Page 58 - Çevre Şehir İklim İngilizce - Sayı 4

Urban Transformation Forecasts with Artificial Intelligence
Based Algorithms

development. A simulation using seven variables for the city of Dongguan
in China shows the results according to planning purposes in the form of
the past development trend, the choice of single-center and multi-center
developments, the choice of agricultural conservation-based development.
Pijanowski et al. (2005) conducted the simulation and calibration of ANN-
based urban change for two metropolitan areas located in the upper Midwest
of America (Detroit and Twin Cities). In this study, a land transformation model
was used, which was developed previously simulating land use change based
on Geographical Information System and ANN.
The urban development of the province of İstanbul was determined by
Özcan (2008) with the ANN and future forecasts were made. With the 1995
master plan, the provinces were divided into digital networks with a resolution
of 0,25x0,25 km, and it was all assumed that ecologically and geologically risky
regions, distance to the 1.,2. and 3. degree city centers, the distance to the
main transportation axes, forest land, population, historical spread rate, areas
with a slope above 20% had an impact on the living. The obtained data were
evaluated through Statistica and designed with a three-layer multi-layer ANN.
According to the model results shown on ARCGIS, it was determined that the
urban sprawl in İstanbul is concentrated in Tuzla and Maltepe districts on the
Anatolian side and in Esenler, Küçükçekmece, Güngören and Silivri districts
on the European side.
Maithani (2009) used constructed and non-constructed area data (basin
photos, remote sensing data produced by the Indian Remote Sensing
Satellite (IRS)-1C and PAN sensor data) produced by the city of Saharanpur
in India between 1993 and 2001 and an urban development model based on
ANN. Facilities of infrastructure, road and access to the city center constitute
the model inputs as important factors. According to the 12 different ANN
architectures modeled, the model reached up to 90.56% training accuracy
and 87.83% test accuracy.
An urban development model was developed by Mohammady et al. (2014)
for the city of Sanandaj in Iran using ANN. The data of the Landsat satellite
between 2000 and 2006 were used in these studies. Basically, the aim of the
study is to determine the change of land use and land cover, to simulate
urban growth under different scenarios, and to provide reference information
for urban planning and land development. In the model where multi-layer
ANN is used, there is one input and output layer and 13 hidden layers. In
the evaluations of the model made by percentage of correct matching (PCM)
and measure of performance (FoM) methods, an evaluation rate of 90.1% and
43.75%, respectively, were found.

Year 2 / Issue 4 / July 2023 47