Multitemporal Land Use and Land Cover Classification from Time-Series Landsat Datasets Using Harmonic Analysis with a Minimum Spectral Distance Algorithm

Abstract

An understanding of historical and present land use and land cover (LULC) information and its changes, such as urbanization and urban growth, is critical for city planners, land managers and resource managers in any rapidly changing landscape. To deal with this situation, the development of a new supervised classification method for multitemporal LULC mapping with long-term reliable information is necessary. The ultimate goal of this study was to develop a new classification method using harmonic analysis with a minimum spectral distance algorithm for multitemporal LULC mapping. Here, the Jiangning District of Nanjing City, Jiangsu Province, China was chosen as the study area. The research methodology consisted of two main components: (1) Landsat data selection and time-series spectral reflectance reconstruction and (2) multitemporal LULC classification using HA with a minimum spectral distance algorithm. The results revealed that the overall accuracy and Kappa hat coefficients of the four LULC maps in 2000, 2006, 2011, and 2017 were 97.03%, 90.25%, 91.19%, 86.32% and 95.35%, 84.48%, 86.74%, 80.24%, respectively. Further, the average producer accuracy and user accuracy of the urban and built-up land, agricultural land, forest land, and water bodies from the four LULC maps were 92.30%, 90.98%, 94.80%, 85.65% and 90.28%, 93.17%, 84.40%, 99.50%, respectively. Consequently, it can be concluded that the newly developed supervised classification method using harmonic analysis with a minimum spectral distance algorithm can efficiently classify multitemporal LULC maps.