Study On Multi-Source And Multi-Scale Driving Mechanism Of Wildfire Based On Cross Classification Multi-Level Model

Wildfire is one of the most important disturbances in natural ecosystems and can be triggered by malicious,accidental or natural factors.On the one hand,wildfire can contribute to biodiversity and influence ecosystem succession;on the other hand,it can have a negative social,economic or environmental impact.Inner Mongolia is located on the northern border of China,with a fragile ecological environment and a large reservoir of combustible material,and is extremely active in wildfires.Due to the large spatial span of the region,the occurrence of wildfires varies significantly within the region,mainly due to the aggregated distribution of wildfires under the combined effect of different scale drivers.To investigate the influence of different scale drivers on wildfire occurrence,we used scale as a basis for dividing levels,with data at lower levels nested within units at higher levels,thus forming a multi-level structure.By using the Inner Mongolia region as the study area to reveal the multi-level driving mechanism of wildfire occurrence,we can solve the problem that traditional regression analysis models cannot handle data with hierarchical structure,and it is also of great practical significance for the protection of ecological environment and the achievement of sustainable economic development.This paper first investigates the spatial and temporal variability of wildfire events and their drivers in Inner Mongolia,and then constructs a multilevel logistic regression model for climate,topography,vegetation and human drivers of wildfire occurrence,and finally selects the explanatory variables that play a significant role in wildfire occurrence in the model,and constructs a multilevel logistic regression model with climate zones,landscape types and county-level administrative regions as the high level and individual wildfires as the low level.A cross-classified multilevel model with climate zones,landscape types and county administrative regions as high levels and individual wildfires as low levels was constructed,and how the high level explanatory variables moderated the effects of the low level explanatory variables on the explained variables was discussed.The main findings of this study are as follows:(1)The wildfire events in Inner Mongolia from 2001 to 2019 are characterized by spatial and temporal aggregation,with large inter-annual variations in time,mainly in 2003 and 2008;significant seasonal variations within years,with a higher concentration in spring and autumn;spatially,wildfire events are characterized by spatial heterogeneity,with significant differences among different climate zones,landscape types,vegetation types and administrative regions.The spatial variability of wildfire events is evident in different climatic zones,landscape types,vegetation types and administrative regions,mainly in the middle temperate semi-arid zone,mountains,meadows and grasslands,and the Hulunbuir Union.The drivers of wildfire occurrence in Inner Mongolia are also characterized by spatial and temporal heterogeneity,mainly in terms of inter-annual and intra-annual variability and differences between different regional types.(2)The construction of multi-level logistic regression models is necessary to analyse the driving effects of climatic,landscape and administrative level elements on wildfire occurrence,and their accuracy and predictive power are much higher than those of traditional logistic regression models.The first level of climatic variables,monthly precipitation and VPD,suppress wildfire occurrence,while seasonal mean temperature and pre-VPD promote wildfire occurrence;the second level of seasonal precipitation suppresses wildfire occurrence by promoting the negative effect of VPD(OR=0.470)on wildfire occurrence;the third level of annual precipitation(OR=2.115)generally promotes wildfire occurrence,but to some extent The third level of annual precipitation(OR=2.115)generally contributes to wildfire occurrence,but to some extent mitigates the positive effect of pre-VPD(OR=1.906)on wildfire occurrence;mean annual temperature contributes to wildfire occurrence by reducing the negative effect of pre-VPD(OR=0.641)on wildfire occurrence and increasing the positive effect of pre-VPD on wildfire occurrence.Of the topographic elements,elevation(OR=0.395)was negatively correlated with wildfire occurrence,and slope(OR=1.152)was positively correlated with wildfire occurrence.The mountainous areas of the landscape types are more likely to have wildfires than the other types.Among the anthropogenic factors,total population at the prefecture level and high school enrollment at the county level contributed to wildfire occurrence in the study area,while GDP at the prefecture level inhibited wildfire occurrence;total population contributed to the positive effect of GDP per capita(OR=1.469),international tourism(OR=1.295)and road mileage(OR=1.633)on wildfire occurrence;GDP weakened the positive effect of GDP on wildfire occurrence.Gross domestic product(GDP)reduced the contribution of the number of high school students(OR=1.069)to wildfire occurrence.(3)The cross-classified multi-level model with climate zone types,landscape types and county administrative regions as high levels and individual wildfires as low levels integrates natural and socio-economic elements,and its predictive and identification capabilities are significantly better than traditional logistic regression models.The results of the cross-classified multilevel model indicate that the first level explanatory variables of early precipitation(OR=0.326)and early VPD(OR=0.369)inhibit the occurrence of wildfires,while distance from railways(OR=1.577)promotes the occurrence of wildfires,and the vegetation type of meadow grassland is more likely to have wildfires than the rest of the vegetation types.Road mileage(OR=1.763),high school enrollment(OR=1.305)and gross domestic product per capita(OR=1.212)at the county level all contribute to wildfire occurrence.Wildfires are more likely to occur in the mesothermal semi-humid zone and the mesothermal semi-arid zone at the climate zone type level.The first level of the explanatory variables,elevation and NPP,had a stronger inhibitory effect on wildfire occurrence in the Mesothermal Semi-Humid and Mesothermal Semi-Arid zones than in the remaining climatic zones.When the landscape type was terraces,both current month VPD and pre-VPD were weaker inhibitors of wildfire occurrence than the other landscape types,while wildfire occurrence in hills was more likely to be influenced by pre-VPD and pre-wind speed.The higher the per capita GDP at the county level,the more susceptible wildfire occurrence was to elevation(OR=1.273)and less susceptible to pre-VPD(OR=0.839).