Small Scale Forest Classification and Evalutaion of Wood Resources in the Ukranian Carpathians

Publikation aus Digital

Hirschmugl M., H. Schmidtke, Granica K., Schardt M., B. Ruff

ForestSat 2007, Montepellier, France , 2007


A project for investors in wood industry was conducted to evaluate the standing timber volume, its geographical distribution and the harvest potential of the Ukrainian part of the Carpathian Mountains. Based on these needs, remote sensing and GIS are suitable tools to derive the required information. The special challenge of this project was the absence of any detailed ground truth information and also no possibility to fall back on aerial imagery as well as the short time frame available. Therefore a procedure had to be designed, which allows to derive useful information for a large area based on satellite data and a very limited field work only. The setup is based on area-wide multi-temporal LANDSAT data and samples of recent IKONOS scenes. LANDSAT data from three epochs were used: images from the beginning 90ies, the ‘main’ data coverage from 2002/3 and a recent coverage of the SLC_OFF LANDSAT scenes from 2004/5. The 2002/3 data was used for the basic classification. A forest mask was derived taking into account all three time frames. IKONOS data (7x7km chips) were selected representatively over the area in order to effectively collect training data for the classification. To be able to derive training sites from the IKONOS scenes, an interpretation key was derived by field work. The result comprises a map of the main forest types and four age classes. The accuracy from the basic classification (2002/3 data) showed an average classification accuracy for the tree species classification of 90%. The classification of young development stage yielded 84% correct and for older stocks 79%. This is not the final quality, since the additional multi-temporal analysis is not yet included. The final classification result was visually compared to the IKONOS images and showed an improved quality compared to the initial mono-temporal classification. Based on the classification and existing data from the literature, the standing timber volume and yield conditions of each forest type could be estimated for each pixel. The whole project was conducted within four months, which was only possible by accelerating the remote sensing component using IKONOS data as ground truth.