The Second APEC Workshop of SAKE

Application of Knowledge-Based Expert System Model for Fishing Ground

Prediction in The Tropical Area
1 Muhamad Sadly, Nani Hendiarti 1, Suhendar I Sachoemar 2, Nurjannah Nurdin3, Yoke Faisal 1, Awaluddin 1
1 Center of Technology for Natural Resources Inventory (P-TISDA),
2 Center of Technology for Agro-industry,
3 Dept. of Marine, Hasanuddin University, Makassar
Agency for the Assessment and Application of Technology (BPPT)
BPPT Bldg. II, 19th Fl., Jl. M.H.Thamrin No. 8, Jakarta 10340

ABSTRACT
A geographical information system (GIS) may be viewed as a database system in which most of the
data is spatially indexed, and upon which a set of procedures operate in order to answer queries about spatial
entities represented in the database. Geo-study deals with answering What, Where, and Why questions.
Despite the fact that GIS is a powerful tool dealing the first two questions, GIS is inferior for answering the
Why question in geo-study (Zhang and Giardino, 1992). One of the possibility way to overcome the
inferiority of GIS for answering the Why question of Geo-Studies is by integrating expert system in a GIS
to form a Knowledge-Based Expert System GIS Model. In this study, we present the result of the application
Knowledge-Based Expert System GIS Model on the prediction of the fishing ground for pelagic fish in the
coastal area of Tomini Bay (Central Sulawesi) and South Sulawesi. As input data, we used and applied a
series satellite data of sea surface temperature (SST), sea surface chlorophyll-a (SSC) and turbidity derived
from Aqua MODIS in period of 2003-2005 to understand the temporal and seasonal variability of the marine
environment of the study area, and identified the oceanographic phenomena, i.e. upwelling, front or eddy. To
generate spatial configuration of fishing ground prediction map, we developed and integrated the result of
Knowledge-Based Expert System into GIS model by using ERDAS Macro Language (EML) of ERDAS
Imagine 9.0 software. To verify this result, a series of the in-situ fishing ground spots data of the study area
were collected for the similar periods and location, and they were then analyzed by using a simple statistical
method. The result shows that fishing ground prediction derived from Knowledge-Based Expert System GIS
Model has a high accuracy level with a range of 80-90 % against the in-situ data. This result has
demonstrated that the Knowledge-Based Expert System GIS Model can be applied to predict, localize and
determine fishing ground spot areas in which their accuracy level will be determined by the completeness of
spatial knowledge of the domain expertise and the sophistication level of the programming utilities being
used.
Keywords: Remote sensing, Knowledge-based, Expert Systems, GIS, Fishing Ground

1. INTRODUCTION
It is simplest form, a geographical
information system (GIS) may be viewed as a
database system in which most of the data is
spatially indexed, and upon which a set of
procedures operate in order to answer queries
about spatial entities represented in the database.
Geo-study deals with answering What, Where,
and Why questions. Despite the fact that GIS is
a powerful tool dealing the firs two questions, GIS
is inferior for answering the Why question in
geo-study.
Expert systems, or knowledge-based systems,
are branch of artificial intelligence (AI). AI is the
capability of a device such as a computer to
perform tasks that would be considered intelligent

if they were performed by a human. An expert

system is a computer program that attempts to
replicate the reasoning processes of experts and
can make decisions and recommendations, or
perform tasks, based on user input. Knowledge
Engineers (KE) construct expert systems in
cooperation with problem domain experts so that
the experts knowledge is available at all times
and in many places, as necessary [1,2,3,4].
On the basis of previous research concerning
the design and implementation of GIS, one may
infer several requirements that a GIS should
satisfy, as well as several principles of design and
implementation that permit the satisfaction of
such requirements.

23-1

Expert System Model for Fishing Ground Prediction

In this essay, we examine both the

requirements and the associated principles, first in
general terms and then in terms of a knowledgebased GIS that has been recently implemented.
One possible way to use GIS for dealing with
reasoning based on knowledge is by incorporating
an expert system concept in a GIS.
In this study, we develop an expert system
integrating with GIS for probable fishing grounds
predictive model of economic pelagic fish in
Tomini Bay, Central-Sulawesi using remote
sensing data. The oceanographical data of Sea
Surface Temperature (SST), Sea Surface
Chlorophyll-a (SSC) and Turbidity in relation to
the oceanographical phenomena i.e. upwelling,
front and Eddy were used as input parameters of
the system to generate fishing ground prediction
map. These input data were than taking account
into the input environmental data of the
relationships established, and transforming the
knowledge bases of spatial configuration in the
form of IF..THEN type of production rules into
macro programs written in ERDAS Macro
Language (EML) of ERDAS Imagine software
[5].
There appear to have been relatively few
expert systems constructed, or envisioned, for
oceanographic and maritime purposes. The
knowledge-based expert system described below
screens near real time incoming sea surface
temperature, chlorophyll and turbidity data. The
use of the expert system allows researchers to
develop the environmental model of fishing
ground prediction for further understanding of the
marine environment phenomenon as a knowledge
base that can be easily reconfigured [6].
2. BASIC THEORY OF SATELLITE
FISHERIES OCEANOGRAPHY
The optical properties in the marine surface
layer are determined by the presence of dissolved
and suspended matter. Under normal conditions,
visible light penetrates marine waters to a depth of
tens of metres. As the concentration of the water
constituents increases, i.e. the water becomes
more turbid, the penetration of sunlight is reduced
as a result of absorption and scattering processes.
Depending on the specific characteristics of the
materials present in the water, i.e. on their spectral
signature, the absorption and scattering processes
will vary with the wavelength of the incident
radiation [7]. Multispectral observations, therefore,
can be employed to estimate the nature and

23-2

concentration of the water constituents. Passive

sensors working in the visible wavelengths are
commonly used to image water colour. Active
sensors providing their own source of illumination,
can also be used but only from aircraft and for
sampling, rather than for imaging purposes.
The sea covers two thirds of the earth's
surface. To a large extent, man is dependent on it
for food species which include fish, shellfish,
marine mammals, turtles, aquatic plants and algae.
To exploit these resources more effectively,
fishermen must catch the most fish possible while,
at the same time, minimizing costs and optimizing
the scheduling of their operations. Reliable
environmental information is required from the
scientific community for these purposes. Remote
observations of the sea surface can provide a
significant part of the information needed to
assess and improve the potential yield of the
fishing grounds. The environmental parameters
most commonly measured from airborne and
spaceborne sensors are as follows: surface optical
or bio-optical properties (diffuse attenuation
coefficient, total suspended matter, yellow
substance, chlorophyll pigments and macrophytes,
commonly grouped under the general term of
ocean colour); surface temperature; vertical and
horizontal circulation features [8]. Several remote
sensing techniques can provide information
regarding surface circulation features of
importance in defining marine fish habitats. These
include the location and evolution of frontal
boundaries, upwelling areas, currents and
circulation patterns in general. Optical and
thermal characteristics of surface waters can be
used as natural tracers of dynamic patterns.
Variations in environmental conditions affect
the recruitment, distribution, abundance and
availability of fishery resources. It is not possible
to measure remotely the entire range of
information needed to assess changes in the
marine environment. Knowledge of particular
conditions and processes affecting fish
populations, however, may often be deduced
using measurements made by remote sensors, e.g.,
concentration of dissolved and suspended matter,
variations in primary production levels,
distribution of surface isotherms, location of
frontal boundaries, regions of upwelling, currents
and water circulation patterns. The parameters
providing information on these environmental
factors may allow a forecast of fish distribution or
more generally the definition of marine fish
habitats [9]. These are often easier to sense

The Second APEC Workshop of SAKE

remotely than the presence of fish. Estimation of a

fishery resource can be assisted by the
measurement of parameters which affect its
distribution and abundance. Much of the research
dealing with environmental effects related to
fisheries are concerned with the correlation of a
single parameter with the spatial and temporal
distribution of fish. It is most likely, however, that
fish respond to the sum total of environmental
factors. Thus, it becomes necessary to correlate a
large number of parameters, obtained by remote
sensing techniques, with fish distribution.

3. METHODOLOGY
3.1 Knowledge-Based Expert System-GIS
Development
A Knowledge-based Expert Systems GIS is
defined as an integrated GIS and Expert System
that is specially designed to answer the three
questions of a geo-study. A module that integrates
these two components is employed (Figure 1).
The GIS component is composed of two
main modules that facilitate answering the
What and Where questions: GIS database and
spatial analysis. The expert system component is
composed of two main modules for facilitating
answering the Why question: knowledge-base
and inference engine.
The knowledge-base of an expert system is
built based on the result of knowledge acquisition
in the form of production rules. A production rule
is composed of sets of heuristics. One of the
typical characteristics of heuristics is the use of
IFTHEN statements that represent knowledge
or guidelines through which a system may be
operated.
An inference engine is a knowledge
processing tool of the expert system component.
Its main task is to merge facts with rules to
develop or to infer or to draw conclusions about
new facts. If the rules of a knowledge base relate
to a specific domain or expertise, those of the
inference engine pertain to more general control
and search strategy for deriving inferences based
on screening, filtering, and pruning mechanism.

Figure 1: The elements of a knowledge-based

GIS
The most important element of a knowledgebased GIS is an integration module that links the
GIS spatial analysis and inference engine through
GIS database and knowledge base. In this
research, ERDAS Macro Language (EML) was
used for the expert system processing.
3.2 Study Area and Data Acquisition
The study area was located in the coastal
area of Tomini Bay, Central-Sulawesi and South
Sulawesi (Figure 2). As input data, we used the
Sea Surface Temperature (SST), Sea Surface
Chlorophyll-a (SSC) and Turbidity data derived
from MODIS satellite data of NASA. While to
verify the result of Fishing Ground Prediction
Model, we collect the in-situ data of fishing
ground spots in both areas, and the simple
statistical analysis was employed to understand
the percentage of their accuracy.

23-3

Expert System Model for Fishing Ground Prediction

Figure 3: A representational structure of the

Knowledge-Based Expert System
GIS Fishing Ground Model (KBESGIS- FG) ontology.

From the above processes, we formulated the

rules to produce the fishing ground prediction
modules as the expert system as described
bellowed :
Figure 2: Study are in the coastal area of Tomini
Bay (Central Sulawesi) and South
Sulawesi.

3.3.1 Sea Surface Temperature (SST)

o

IF SST (range : 240c - 270c) with different

1.50c, AND diamaeter > 10km, AND
located > 4 mill sea line with area 100 km2
AND deep > 100m, THEN this location is
front.

IF front have V (speed value) = 5 cm/s,

THEN this location is Fishing Ground.

IF SST (240c - 270c) with different 20c,

AND diameter > 30km, and located > 4
mill sea line with area 100 km2, deep >
100m, AND different 50c AND length >
10km,
THEN this location is Eddy Current.

3.3 Design of Knowledge-Based Expert System

GIS (KBES-GIS) Fishing Ground
In this research, we used 3 (three)
oceanographic parameters (SST, Chlorophyll-a,
and Turbidity) as input data of the KnowledgeBased Expert System GIS Fishing Ground Model
(KBES-GIS-FG) to define potential fishing
ground. These parameters were then also
processed to investigate and identify the
oceanographic phenomena (upwelling, front, and
eddy) in the study area that suspected has strong
correlation with the potential fishing ground. In
the application of the KBES-GIS-FG, we used the
daily data of SST, Chl-a, and Turbidity images as
input variables to generate the daily information
of the potential fishing ground formation. The
formulation process of the KBES-GIS-FG
ontology is presented in Figure 3.

IF area with UPWELLING criterion,

THEN predicted area as the Potensial
Fishing
Ground.
o

3.3.2 Chlorophyll
o

23-4

IF Chl-a concentration in the range of 0.3 to

2.5 mg/m3, AND that area is located more

The Second APEC Workshop of SAKE

than 4 mill sea line, AND deep more than

100m, AND acreage is 100km2 , THEN
Upwelling area.
o

Pattern
Recognition:
- FG, non FG
- Ocean Physical
Condition

IF area is called as Upwelling, THEN

Fishing Ground area.

OUTPUT
INPUT

IF Chl-a concentration < 0.3 AND more

than 2.5 mg/m3, THEN Non Fishing
Ground

3.3.3 Turbidity
o

IF turbidity value less than 10mg/lt, AND

located in more than 4 mil sea line, AND
deep more than 100m, AND have acreage
more than 100 km2, THEN Upwelling area.
IF the area is UPWELLING, THEN the
area is Fishing Ground.

IF turbidity have the chlorophyll

concentration between 0.3 to 2.5 mg/m3 ,
AND located more than 4 mill sea sea line,
AND deep more than 100m, AND have the
coverage area more than 100 km2 ,THEN
Upwelling.

IF area is Upwelling, THEN the area is

Fishing Ground.

Prediction Model
(Fishing Ground)

Observation

Analysis and
Interpretation

Modeling and
Testing/Verification

- FG, non FG proportion

- acreage of FG
- productivity
Knowledge-based Expert System

Figure.4. Prediction model of Fishing Ground.

(A proposed cyclical modeling approach)
From this model, we will understand that the
proven of the model performance will depend on
the feedback data from the in-situ field
observation and the knowledge base of the expert
system in which further investigation can be
carried out and adjusted to predict more
accurately fishing ground events and the
oceanographic phenomenon.

4. EXPERIMENTAL DESIGN
3.3.4 Predictive model for Fishing Ground (A
proposed cyclical modeling approach)

The experimental design in this research is

conducted as follows:

The modeling concept used in this research is

based on the cycling model approach (as
illustrated in Figure 4). This model consists of
three (3) stages, as described bellow :

(a). Preparation of the Knowledge-Based Expert

System GIS Fishing Ground Model (KBESGIS-FG) systems.

(a). Observation
stage:
to
define
the
characteristics of fish behavior, physical
condition of ocean and use the pattern
recognition method to cluster fishing
ground and non fishing ground.
(b). Analysis and interpretation of data (SST,
Chl-a, Turbidity images.
(c). Modeling and testing/verification: using a
knowledge-based expert system GIS model
for fishing ground prediction.

(b). Building the knowledge-base and parameter

settings.
(c). Data preparation: Satellite data (SST, Chl-a,
Turbidity), field observation data of fishing
round.
(d). Pre-processing data.
(e). Run model and validate.

5. RESULTS AND DISCUSSION

The scenes result of the KBES-GIS-FG
model in the coastal area of Tomini Bay (Central
Sulawesi) and South Sulawesi and their statistical
analysis were illustrated in Figure 5 and 6,
respectively. Three (3) input data of SST,
Chlorophyll-a, and Turbidity are used in the
model to generate predicted potential fishing
ground. The analysis processes were used the
Knowledge-Base Engineer of ERDAS software.

23-5

Expert System Model for Fishing Ground Prediction

the near border of the different water environment

(front).

A
--------------------------------------------------------------

To understand the accuracy level of the result

model, the simple statistical analysis was
employed by comparing the daily result model
and the daily field observation fishing ground data
in the similar time of the data acquisition and
observation. The result as shown in Figure 6,
indicated that the average percentage of level
accuracy of the model result in both areas of the
coastal area of Tomini Bay, Central Sulawesi and
South Sulawesi was relatively high with accuracy
level of 86 %. While the daily result of the
prediction model in the coastal area of Tomini
Bay, Central Sulawesi and South Sulawesi is
within a range of 68 % to 95 %. As mentioned
above, the variability of the accuracy level of the
model result in the prediction potential fishing
ground seems to be strong corresponded with the
cloud ness level (could cover) in which in the
tropical area to be a problem. To eliminate and
increase the accuracy level in the satellite data
acquisition, the development of technical
processes of the data satellite for the tropical area
was recommended.

Figure.5. Fishing Ground Map produced by the

KBES-GIS-FG Model in the coastal
area of Tomini Bay, Central Sulawesi
(A) and South Sulawesi (B)
The result in Figure 5 shows that the fishing
ground map which produced from the model can
be divided into two (2) categories. The first
category is potential FG area that is marked by red
color dots, and the second category is semi
potential FG area marked by green color dots.
While the black color dots is non FG area. The
KBES-GIS-FG model was applied and run by
considering the marine environmental information
identified by three (3) parameters of daily SST,
Chlorophyll-a, and Turbidity as input data. The
daily and variability result of the model in
generating predicted potential fishing ground area
was strong depend on and governed by the level
of cloud ness (cloud cover) as well as the
meteorological situation. The result also shows
that the potential fishing ground area was mostly
concentrated in the near border (front) of the high
and low concentration level of chlorophyll-a.
This result indicated and well agreement with the
most previous results that shows the potential
fishing ground is well correspond with the area in

23-6

Figure 6. The statistical analysis result of the

model and field data of fishing round.
6. CONCLUSIONS
In this paper, a knowledge-based expert
system GIS model for prediction of fishing
ground was introduced. We developed an expert
system integrated with the GIS model for
prediction of the potential fishing ground of the
economic pelagic fish in the coastal area of
Tomini Bay, Central-Sulawesi and South
Sulawesi by using Sea Surface Temperature
(SST), Sea Surface Chlorophyll a (SSC) and

The Second APEC Workshop of SAKE

Turbidity data derived from Modis satellite. The

results demonstrated that the level of the success
of implementing any knowledge based GIS model
is determined by the completeness of spatial
knowledge concerning the domain expertise and
the sophistication level of the ERDAS macro
programming. Moreover, it was found that the
use of the expert system has allowed the
development of fishing ground model to
understand further of the ocean phenomenon in
relation to the fishing ground characteristic. By
using the knowledge base expert system, the
prediction of fishing ground automatically would
be more easy and accurate.

ACKNOWLEDGEMENT
We would like to acknowledge The
Ministry of Research and Technology (KMNRT)
Republic of Indonesia through the Incentive
Research Program (Program Riset Insentif) for
funded this research. Also we would like to thank
to all members of the Center of Technology for
Natural Resources Inventory (P-TISDA), BPPT
for their support of the project.

Report P-TISDA, BPPT. 2005.

Hendiarti, N., Siegel, H., Ohde, T., 2004.
Investigation of different coastal processes in
Indonesian waters using SeaWiFS data, Deep
Sea Research Part II 51 : 85-97. (2004)
Venegas, R., P.T. Strub, E. Beier, Letelier, T.
Cowles, and A.C. Thomas, Assessing
satellite-derived variability in chlorophyll
pigments, wind stress, sea surface height, and
temperature in the northern California Current
System, J. Geophys. Res. In Press, 2007.
Laurs, R.M. et al., Albacore tuna catch
distributions relative to environmental features
observed from satellites, Deep-Sea Res.,
31(9):108599, 1984.
Kemmerer, A.J., Environmental preferences and
behavior patterns of Gulf menhaden
(Brevoortia patronus) inferred from fishing
and remotely sensed data, ICLARM
Conf.Proc., (5):34570, 1980.

REFERENCES
P. M. Atkinson and A. R. L. Tatnall, Neural
networks in remote sensing, Int. J. Remote
Sensing, Vol. 18, No. 4, 1997.
Ignizio, J.P. Introduction to Expert Systems: the
development and implementation of rulebased expert system. NewYork: McGrawHill, Inc.
Poul Degnbol, The Knowledge base for fisheries
management
in
developing
countries:
alternative approaches and methods, Institute
for Fisheries Management and Coastal
Community
Development, Bergen Norway
Published, 2004.
James C. Hendee, An Expert System for Marine
Environmental Monitoring in the Florida Keys
national Marine Sanctuary and Florida Bay,
Proceedings of the Second International
Conference on Environmental Coastal
Regions, ed. C.A. Brebbia, Computational
Mechanics
ublications/WIT
Press.
Southampton, pp. 57-66, 1998.
Sadly, Muhamad, Assessment and Applicfations
of the Knowledge-based Expert System in
Natural Resources Management, Technical

23-7

Expert System Model for Fishing Ground Prediction

23-8