Research On The Accessibility Of Artificial Fire Sources In Grassland Based On The Potential Model Theory

As one of the most important ecosystems,grassland covers nearly one-fifth of the global land surface.However,in the area of grassland,fire has a great impact on the local economic and ecological conditions,as one of the most important disturbing factors.China is with abundant grassland resources,the grassland area accounts for about 40% of the total land area,which makes it the largest inland ecosystem in China.Once a fire breaks out,it would greatly damage the lives and property of the people in the pastoral area and the local ecosystem.Of all the grasslands in China,the areas where fires frequently occurred account for 1/6 of them,most of which were caused by man-made fires.Therefore,we need to study the driving mechanism of artificial fires on grassland fires.At present,in the studies of grassland fire caused by artificial fire sources,most scholars were limited to exploring the relationship between artificial fire sources and fire occurrence through simple administrative district statistics.Few related studies had explored the impact of artificial fire sources on grassland fire occurrence from the perspective of artificial fire sources accessibility.In our research,the scale and spatial distribution of residential areas,roads and cultivated land were applied to represent the intensity and spatial distribution of human activities.Based on the potential model theory,the two attributes of artificial fire sources were both considered,which made up for the shortcomings of previous studies.In this paper,Chenbaerhu Qi of Inner Mongolia Autonomous Region was selected as the study area.Based on the potential model theory,a fire sources potential model was constructed to estimate the total capacity and friction coefficients of artificial fire sources.By further improving the fire sources potential model,a fire sources accessibility model was constructed as well.In addition,nonlinear regression,neural network(Multilayer perceptron),multiple linear regression,correlation analysis and other methods were applied to study the distribution of the comprehensive artificial fire source accessibility in the study area.The results are as follows:(1)Combining Landsat 8 and Google satellite imagery,the land use data covering the research area was interpreted,which was used to extract the MODIS fire products from 2000 to 2016.Through analysis,it was found that there were 568 grassland fires in the study area.Among them,Ewenkizu Sumu contained most of the fire points,which was 375;The density of fire point in the study area was 0.0305i/km2,which was highest in Tenihe Farm(0.1061);The frequency of grassland fire in the study area was 33.35times/year,among which,2014(84 times)was the highest.(2)In order to calculate the impact of fire sources on grassland,various land use types and road levels that assigned different speeds were combined with slope data,and the transit time cost raster layer of the study area were obtained.The analysis of the layer showed that the high value of the time cost raster overlapped with the high value of the slope layer to a certain degree.The overlapped areas were mainly located in the east and northwest of the study area.In addition,in the areas covered by wetlands and water bodies,the time cost was also relative high.While low value was mainly located in the area which covered by high-grade roads and high-density road network.(3)Three types of primary fire sources(residential land,road and cultivated land)were classified into ten secondary fire sources according to their attributes(length,area).These ten types of secondary fire sources were subjected to nonlinear regression fitting according to the fire source potential model.And 6 of the 10 types of secondary fire sources were fitting well,which were level-2,level-3 residential land,level-4,level-5,level-6 roads and cultivated land.Among them,The M and values corresponding to the level-3 residential land were greater than the level-2 residential land's,and the M and values corresponding to the level-6,level-5,level-4 road were also decreased in turn,indicating that the total capacity and the influence of the cost distance of lower level residential land and road fire sources were larger.The six types of secondary fire source layers were overlay based on the primary fire sources,forming the accessibility layers of the three fire sources.Through analysis,we find that each type of fire source accessibility had a certain explanatory effect on the accessibility distribution of the artificial fire source.(4)The multilayer perceptron(MLP)was used to obtain the weights that three fire source accessibility layers overlaid.By evaluating the accuracy of MLP model,the AUC value was 0.822,which performed well.By analyzing the comprehensive artificial fire sources accessibility index generated by overlaid,we can see that the average value of the index in the study area reaches 0.369,which indicates that the accessibility risk of artificial fire sources in this area is relatively high.(5)The comprehensive artificial fire source accessibility index was divided into five zones by the equal interval classification method,and the number of verification fire points(2014-2016)included in each zone was separately counted.Results showed that the extremely low,low,medium,high and the extremely high index zone contained 2.94%,18.82%,20.01%,22.35%,and 35.88% of the verification fire points,respectively.In addition,the result of Kernel density validation showed that the Kernel density of the fire points was significantly positively correlated with the comprehensive artificial fire source accessibility index at the 0.01 level.Both verification results indicated that the index layer could represent the distribution of artificial fire sources accessibility to a certain extent.(6)According to the obvious difference of the index between administrative units,the study area was divided into three parts: the low-value area in the west,the middle-value area in the middle and the high-value area in the east.The average index of the low-value(west)area was 0.196,the average index in the middle-value(middle)area was 0.333,and the average index in the high-value(east)was 0.553.Multiple linear regression was used to investigate the influence weights of these three types of fire source in the comprehensive artificial fire source accessibility index.It was concluded that the fire source accessibility layer was the main influencing factor of this index in the whole study area,whose weight coefficient is 0.478;in the low-value(west)area and the high-value(east)area,the cultivated land fire source accessibility was also the dominant factor of the corresponding index area;while in the middle index(central)area,the road fire sources accessibility and the cultivated land fire sources accessibility weight coefficients were basically the same,which were together as the dominant factor of this index area,and the accessibility of the residential land fire sources was the smallest.