Wetland Maps of Central Canada based on L-band SAR Imagery

Abstract:

Northern wetlands have the potential to become major sources of greenhouse gases. Detailed and accurate maps of the locations, types, and extents of these wetlands are therefore essential to the development of accurate carbon budgets. However, due to their vast extent and inaccessibility, most northern wetlands remain unmapped.

We have been constructing high-resolution (100 m) thematic maps of boreal wetlands, with current focus on Canadian wetlands. The maps are developed using spaceborne synthetic aperture radar (SAR), which efficiently collects high-resolution imagery over extensive regions and, unlike optical sensors, is unimpaired by clouds or lack of sunlight. Spaceborne L-band (~1.3 GHz) SAR, in particular, records scene characteristics imperceptible to optical sensors such as vegetation structure and moisture content, soil moisture and roughness, and canopy-obscured surface waters. These attributes make it the best single tool for mapping boreal wetlands.

Two L-band SAR-based wetland maps are being assembled: one using HH-polarized imagery from the JERS-1 satellite collected in the winter and summer of 1997-1998, and a second using dual-polarized (HH and HV) imagery from the PALSAR sensor of the ALOS satellite collected in the summer of 2008. Ancillary data layers such as image texture, topographic slope, and proximity to water are also generated, and a training/testing data layer is formed by merging polygons from the Canadian Wetland Inventory (CWI) with other land cover databases. A Random Forests decision tree classifier takes as input the SAR, ancillary, and training/testing data layers and uses them to produce thematic wetland maps. The accuracy of each map is quantified via producer and user error statistics. Finally, the SAR-based wetland maps are compared to form a 1998-2008 wetlands change map.

Recent advances include a powerful new software suite developed to handle huge volumes of data and much-improved JERS-1 registration. Challenges, including non-GIS-ready CWI data and nonstructural CWI class definitions, are being addressed.

This work was carried out in part within the ALOS Kyoto & Carbon Initiative at the University of Southern California and Jet Propulsion Laboratory (JPL) under contract to NASA. JERS-1 mosaics were generated by JPL. PALSAR mosaics were provided by JAXA/EORC.