【作者】 曹姗姗；

【导师】 唐小明；

【作者基本信息】 中国林业科学研究院 ， 森林经理学， 2014， 博士

【摘要】 森林火险是多种自然因素和社会因素综合作用的结果，其等级的高低用于衡量林火发生概率、蔓延速度、释能强度、难控程度和损失大小。森林火险预测预报是林火管理的基础性工作之一，其准确性和时效性决定了森林防火工作的针对性和科学性。我国现有的森林火险等级预报方法是在大尺度上基于气象因子发布国家级森林火险气象等级，在中小尺度上单方面基于气象因子、可燃物含水率或可燃物燃烧性预测林火发生的可能性。可燃物因素、火环境因素和火源因素是森林燃烧的三个基本条件，任一因素的缺失都会局部性增强或弱化森林火险等级，进而影响森林火险预测预报的精度，综合考虑上述三方面因素建立森林火险等级预报模型是提高林火预报准确性的有效方式。由于我国森林资源分布具有明显的区域性和复杂的空间结构，自然和社会环境差异显著，亟需开展综合考虑林火燃烧三个基本条件的以小空间范围为研究单位的小尺度森林火险等级预报研究，以提高森林防火工作的精细化管理水平、有效利用森林防火资源、提高森林防火工作效率。随着大数据时代的来临，面向物联网的无线传感器网络为小尺度下各类林火影响因子数据的获取提供了新的途径，有利于提升森林火险预测预报的精准性和时效性。本研究从火环境因素、可燃物因素和火源因素三个方面归纳并分析各因子对森林火险的影响及因子间的相互关系，并以小尺度森林火险等级预报为指导分层构建了可燃物分类方法，研究了各火险影响因子的估测模型，提出了小尺度森林火险等级预报模型的构建方法，并以北京市九龙山区华北林业实验中心为研究区域，实现了小尺度森林火险等级的预报。主要研究结果和结论有：（1）森林火险的影响因子众多，因子的选取从本质上决定着所建立的火险等级预报模型的准确性。本研究对森林火险的影响因子进行分类，从机理方面详细分析各因子对森林火险的影响方式，并探讨了各因子间的相互作用关系，为小尺度森林火险等级预报模型构建的因子选取提供理论依据。（2）火环境因子即是林火的孕灾环境，又是林火其他影响因子的差异性基础。本研究基于DEM数据及森林资源二类调查数据，综合应用GIS、RS和数理统计理论建立小尺度地形信息提取方法。利用ENVIS梯度气象站采集的气象数据，结合获取的地形因子数据，修订山地小气候模拟模型MTCLIM中的关键参数，并应用该模型模拟山地小环境的温度和湿度；应用WindNinja软件的模拟算法获取小班的平均风速。实现了山地小环境气象信息的模拟，为火环境因子的计算提供方法依据。（3）建立森林可燃物的分类方法，为小尺度森林火险等级预报模型的因子计算提供依据。提出基于森林可燃物理化性质的内特性的小班燃烧性指标计算方法，并测定可燃物的初始引燃含水率，为小尺度森林火险等级预报模型提供因子数据。（4）建立了基于多元回归分析的乔木型灌木生物量估测模型、地表枯落物未分解层和半分解层载量估测模型，基于遗传算法优化的BP（GA-BP）神经网络构建了典型灌木的生物量估测模型，实现了大范围灌木生物量的快速估测，有效避免了灌木生物量估测模型构建时自变量选取和模型选取的繁琐工作。提出了小班乔木层、灌木层及地表枯落物未分解层、半分解层载量的计算方法，为小尺度的森林火险等级的可燃物载量计算提供方法依据。（5）基于时间序列分析理论，应用多元逐步回归分析方法分树种构建地表枯落物未分解层和半分解层的含水率预测模型，提出小班可燃物含水率计算方法，结合山地小环境气象数据实现可燃物含水率动态预测。（6）建立小尺度森林火险等级预报模型构建方法。综合考虑小尺度森林火险影响因素，选取小尺度森林火险等级预报模型因子，应用因子分析方法消除因子多重共线性并提取因子主成分，基于主成分得分函数计算各小班的森林火险指标值，应用聚类分析将各小班的森林火险等级聚类，应用GA-BP神经网络建立研究区域的小尺度的森林火险等级预报模型。更多还原

【Abstract】 Forest fire is a damage which is comprehensively caused by a variety of natural and socialfactors, and its grade level is utalized to measure the mutiple characteristics of the forest fireincluding the probability of occurrence, spread rate, release energy intensity, the difficultdegree of control and the damage loss. Forest fire forecasting is a fundmental work of forestfire management, and the direction and scientificity of forest fire prevention are always decidedby the accuracy and timeliness of forest fire forecasting.The features of the existing forest fire danger rating prediction methods are: the nationalforest fire weather rating is forecasted based on the meteorological factors data on thelarge-scale, the likelihood of fires occurring is predicted by exploiting meteorological factorsand wildland fuel moisture content and the forest fire behavior is prognosised by unilaterallyconsidering the combustion characteristics of fuel on the mesoscale and the small-scale.According to statistics, the man-caused fire is one of the main reason in forest fires, and thesmall area of forest, which are always distirbuted near the human settlesments and withfrequent human activities, are the places of the high incidence of forest fires. However, it is noteffective to deduct the results of the large-scale and mesoscale forest fire danger ratingpredition to the small-scale, meanwhile the difference of natural and social environment issignificant in forest resource subcompartments which is a typical representative of thesmall-scale. Forest fire impact factors in forest resource subcompartment includingmeteorology, fuel combustion resistance and moisture content, terrain and the man-made factorare tightly related, and the absence of any one factor will locally strengthen or weaken theforest fire danger rating, thereby the accuracy of prediction of forest fire will be affected. Withthe advent of the era of big data, it is a new way for the data acquisition of various forest firesfactors on small-scale by making use of the wireless sensor networks for the internet of thingsto help improve the accuracy and timeliness of forest fire prediction. Fire environmental factors, combustible materials factors and fire sources factors weresummarized, and the influence of these three factors acting on the forest fire danger rating andthe relationship between factors were analyzed. Then, the wildland fuel classification methodwas hierarchically builded to meet the requirement of small-scale forest fire danger ratingprediction. Meanwhile, the estimation model of various fire impact factors were constructed,and then the construction method of a Small-scale Forest Fire Danger Rating PredictionModel(SFFDRPM) was proposed. Finally, Jiulong mountain district forestry experiment centerof north China in beijing city was taken as the study area, and a small-scale forest fire dangerrating prediction of which was achieved. Thus, the main results and conclusions are as follows:(1) There are a great number of forest fire impact factors, and the accuracy of the firedanger rating prediction model was essentially determined by the selection of factors. Theimpact factors on forest fire were classified, and the way of these factors impacting on theforest fire was analyzed in detail. Then, the interaction relationship between each factor wereexplored. Thus, the factors selection theoretical basis for the SFFDRPM was provided.(2) Not only are forest fire environmental factors the pregnant disaster environment, butalso the differences basis of other forest fires impact factors. The small-scale topographicalinformation extraction methods were established by the integrated application of GIS, RS andstatistical theory based on the DEM data and forest resource inventory data. At the same time,the meteorological data of ENVIS gradient stations were collected and the acquiredtopographical factors data were combined to amend the key parameters in MTCLIM(MountainMicroclimate Simulation Model), which was utalized to simulate the small mountainenvironment temperature and humidity. Meanwhile, the average wind speed in forestsubcompartment were obtained by applying the simulation algorithm of WindNinja software.Ultimately, the weather information in mountain microenvironment were simulated to providethe method basis of the fire environment factors calculation.(3) The forest wildland fuel classification method was established to provide the basis ofthe factors calculation in the SFFDRPM.Then, the combustion index calculation method in forest subcompartment was proposed based on the intrinsic properties in physical and chemicalproperties of forest wildland fuel, and the moisture content of the initial ignition of forestwildland fuel was measured to provide factor data for the SFFDRPM.(4) The tree shrub biomass estimation model, surface litter undecomposed layer loadestimation model and semi-decomposed layer load estimation model were builded based onmultiple regression analysis. Meanwhile, the typical shrub biomass estimation model wasconstructed to achieve a quick biomass estimation in a wide range of shrub and effectivelyavoid the tedious process of independent variables and models selection in shrub biomassmodeling based on Genetic Algorithm optimized Back Propagation (GA-BP). And then, theload calculation methods of tree layer, shrub layer, surface litter undecomposed layer andhalf-decomposed layer in forest subcompartment were proposed for the fuel load computingmethods reference in the SFFDRPM.(5) The moisture content models of surface litter undecomposed layer andsemi-decomposed layer were proposed by exploiting multivariate stepwise regression analysisbased on time series analysis theory for each tree specie, and the wildland fuel moisturecalculation method in forest subcompartment was constructed to achieve the fuel moisturedynamic prediction by combining with meteorological data in mountain microenvironment.(6) The construction method of the SFFDRPM was established. The small-scale forest fireimpact factors were comprehensively considered to select input variables in the SFFDRPM,and the forest fire danger indexs were calculated based on the the principal component scorefunction, which was fitting by using factor analysis method to eliminate factorsmulticollinearity and extract principal components. Then, the forest fire danger rating in eachforest subcompartment was clustered by making use of cluster analysis method. Finally, theSFFDRPM in the study area was created by applying the GA-BP.更多还原