Classification of forested areas by multispectral data

Abstract

The question of the combination of modern open geoinformation systems and remote sensing data of the Earth in the problems of forest management is investigated in the article. The purpose of this work is to develop a methodology for obtaining and prompt processing of high quality multispectral remote sensing data for mapping and monitoring of forest environments. The multispectral data of the remote sensing of the earth indicate that the waves of the visible and near infrared bands have a great perspective in solving the classification of forests. Accurate identification of forest types is essential for sustainable forest management and planning, for estimating timber reserves and for modeling species and habitat distribution. The object of the study was a fragment of the forest massif within the Osiv and Verbichan forests, located in the Turiysky district of Volyn region, whose area is 65.61 km² On the basis of existing afforestation plans, classifiers of the forest species of the study site were developed, taking into account age groups and without taking them into account. In this paper, we used the method of increasing the area to form training samples. Image classification was performed using the maximum likelihood algorithm, which calculates the probability distribution for the classes according to the Bayesian theorem, estimating the pixel belonging to the land cover class. The most common method of estimating accuracy is to calculate an error matrix that compares the image data obtained with the control data for the corresponding number of classification units. According to the results of the accuracy assessment, the coefficient κ for the classifications according to the training sample, taking into account the age groups and without them, is 90.3% and 91.4%, respectively.