Multitemporal classification of TerraSAR-X data for wetland vegetation mapping

Abstract

This paper is concerned with wetland vegetation mapping using multitemporal syn- thetic aperture radar imagery. Although wetlands play a key role in controlling flooding and nonpoint source pollution, sequestering carbon and providing an abundance of ecological ser- vices, knowledge of the flora and fauna of these environments is patchy, and understanding of their ecological functioning is still insufficient for a reliable functional assessment on areas larger than a few hectares. The aim of this paper is to evaluate multitemporal TerraSAR-X imagery to precisely map the distribution of vegetation formations considering flood duration. A series of six dual-polarization TerraSAR-X images (HH-VV) was acquired in 2012 during dry and wet seasons. One polarimetric parameter, the Shannon entropy (SE), and two intensity parameters (σ° HH and σ° VV), which vary with wetland flooding status and vegetation roughness, were first extracted. These parameters were then classified using support vector machine techniques based on a specific kernel adapted to the comparison of time-series data, K-nearest neighbors, and decision tree (DT) algorithms. The results show that the vegetation formations can be identified very accurately (kappa index ¼ 0.85) from the classification of SE temporal profiles derived from the TerraSAR-X images. They also reveal the importance of the use of polarimetric param- eters instead of backscattering coefficients alone (HH or VV) or combined (HH and VV). © 2014 Society of Photo-Optical Instrumentation Engineers (SPIE) (DOI: 10.1117/1.JRS.8.083648)