【作者】 仝慧杰；

【导师】 冯仲科； 罗富和；

【作者基本信息】 北京林业大学 ， 森林经理学， 2007， 博士

【摘要】 论文的研究目的是通过遥感的波段信息及其派生数据、地形数据、气候数据，建立森林生物量的遥感反演模型。目前国内外的研究基本上有三种建模方法。第一种是基于统计模型，利用回归算法建立生物量与遥感信息的回归模型，这种模型只适合一时一地一事的情况。不能解释机理，参数之间缺乏逻辑性。对于不同地区、不同条件，往往可以得出多种统计规律。不能进行时间和空间的外推。第二种方法，是建立神经网络模型。应用神经网络方法的“黑箱”操作，可能精度比回归方法高，但同样不能解释机理，也只适合一时一地一事的情况，不能进行时间和空间的外推。第三种方法是基于机理的过程模型，这种模型参数众多，难以获取，方程复杂，实用性较差。对参数进行简化后使用，甚至全球使用某一固定值，与实际值差距较大。为了解决以上问题本研究提出了根据地理相似理论与相似准则建立生物量、遥感信息、地貌因子、气候因子的建模方法。解算数据来自2001年的森林清查北京北部山区部分固定样地数据。将样地调查的活立木总蓄积利用连续系数法转换为活立木生物量作为模型的解算值。对试验区的多期TM图像利用DEM数据进行了正射校正，利用测区的降水等值线图和测站点的观测值生成了降水影像，从而精准的获取了建模信息，建立了森林特征的独立因子团。利用地理相似准则建立了生物量的遥感模型。计算了模型参数：地理指数、地理系数在时间和空间上的分布，预测了3年后的试验区生物量。同时用逐步回归的方法建立了统计回归模型并对这两种模型进行了精度比较。在模型的建模方法上创新，第一次将地理现象的相似准则用在森林生物量遥感建模；使森林生物量遥感建模有了新的方法，这种方法不仅利用了已知的规律，而且兼顾了随机性和模糊性。并且和森林生物量的维量分析法达到了统一。建立了不同树种幼龄林、中林龄和近熟林的遥感生物量模型。在森林特征因子建立和选择上进行了创新；选择了植被覆盖度、建立了TM吸收反射与反照度因子、年龄因子等具有生态意义的森林因子团。结合NASA-CASA和马蔼乃的NPP模型创造性地构建了光合作用因子团。在模型中首次使用植被覆盖度f_g这个遥感因子代替外野调查的郁闭度。在模型的时间预测上进行了创新，利用NASA公布的MODIS植被指数年变化图，对不同时期的遥感影像的植被指数进行了换算；模型最小单元是像元，对影像可以进行像元计算，推进了定量遥感在森林生物量估测方面的应用。更多还原

【Abstract】 The research aims to construct forset biomass model based on remote sensing information, topogaphy and climate data according to the theory of geographical similitude.At present the methods of modeling forset biomass are three kinds. The first one is using statistic method by correlation analysis and regression to obtain linear equation. This kind models can not explain the mechanism and lack of logic between the parameters. Many statistic rules have been gotten from different conditions and different regions. The models can not scaling in spatial and time domain, So that they only apply to the certain region, certain time and certain thing.The second is neural network model which is complete black system and no help for explaining the mechanism and has the same shortcomings with the first kind. The third is the mechanism model, ie process-based model, which simulate the bionomics process. This kind models have many parameters that are difficult to get, and complex equation, so the practicality is limited. This research adopts geographical similitude standards for modeling to solve the problems mentioned above.The biomass data and forest information used for model computed are converted from the sixth forest inventory 199 pieces sample plots investigated in 2001 located in mountain area north beijing. The remote sensing information derived from landsat5 TM image captured on Aug 31th 2001and Sep 8th 2004 and rectified to orthoimage . the topographic data such as slope, aspect computed from the DEM 1:250,000. The climate data that are processed to image in GRID format are the 30 years average value from 1971 to 2001 observed by beijing climate center. The independence factors groups and biomass model are established according to geographical similitude standards . The geographical indexes and geographical parameter are computed. The biomass of the plots 3 years later is predicted by the model through calculating the remote sensing information in 2004. while trying the new modeling method, using statistic method by correlation analysis and step regression, linear equation was gotten and compared with the model established based on theory of geographical similitude in model accuracy. This research firstly establish the forest biomass remote sensing model based on the theory of geographical similitude phenomena. This modeling method not only uses the rules discovered but also considers the random and fuzzy of the factors. The model is consistent with the growth equation used in forest for many years coincidently. The independence factors groups in forest specialty and remote sensing were derived, such as age, Photosynthesis, forest absorbing reflection ratio of TM image. Especially Photosynthesis independence factors group is combined from NASA-CASA model and MaAinai NPP model. The vegetation covering ratio f_g is firstly used in the remote sensing biomass model. The different tree species in different age class models were established which were used to predicted 3 years later biomass of the plots. The vegetation index (NDVI) was transformed successfully between different time according to MODIS vegetation index season changes and used in prediction computed. The model is based on image pixels, so can calculate image and produce biomass image.更多还原