New Eyes in the Sky: Cloud-Free Tropical Forest Monitoring for REDD With ALOS/PALSAR

Josef Kellndorfer, The Woods Hole Research Center, josefk@whrc.org (Presenting)
Wayne Walker, The Woods Hole Research Center, wwalker@whrc.org
Claudia Stickler, The Woods Hole Research Center, cstickle@ufl.edu
Katie Kirsch, The Woods Hole Research Center, kkirsch@whrc.org
Masanobu Shimada, Japanese Aerospace Exploration Agency (JAXA), shimada.masanobu@jaxa.jp
Ake Rosenqvist, European Commission - Joint Research Center (EC-JRC), ake.rosenqvist@jrc.it
Daniel Nepstad, The Woods Hole Research Center, dnepstad@whrc.org
Paul LeFebvre, The Woods Hole Research Center, paul@whrc.org

1. INTRODUCTION
Much of the discussion at the recent United Nations Climate Change Conference in Bali, Indonesia, focused on monitoring tropical deforestation and the critical role that remote sensing systems will play in the development of REDD (Reduced Emissions from Deforestation and Degradation) mechanisms - policies designed to compensate rainforest nations for avoiding deforestation. Most of the world&rsquos tropical forest countries still lack high-quality maps of forest cover for multiple reasons including chronic cloud cover, and run the risk of being excluded from REDD. In January of 2006, the Japanese Space Agency (JAXA) launched their newest space-borne Earth observing platform, the Advanced Land Observing Satellite (ALOS) featuring PALSAR (Phased Array L-Band Synthetic Aperture Radar) - one of the most advanced imaging radar sensors currently deployed for civilian Earth observation, and well suited to global forest observation. The work presented here represents one of the first successful attempts to generate large-scale forest/non-forest and change detection map products from ALOS/PALSAR image data. The products span some 400,000 square kilometers of southeastern Amazonia, representing the state-of-the-art in regional to continental-scale radar-based forest monitoring.

2. IMAGE MOSAIC GENERATION
Using synthetic aperture radar (SAR) data acquired between 6 June and 27 July 2007 by the ALOS/PALSAR, we have successfully generated one of the first wall-to-wall, cloud-free image mosaics at 25 m resolution for a large portion of the Amazon basin focusing on the Xingu River headwaters in the Brazilian State of Mato Grosso. The mosaic was produced using 116 L-band (~23 cm) SAR scenes acquired in Fine Beam Dual (FBD) mode, i.e., dual polarization HH (horizontal-horizontal) and HV (horizontal-vertical), at an incidence angle of 39 degrees and a pixel spacing of 9.4 m (range) by 3.2 m (azimuth). Orthorectification (SRTM DEM-based) and radiometric calibration were performed using the SARscape (ITT-VIS/SARMAP) SAR processing package installed on a five-node high-performance computing cluster running Enterprise Red Hat Linux. Image mosaicing was performed using software routines written in-house. All processing, i.e., from raw imagery to final mosaic, was completed in less than three days.

3. LAND COVER MAPPING/CHANGE DETECTION
The mosaic is used to generate a land cover classification focused on the accurate delineation of forest and non-forest regions. Data from 879 field plots representing 12 land cover classes (i.e., forest, degraded forest, regenerating forest, riparian forest, cerrado, degraded cerrado, regenerating cerrado, open pasture, degraded pasture, agriculture, wetland, and open water) are used for calibration and validation of the image-object (eCognition) classification. Algorithms based on classification trees and support vector machines are compared as part of the classification analysis. Additionally, a map of land cover change focused on deforestation/regeneration is being produced for the period 1996-2007 using the 2007 ALOS/PALSAR mosaic and the 1996 Japanese Earth Resources Satellite (JERS) acquisition from the same region. Given the excellent positional accuracy of ALOS and the availability of advanced processing methods, value-added products such as these are certain to strengthen existing global monitoring efforts.