Publications

Tai, Xiaonan; Wang, Le (2014). Develop an Ensemble Support Vector Data Description method for improving invasive tamarisk mapping at regional scale. INTERNATIONAL JOURNAL OF REMOTE SENSING, 35(19), 7030-7045.

Abstract
Non-native species threaten an ecosystem by interfering with ecosystem processes. Tamarisk is a problematic invasive species in the southwestern USA that has replaced a large amount of the native cottonwood species and altered the water regimes. Quantifying the spatial patterns at regional scale is essential for the in-time monitoring of invasion, given its fast-spreading characteristics. Suitability mapping is useful for estimating the potential distribution of a specific species at the regional scale, but is limited in providing real-time updates since it relies on the generation of related environmental factor products. One-class mapping techniques and public volunteered species occurrence locations together make remote-sensing imagery very suitable for the timely mapping of the spatial distribution of a species. However, the quality of volunteered locations has rarely been fully examined before using them as the ground truth for classification. This research develops a new method - Ensemble Support Vector Data Description (SVDD) - to incorporate volunteered data evaluation into one-class mapping. It can quantify the uncertainty associated with each volunteered sample based on a deviation index and improve the user's accuracy of one-class mapping. This method is applied to map the distribution of tamarisk at the regional scale using Moderate Resolution Imaging Spectroradiometer (MODIS) imagery. Results indicate that this method can generate producer's accuracy comparable to the original SVDD procedure, while providing higher user's accuracy.

DOI:
10.1080/01431161.2014.965283

ISSN:
0143-1161