Change detection of Phragmites australis distribution in the Detroit Wildlife Refuge based on an iterative intersection analysis algorithm

Abstract

Satellite data have been widely used in the detection of vegetation area changes, however, the lack of historical training samples seriously limits detection accuracy. In this research, an iterative intersection analysis algorithm (IIAA) is proposed to solve this problem, and employed to improve the change detection accuracy of Phragmites area in the Detroit River International Wildlife Refuge between 2001 and 2010. Training samples for 2001, 2005, and 2010 were constructed based on NAIP, DOQQ high-resolution imagery and ground-truth data; for 2002-2004 and 2006-2009, because of the shortage of training samples, the IIAA was employed to supply additional training samples. This method included three steps: first, the NDVI image for each year (2002-2004, 2006-2009) was calculated with Landsat TM images; secondly, rough patches of the land-cover were acquired by density slicing using suitable thresholds; thirdly, a GIS overlay analysis method was used to acquire the Phragmites information in common throughout the ten years and to obtain training patches. In the combination with training samples of other land cover types, supervised classifications were employed to detect the changes of Phragmites area. In the experiment, we analyzed the variation of Phragmites area from 2001 to 2010, and the result showed that its distribution areas increased from 5156 acres to 6817 acres during this period, which illustrated that the invasion of Phragmites remains a serious problem for the protection of biodiversity.