【作者】 张海玉；

【导师】 范文义；

【作者基本信息】 东北林业大学 ， 森林经理学， 2011， 博士

【摘要】 森林是地球上最重要的资源之一,是人类赖以生存的自然资源,是陆地上面积最大、分布最广、组成结构最复杂、物质资源最丰富的生态系统,是陆地生态系统的主要碳汇。然而,七十年代、八十年代人们对森林进行了大量的采伐,导致生态系统环境的严重破坏。随着天然林保护工程的启动,从根本上遏制了生态环境恶化,保护了生物多样性,促进了社会、经济的可持续发展。在天然林保护期间,森林生物量碳储量发生了很大的变化,对森林生态系统有着积极的作用。因此,研究森林生物量变化特别是天然林保护期间森林生物量及变化意义重大。我国东北森林是世界上3大块温带森林之一,占全国森林面积和森林蓄量积的1/3以上,在我国和全球碳循环、林业和生态环境建设中起着举足轻重的作用。然而东北森林碳循环的研究尚不全面,在我国和全球碳循环评估、建模和预测中还急需来自该区域的研究成果。本研究结合遥感技术和地理信息系统技术,分析和探讨了张广才岭和完达山地区森林生物量的估算方法和森林生物量时空变化规律。本研究采用各种与生物量相关的遥感因子、GIS因子、气象因子、社会经济因子建立估算张广才岭和完达山地区的生物量模型,并且分析生物量的时空变化信息。研究生物量变化的自然驱动因子、人为驱动因子和社会经济驱动因子,并作定量的研究,分析各因子对生物量变化的驱动贡献量。与以往的定性分析和线性定量分析比较,找出最适合模拟这种驱动模型的方法,对生物量变化及驱动因子做非线性的定量分析。为研究此区域的森林产量、森林碳收支及碳循环模型等提供参考依据。针对上述问题展开了如下研究并得出相应的主要结论：1、大区域和大时间尺度的遥感影像由于受大气条件、光照条件、地表起伏等因素的影响,造成同一地区所成的影像具有较大的辐射差异。这种差异对于后续定量提取信息具有重要影响。只有在进行生物量定量反演以及时空变化规律分析之前进行几何校正、辐射校正、大气校正、太阳高度角的校正以及多时相遥感影像的辐射归一化,才能保证精度要求。2、通过研究多元逐步回归方法、线性和非线性偏最小二乘方法、BP神经网络及改进Erf-BP神经网络建模方法,发现以改进的Erf-BP神经网络模型模拟和预测结果最好,预测精度达到了84.91%,列于几种方法之首。其次是非线性偏最小二乘模型,模拟精度达到85.8%,预测精度为83.08%,线性偏最小二乘模型模拟精度为83.08%,预测精度为83.09%;传统的BP神经网络位于其后,网络预测精度为80.97%；传统的多元逐步回归分析建立的生物量模型模拟精度仅为76%,而预测精度为75%。但受制于运算时间,导致偏最小二乘模型和神经网络模型很难推广应用。3、分析70年代到现在森林生物量变化受自然因素、人为因素的影响,并做定量研究。利用偏最小二乘的线性和非线性方法计算了气候变量、经营措施变量和经济因素变量的VIP值,通过VIP值的大小进行森林生物量变换的驱动力和驱动机制的分析。森林生物量变化在70—80年代经营措施是主要驱动因子,80—90年代3类因子都起到重要的影响作用,90年代末到现在经营措施类因子的重要性大幅度降低,自然因素和社会经济因素的影响升高,说明天然林保护工程初见成效。结果表明,本文所使用的方法正确,能够准确预算出研究区域的四期生物量,并且对于研究生物量的变化、变化机制和驱动力作出正确的分析和判断。更多还原

【Abstract】 Forest, the important resource on surface, is the natural resource on which human survival depends. It is not only an ecosystem with the biggest area, the most widely distribution, complex composition and abundant material resources, but also a main carbon sink of the terrestrial ecosystem. However, people had cut large numbers of woods during 70s and 80s that led to heavy destruction of the ecosystem. During the period of natural forest protection, forest biomass and carbon storage have undergone great changes that have a positive effect on forest ecosystems. Therefore, it is significant to study changes of forest biomass, especially in natural forest protection time.Northeast forest in China, one of the large temperate forests in the world occupied over a third of national forest area and volume, plays an important role in global carbon cycle and entironment constructions. However, researches on forest carbon cycling in Northeast are not comprehensive. Carbon cycle estimating, modeling and forecasting in the region are also needed in Chinese and the global research. This paper analyzes and discusses some estimating methods of forest biomass and its rules in temporal and spatial changes, combining with RS and GIS technology.This paper develops biomass estimation models of Changbai mountain using many relevant factors, such as remote sensing factors, GIS factors, meteorological factors and economic factors, then analyses information of temporal and spatial changes. It also focuses on natural driving factors, factitious driving factors and economic driving factors, and analyses each factors for biomass changes quantitatively. Comparing with formerly qualitative and linear quantitative analysis, this paper finds the most proper method to simulate the driving mechanism and makes nonlinear analysis of biomass changes and driving factors, which provides references for investigating forest productions, carbon budgets and carbon cycling models.Some conclusions were made according to above problems:1. Due to atmospheric conditions, lighting conditions, surface fluctuations and other factors of remote sensing image in large area and large time scales, images in the same region have large radiative differences. These differences have important quantitative effects on extracting information. Only when geometric correction, radiometric correction, atmospheric correction, sun elevation angle calibration, radiation normalization of multi-temporal remote sensing images are made before forest biomass estimation and tempory-spatial analysis can precision have good results.2. Models for estimating forest biomass quantitatively were discussed and established by using multivariate stepwise regression, linear and nonlinear partial least-squares (PLS) method, back-propagation (BP) neural network model and back-propagation neural network based on gaussian error function (Erf-BP) model. The best predicted model among them was Erf-BP neural network with the precision of 84.91%. The second one was nonlinear PLS model with the fitted and predicted precision of 85.8% and 83.08% respectively. The third was linear PLS model with the fitted and predicted precision of 83.08% and 83.09%. The fourth was BP neural network model with the precision of 80.97%. The last one was multivariate stepwise regression with the fitted and predicted precision of 76% and 75%. Because of the limitations of calculating time, PLS and neural network models were hard to popularize and apply.3. The impacts of some factors on biomass variations, such as natural factors and man-made factors were analysed quantitatively in this study. Linear and nonlinear PLS models were used to calculate variables’VIP values of atmosphere, management and economy. These VIP values were useful for analyzing driving mechanism of biomass changes. The results indicated that the main driving factors of biomass changes between 70s and 80s,80s and 90s,90s and 2000s were management factors, all of the three factors, natural and social factors respectively. This result showed the effects of Natural Forest Protection Project.In this paper, not only correct methods were used for predicting biomass of the study region, but also biomass changes, changes mechanism and driving factors were identified and analyzed.更多还原