The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing2008
1039
resulted from comparing output of the forecasted model against
known land use changes. GIS has been used to overlay model
forecasted and known land use changes to calculate the
percentage of cells that the model correctly identifies as
transition to urban area.
3.1 Variables of land use change in GIS environment
After preparation of land use maps for Tehran Metropolitan
Area between 1980 and 2000 years, selection of appropriate
parameters by expert people were performed. The first step in
assessing the variables is to determine the factors affecting the
suitable land use change on the basis of an analysis of existing
studies. Input layers represent phenomena which may influence
the model. In our case, we assume that the following 5 drivers
will influence land use change in Tehran Metropolitan Area
including: slope, proximity to residential area, commercial
place, service centre and roads. Variables also include average
yearly precipitation and elevation.
These parameters were inserted to ArcMap software. Distance
function has been used for road parameters. After these
calculations, different layers in ArcMap have been stored.
Exclusionary cells are cells which are not going to be included
in the analysis. These driving predictor variables and the
exclusionary zones were compiled in Arc/Info Grid format.
Effective parameters require main considerations and criteria
listed as follows:
Absorbing Excursion Spaces: Absorbing excursion spaces
contain distance from administrative and service centres. The
distance each cell was from the nearest absorbing cell was
calculated and stored as a separate variable grids. It is assumed
that the costs of connecting to current absorbing services
decrease with distance from urban areas.
Transportation: It is another important factor which the
distance of each cell had from the nearest road cell calculated
and stored in separate coverage. The value of driving variable
grids represented the potential accessibility of a location for
new development.
Landscape Features: Landscape topography is an influential
factor contributing toward residential use. The amount of
topographic variation surrounding each cell was estimated by
calculating the standard deviation of all cell elevations within a
4 km square area. Cells containing larger values reflect
landscapes that contain a greater amount of topographic relief
around them.
Exclusionary Zones: This group includes limitations
considered for land use change modelling in Tehran, which
include existing urban areas, urban expansion plan, green
spaces, historical and cultural centres, specific buffer for
hospital and mosques, and other pious legacies mentioned in
the comprehensive plan of Tehran.
3.2 Neuro-fuzzy approach to simulate urban land use
change
This part shows that how the GIS based neuro-fuzzy approach
has been used to simulate urban land use change. Output layers
of previous stage were used as input for this stage. After the
extended variables were introduced in different layers, we can
apply fuzzy function in Spatial Analysis Tools for each layer
separately. The main objective of this stage is fuzzification of
layers that had been prepared in previous stage. Definition of
fuzzy function which was used for fuzzification was performed.
A new Toolbox in ArcGIS has been developed to perform the
requested calculations. Accordingly, outputs from GIS
calculation are the input files for the neural networks, the same
spatial features and spatial rules were applied. For definition of
fuzzy function, the views of some experts in addition to the
parameters considered have been taken into account. For
example, according to distance parameter, near distance fuzzy
function was defined in ArcMap in spatial analysis tools while
fuzzy driving variable grids are shown in Figure 3.
Fuzzy Distance to Road Fuzz y Distance “>
Administrative Area
Fuzzv Slope
Figure 3. Fuzzy driving variable grids produced by the GIS
In order to perform prediction in neural networks, training and
testing phase should be done carefully. In training phase
presenting input values and adjusting the weights applied at
each node has been considered. The testing presents a separate
data set to the trained network independently to calculate the
error rate. ANNs were applied to the prediction of land use
change in four phases: (1) design of the network and of inputs
from historical data; (2) network training using a subset of
inputs; (3) testing of the neural network using the full data set
of the inputs; and (4) using the information extracted from the
neural network to forecast changes. In this study, the proposed
neural networks have only three layers each - the input layer, a
hidden layer and the output layer which Simple Back
propagation algorithm was used as the learning process. SNNS
(Stuttgart’s Neural Network Simulator) version 4.2 was used
for design; training and prediction of the ANN (Zell et al., 1996)
shown in figure 4. Difficult learning tasks can sometimes be
simplified by increasing the number of hidden layers, but
according to Gong (1996), a three-layer network can form any
decision boundaries. The neural network has flexibility to
choose optional number of inputs depending on the number of
predictor variables presented to it, an equal number of hidden
units as input units and a single output.