 |
 |
 |
| All about Russian forests | Russian NGOs Forest Club | Useful links | Site map | Site search |
|
| |||
| ||||
| ||||
|
|
Preliminary identification of large intact forest-dominated areas and classification of their tree species compositionIn a special study for Global Forest Watch, R&D Center ScanEx has tested a method for identification of large blocks of intact, forest-dominated areas. In the process of this study, the tree composition of these forests was classified according to the criteria used in the map Vegetation of the USSR edited by B. V. Sochava (1957). These criteria were selected because they have been used to develop other systems for country-wide classification of vegetation and have been found well suited for this purpose.The purpose of the study was to test a method for direct identification of large intact forest-dominated areas (landscape mosaics with a minimum of 50 percent in forest). The approach was to first identify a set of “typical” intact forests of different composition and then use an automatic algorithm to search for identical forests across the landscape. Medium-resolution imagery was used. The result shows that this approach is sensitive to small differences in forest composition and to the quality of the satellite images used. Primarily closed forests tend to be identified. The study has produced maps that shows the location and tree species composition of large blocks of apparently closed forest that fitted the search criteria (i.e. the legend of the map). These maps are shown in the thematic section of this Atlas, beginning on page 147. A more technical description of the work is given below. Medium-resolution satellite images from the Russian satellites Resurs-O1 series from different seasons were used. The images were geometrically transformed with the ScanEx Transformer software, then arranged into a mosaic with 6-degree zones in the Gauss-Kruger projection. The images were brought into a uniform resolution of 150x150 meters, and geo-referenced to the 1:1 million scale topographical map. The thematical analysis was conducted in the ScanEx NeRis software, using the Kohonen algorithm for self-organizing nets. Blocks of closed forest assumed to be typical of intact forest were selected and used for training of the neural nets employed by this software. The quality and characteristics of the representation of these forests in medium resolution imagery was assessed with high-resolution Resurs-O1 MSU-E images (35x45 meter resolution). The trained neural nets were used to perform a multi-channel spectral analysis and the resulting layer was then analyzed for contextual features. The classification of textural and contextual characteristics resulted in an 8-bit raster layer. Based on expert decision, a color coding table was designed and used to vectorize the result. The vector layers were then overlayed with the original imagery and with fragments of high resolution imagery for visual control.
| |||||
|
|
