Assessment of biomass of forest species using satellite images of high spatial resolution (on the example of the forest of Khanty-Mansi Autonomous Okrug)

• Published in: Forest science issues
• Main Authors: E.N. Sochilova, N.V. Surkov, D.V. Ershov, V.A. Khamedov
• Format: Article in Journal/Newspaper
• Language: English; Russian
• Published: Russian Academy of Sciences, Center for Forest Ecology and Productivity 2018
• Subjects: stand biomass; wood stock volume; remote sensing data; landsat-8; forest classification; random forest; forestry; SD1-669.5; khanty; khanty-mansi; Mansi
• Online Access: https://doi.org/10.31509/2658-607X-2018-1-1-1-23; https://doaj.org/article/68dac1f9c57343e485aa930d5889fb3d

The paper describes assessment of spatial biomass of top wood layer based on combination of high-resolution Landsat-8 satellite images and selected ground forest inventory data measurements. Test area is one of forestry of Khanty-Mansiysk region. Segmentation of satellite images for spectral homogeneous land sites (segments) mapping is applied. Land category, dominated specie, age and wood stock volume for these sites are defined. Ground forest inventory data and segments used for selection of segments for dominated specie classification and validation of obtained map. The first, nine types of land cover are classified, four of them belong to forest cover with dominating of pine, spruce, cider and birch. The reference sample is updated by segments of such non-forest classes as fires, cuts and other non-forested lands, swamps, water internal bodies. Twelve spectral metrics are used for classification: reflectance in blue, green, red and near-infrared bands of Landsat-8. There are following vegetation seasons: and of winter, beginning of spring and middle of summer. The most significant informative metrics are the reflectance in the NIR band of the spring image, also green and red bands of the summer image. Random Forest algorithm is applied for training classification. The total accuracy of land categories and dominated species classification is 86,3%. Cross-validation of the classification based on the control sample was 0.712. In the second stage, we used regression models to relate the reflectance in the red band of the winter image with the taxation characteristics of the wood stock and age of the forest species in the selected reference segments. The level of relationship between the reflectance and wood stock values were equal to 0.80 for pine, 0.56 for dark coniferous species and 0.73 for birch. Between the reflectance and the specie height is following 0.75 for pine, 0.61 for birch and 0.64 for dark coniferous species. A check with control data showed that the error in estimating the wood stock above 250 ...