Discrimination of peatlands and mineral soil lands using multisource remote sensing data

he discrimination of peatlands and mineral soil lands and different peatland canopy types was studied using Landsat TM-, Landsat MSS-, NOAA A VHRR-images, aerogeophysical data and data derived from a digital terrain model. The study area, centered at 63° 28' N, 26° 14' E was located in the Middle Finnish Boreal forest. The field data were 2126 temporary sample plots of the National Forest Inventory. The analysis methods were discriminant and clustering analysis and Tukey's studentized range tests. The classifications, based on discriminant analysis (maximum likelihood classification) were tested using external ground truth data. Peatlands and mineral soil lands were separated in an accuracy of 76.6 percent. The best image variables to discliminate peatlands and mineral soil lands were geophysical variables (Gamma ray intensity of Potassium (K40) and out-of-phase component of electromagnetic data). Without geophysical data the classification accuracy was more than 10 percent lower. Open bogs and poor mineral soil lands were separated well but the peatlands and mineral soil lands with abundant growing stock were mixed. The subgroups of peatlands were separated best using Landsat images only. The proportion of correctly classified field plots was 70 percent. Open bogs and spruce dominated peatlands were separated best. The percentage of correctly classified pixels was 67.4 percent, when peatlands were discriminated on basis of peat type (Sphagnum, Carex). The discrimination accuracy of peatlands on basis of ditching stage was 63 percent.