Ireland, Peatlands, Peat extraction, Landsat
Eight Landsat TM images, of spatial resolution 30m, were used to produce the national scale exposed peatlands map. Images from three swaths dated 8th August 2003, 20th October 2003 and 17th July 2006 were mosaiced, with each date mosaic classified independently using a Maximum Likelihood classification procedure on the raw Landsat TM bands 1 to 5 and 7 to identify areas of exposed peatland, possible vegetated peatland and non-peat. The three mosaics were then combined to produce a single map which was cleaned to remove spurious areas of misclassification, a cloud mask was applied and the final product reprojected into UTM zone 29N co-ordinates. While these images were accepted to be non-optimal in terms of their historic date and different season of acquisition, they represented the best cloud-free data
available for the whole of the country. A visual accuracy assessment using 500 points selected on the Landsat peatlands map compared with Google Earth imagery showed the peatlands map to have an average accuracy of 82%, with over 97% of exposed peatlands correctly identified. When compared with the CORINE 2006 dataset and the Derived Irish Peatlands Map (Connolly et al., 2007) the accuracy was lower although still typically in excess of 75%, however these sources have their own inherent errors and are not derived from ground surveys, as well as being of a lower spatial resolution and based on more historic data. It was concluded that the Landsat data allow for an accurate, national scale exposed peatlands map to be produced, but there will be difficulties experienced in annually updating this baseline using Landsat data prior to the launch of Landsat 8. There are no other freely available data of the same spatial resolution and swath width as Landsat, thereby necessitating additional expense, e.g. DMC imagery, costing in excess of €7000 for national coverage. No suitable summer cloud-free SPOT images from 2009 were identified as suitable for classification, and the March 18th image proved too early in the season for adequate discrimination of vegetative habitats. SPOT images of the Irish midlands were acquired from 7th June 2007 and 12th May 2008, however both had some cloud cover, so selected areas were
subset to explore the potential and accuracy of regional peat mapping. The same
classification, cleaning and accuracy assessment procedures were applied as for the Landsat images. 5m panchromatic and 10m multispectral images were fused together to give higher resolution information in both domains, increasing the accuracy with which generic classes of intact, semi-vegetated and exposed peat could be classified at both 5 and 10m resolution. Note that with only four spectral bands SPOT did not prove capable of discriminating habitats to species level. The 5m fused image had an overall accuracy of 84.4%, with few errors of omission for individual peatland classes, however some errors of commission were present where non-peatland classes were inadvertently labeled as such. Two SPOT images of the same location acquired 11 months apart were classified to show changes over that time, demonstrating the value of SPOT for monitoring change in specific peatland areas over time. On that basis, it can be concluded that high spatial resolution satellite images are very beneficial to the identification of small scale, marginal changes in particular areas of interest at a cost of up to €5000 per 60x60km image, although prices are significantly cheaper for research purposes. It was suggested that microwave images from TerraSar –X might provide an interesting avenue for future work.