【作者】 曲智林；

【导师】 胡海清；

【作者基本信息】 东北林业大学 ， 生态学， 2007， 博士

【摘要】 黑龙江省是我国森林大省，也是林火多发区域，林火对黑龙江省的林业资源造成了巨大的破坏，是全国森林防火的重点省份，年均森林过火面积居全国之首，是火灾危害最严重的地区，潜在林火危害预测的研究对于黑龙江省的林火预测预报具有重要的意义。本文主要研究和结果以下：(1)根据黑龙江省森林植被特点和气候特点，将黑龙江省划分成两个研究区域：寒温带针叶林区和温带针阔混交林区；研究时段：寒温带针叶林区为4月份至6月份和10月份，温带针阔混交林区为3月份至6月份和10月份，这些时段是黑龙江省森林火灾的多发期。不同区域和不同时段的森林植被特征有很大差异，对于林火发生与林火蔓延是不同的，说明按不同区域和不同时段分别进行研究是合理的。(2)通过1980年至2003年黑龙江省森林火灾统计数据和气象数据，利用多元回归理论建立了森林火灾面积预测模型，该模型是非线性模型，选取的自变量为当日的平均风速、相对湿度和当日平均温度，该预测模型是在降雨量为0mm的情况下适用的，模型中因变量选取的不是火灾面积，而是林地过火面积等级，通过林地过火面积等级可以反映林火的危害程度，并用2001年和2003年的黑龙江省森林火灾统计数据检验，正确率达63.3％。说明用林地过火面积等级来预测潜在林火危害程度是可行的，因此该模型可以作为潜在林火危害等级预测模型。(3)给出了林火危害强度的概念；通过1980年至2003年黑龙江省森林火灾统计数据和气象数据，利用微分方程理论建立了林火危害强度预测模型，该模型是在降雨量为0mm的情况下，选用的自变量是当日的平均风速、相对湿度和当日平均温度。用林火危害强度可以真实反映林火危害程度，通过检验，寒温带针叶林区正确率为65％，温带针阔混交林区正确率为72.7％；并给出了林火危害强度等级划分标准，因此用林火危害强度预测模型可以预测黑龙江省短期潜在林火危害程度。(4)建立了一个由输入层是4个神经元、隐层是9个神经元和输出层是1个神经元的BP神经网络，该网络较好地模拟出林地过火面积与当日的平均风速、相对湿度、当日平均温度和燃烧时间之间内在关系，并由此给出基于BP神经网络的林火危害强度模型，说明神经网络技术在林火预测预报中是一个重要的工具。(5)通过统计分析可知春季防火初期的林火危害强度与之前冬季平均风速、降雪量、平均相对湿度和平均温度之间有一定的相关性，本文利用多元回归理论建立了它们之间关系的线性模型，这里寒温带针叶林区冬季选择的是10月份至3月份，温带针阔混交林区冬季选择的是11月份至2月份。以及建立了春季防火期各月份林火危害强度之间的关系模型，这些模型整体构成春季潜在林火危害强度模型，实际验证预测结果与实际值较为接近，说明本文选择的变量和选择的模型是合理的，可以用来做长期潜在林火危害强度预报模型。(6)本文利用地理信息系统技术建立了“黑龙江省森林火灾危害等级预报系统”，该系统把气象数据库和林火危害强度模型相连接，建立识别模型，通过输入全省各县级行政区域防火期内当日的气象数据，就可预测当日潜在林火危害强度等级，并使之可视化。更多还原

【Abstract】 Heilongjiang province has obvious advantage at forest resource in china, but oftencatches forest-fire which brings enormous breakages. Thus Heilongjiang has the largest forest-fire area in China annually, and is the seriousest area on fire disservice. So reseach on latentforest-fire harm prediction is very important to the prediction of forest-fire of Heilongjiangprovince.The main content and results of this paper are as below:(1) Based on the characteristics of vegetation and climate in Heilongjiang, Heilongjiangwas partitioned into two research regions: Frigid temperate coniferous forest region andTemperate broadleaved and conifer mixed forest region; research periods: they are from Aprilto June and October in Frigid temperate coniferous forest region, they are from March to Juneand October in Temperate broadleaved and conifer mixed forest region, these periods are themost frequent periods of forest-fire. Different districts and different periods lead to largedifferences in characteristics of forest vergetation. So the occuring and spreading of forest-fireare different. These showed that it is logical to research separately according to differentdistuicts and different periods.(2) By the statistical data of forest-fire and the weather from 1980 to 2003 in Heilongjiang, using pluralistic regression theory a prediction model of forest-fire area was establishedThe model was non-linear.The independent variables were intraday average wind speed,relative humidity and intraday average temperature. The prediction model was effectual whenthe rainfall was 0 mm. In the model the dependent variable was not the forest-fire area, butthe grade of forest-fire area, by which we could know the forest-fire harm degree. Checkingup the model using the statistical data of forest-fire from 2001 to 2003 in Heilongjiang, theaccurate rate was 63.3%. This showed that it was feasible to predict the forest-fire harm degreeby the grade of forest-fire area, so the model could be used to predict the latent forest-fireharm grade.(3) The concept of Forest-fire harm intensity was defined ; Using differential equationstheory a prediction model of forest-fire harm intensity was established by the statistical data offorest-fire and the weather from 1980 to 2003 in Heilongiiang. The prediction model waseffectual when the rainfall was 0 mm.The independent variable were intraday average windspeed, relative humidity and intraday average temperature. Forest-fire harm intensity couldfactually reflect forest-fire harm degree. By proof-test, the accurate rate was 65% in Frigidtemperate coniferous forest region, 72.7% in Temperate broadleaved and conifer mixed forestregion, the divisioal standard of forest-fire harm intensity grade was also given, so the model of forest-fire harm intensity prediction could predict short-term latent forest-fire harm degreein Heilongjiang.(4) Using BP Neural Network (short for NN) a prediction model of forest-fire harmintensity was established. Its import tier had four nerve cells. Its hiddent tier had nine nervecells. Its export tier had one nerve cell. The network simulated preferablely the interconnectionbetween forest-fire area and intraday average wind speed, relative humidity, intradayaverage temperature, time of burning, and thence gave out forest-fire harm intensity basis onthe BP NN. This showed that NN technique was an important tool on prediction of forest-fire.(5)By statistic analysis we knew that there existed some pertinency between forest-fireharm intensity of the prophrase springtime fireproofing and the winter average wind speed,snow fall, average relative humidity and average temperature in last winter. This paper usingpluralistic regression theory established the linear relational model, in which the winter ofFrigid temperate coniferous forest region it chose time from Oct. to Mar. and the winter ofTemperate broadleaved and conifer mixed forest region it chose time from Nov. to Feb.therelation model between months was also established in spring-fireproofing-time. All thesemodels constitute the model of latent forest-fire harm intensity in spring, and the proof-testingshowed that the prediction result approximated to actual value. So the variables and model thatthis paper chose were logical, which could used in prediction of long-term latent forest-fireharm intensity.(6) In this paper, "The forest-fire forecast system of Heilongjiang province" was settedup by using the teetmology of geography information system. This system connected weatherdata-base and forest-fire harm intensity and established identification model. By inputting theintraday weather data of all prefectural districts in fireproofing period, the intraday latentforest-fire harm grade could be forecastedand be made visuable.更多还原