Determination of the state of forests based on a regular grid of ground-based sample plots and Sentinel-2B satellite imagery using the k-NN (“nearest neighbor”) method

Abstract

In the paper on the base of regular grid of sample plots and remote sensing data from Sentinel – 2B the tree stands health status was estimated in continuous way for the study area. This study used a Sentinel-2B image (date of shooting July 31, 2018). The following spectral channels were selected: 8 – Near Infrared, 4 – Red, 3 – Green, 2 – Blue. After the initial processing, a single multi-zone image with four channels was formed, which was later used for classification. The following settings of the kNN classifier plugin were used to classify the territory based on the state of the stands: the number of nearest neighbours k = 5, the number of spectral channels – four, and the classification mode – for continuous values. The quality of determining the state of tree stands based on the k-NN classification of the territory was determined by comparing its results with the data of the control sample plots by calculating systematic and random errors. As a result of the calculations the following values of errors were obtained: the systematic error was equal to zero meanwhile random appears to be equal 1 damage class, both may be considered as satisfactory and used classification method as practically applicable.