he dissertation is devoted to the development of a scientific and methodological apparatus for forecasting the occurrence of forest fires based on a statistical model in integration with geo-applications to support management decisions.
The aim of the dissertation is to increase the efficiency of geodata processing to minimize the risks of forest fires based on the Bayesian statistical model to support management decisions.
Among the mathematical models aimed at predicting events, an important place is occupied by predictive models, which are used both to assess the current state of natural ecological systems and to predict the dynamics of anthropogenic influence, which leads to negative consequences (deforestation, fires, floods, etc.). The main task of predictive models is to determine the outcome of events. Ecological forecasting, in particular the prediction of events in forest ecosystems, is based on statistical data. Ecological forecasts include a large number of characteristics, both biotic, abiotic and statistical by year, which make it possible to predict the state of the system as a whole as accurately as possible or to identify a potential threat in the future. Therefore, in modern conditions, an important task is to minimize the risks of forest fires on the basis of the Bayesian statistical model to support management decisions. It is advisable to divide this complex task into a number of partial tasks, one of which is the creation of a mathematical model for forecasting the occurrence of forest fires. Existing physico-mathematical models for the study of the spread and neutralization of forest fires consider the consequences of fire spread, and not the causes themselves. Most models have certain disadvantages that prevent them from being universal. However, thanks to automated systems that include mathematical apparatus, the models are simplified.
The scientific novelty of the obtained results is as follows.
For the first time, the software architecture of the forest fire forecasting system based on the Bayesian statistical model was developed, which differs from the existing ones by using a mathematical model for assessing the effect of environmental temperature on the probability of forest fires, a method for deciphering satellite images, and a mathematical model for forecasting the occurrence of forest fires. Using the specified software allows you to develop an information system for forecasting forest fires.
For the first time, a mathematical model for assessing the impact of ambient temperature on the probability of forest fires has been developed, which is based on the analysis of long-term climatic statistical data using Data Science. The model makes it possible to study the influence of global temperature changes on the occurrence of forest fires.
The method of deciphering satellite images for identifying fire-hazardous places and determining areas affected by fires, which is based on the spectral analysis of brightness temperatures, has been improved. The specified method during decoding allows to exclude from the images fragments that are covered by clouds and occupied by water bodies to establish spatio-temporal characteristics of fires. The implementation of this method will also make it possible to establish areas affected by fires and determine their fire hazard class.
For the first time, a mathematical model for forecasting the occurrence of forest fires was developed based on the Bayesian statistical model, which is based on the assessment of posterior probabilities of the taxa characteristics of forest allocations. The specified mathematical model is the basis for the development of software for forecasting the occurrence of forest fires and increases the accuracy of estimating the specified posterior probabilities by 12-18% on average.
The method of assessing the consequences of fires based on the data of remote sensing of the Earth has been improved, which, unlike the existing ones, is adapted to the processing of low-resolution images and is based on the establishment of the fire index. The implementation of the specified method will allow to increase the accuracy of the assessment of the species composition and the area of the affected areas of forest lands by an average of 8-12%, as well as to increase the efficiency of solving tasks by 25-30 times compared to traditional methods.
According to the results of modeling based on the use of the Bayesian statistical model, an increase in the accuracy of forecasting the occurrence of forest fires was achieved, which ensures the reliability of solving emergency situations and allows us to talk about increasing the reliability of management decision-making by 15% to the use of the created software complex in the process of the occurrence of catastrophic situations caused by forest fires.
