International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004
Data driven weighting has less blunders than Knowledge
driven, but its correct operation depends on primary existent
sites accuracy. In our evaluation, because of inaccuracy
locations in the primary industrial estates layer, we used data
and knowledge driven weighting simultaneously. The results of
data weighting are shown in Table 1.
3.5 Execution of Models
At first, weighted factor maps must be integrated by using
integration models. For this purpose, four programs were
written and user interface was customized for running Index
Overlay, Fuzzy Sum, Fuzzy y and Genetic models.
Index Overlay execution : This model was executed in two
stages. First, in each class, factor maps were integrated with
respect to Table 1 and were resulted in four class factor maps.
Then, output factor maps were integrated using designed
interface.
Fuzzy Sum execution : Fuzzy Sum model was executed based
on Fuzzy Sum operator described in section 2.3.2 similar to
Index Overlay model.
Fuzzy y execution : Fuzzy y model was executed based on its
operator described in 2.3.2. Important problem in its operator
was determining correct y. For this purpose, we wrote an
auxiliary program that for each input y it produced output map
and computed dependency value between output map and
existent industrial estates layer using Equation 9. Then,
Convenient y was determined by changing y value between
range 0 to 1 and comparing dependency values (Figure 3).
1
og 1 Dependency value
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1 :
BEL,
0 01 02 03 04 05 06 0.7 0.8 09 4
Figure 3. Dependency Value diagram
According to Figure 3, the best y occurred in 0.74 with
maximum dependency value. Finally, factor maps were
integrated by entering y 70.74 and running Fuzzy y program.
Genetic execution : In this model, first, primary solutions zone
was defined. Then, Fitness function was introduced as below.
IGwy-Gwq.i«- 0.174 (10)
where Gwu= The weight of genes in the new generation
Gw.;y= The weight of corresponding genes
In the previous generation
In Equation 10, 0.174 was determined based on subjective
experiences and amount of dependency between factor maps
and primary selected solutions.
3.6 Evaluation of Models
In this stage, models execution results are evaluated and the
optimum model is determined. Performance and accuracy are
two important parameters that usually influence on GIS models
execution routine. Model performance depends on operation of
reference programs algorithm that is used for running model
and entering data volume capacity. Because of similarity of
entering data volume capacity and using one programmer and
programming language, running time was selected for
evaluating models execution routine. Then, running time was
computed using a precise chronometer and doing accurate
observations (Table 2).
Table 2. Running time for each model
i Running E "Running
i Model "lime(see) Model "Time(sec)
Index overlay 36.33 Fuzzy Y 196.64
Fuzzy Sum 194.39 Genctic 128.61
According to Table 2:
- Index Overlay has the least running time with
comparing to other models. The reason for this can be
originated from its operator linear operation.
- . Fuzzy models are slower than Index Overlay because
their operators have more complexity calculation and
non-linear properties.
- Genetic model is slower than index overlay because
its algorithm depends on entering factor maps in each
repetition step.
- . Fuzzy y is slower than Fuzzy Sum because of its less
computing operations.
Therefore, Index overlay has the best running time for industrial
estates location. It's necessary that mentioned, in MIMGIS
software, it is important that analysis have had minimum
running time value because of MIMGIS software is run on the
network workstations and clients must execute models in the
minimum executing time.
Accuracy of models is another parameter that is important for
selecting the optimum model. Therefore, we need a criterion
for comparing accuracy of output model results. In this case
study, existent industrial estates were selected as comparing
criterion and their location appropriateness were determined.
By paying attention to experts' ideas for each existent industrial
estate in study area, status location value was determined based
on its annual efficiency and position affects on this parameter.
Status location value was defined as very good (A), good (B),
bad (C) and very bad (d). After that, with respect to experts
ideas each output map was classified in four classes as A (0.7-
1), B (0.5-0.7), C (0.3-0.5) and D (0-0.3). Then, by comparing
location status value and output maps estimated status, each
correct estimating was identified by "+" and each incorrect
estimating was identified by "—" (Table 3).
W
ne
In
