Examining the Influence of Subarctic Boreal Ground Condition on C-Band RADARSAT-2 POLSAR Variables for Target Separability and the Application of a Support Vector Machine
Accelerated permafrost thaw is being observed in northern peatlands, with profound effects on local hydrology and ecosystem energy balances. Synthetic aperture RADAR (SAR) has shown the capability to monitor these landscape alterations. In this research RADARSAT-2 scenes were acquired over a Northwest Territories study site for examining the sensitivity of ground conditions on C-band polarimetric variables. Soil moisture produced the strongest relationship to σ°HV/HH with low incidence angle imagery (R2=0.66, θ=23.1°), and canopy height to σ°HV with high incidence angle imagery (R2=0.85, θ=48.1°). Peatland landscapes were then tested for separability within SAR imagery using a bootstrapping approach. Results indicated σ°RL (θ=19.4°) best discriminated bogs from fens, and plateaus from uplands. Optimal polarimetric variables were then used for land-surface classifications with a support vector machine (SVM) classifier. A multi-angle approach produced the highest reported accuracy of 92% (θ's=23.1° and 45.8°). Results underline the potential for monitoring permafrost thaw-induced land surface change with SAR imagery.