Modeling of forest fires based on the Bayesian forecast model and geoinformation technologies

Abstract

Today, a variety of information about forest ecosystems can be obtained using remote sensing methods. The use of space data for forest monitoring is cost-effective because it allows you to quickly obtain the objective information needed by foresters to solve practical problems. Satellite data provide wide coverage of forest lands, high accuracy of results, as well as high frequency of data obtained. Space images of the Ovruch district of the Zhytomyr region of Ukraine in the summer of 2020 were selected for the study. Determination of breed composition was carried out by the methods of controlled classification, namely the Bayesian classifier. It was found that 70 % of forests are pine, less aspen, hornbeam, birch, alder and ash tree species. According to statistics, during 2000-2020, 51.4 thousand hectares of forest plantations in Ukraine were damaged and destroyed by forest fires. Therefore, objective and timely information on the consequences of fires is needed to solve a wide range of applied problems of forestry. An important task in assessing the environmental and economic damage caused to forestry as a result of forest fires is to determine the area of damaged forests. The paper considers technologies for determining the area of the forest where the fire took place, using space images of the Landsat 8 satellite. The normalized NBR fire index before and after the fire and the DNBR index are used to identify areas burned by fire and impression levels. To predict forest fires, a mathematical model based on Bayes' theorem was created and a thematic map with fire hazard classes on a quarterly basis was created. To check the accuracy of the results of the created forecast model, the thematic map was combined with a layer of defined fire areas. This software product is quite flexible and versatile, it can be easily adapted for use not only to identify burned forest lands, but also for other areas.