【作者】 曹云生；

【导师】 杨新兵； 鲁绍伟；

【作者基本信息】 河北农业大学 ， 水土保持与荒漠化防治， 2011， 硕士

【摘要】 森林是陆地生态系统的主体,是人类赖以生存的重要自然资源,也是应对全球温室效应、生物多样性丧失、生态失衡等诸多环境问题的重要基础保障。森林的健康状况将直接关系到全球生态安全和人类社会的可持续发展。本研究以生态学、统计学理论为依据,结合样地调查和林场最新调查数据,建立冀北山地典型森林生态系统健康评价和预警指标体系,利用Matlab软件构建了基于支持向量机(SVM)的森林生态系统健康评价和预警模型,对冀北山地森林进行了健康评价和预警,旨在为冀北山地森林健康可持续经营提供理论依据。研究结论如下:支持向量机(SVM)是建立在统计学理论基础上的一种小样本机器学习方法,广泛应用于分类和回归问题。SVM的最终决策函数只由少数的支持向量所确定,可以抓住关键样本、剔除大量冗余样本,该方法不仅算法简单,而且具有较好的推广和泛化能力,能以较小的样本预测较大的范围,同时,能保证较高的准确率。本研究以蒙特利尔进程中提出的指标为基础,参考国内外相关研究成果,构建了基础性指标、结构性指标评、抗干扰性指标和生态服务功能指标的健康评价指标体系,以样地调查所获取的21个典型森林生态系统类型的数据作为学习样本,通过SVM的自适应学习,并对分类器性能进行评价,同时对给出反馈信息进行修正,建立新的森林生态系统健康评价模型,运用该模型对研究区进行健康预测。对全区225个小班的预测结果表明:优质的森林小班面积占研究区总面积的5.7%,健康的森林小班占61.5%,亚健康的森林小班面积占28.1%,不健康的森林小班面积占4.7%,总体上健康状况良好。参考本研究建立的森林健康评价指标体系,结合北京森林健康预警指标体系,确定了本研究中的预警指标体系。采用北京森林健康预警基值和警兆指标值作为训练集和预测集进行训练,并对模型的分类准确率进行检验,通过训练好的模型对研究区森林生态系统健康状况进行预警,根据森林健康警限来划分警度。研究区225个小班的预测结果为:无警森林小班面积占研究区总面积的35.37%,轻警占58.51%,中警占0.45%,重警占5.66%。最后,根据森林健康评价和预警结果以及实地考证,提出了冀北山地森林健康经营技术。更多还原

【Abstract】 The forest is the subject of terrestrial ecosystem, not only to the survival of humans important natural resources, but also to cope with global greenhouse effect, the loss of biodiversity, ecological balance destroyed and so on. Forest is the important basis of well-protected ecological environment and environment problem.Forest health will directly related to the global ecological security and the sustainable development of the human society. According to the ecological and statistics theory , based on the data from survey of permanent sample plots and the the latest survey data. This research constructed the health assessment and early warning system for northern mountain of Hebei.Use Matlab established based on SVM health assessment and early warning model in the typical forest ecosystem of northern mountain of Hebei. From the perspective of forest ecosystem for health assessment and early warning, aiming at provide theory basis for the forest sustainable management. The main research conclusions are as follows:This study based on Montreal process proposed indexes,reference related research achievements at home and abroad,constructed forest health evaluation index system based on support vector machine (SVM). In this study,to investigate the 21 typical forest ecosystem data as learning samples, through the SVM adaptive learning to build new classification evaluation model, and with the classification and evaluation model to evaluate the performance of the classifier, meanwhile give feedback information for learning, establishing forest health classification model. Applying this model combining the latest survey data of study area to predict the rest subcompartment,obtained the predicted classification results.In this study, SVM achieved good prediction for evaluation of forest health. For 225 predictive results of subcompartment of quality classes have 21, accounting for a total area of 5.7% of study area, Healthy forest subcompartment accounting for 61.5%; Sub-health forest subcompartment accounting for 28.1 %, Unhealthy forest subcompartment accounting for 4.7 %.This study applies to the Beijing forest health base value and warning signs index for training set and prediction set,and established early warning model.Get this research areas of the relevant early-warning index, warning the forest ecosystem health through the trained model in 2010 , and differentiate early warning results warning degrees according to the forest health condition .To predict the results of the research region:No warning forest subcmpartment accounting for 35.37%;Slight warning forest subcompartment accounting for 58.51%; Middle warning subcompartment accounting for 0.45%, Heavy warning forest subcompartment accounting for 5.66%. Make corresponding measuresa acording to the forest health assessment and warning results,Put forward the forest health management technology.更多还原