An intelligent method to identify land use changes in an urban watershed in Iran
摘要
Urban growth and its monitoring are some of the most important considerations of cities and municipalities spatially in urban watershed management. Changes in land use are essential factors in the relationship between precipitation and runoff. Remote sensing and related technologies are new tools recently used to monitor urban growth and land use changes. This study aims to identify changes in four classes of land use, residential, rangeland, barren land, and irrigated agricultural land in the Gonabad urban watershed in the period 1987 and 2017 using Landsat images. Some known areas in both the 1987 and 2017 images are selected by specialists and used as ground truth. Different algorithms are applied to the ground truth data to find the best classification model. Without losing the generality, 50% of the pixels of each class are selected randomly for training. By analyzing the results, Support Vector Machines (SVM) are found to be the most powerful algorithms. Among them, Optimizable SVM is the most suitable model with an accuracy of 97.1 to 99.3%. In the next step, several SVM models by different randomly selected values are trained by ground truth data and asked to predict the entire pixel classes. According to the results, the greatest increase has occurred in the land use of barren land (111.36%). Residential areas have also increased by 39.91% after 30 years. But the area of irrigated agricultural lands and rangelands has decreased. Based on the results, the impermeable surface area increased from 7.01 km2 in 1987 to 9.82 km2 in 2017. Most changes in the land use of irrigated agricultural lands have occurred and have been converted into residential. Therefore, the irrigated agricultural lands are more vulnerable to converting residential into urban development in the study area.