Rule-based land cover classification and erosion risk assessment of the Krkonose National Park, Czech Republic

Abstract: The Krkonoše National Park is located in the Sudetes mountain range at the Czech -
Polish border. It is characterized by large variations in altitude and a mosaic vegetation
pattern. The complex landscape raises the issues of appropriate classification techniques
and scales. The first aim of this study is to produce a land cover map using 21 classes
based on a rule-based classification system and to evaluate classifications based on this
method and a maximum likelihood algorithm based on Landsat 7 ETM+ data. Ancillary
data and prior probabilities are used in the rule-based system.
The rule-based classification (21 classes) yields an overall accuracy of 61.5%. A higher
accuracy is reached if 11 land cover classes are used (overall accuracy: 74.9%). This
indicates that the result of the accuracy assessment of the land cover map with 21 classes
is strongly influenced by the rather low accuracy of the more infrequent and complex
classes. Considering the differences in altitude, the mosaic vegetation and the large
number of classes the result of the rule-based classification system is satisfactory,
especially when the number of classes is reduced. This study shows that a rule-based
classification system using ancillary data and prior probabilities clearly enhances a
maximum likelihood classification based solely on spectral data. An interpretation of
satellite data based exclusively on spectral information does not produce a satisfactory
result for this region. To achieve an improved classification the use of ancillary data and
prior probabilities in a rule-based classification system seem to offer a promising
solution.
The second aim of this study is to assess the erosion risk in the National Park. Heavy air
pollution has been deposited in this region during the last decades causing soil
acidification, decreased biodiversity, deforestation and soil erosion. The erosion risk
varies within the park and it is therefore essential to make an assessment of which areas
are in danger of becoming eroded to prevent actual erosion. Two classes are used in the
erosion risk assessment; erosion versus no erosion risk areas. It is based on statistical
analyses of field data, consisting of GPS points including information on land cover and
the presence of erosion /no- erosion, digital data on soil type, topographical form, slope
gradient, aspect and altitude, and on literature studies. Accuracy assessments yield an
overall accuracy of 86.4%. This indicates that assessments of this type can be made with
satisfactory results.