RANDOM FORESTS FOR CLASSIFYING MULTI-TEMPORAL SAR DATA

The accuracy of supervised land cover classifications depends on several factors like the chosen algorithm, adequate training data and the selection of features. In regard to multi-temporal remote sensing imagery statistical classifier are often not applicable. In the study presented here, a Random Forest was applied to a SAR data set, consisting of 15 acquisitions. A detailed accuracy assessment shows that the Random Forest significantly increases the efficiency of the single decision tree and can outperform other classifiers in terms of accuracy. A visual interpretation confirms the statistical accuracy assessment. The imagery is classified into more homogeneous regions and the noise is significantly decreased. The additional time needed for the generation of Random Forests is little and can be justified. It is still a lot faster than other state-of-the-art classifiers. 1