Search

The paper presents a method of Landsat 5 Thematic Mapper satellite image processing to assess the condition of forests in the Tatra National Park (southern Poland). Selected images were acquired on 1987/09/01, 2005/09/02 and 2011/09/03 from the same sensor with maximum time interval for the first and last scene and from similar phenological period. Firstly, the data were radiometrically corrected using the ATCOR 2/3 software and Digital Terrain Model from the ASTER mission. Quality of the correction was assessed calculating RMSE for reflectance values from images and resampled spectral characteristics collected in terrain. RMSE was in range 3−10%. Next, basing on Landsat images, Normalized Difference Infrared Index (NDII) and a Maximum Likelihood supervised classificatory, following dominant land cover types were identified: forests (including dwarf pine), grasslands, rocks, lakes, shadows (additionally clouds were distinguished on data from 1987/09/01). It allowed to select forest areas with producer accuracy not worse than 97.69% and user accuracy not worse than 98.31%. On corrected Landsat images Normalized Difference Vegetation Index (NDVI, an overall vegetation state) and Moisture Stress Index (MSI, canopy water content) were calculated. Vegetation indices discriminated forest state using the decision tree method. The worst overall condition was observed for the 1987 (about 21% of forest stands were in the worst condition and 87% were in medium condition), while the best one in 2005 (75.51% forest stands were in good condition and 10.66% were in the best condition). In case of 2011, the overall condition was quite good, but there were large areas with poor condition caused by bark beetle outbreaks. Proposed method allows for a fast and objective assessment of forest condition. It is possible to detect damaged areas or stands in poor condition. It can be complement for traditional methods of monitoring and management in forestry and nature protection., Sylwan 160 (3): 256-264

We used hyperspectral data from APEX scanner (288 spectral bands in 380−2500 nm spectral range; 3,5 m spatial resolution) to classify five tree species occurring in the area of Mt. Chojnik in the Karkonoski National Park (south−western Poland). Data used to delimit learning and verification polygons were acquired during field research in August 2013, when ground truth polygons were acquired using device equipped with GPS receiver. Raw APEX data went through radiometric and geometric correction at VITO office. To reduce processing time, 40 most informative bands were selected using information content analysis. The Support Vector Machines (SVM) algorithm was used for classification of the following tree species: Fagus sylvatica L., Betula pendula Roth, Pinus sylvestris L., Picea alba L. Karst and Larix decidua Mill. Final classification had 78.66% overall accuracy with Kappa coefficient equal to 0.71. The best classified species included beech (87.09%) and pine (83.96%), while the worst results were obtained for larch (60.29%). Low accuracy for larch could be caused by the fact that most of larch trees in the research area grow in small patches, which made it hard to specify large enough sample of training data. All classified tree species had producer's accuracy of at least 60%, with the highest value reaching 87%. User's accuracies were from 53% for pine to 85% for beech. It is possible to classify tree species using hyperspectral data with moderate to high accuracy even if the data used lacked atmospheric correction. Further work will focus on improving the classification accuracy and use of neural networks based classification methods. Results from this paper will serve as basis for tree species map of the Karkonoski National Park., Sylwan 159 (7): 593-599