ESA       
   workshops       
envmail@esa.int

 

Evaluation of ALOS PALSAR Data for Forest Classification

David Goodenough(1), Hao Chen(1), Andrew Dyk(1) and Ashlin Richardson(1)

(1) Natural Resources Canada, 506 Burnside Road West, V8Z 1M5, Victoria, BC, Canada

Abstract

Multiple ALOS PALSAR L-band quad-pol images were acquired in the summer of 2007 over the study area in the mountainous high biomass temperate coastal rainforest of Vancouver Island in British Columbia. The rugged topography of the study area makes forest landcover classification of the PALSAR images a challenging task. The polarization and polarimetric information in the data can be processed in many ways and multiple approaches exist for classifying the PALSAR imagery for forest landcover mapping. This paper is concentrated on the evaluation of several processing and classification approaches with the PALSAR quad-pol data for landcover mapping. The polarimetric SAR processing techniques to be examined include the Faraday rotation correction, quad-pol data compensation for terrain azimuth slope variation, quad-pol data filtering and decompositions. Due to the difficult environment of rough topography relief in the study area, the processing emphasis was on integration of an Eigen decomposition filtering technique introduced by Shane Cloude to isolate three types of land cover (water surface, vegetated surface and forest cover). The PALSAR quad-pol data classification approaches include applying different classifiers, such as the Maximum Likelihood classifier, non-parametric LOGIT classifier and Wishart classifier for an improvement of PALSAR quad-pol data classification results. The accuracy of these classification approaches is demonstrated on multitemporal PALSAR quad-pol data sets (level 1.1 data, Pauli RGB and filtered imagery). Radarsat 2 quad-pol images were acquired in 2008 over the same test site and classified using methods similar to PALSAR. The classification results for the various data sets are compared with the ground truth measurements and our hyperspectral classification results.

 

Symposium presentation

 

  Higher level                 Last modified: 07.05.06