Random forest classification of urban landscape using Landsat archive and ancillary data: Combining seasonal maps with decision level fusion

Abstract

Mapping landscapes in a rapidly urbanizing region can contribute significantly to quantifying, monitoring and understanding the complex process of urbanization. However, mapping such urban areas is a challenging task due to issues of spatial heterogeneity and dynamic land use practices. In this study we propose an operational mapping algorithm using multi-season Landsat and ancillary data with minimum image pre-processing and limited training samples. The methodology was applied to produce a detailed land use land cover (LULC) map of National Capital Region of India. Seasonal maps (with nine LULC classes) were produced by using Random forest (RF). A second classification involving seasonal maps with decision level fusion based on expert knowledge resulted in an annual composite map with increased number (eleven) of LULC classes. These detailed maps have moderately high (>60%) overall accuracies. The maps generated over different seasons are especially significant in identifying areas with mixed land use practices (like agriculture) occurring over an annual cycle. The annual map as the end product of the decision fusion summarizes the LULC dynamics of the study area with the help of eleven LULC classes. The significance of this work lies not only in generating accurately classified LULC maps, but also in detecting the seasonal dynamics of land use practices in a complex urbanizing landscape. Furthermore, reproducibility of the developed methodology will aid the extension of research for different time periods and with newer sensors in investigating the patterns and dynamics of land use and urban planning activities.