Tree species and age class mapping in a Central European woodland using optical remote sensing imagery and orthophoto derived stem density – Performance of multispectral and hyperspectral sensors

Optical remote sensing imagery and orthophoto derived stem density were classified to map coniferous forest cover for a study area located in the Western Hunsrück, Germany. Our objectives were (i) to investigate if hyperspectral data contain more information relevant to classification of species and age classes than multispectral imagery and (ii) to test in what way classification results can be improved through the integration of orthophoto derived stem densities into the classification process. Airborne hyperspectral data (HyMap) covering the entire test site has been acquired for July 1999. Subsequent to radiometric and geometric correction, data reduction and enhancement was performed by a Minimum Noise Fraction transformation. Multispectral data (TM) was simulated through degradation of the spectral and/or spatial information of the HyMap data. Three different synthetic datasets were created: spectrally degraded, spatially degraded and a combination of both. Stem density information were added to the original HyMap imagery for classification purpose. Stem density had previously been derived from black/white orthophotos by an automatic method. Classification was achieved by the Spectral Angle Mapper algorithm. Our results show that mapping of coniferous forest cover is improved by the use of hyperspectral imagery compared to multispectral data and that classification accuracy is greater using 30 m spatial resolution data compared to 5 m resolution data. Integration of orthophoto derived stem density into the classification process resulted in slightly better performance compared to the results obtained with image data alone.