Research On Forest Fire Risk Prediction Model In Panxi Region Based On Machine Learning

In recent years,due to human activities and climate change and other reasons,forest fires have caused significant losses on a global scale,making the global forest ecosystems face serious threats.The focus of forest fire risk prediction research is to analyze the driving factors that affect forest fires,summarize the temporal and spatial distribution pattern of forest fires,classify forest fire risk levels in different regions,provide decision-making support for forest fire prevention and control to relevant departments,and reduce losses caused by forest fires.Panxi region is one of the most serious areas of forest fires in Sichuan Province,in 2019 and 2020,severe forest fires occurred in Panxi region.Therefore,in order to predict the occurrence of forest fires better,this thesis takes the Panxi region as the research area,based on spatial analysis tools and scientific statistical language such as Arc GIS,python,using traditional regression model and the current international cutting-edge machine learning models.From two different perspectives,conduct in-depth research and analysis on the temporal and spatial distribution of satellite monitoring forest fire data,forest fire occurrence driving factors and forest fire risk prediction in Panxi region.The specific research content and research results are as follows:(1)Using MOD14A1 and MCD12Q1 time series data,a total of 1,725 forest fires in the study area from 2009 to 2019 were obtained.The same number of random points were extracted at a ratio of 1:1,combine meteorological data,topography data,combustible data and social basic data,established the forest fire GIS database of the study area.(2)Research on the temporal and spatial distribution characteristics of forest fires in Panxi region,analyzed the temporal change characteristics of the number of forest fires on the interannual,monthly and daily scales,and analyzed the spatial distribution characteristics of forest fires from the perspective of administrative division and spatial autocorrelation analysis.The results show that the interannual changes in the number of forest fires in Panxi Region from 2009 to 2019 fluctuate and then stabilize;from a monthly perspective,the number of forest fires is mainly concentrated from December to June of the following year,of which January to May is the period of high incidence of forest fires;from an daily perspective,the dates with the highest number of forest fires are February 1,February 15,April 16,and April 24 in non-leap years;from the perspective of the spatial distribution of forest fires,Yanbian County,Miyi County and Huili County are not only the areas with high incidence of forest fires in Panxi,but also show clustered distribution in spatial autocorrelation analysis.(3)Explore the predisposing factors of forest fires.14 variables including weather,topography,combustibles,and human activities are selected as candidate driving factors,the correlation between each driver and historical forest fire points was statistically analyzed to reveal the changes of forest fires with different drivers in Panxi region in the past 11 years.(4)Logistic regression,random forest,support vector machine and deep neural network are selected for forest fire risk prediction research.Based on the prediction results of the four forest fire risk prediction models,the forest fire risk zoning of Panxi region from 2009-2019 were established respectively,and the overall trend shows that the middle and high and extremely high fire risk areas are mainly concentrated in the central and southwestern parts of Panxi region,and the extremely low and low fire risk areas concentrated in the northeast and northwest of Panxi region.Since the deep neural network has the highest performance on the full sample,this thesis uses the model to establish the 2009 and 2019 forest fire risk zoning of the Panxi region based on the full sample data.Through the establishment of a forest fire risk area transfer matrix,it is found that the overall fire risk in Panxi region is on a downward trend.