Aerial high resolution hyperspectral data for validation of the Edale upland peat moorland burn scar derived by SAR and Optical satellite imagery

Systematic detection and monitoring of upland peat moorland wildfires in the UK is critical to conservation groups to understand the extent of damage and to monitor natural post-fire recovery and peatland restoration measures. The collection of burn scar perimeter data and the rate of vegetation recovery in the field is labour-intensive and time-consuming, limiting the spatial coverage of regular monitoring. Since upland peat moorland areas in the UK are regularly covered by cloud, burned area characterization based on existing optical satellite data is limited. With the launch of Sentinel-1A, Sentinel-2A and upcoming hyperspectral missions (e.g. PRISMA, EnMaP), regular monitoring of moorland areas for burn scar detection is possible using SAR and optical data in synergy.
The aim of this study is to use aerial high resolution hyperspectral data to validate multi-temporal and multi-sensor techniques to compare and contrast the ability of optical (Landsat 7 ETM+) and radar (ASAR and ERS-2) data to detect the small upland peat moorland burn scar located in the Peak District National Park (PDNP), in northwest England. A wildfire on 26 May 2008 resulted in a 0.10 km2 burn scar on the Kinder plateau near Edale, PDNP. The fire moved quickly across dwarf shrubs and moorland grasses, driven by easterly winds and burnt into the peat soil in places (McMorrow et al., 2010). Aerial hyperspectral images from the twin Eagle-Hawk sensors was collected five weeks after the wildfire consisted of imagery by the Natural Environment Research Council (NERC) Airborne Research and Survey Facility (ARSF). These imaging spectrometers operate in the VNIR-SWIR spectral range with a spatial resolution of 1.5m/pixel for Eagle. Unfortunately, atmospheric correction and mosaicking of the airborne imagery was not sufficient for a differenced Normalized Burn Ratio (dNBR) calculation. We carried out a supervised classification of the hyperspectral data by using a Support Vector Machines (SVM) algorithm, as implemented in the EXELIS-HARRIS ENVI 5.1 software to validate Edale burn scar products derived from ASAR and ERS-2 (25m/pixel) and Landsat 7 (30m/pixel). Four classes –burned area, vegetation, bare peat and cloud-covered vegetation –were selected to train the algorithm. A modal filter was applied to the classification result (Figure 1) and prior comparison with both satellite data types (Figure 2). Preliminary results show the feasibility of using SVM classification of hyperspectral image to validate burn scars in upland peat moorland environments delineated by ASAR, ERS-2 and Landsat 7 ETM+ satellite data. Data fusion of optical and radar data is explored for burn scar enhancement are now being explored.