Land-cover classification and analysis of change using machine-learning classifiers and multi-temporal remote sensing imagery

Abstract

Frequent human activity and rapid urbanization have led to an assortment of environmental issues. Monitoring land-cover change is critical to efficient environmental management and urban planning. The current study had two objectives. The first was to compare pixel-based random forest (RF) and decision tree (DT) classifier methods and a support vector machine (SVM) algorithm both in pixel-based and object-based approaches for classification of land-cover in a heterogeneous landscape for 2010. The second was to examine spatio-temporal land-cover change over the last two decades (1990–2010) using Landsat data. This study found that the object-based SVM classifier is the most accurate with an overall classification accuracy of 93.54% and a kappa value of 0.88. A post-classification change detection algorithm was used to determine the trend of change between land-cover classes. The most significant change from 1990 to 2010 was caused by the expansion of built-up areas. In addition to the net changes, the rate of annual change for each phenomenon was calculated to obtain a better understanding of the process of change. Between 1990 and 2010, an average of 4.53% of lands turned to the built-up annually and there was an annual decrease of about 0.81% in natural land. If the current trend of change continues, regardless of the actions of sustainable development, drastic declines in natural areas will ensue. The results of this study can be a valuable baseline for land-cover managers in the region to better understand the current situation and adopt appropriate strategies for management of land-cover.