Authors: Hang Li*, Indiana State University -Department of Earth and Environmental Systems , Qihao Weng, Indiana State University- Department of Earth and Environmental Systems
Topics: Land Use and Land Cover Change, Land Use, Remote Sensing
Keywords: Land cover, Time series, Landsat, Change detection
Session Type: Paper
Start / End Time: 8:00 AM / 9:40 AM
Room: Taylor, Marriott, Mezzanine Level
Presentation File: No File Uploaded
Monitoring land use and land cover change (LULC) in urban area is crucial to urban planning and policy making. Previous studies were inclined to just use images with high quality as base images and discard a large number of images with cloud contamination. Except the spectral features, the temporal feature for each land cover like seasonal component and trend component can also be selected as indicators to improve classification accuracy. This research chose Indianapolis, IN as our study area which had an obvious phenology and applied Continuous Change Detection and Classification algorithm (CCDC) to do land cover classification using all available Landsat data from 1998 to 2017. The algorithm could extract the seasonal, trend components and abrupt changes from 419 images. And the Random Forest (RF) classifier could generate annual land cover images with chosen features from base images. However, the results showed that classification accuracies were not good when images were from Landsat 8 sensor. The wave ranges for each band in Landsat 8 could not totally match those in Landsat 5 and Landsat 7. After rectifying the reflectance of Landsat 8, the improved surface reflectance was put back into CCDC algorithm again. Finally, the classification accuracy with improved reflectance outperformed the old one without being rectified. And the research could extend the use of CCDC on Landsat 8 sensor.