samins, 2007).
The difference is typically based on pixel color,
intensity, texture, and location, or a weighted Artificial Neural Network
combination of these factors. K can be selected Based Classifiers (Back
manually, randomly, or by a heuristic (Pal & Pal, Propagation Algorithm)
1993).
This algorithm is guaranteed to converge, but Many adaptive, non-parametric neural-net classi-
it may not return the optimal solution. The quality fiers have been proposed for real-world problems.
of the solution depends on the initial set of clusters These classifiers show that they are capable of
and the value of K. achieving higher classification accuracy than
In statistics and machine learning, the k-means conventional pixel-based classifiers; however,
algorithm is a clustering algorithm to partition n few neural net classifiers which apply spatial in-
objects into k clusters, where k < n. It is similar formation have been proposed. The feed-forward
Figure 1. Satellite image (Google Earth Oct 2010 GeoEye) of study area
287
Figure 2. Distribution of digital numbers (pixel values) in 3 layers of red, green, blue in image of figure 1
multilayer neural network has been widely used be used to perform the image classification (Rui
in supervised image classification of remotely & Wunsch, 2005).
sensed data. A back propagation Feed forward The back propagation algorithm is a generaliza-
multilayer network is an interconnected network tion of the least mean square algorithm that modi-
in which neurons are arranged in multilayer and fies network weights to minimize the mean squared
fully connected. There is a value called weight error between the desired and actual outputs of
associated with each connection. These weights the network. Back propagation uses supervised
are adjusted using the back-propagation algorithm learning in which the network is trained using
or its variations, which is called training the neural data for which inputs as well as desired outputs
networks. Once the network is well trained, it can are known. Once trained, the network weights are
frozen and can be used to compute output values

for new input samples (Ge, Wang & Liu, 2006).
Start with randomly chosen weights.
Figure 3. Training classes in image of figure 1
While classification is unsatisfactory and
computational bounds are not exceeded, do for
each input pattern:
• Compute hidden node inputs ( net p, j )

• Compute hidden node outputs ( x p, j )

Figure 4. Output of classification of figure 1 using MLC
• Compute inputs to the outputs nodes CLASSIFICATION

( net p,k )
• Compute the network outputs (op,k ) Input Data
• Compute the error between output dp,k and
The satellite data acquired is a Google Earth
desired output op,k
image of a section of Haridwar district, state Ut-
• Modify the weights between hidden and tarakhand, India lying between latitude 29°57’3”E
output nodes: to 29°57’46”E and longitude 78°9’20”N to
∆wk , j = η(dp,k − op,k )δ(net p,k )x p, j 78°10’49”N and elevation above mean sea level
• Modify the weights between input and hid- from 289m to 457 m. Apart from the satellite
den nodes: image, there were data collected on field in the
w j ,i = η ∑ ((d p,k − op,k )δ(net p,k )wk , j )δ(net p, j )x p,i region for validating the results. The aforemen-
• End for tioned region is shown in Figure 1. This image
• End While
Table 1. Statistics from classification using MLC
Distinct classes Number of pixels (array locations) Misclassification

percentage
Testing Samples Output from MLC
from each Class for testing samples
Forest 3247 2517 24.1
Rock 1011 778 20.2
Water 801 613 19.4
Built-Up Area 713 499 27.3
Dry river bed 650 330 26.5
289
Figure 5. Output of classification of figure 1 using ISODATA
is in TIFF format which is fed as input data for dry river beds as well. So the prominent classes
classification using the various algorithms. included for classification of the image are forest,
Figure 1 captures a region which is hilly and rock, water, built-up area, and dry river bed. In
rocky with forests. The river Ganges and tributar- Figure 2 the histograms show a collective and
ies flow through and there are lots of built-up uniform distribution of all pixel values present in
areas around the river banks. There are areas of the raw image of Figure 1 but the histogram ap-
Figure 6. Output of classification of figure 1 using fuzzy k-means


Table 2. Statistics from classification using ISODATA

percentage
Testing Samples Output from Isodata
Forest 1219 997 19.3
Rock 1211 998 18.9
Water 801 751 16.2
pearances are devoid of any indication of the back-propagation were executed. MLC, Isodata
different classes present in the image. To segregate and Fuzzy classification are made available by
the classes it is imperative that the image be clas- ERDAS. The TIFF image was 300 x 300 pixels
sified using classification algorithms and the re- and housed in an integer array of appropriate size.
sulting histograms studied.
The algorithms were fed with supervised selec- Using MLC
tion of pixels data from each of the classes as is
shown in Figure 3, which shows the classes of After training MLC with the sample pixels (Figure
the training sets selected. 3), it was run on the complete image to classify
the same. The resulting classified image is shown
System Requirement in Figure 4.
The classified image shows that MLC ex-
In order to implement the various algorithms, tended pixels in rock class over a wider area,
the setup utilised was a platform having instal- occupying what should ideally have been forest
lations of digital image processing package ER- in some cases. Also the classes of dry river bed
DAS Imagine 9.1, and MATLAB 7.5. Wherever and built-up areas were mixed. The water body
necessary these were appended with customized has been classified to a very limited extent. The
algorithms written in C++ as additional modules variability of pixels in actual image and classified
in case the technique was not available inbuilt. image is listed in Table 1.
In Matlab the C++ codes implementing ANN
Table 3. Statistics from classification using fuzzy k-means

percentage
Testing Samples Output from Fuzzy k-means
Forest 1267 997 19.3
Rock 1011 798 18.9
Water 801 551 21.2
291
Figure 7. Output of classification of figure 1 using ANN
Using Isodata Using Fuzzy K-Means
The classified output image (Figure 5) shows The classified output image (Figure 6) shows that
that Isodata has been better than MLC as far as Fuzzy k-means has been better than MLC and
classification of water body is concerned. It has Isodata in proper classification of water body and
also identified built-up ares in hills which was forests. But it has missed rock and hilly areas in
missed by MLC. The overall statistics are indi- some cases. The overall statistics are indicated
cated in Table 2. in Table 3.
Figure 8. Well defined extents for each of the 5 classes shown in histogram of the classified image

Table 4. Statistics from classification using ANN

percentage
Testing Samples Output from ANN
Forest 2238 1997 9.3
Rock 1011 998 8.9
Water 801 751 6.2
Using ANN with the next method correctly classified samples:

82, misclassified: 18, so accuracy: 82% ; so on.
The classified output image (Figure 7) shows that This way the comparative study of the different
Fuzzy k-means provided classification commen- methods yields the efficacy of each, and for each
surate with Fuzzy k means and has been better class involved. As is shown in Tables 1, 2, 3, 4 and
than MLC and Isodata in classification of forests 5, the best results came in through the robustness
and rocky regions. But it has over classified water of ANN, Fuzzy algorithms followed by Isodata
bodies in some cases. The overall statistics are and MLC.
indicated in Table 4 below. The evaluatory criteria can be understood as the
resources and effort being put into the techniques of
the study for desired output. For the improvement
RESULT ACCURACY of the generated output, the study calculated the
INTERPRETATION AND consumed-time by each technique, the amount of
DISCUSSIONS computer memory occupied, resultant accuracy
of output, and further scope of improvisation al-
The method for accuracy assessment of classifica- lowed by the concerned technique in the form of
tion results was adopted as first selecting pixels upgradability quotient.
from known areas established from ground truth It was seen that while ANN produced output
with known classes and then let the methods of commendable accuracy and offered lots of
‘identify’ what these samples are. Suppose with improvisation and hybridization opportunities, it
100 selected samples of a given class, results came consumed the most resources of the four compared.
correctly classified samples in a given method: Fuzzy logic based technique was commendable in
72, misclassified: 28, hence accuracy: 72% ; say, most of the criteria, it required strenuous training
Table 5. Comparison of the classification techniques over the parameters in this study
CRITERIA Time consumption Memory consumption Accuracy Upgradability

TECHNIQUES
M.L.C. Fastest 2nd best 85% Less scope
Isodata 3 fastest
rd
3 best
rd
87% Less scope
Fuzzy 2 fastest
nd
Least 78% Much scope
ANN 4th fastest 4th best 89% Much scope
293
strategy and hence the output accuracy suffered. The developer can mould the algorithmic require-
The traditional methods of MLC and Isodata ments according to the classes in use, resources
lived upto their reputation of much prevalent use available and accuracy desired.
and demonstrated performance commensurate
with algorithmic capability. The future scope of
inherent tweaking in the two techniques was not REFERENCES
much. It can be understood that looking at the
above Table 5, a judicious choice can be made Ballard, D. H., & Brown, C. M. (1982). Computer
as per the required application at hand. vision. Prentice-Hall Inc.
Barbieri, A. L., Arruda, G. F. D., Rodrigues, F.
A., Brunoa, O. M., & Costa, L. D. F. (2011).
CONCLUSION An entropy-based approach to automatic image
segmentation of satellite images. Physica A, 390,
The study has put forward a novel way of com- 512–518. doi:10.1016/j.physa.2010.10.015
parison of image classification and segmentation
algorithms based on the criteria of execution speed Bishop, C. M. (2006). Pattern recognition and ma-
(time complexity), efficiency (space complexity), chine learning. New York, NY: Springer-Verlag.
results correctness (accuracy), and future scope
Comaniciu, D., & Meer, P. (2002). Mean-shift:
of enhancements (upgradability). These are the
A robust approach toward feature space analy-
parameters that developers and scientists are most
sis. IEEE Transactions on Pattern Analysis
concerned about while evaluating the feasibility
and Machine Intelligence, 24(5), 603–619.
of application of any given algorithm to a task
doi:10.1109/34.1000236
similar to the one considered in the study. It was
shown that the specificity of MLC and Isodata Ge, F., Wang, S., & Liu, T. (2006). Image segmen-
for classification of data came out with better ef- tation evaluation from the perspective of salient
ficiency, and the versatility of fuzzy techniques and object extraction. Proceedings of IEEE Conference
ANN rendered customization and improvisation on Computer Vision and Pattern Recognition, Vol.
possible. The application had multiple classes for 1, (pp. 1146-1153).
which the respective results were studied and it
Hord, R. M. (1982). Digital image processing of
was recommended that for demarcation of forests
remotely sensed data. New York, NY: Academic
from other classes fuzzy approach got better ac-
Press.
curacy results, for rock class fuzzy and ANN both
should better than the others, for water ANN and Lillesand, T. M., & Kiefer, R. W. (1994). Remote
MLC performed poorly whilst fuzzy and Isodata sensing and image interpretation (3rd ed.). Wiley
gave accurate results, for built-up area Isodata & Sons.

gave best accuracy followed by MLC whilst fuzzy
Meyer, F., & Beucher, S. (1990). Morphological
and ANN lagged, and finally for dry river bed,
segmentation. Journal of Visual Communica-
ANN and MLC did well but fuzzy and Isodata
tion and Image Representation, 1(1), 21–46.
misclassified to a larger extent. Overall it is seen
doi:10.1016/1047-3203(90)90014-M
that the consideration is between resource provi-
sion and algorithmic power. MLC and Isodata Pal, N. R., & Pal, S. K. (1993). A review on image
can work with constrained environments with a segmentation techniques. Pattern Recognition, 26,
compromise but fuzzy and ANN show resource 1277–1294. doi:10.1016/0031-3203(93)90135-J
hungriness. It has to be a tradeoff between the two.

Rui, X., & Wunsch, D. (2005). Survey of Unnikrishnan, R., Panpofaru, C., & Herbert, M.
clustering algorithms. IEEE Transactions on (2005). A measure for objective evaluation of
Neural Networks, 16(3), 645–678. doi:10.1109/ image segmentation algorithms. IEEE Computer
TNN.2005.845141 Society Conference on Computer Vision and Pat-
tern Recognition, San Diego, CA, USA, Vol. 3,
Sabins, F. F. (2007). Remote sensing: Principles
(pp. 34-44).
and interpretation. Waveland Press.
Zhang, Y. (1996). A survey of evaluation
methods for image segmentation. Pattern Rec-
ognition, 29(8), 1335–1346. doi:10.1016/0031-
3203(95)00169-7
295

Chaos and Complexity Theory For Management Nonline... - (Section 4)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chaos and Complexity Theory For Management Nonline... - (Section 4)

Uploaded by

Copyright:

Available Formats

Section 4

Nonlinearity in Water and Land

Manuel Francisco Pacheco Coelho Maria Isabel Pedro

Chaos and Complexity 272

happen over time. Equations for discrete time x 1 , x 2 , x 3 , ..., x n

in the water. However, a microscopic observation

Chaos and Complexity 274

Chaos and Complexity 276

eco-niche, the various ports and their discharges, MODELLING FISHERIES’

Chaos and Complexity 278

Chaos and Complexity 280

Grabinski, M. (2004). Is there chaos in manage- Rasband, N. S. (1990). Chaotic dynamics of

the mutual information or redundancy.

Chaos and Complexity 282

INTRODUCTION the techniques wherein the time taken by each

Chaos and Complexity 284

Maximum Likelihood Classifier Covc = the covariance matrix of the pixels in

covariance matrix (Bishop, 2006).

Chaos and Complexity 286

The K-means algorithm is an iterative tech- to the expectation-maximization algorithm for

frozen and can be used to compute output values

• Compute hidden node inputs ( net p, j )

Chaos and Complexity 288

Figure 4. Output of classification of figure 1 using MLC

• Compute inputs to the outputs nodes CLASSIFICATION

Table 1. Statistics from classification using MLC

Distinct classes Number of pixels (array locations) Misclassification

Figure 5. Output of classification of figure 1 using ISODATA

Figure 6. Output of classification of figure 1 using fuzzy k-means

Chaos and Complexity 290

Table 2. Statistics from classification using ISODATA

Distinct classes Number of pixels (array locations) Misclassification

Table 3. Statistics from classification using fuzzy k-means

Distinct classes Number of pixels (array locations) Misclassification

Figure 7. Output of classification of figure 1 using ANN

Using Isodata Using Fuzzy K-Means

Chaos and Complexity 292

Table 4. Statistics from classification using ANN

Distinct classes Number of pixels (array locations) Misclassification

Using ANN with the next method correctly classified samples:

CRITERIA Time consumption Memory consumption Accuracy Upgradability

gave accurate results, for built-up area Isodata & Sons.

Chaos and Complexity 294

You might also like