benefits of considering land cover seasonality in multi-spectral image classification

Effects of incorporating multi-seasonal information into image classifications for large-scale land cover mapping are investigated. Data from four Landsat7 ETM+ scenes (March, May, June 2002, September 1999) were included step-wise into classifications by discriminant analysis to document their relevance for classification accuracy. The classification using all four images reached a maximum accuracy of 69.2%, significantly higher compared with single-date classifications and showing less fluctuations in classification accuracy. Considering all four images produced similar accuracies for pixels affected and not affected by temporary snow cover, so that data limitations by partial snow cover can be resolved using multi-seasonal data. As some land cover classes were only poorly discriminated, the approach still cannot provide a level of accuracy sufficient for landscape-scale studies. Further research is suggested on the separability of land cover classes to achieve an adequate thematic resolution for regional assessments of land cover and land cover change.