Land–Use and Land-Cover Change Detection Using Dynamic Time Warping–Based Time Series Clustering Method

Abstract

Accurate and timely monitoring of urban land-use and land-cover (LULC) change is useful for understanding the various impacts of human activity on the urban environment. In order to demonstrate the advantage of time series imaging for urban LULC change detection, we selected time series Landsat images over a two-year period to detect inter-annual changes. A time series trajectory for each pixel was developed by the biophysical composition index (BCI) and normalized vegetation index (NDVI) values extracted from Landsat images. Considering that temporal length of trajectories in different years might be different, the dynamic time warping (DTW) measure was selected as the LULC change magnitude indicator. After DTW-based change/unchanged detection, the DTW-based clustering method was used in LULC change type extraction. Finally, the overall accuracy of change/unchanged detection result was 92.3%, and the overall accuracy of all change types was 71%. Some change types that are difficult to extract by bi-temporal images were detected, such as inter-class changes between farmland and forest, and intra-class change of farmland, indicating the advantage of time series information in LULC change detection.