Study On Driving Factors And Forecasting Models Of Forest Fire In The Great Xing'an Mountains Based On Geographic Information System

Forest fires are major disasters that affect the world every year.Especially in recent years,climate change and other reasons have caused global forests,grasslands and other ecosystems to face serious threats.Therefore,how to better predict the occurrence of forest fires and the driving factors of forest fires is the key to formulating appropriate fire prevention measures and rationally allocating limited fire protection resources.Based on spatial analysis and mathematical statistical methods such as ArcGIS,SAS,and R,this paper uses traditional regression models and the current international cutting-edge spatial regression models from two different data processing perspectives.The spatial distribution of forest fire data monitored by satellite,the driving factors of forest fire occurrence,and the prediction model of forest fire occurrence are studied and analyzed in depth.This paper firstly makes a preliminary statistical analysis of the record and satellite monitoring data of forest fires in the Great Xing'an Mountains,and obtains the spatio-temporal distribution and spatial distribution trend of forest fires in the Great Xing,an Mountains.Ripley's was used to conduct multi-distance spatial clustering analysis on the recorded forest fire data and satellite monitored fire point data,and the analysis results showed that both types of data showed significant spatial aggregation distribution.Using Arcgis to explore the data trend analysis results show that the overall distribution of recorded fire point data is average,while satellite monitored fire point data have an obvious increasing trend from north to south and from west to east in terms of the number of fire points and forest fire density.The global model and the geoweighted regression model were used to record the fire point data from 2000-2014 and the satellite monitoring forest fire data from 2001-2016 respectively.2)grid-mode fire point density;3)model fitting for the number of fire points in grid mode:(1)Fire-point mode(i.e.forest fire probability)Under the comprehensive factor analysis,the results based on global Logistic regression model and random forest classification showed that meteorological factors(monthly average temperature,daily average temperature,monthly cumulative precipitation and vegetation coverage)had a greater impact on forest fire than other environmental human factors.The results of satellite data show that meteorological factors(daily cumulative precipitation,daily average temperature,daily average relative humidity,monthly cumulative precipitation and topographic factors(slope,altitude)play a dominant role in influencing the occurrence probability of forest fire.The fitting results of Geographically Weighted Logistic regression show that various factors show spatial stability to the recorded forest fire data,among which only monthly average temperature,daily average temperature,monthly average precipitation and vegetation coverage have globally significant influence on the occurrence of forest fire.The fitting results based on satellite fire point data show that all variables have different degrees of significant influence on forest fire occurrence probability,among which only vegetation coverage,daily average temperature,monthly average relative humidity and monthly average temperature show spatial non-stationariness.It can be seen from the fitting results of global Logistic regression model and random forest classification that the two models have little difference in prediction and residual results.The prediction of random forest classification was slightly improved compared with the global Logistic regression model.According to the prediction and residual results of the model,the prediction of the Geographically Weighted Logistic regression model was slightly improved compared with the global Logistic regression model,and the spatial regression model provided more comprehensive spatial information on the explanatory ability of environmental factors on the impact of forest fire occurrence on forest fire.(2)Grid modeThe analysis and fitting results of the three grid scale data show that:under the 2×2km grid system,the data range is smaller than that under the 4×4km and 8×8km grid system.In addition,the data dispersion under this scale grid system is too large,which is not conducive to model fitting.All model fitting results of the grid system at 8×8km scale showed that the significance level of Moran 'i was greater than 0.05.That is to say,under the grid system of this scale,the spatial differences between the data were not obvious,so it was not suitable for data fitting as a spatial model.In addition,the size of the 8km grid is too large and the statistical accuracy of the data is too rough,resulting in the loss of information,resulting in a large error in the model fitting results.Therefore,the 4×4km scale grid system is more suitable for modeling and data analysis.(3)Fire density in grid modeDifferent from the results of point data,FVC plays an obvious leading role in the prediction of forest fire density,along with the meteorological factors such as annual cumulative precipitation,annual average temperature,annual average temperature and annual average sunshine hours,and the influence of human activities and topography is relatively weak.According to the results of Geographically weighted OLS regression fitting,only four factors show spatial non-stationarity in the impact of forest fire recording:annual average relative humidity,annual average wind speed,slope and slope direction.Among them,except the annual average relative humidity,the other three variables have a globally significant influence on the occurrence of forest fire.The fitting results based on the satellite fire point data show that all the variables have spatial non-stationarity on the forest fire density,and all have different degrees of significant influence.According to the fitting results of the model,the explanatory rate of global OLS regression model is relatively low,while that of random forest is relatively high.However,the inclusion of geographical weighting technology did not bring much improvement to the fitting of OLS regression model.The fitting results of forest fire density and environmental factors were still not ideal,and the model interpretation rate was also very low.Therefore,the random forest regression algorithm has more advantages under this scale data.(4)Number of fire points in grid modeThe fitting results of global Poisson and negative binomial regression models based on the recorded data show that the annual average precipitation and vegetation coverage have a significant influence on the number of forest fires.Based on satellite data of global Poisson,Negative Binomial regression model fitting results show that the four meteorological factors(annual average temperature,annual average relative humidity,annual average wind speed,the average annual sunshine hours)has significant effect on the fire,and terrain factors of vegetation,vegetation coverage,vegetation moisture index,altitude,slope)and human factors(residential distance,distance to road,GDP)has also had a significant effect on the forest fire.The results of Geographically Weighted Poisson and Negative Binomial regression model fitting show that the effects of all factors on forest fire show spatial stability.Satellite-based point data of Geographically Weighted Poisson fitting results showed that all variables influence on fire density show that the spatial non-stationarity,and Geographically Weighted Negative Binomial regression results show that the average annual precipitation,annual average relative humidity,annual average sunshine time,slope,slope and vegetation coverage of forest fire happened both in spatial non-stationarity and show the influence of the quantity has the significant effect of different level.In addition,the meteorological factors(annual average temperature,annual average relative humidity,annual average sunshine hours),vegetation terrain factors(vegetation coverage,vegetation moisture index,altitude and human factors,population density,distance,distance to road)to the neighborhood to fire the influence of the number of significant scope is larger and even the entire study area.Based on point record point data and satellite data of Geographical Weighted Poisson,Negative Binomial regression model fitting results can be seen that the two models on the point of prediction of the number of basic have similar results,overall similar to global regression model,Geographical Weighted Negative Binomial regression model fitting results are slightly better than Geographical Weighted Poisson regression model.On the other hand,similar to the global model,the spatial autocorrelation results of the residuals in the Geographical Weighted model for the fitting results of recorded data show that the spatial correlation is not significant,while the fitting results of satellite data show the opposite.In addition,according to the prediction results of forest fires by various models,Arcgis software was used to obtain the corresponding prediction map and residual spatial distribution map of forest fires in the Great Xing'an Mountains.In conclusion,under different data models,different models have basically the same discriminations on the driving factors of forest fire in the greater hinggan mountains.Forest fires in the greater hinggan mountains are mainly affected by temperature,precipitation,relative humidity,vegetation coverage and vegetation moisture index.In addition,the global model does not fully describe the relationship between forest fire occurrence and potential impact factors,while the geoweighting technique enhances the explanatory power of the global model as a whole.The research results show that geographically weighted regression model is a complement to a global regression model,help to overcome the problem of spatial nonstationarity,thus better able to predict the details of the geographical area of forest fires happen,for improving the management of the forest fire and fire fighting resources reasonable allocation have the important times value,for forest fire management and prevention of precise ShiCe provides certain theoretical basis.