Mapping of forest types of the tested Reshetka river watershed
In order to estimate the spatial distribution of irrecoverable loss values on the canopy of tree stands growing in the watershed area of the Reshetka river, tree species mapping was performed on the basis of the enlarged features: deciduous, coniferous (the share of other species is less than 5% of the forest-covered area), mixed with the prevalence of one of the species (more than 75% of the forest-covered area).
The Sentinel-2A satellite image captured on the 5th of June, 2018 was used as the initial data. The choice of the season stems from the fact that the first half of May is the period prior to active growth of deciduous phytomass, so it is the best way possible to determine the share of coniferous trees in the tree stand structure from the image. Only spectral channels with spatial resolution of 10 m (near infrared, red, green) were used for classification. The blue channel was excluded because it is strongly influenced by atmospheric conditions.
Data processing was performed with the use of the licensed software package ArcGIS 10.4 and ToolBox application. Sentinel-2A satellite was launched as a part of Copernicus programme by the European Space Agency in June 2015. (ESA Introducing Sentinel-2, http://www.esa.int/Our_Activities/Observing_the_Earth/Copernicus/ Sentinel-2/Introducing_Sentinel-2). This satellite is equipped with an optoelectronic spectral sensor for remote surveying of the Earth with a resolution from 10 to 60 m in the visible, near infrared and short-wave infrared spectral zones, including 13 spectral channels. It also allows repeated surveys every 5 days and makes a 290 km wide swath. These remote sensing data are actively used in mapping and monitoring of forest species composition (Kurbanov et al., 2018).
To create a training set of samples, the original image in the synthesis of channels ”red colors” (8-4-3; near IR, red, green) was used. Training samples were collected using the toolbar ”Image Classification” of ArcGIS 10.4 software product (developer - ESRI (Environmental Systems Research Institute, licensed version) and 4 classes for forest cover and 5 classes for non-forest areas were identified (Fig. 5). From 5 to 15 training samples were selected from different parts of the image for each class. The estimation of class separability by spectral features was carried out using scattering diagrams. After receiving the training set of samples, a signature file containing the distribution of pixel intensity for each class was created. Then maximum likelihood estimation was applied for performing the classification on the basis of the obtained signature file. Furthermore, a classification raster was created and generalized using the majority filter. The resulting map of vegetation is shown in Fig. 5а.