Modeling dynamic changes of Land Use with Object Based Image Analysis and CA-Markov approach (Case study: Shiraz city)

Land use is one of the most important indicators of economic and social development in urban areas, and has resulted in extensive changes in available structures and procedures of these areas. Therefore, human activities are known as one of the main principles and components of change in land use. Generally, land use changes are inevitable product of interactions between human activities and environmental elements. Remote sensing technology with capabilities such as providing update and reliable information about natural and urban areas, digital processing of satellite imageries, providing the possibility of temporal and spatial comparing of different phenomena, diversity of products, and etc. is considered to be a powerful tool in improving the efficiency of urban management. Consequently, remote sensing data are used to determine type, quantity and location of land use changes. Moreover, remote sensing technology is used extensively in land use maps all over the world. Many models have been applied to predict land use changes, which due to the complex, dynamic, and non-linear nature of the issue gained little attention. However, CA-Markov model, which is a combination of Markov chain and cellular automata, is commonly considered to be an appropriate and good method for spatial-temporal modelling of land use changes. In the present study, land use changes were investigated for a 15-year period in Shiraz using object- based image analysis. Then, a land use map was produced using cellular automata-Markov (CA-Markov) model to predict land use changes in the study area in 2020.
Material & Methods: The present study includes two main phases. In the first phase, land use map of Shiraz was produced using Fuzzy object based analysis of satellite imageries. In the second phase, modeling and predicting of land use changes in 2020 were performed. Landsat imageries of the study area in 2005, 2010 & 2015 were used in this research. After preprocessing and preparing the imageries, segmentation procedure was performed as the first stage of object based classification using multiresolution segmentation algorithm.  The nearest neighbor algorithm was used for object based classification of satellite imageries. Classification conditions were defined in accordance with each class properties, and classification was performed based on fuzzy operators of the classification operation. In CA-Markov model, the possibility of changing from one class of land use to another was calculated using transfer matrix table. Then, land use map of future years will be predictable in accordance with the transfer probability matrix, and desired time interval.
Result & Discussion: In this study, scale parameter of 10, shape index of 0.4, and compactness index of 0.2 were extracted as the optimum conditions for segmentation. Apart from spectral data, information regarding the location, context, texture, normalized difference vegetation index, enhanced vegetation index, and digital elevation model were also used to improve the efficiency of classification phase. The results of model validation shows an overall accuracy of 89% and kappa coefficient of 0.87. Therefore, the results of CA-Markov model shows a very good potential for predicting land use changes in Shiraz. Thus with the adjustment and calibration of model parameters and based on land use maps of 2010 and 2015, Shiraz land use in 2020 was predicted. Due to the complexity of modeling dynamic changes in urban land use, utilizing efficient and update methods of data analysis is crucial. Therefore, satellite imageries and object based image analysis techniques were used to prepare land use map of Shiraz based on the data collected over a 15 year period. By considering the defined land use classes (residential area, barren lands, street network and urban green space), optimum image segmentation parameters were found. Then, classification conditions were defined for each class using the nearest neighbor algorithm and fuzzy operators. In this way, image classification was performed. By analyzing land use changes during the 20-year period, we understand that residential area has increased from 38 square kilometers in 2005 to 142 square kilometer in 2020. Additionally, green space area faced a reduction of 4 km in the first 5 years of the period, while in the next 15 years green space area shows an increasing trend.