Comparative tests on methods for automatic mapping of land cover dynamics from multi-year images

摘要

Most land use/land cover (LULC) mapping methods require us to collect ground reference data at the time when the remotely sensed data are acquired. However, the high cost of the data collection limits the production of annual LULC maps in a short time span. In this study, in order to reduce the mapping cost and improve the timeliness of the map products, several simple and effective methods were tested and compared with same-year mapping (SY), including cross-year mapping (CY), multiple-indices automatic classification (Ml), and spectral transfer (ST) and index transfer (IT). For the study years of 2001-2016, SY yields accuracies higher than 93%, the mean CY accuracy is about 83%, and the Ml accuracy reaches 88%. Both the overall accuracies (above 90%) and the detailed accuracy indicators of the ST and IT are higher than those of CY and Ml, and very close to the SY accuracy. It can, therefore, be stated that the methods are able to map urban LULC variations automatically, withsatisfactory performance.