After the first introductory posting we are now presenting our first classification results of County Longford (Figure 1). The maps below indicate the land-cover of Longford, subdivided into five different classes: Grassland, Forest, Water, Settlement and Peatland. A 7-year MODIS 13Q1 time-series on the left, as well as a 1-year Envisat ASAR time-series on the right were used in the analysis. The influence of the pixel size (250 meters and 75 meters) can be clearly seen between both maps (see Figure 2 for a subset of the classification maps). It can also be seen that Peat tends to be overestimated in the MODIS map compared to the ASAR map. However, the Radar data misclassifies Peat as Settlement or Grassland in some instances.
The overall classification results are promising. MODIS achieved an overall accuracy of 93.8 % and a kappa of 0.904 in a 5-fold cross-validation. ASAR reached an overall accuracy of 91.7 % with a kappa of 0.887. In both cases Random Forest was used as the classification method.
The next steps of our analysis are focused on the discrimination of “improved” and “semi-improved” grasslands in County Longford. Furthermore, the currently pixel-based approach will be adapted to an object- or parcel-based classification and reporting scheme.