Farber S. K., Kuz’mik N. S., Bryukhanov N. V. Errors of Forest Interpretation in Angara River Region by the Method of Satellite Scene Pixel Classification
1 V. N. Sukachev Institute of Forest, Russian Academy of Sciences, Siberian Branch
Akademgorodok, 50/28, Krasnoyarsk, 660036 Russian Federation
2 Federal State Budgetary Enterprise «Roslesinforg» «Vostsiblesproekt»
N. K. Krupskaya str., 42, Krasnoyarsk, 660062 Russian Federation
E-mail: email@example.com, firstname.lastname@example.org, email@example.com
How to cite: Farber S. K.1, Kuz’mik N. S.1, Bryukhanov N. V.2 Errors of forest interpretation in Angara river region by the method of satellite scene pixel classification // Sibirskij Lesnoj Zurnal (Siberian Journal of Forest Science). 2016. N. 4: 56–67 (in Russian with English abstract).
© Farber S. K., Kuz’mik N. S., Bryukhanov N. V., 2016
A purpose of research is to identify errors in interpretation of satellite images of forests based on image pixel classification by spectral brightness. The Landsat 5 satellite image was used (August, 2005). The results of interpretation were compared with data of forest estimation, i.e. descriptions of forest plots and maps of dominated species. The forest area made up 80.8 thousand ha; quantity of plots was about 2700, including 573 sampling plots; specified number of clusters classification image – 10. As a result, there were intolerable errors in land categories, forest formations and dominated species on the level of forest plot generalization. Thus, interpretation of forest land images having applied a method of classification of spectral brightness pixels could be applied for small scale mapping only. It is supposed that inclusion of spatial analysis of relief digital simulation in the process of interpretation will improve a quality of performance. Stratums of locations were formalized by means of registration of absolute altitudes, slopes, and exposures. Spatial analysis was carried out on the base of Shuttle Radar Topographic Mission database. Errors of forest stand density, average ages and heights of trees exceed norms, which were specified for the least detailed third category of forest inventory. In such a case, there is not error reduction considering single stratums of locations. Categories of forest lands and variation of forest estimation indicates do not depend on a picture of satellite images. Therefore, achieving required accuracy of interpretation having applied methods of imagery classification and transformation, i.e. by use of the normalized vegetative index, does not seem possible. Consequently, applying the actual methods of satellite image classification in forest inventory cannot be recommended.