Urban expansion modeling using Logistic regression method based on Geomod model (Case study: Kordkuy city)

Nowadays, land use change is one of the major challenges that different countries of the world are faced with. Land use can be expressed as how humans use natural resources to meet their needs. Among all land uses, urban land use is considered as the main cause of changes in natural landscapes, which result in devastating environmental effects, such as the loss of agricultural and forest lands, water and air quality reduction, an increase in the volume of runoff, and so on. Therefore, monitoring and evaluating temporal and spatial changes in this land use category can be considered as an important step towards achieving sustainable development goals and reducing destructive effects on ecosystems. In this way, decision makers in the field of natural resources management as well as urban planners to expand sustainable development strategies need to be aware of the extent of land use changes and urban growth in the past, and predict the pattern of land use change in future periods to prevent some unpleasant changes on the environment. One of the most important strategies for understanding and analyzing spatial and temporal dynamics is the use of various models to identify, analyze and predict the behavior of these land uses over time. The Geomod model is one of applied models to predict land use changes.
Methods and Material: Kordkuy city is located between longitude 54 6 53 E and latitude 36 47 51N. In order to simulate land use change using Geomod model, three steps are considered: Land use change detection; Modeling of Transition Potentials; and Change Prediction and Modeling.
In the first step, Landsat satellite images were prepared in three periods of 1987, 2000, and 2015. Then, the geometric correction method with ground control points was used according to non-parametric method. In the next stage, the training samples were selected with high accuracy and appropriate distribution for the supervised classification based on the maximum likelihood algorithm. Then, the accuracy of the generated maps was determined. In the second step, it was necessary to estimate the potential for the transition of other land use to urban land use according to logistic regression model. Logistic regression is a statistical model that investigates the relationship between the dependent and independent variables. In this research, the dependent variable is urban growth areas in the two study periods. Independent variables include digital elevation model, slope, distance from agricultural land, distance from natural forest, distance from the road, distance from the river, distance from industrial areas, distance from planting forest, distance from the edge of city, distance from village, and distance from city center. Also, to assess the goodness of fit for logistic regression model, Pseudo-R2 and ROC statistics were used. In the third step, it is necessary to predict the pattern of urban growth in future periods based on Geomod model. The Geomod is a land use change model that simulates one-way change of a land use category into another land use category. There are two components in this model: the amount of change in land use and where the change in land use occurs. In order to obtain land use change, multi-temporal satellite images are used, and to model the locations of change, the relationship between the effective factors of degradation and land use change is investigated. Then, the Geomod model simulates the land use change location based on places that have the highest probability of transition.

In order to produce land use maps of the study area, satellite images were prepared for three periods of 1987, 2000, and 2015. In the next step, ETM, TM and OLI sensors are geometrically corrected with RMSE of 0.37, 0.26, and 0.17. Then, land use maps were prepared in four categories including forest land, agricultural land, planting forests, and residential areas using maximum likelihood algorithm. The Kappa coefficient of the land use maps of 1987, 2000, and 2015 was 0.74, 0.73, and 0.82, respectively which was totally acceptable. According to the results of the land use change detection, changes in land use have led to the growth of residential areas in over the studied period, so that 306 hectares of agricultural land and 27 hectares of forests land have changed into the residential areas. Also, the results of logistic regression model showed that in both periods of 1987-2000 and 2000-2015, the distance from agricultural lands had the greatest effect on urban growth. Also, assessment of goodness of fit of model using two Pseudo R2 and ROC statistics with values of 0.37 and 0.93 for the first period, 0.31 and 0.92 for the second period, and 0.39 and 0.93 for the over studied period indicates a good fit of regression relationship. Then using the probability map of the urban growth in the second period and land use map in 2015, urban growth was predicted for 2040 using the Geomod model. Probable prediction of the Geomod model in 2040 indicates a decrease in forest land (5 hectares) and agricultural land (291 hectares), and increase in urban areas.

In this research, the urban growth pattern in Kordkuy city in 2040 was predicted using the Geomod model. The results of this prediction showed that agriculture land is declining at a rapidly and consequently rate, residential areas will increase with high rates, so that 17.6% of agricultural lands have changed into the residential areas from 1987 to 2040. The results showed that the dispersed trend in Kordkuy city has had a profound effect on the destruction of agricultural lands. The issue can be due to the increase in urban population density, the limitation of the spatial arenas of city, and the low price of city's marginal lands, which makes people tend to inhabit in these lands. But the important issue is that the emphasis on the trend of urban sprawl in the agricultural and forestry lands that could bring environmental and economic irrecoverable damage to the region in the future.