A novel cover-to-use method to map rural settlements from Landsat images incorporating semantic information

Keywords: Landsat, rural settlements, classification, land use, land cover, mapping
Abstract Type: Virtual Paper Abstract

Abstract

Mapping rural settlements accurately and timely is of great significance to understanding the historical evolution of rural communities and making policies for sustainable rural development. Landsat provides us with the longest archives of satellite images at a spatial resolution of 30m, which is the only remote sensing data source for monitoring the trajectory of rural settlements in the past 40 years. However, the existing 30m land cover products based on Landsat images perform poorly in rural areas, thus we face an urgent need of developing an accurate rural settlement mapping method based on Landsat images. Therefore, this paper proposed a novel cover-to-use (C2U) approach for detecting rural settlements from Landsat images and tested it in China. The key innovation of the method is to convert pixel-based land cover to object-based land use (i.e., rural settlement or non-rural settlement) by incorporating semantic information, such as spatial topology of nearby land cover objects. Our results show C2U method performed well at three experiment sites with F1 scores of 0.89, 0.84, and 0.87, which is much better than existing free-accessed Landsat-based land cover products. Importantly, our approach can map rural settlements as spatial-integrity objects rather than a few pixels within rural settlements in existing products, which follows the definition commonly accepted by geographers. Moreover, C2U has good spatial and temporal transferability. This study suggests that 30-m Landsat images have great potential for mapping rural settlements and the proposed C2U method can support the study of rural changes over large areas and long periods.