1038
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing2008
change of Tehran Metropolitan Area over a study period of two
decades. Neural networks with SNNS software were trained
using two datasets; and then predictions were made and the
simulation results were compared to the real condition visually.
For prediction, years 1980 and 2000 were used to train the
network and then predicted land use change for the year of
2020 was accomplished. A series of simulation outputs for the
case study town in the period 1980-2000 were produced.
This paper presents the development of land use change model
by illustrating an application of remote sensing imageries,
neuro-fuzzy and GIS as the inputs of this model. The use of
neural networks has increased substantially over the last several
years because of the advances in computing performance
(Skapura, 1996) and the increased availability of powerful and
flexible ANN software. The main objective of this paper is to
exhibit how GIS and neuro-fuzzy can be used to forecast land
use changes over selected region at a specific period. GIS is
used to develop spatial data layers to be used as inputs to the
ANNs while basic principles of neuro-fuzzy has been used for
land use change modelling. The scheme of this work starts
with design of the neural network and identify the inputs using
a historical data, using subset of the inputs, the network was
trained, then neural network testing was performed using the
full data set of inputs and the final stage was to use the
information extracted from the neural network to forecast land
use changes. In this paper the development of land use change
model for Tehran Metropolitan Area based on neuro-fuzzy was
undertaken.
2. DATA PREPARATION
2.1 Study area
Tehran Metropolitan Area located in North of Iran was selected
for the purpose of this study. The city is considered as the
largest and main economical city of Iran. Tehran is one of the
most rapidly populated area and land use change regions in Iran.
From 1980 to 2000, resident population in the Tehran nearly
doubled. Tehran located on latitude 35° 45' N and longitude of
51° 30' E. Data on land use, transportation, natural features,
public lands and digital elevation were integrated using
Arc/Info 9.2 software.
2.2 Data Sources
Two Landsat TM images of the Tehran Metropolitan Area and
surrounding areas were taken at 1980 and 2000. The images
were geometrically registered and normalized. National
Cartographic Centre (NCC) database developed around 2000
was used as the source of topographic data. NCC topographic
data were integrated with our database to represent the
transportation network and locations of roads to provide the
appropriate inputs to the GIS-based model.
2.3 Data Pre-processing
The historical images were geometrically rectified and
registered to the same projection namely, Universal Transverse
Mercator (UTM) WGS 1984 to lay them over each other.
Figure 1 shows the rectification and registration results as an
example for the 1980 and 2000 images. Registration errors
were about 0.5 pixel.
1980 2000
Figure 1. Rectification and Registration results for 1980 and
2000
2.4 Image Classification
The initial (1980) and final (2000) Landsat imageries were
subjected to a classification of zones. Supervised classification
was utilized to classify the images to different land use
categories. In order to classify the rectified images, four classes
of interest were specified in the images namely, road,
residential area, service centre and administrative area. These
classes were identified using sets of high resolution
orthophotographs over the area and USGS land classification
map as ground references. The overall testing accuracy for the
classification of Landsat TM image (1980) was 82.14%, while
it was 86.46% for Landsat TM image (2000) which Figure 2
shows the image classification results.
Figure 2. Image classification results for 1980 and 2000
3. SIMULATING URBAN LAND USE CHANGE USING
NEURO-FUZZY APPROACH
Land use change for Tehran Metropolitan Area has been
modelled using two urban maps, one from 1980 and the other
from 2000. The land use change model follows four sequential
steps including: (1) processing/coding of data to create spatial
layers of predictor variables; (2) applying fuzzy logic for
spatial layers; (3) integrating all input grids (4) analysis of the
difference between model outputs and real change.
In Step 1, processing of spatial data, generation series of base
layers were established within a GIS of exhibiting different
land uses. In step 2, generation of inputs to neural network,
achieved from fuzzifications of spatial layers that was prepared
in previous stage. Step 3, integration of predictor variables is
required using ANN method. Step 4, spatial error analysis