A Decision Tree Classifier for the Monitoring of Wetland Vegetation Using Aster Data in the Poyang Lake Region, China

This paper examines the applicability of binary decision tree (DT) classifier and ASTER data for the monitoring of wetland vegetation at plant family level (eight dominant plant families in the study area, bare soil, marshland, residential area, and waterbody) in the Banghu Lake, a seasonal lake in the Poyang Lake region. Two sets of ASTER Level-1B Registered Radiance at the Sensor products, ASTER On-Demand L2 Surface Kinetic Temperature products, and ASTER Digital Elevation Model products on April 17, 2006 and July 3, 2005 were used in this study. In addition to the reflectance of VNIR and SWIR bands, environmental variables which can be derived from ASTER products such as vegetation indices, water indices, topographic information, land surface temperature, and principal components were selected as the inputs of DT classifier. Field data collected in December, 2007 is grouped into training and testing samples for DT classification. DT performed poorly compared to those of maximum likelihood classification and support vector machines with the reflectance of VNIR and SWIR bands in single date. The classification accuracy was slightly improved by adding environmental indices and variables derived from ASTER products. In particular, topographic information, such as elevation, slope, and aspect, increased the classification accuracy the most. The classification accuracy is dramatically refined with the combination of multi-temporal ASTER inputs. However, mainly due to the problems with the training pixels for each class, overall and individual class accuracies remain low.