Boreal Forest Classification Employing Dual-pol PALSAR Imagery
Tom Ainsworth(1), Don Atwood(2) and Franz Meyer(2)
(1) Naval Research Lab, Code 7263, Washington, DC 20375, United States
(2) Alaska Satellite Facility, University of Alaska, Fairbanks, AK 99775, United States
Analysis of dual polarimetric Synthetic Aperture Radar (SAR) imagery has taken on new importance with the recent launches of ALOS PALSAR, TERRASAR-X, and RADARSAT-2. While these spaceborne, polarimetric SARs can provide full polarimetric imagery, the majority of the collected data will be restricted to single and dual polarimetric imaging modes.
This paper focuses on the ALOS PALSAR dual-pol imaging mode, which transmits a horizontally (H) polarized signal and coherently collects both the vertical (V) and horizontal polarized returns. In particular, we develop dual-pol analysis methods and apply polarimetric PALSAR imagery to the problem of boreal forest classification. The boreal biome is environmentally important, yet significantly under-sampled. Boreal forests cover 9% of the earth’s land surface and contain a disproportionate share (23%) of the world’s total carbon. Thus characterization by remote sensing remains a significant challenge for understanding the Earth’s carbon budget.
We investigate dual-pol imagery acquired over a well-characterized region of interior Alaska. Field observations from road-accessible regions along with optical imagery will provide the ground truth, while PALSAR quad-pol imagery provides a baseline forest classification against which the dual-pol classifications are contrasted and compared.
Since the inherent information content of dual-pol data is less than that of quad-pol data, one expects that dual-pol classifications will not be as robust as full quad-pol results. Due to incomplete information content, comprising only HH and HV information, PALSAR dual-pol imagery does not allow a clean distinction between odd-bounce (surface) scattering and even-bounce (dihedral) scattering. However, this distinction is less critical for forest classification, provided that forest scattering depends primarily on the relative strengths of the canopy (volume) backscatter and the ground return. Using a mathematical formalism appropriate for dual polarization, we will demonstrate that PALSAR dual-pol SAR imagery may be sufficient for general forest mapping and forest classification applications in the boreal forest biome.