【作者】 李明泽；

【导师】 范文义；

【作者基本信息】 东北林业大学 ， 森林经理学， 2010， 博士

【摘要】 全球气候变化已经是一个无可争议的事实,极端天气和自然灾害频发已经严重的影响了人类的生产和生活。哥本哈根会议虽然未能就减排等问题达成一项具有约束力的协议,但发展低碳经济促进节能减排已经成为各国的一个基本共识,同时也可以看出气候变化是个极其复杂的问题,远远超出环境的传统范畴,它涉及政治、经济、国际法律等诸多复杂问题。因此全球碳循环也就成为科学界广泛关注的研究课题。陆地生态系统中大约77%的植被碳储藏在森林生物量中,因此森林生物量是陆地生态系统碳循环过程中最主要的参数,它直接反应森林的碳储量。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。本研究立足于相关的两个科学问题,一是如何准确的获取大尺度的森林生物量,二是森林生物量的时空变化及其驱动定量分析。遥感和地理信息系统是解决问题的关键技术,针对上述问题展开了如下研究并得出相应的主要结论：1、遥感数据的处理,包括几何校正和辐射校正,研究大尺度问题时需要多期不同时相的遥感数据的辐射归一化,这是正确利用遥感信息的基础工作。2、将东北林区按植被地理分布划分为大兴安岭,小兴安岭和长白山,结合行政区划将长白山分为吉林长白山和黑龙江长白山,四个区域分别估算森林生物量。3、地面生物量模型的研建。这是遥感估算森林生物量的基础,本研究采用两种方法建立地面生物量模型,一是常规统计模型,应用于大兴安岭和长白山林区,包括大兴安岭7个主要树种,长白山林区18个主要树种。二是统一生物量模型,本次研究提出了“切比雪夫(chebyshev)正交多项式配合偏最小二乘建立地面统一生物量模型”,把生物量模型看作是连续函数空间中的一个元素,找到这个空间的一组基,则树种生物量统一模型可表示为这组基的线性组合。模型提出的过程,数学推导严密,参数用偏最小二乘方法解算,模型结果与已有的生物量模型结果比较并用地面实测数据进行验证,整个建模过程科学严谨,该模型具有很好的通用型,在地面生物量模型方面是一种新的方法。该模型应用于小兴安岭林区,建立了16个主要树种以胸径为自变量的统一生物量模型。统一生物量模型比常规统计模型平均精度高出5%以上。4、根据不同区域的乔、灌、草生物量模型计算森林资源清查样地的生物量,以此作为建立遥感模型的基础数据。5、采用逐步回归、BP神经网络和Erf-BP神经网络常规手段建立各个区域的森林生物量估算模型,结果表明逐步回归模型精度较低,为75%左右,难以达到精度要求；Erf-BP神经网络模型精度较高,达到80%以上,但受其自身算法特点的约束难以在大区域进行推广应用。6、偏最小二乘模型估算森林生物量是一种新的方法,使用该方法估算森林生物量与回归模型比较精度有较大提高,达到80%以上,特别是非线性偏最小二乘模型效果更好,但由于该方法算法复杂,程序运行时间长,非线性模型形式不确定,给大区域的遥感估算带来问题,因此该方法也仅停留在小区域实验上7、联立方程组与度量误差模型估算森林生物量,这是新型统计方法在遥感提取地物参数方面新思路,研究结果表明,该方法比常规统计方法更具合理性。联立方程组与度量误差模型在林分生长模型中有较好的应用,但尚未应用于遥感提取专题信息。本研究将联立方程组与度量误差模型引入到森林生物量的遥感估算中,一方面探索了新的遥感估算模型,另一方面为多传感器的联合估算专题信息提供了新的方法。生物量与叶面积指数的联立结合了多角度遥感,采用物理模型与统计模型相结合的算法,针叶模型平均检验精度83.3%,RMSE=17.72,阔叶模型平均检验精度83.0%,RMSE=20.28;生物量与树高联立结合了激光雷达,平均检验精度81.0%,RMSE=15.19;生物量与后向散射系数联立结合了微波遥感,对主被动遥感联合进行森林生物量估算的方法进行尝试,平均检验精度83.9%,RMSE=20.36。这些方法均能够在一定程度上提高森林生物量的估算精度,最后考虑生物量估算的可实现性选择了郁闭度联立方程组模型作为本研究的最优模型,该模型平均检验精度83.1%,RMSE=20.01。8、森林生物量时空变化分析。在GIS空间分析和地统计分析的基础上,广泛收集了研究区域几十年的气象数据、经营活动数据和社会经济数据,采用典型相关分析、主成分分析和偏最小二乘相结合的算法,开发了区域森林生物量变化驱动因子分析的计算程序,定量地计算出每个因子对于生物量变化影响的重要性值。根据上述研究得出结论,逐步回归模型精度难以满足要求,神经网络模型和偏最小二乘模型精度较高但算法复杂,仅能用于小区域实验研究,联立方程组和度量误差模型精度较高,算法简单适合大区域遥感估算。对森林生物量的时空变化分析表明森林生物量变化在70—80年代经营措施是主要驱动因子,80—90年代从各个影响生物量变化的因子重要值看,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。更多还原

【Abstract】 Global climate change is an indisputable fact, extreme weather and frequent natural disasters have seriously affected human production and life. Though it does not reach a binding agreement on emission issues in Copenhagen conference, the development of low-carbon economy and the promotion of energy conservation are becoming a basic consensus of all countries. Meanwhile, it shows that climate change is an extremely complex issue, far beyond the traditional scope of the environment. It involves political, economic, international law and many other complex issues. Therefore, global carbon cycle becomes a widespread concern research topic in scientific community. About 77% of the vegetation carbon stores in forest biomass in terrestrial ecosystems. So forest biomass is the most important parameter in terrestrial ecosystem carbon cycle, which directly reacts to forest carbon stocks. Northeast forest of China, one of the world’s three large temperate forests which occupied above 1/3 of the country’s total forest area and volume, plays an important role in China and global carbon cycle, forestry and ecological environment construction. However, forest carbon cycling in the northeast forest is not yet comprehensive, carbon cycle assessment, modeling and forecasting of our country and global are still needed research results in this region. This study is based on two scientific northeast forest-related issues, one is how to accurately obtain the large-scale forest biomass, and the other is analysis of spatial and temporal changes in forest biomass and quantitative analysis of driving. Remote sensing and geographic information systems are key technologies to solve the problem. The researches and corresponding conclusions according to the above issues are as followed:1、Remote sensing data processing, including geometric correction, radiometric correction and radiation normalized at different phases of large-scale remote sensing data is the right base work for using remote sensing information.2、The northeastern forest was divided into Daxing’an Mountain, Xiaoxing’an Mountain and Changbai Mountain by the geographical distribution of vegetation. Changbai Mountain is divided into Jilin and Heilongjiang Changbai Mountains combined with the administrative divisions. Forest biomass of four regions was estimated separately.3、The establishment of ground biomass model is the basis of estimating forest biomass. Two methods were used to establish ground biomass model in this study, one is conventional statistical model used in Daxing’an Mountain and Changbai Mountain, including 7 main tree species of Daxing’an Mountain and 18 main tree species of Changbai Mountain, the other is uniform biomass model. "Chebyshev orthogonal polynomial with partial least square to establish ground biomass unified model" was proposed in this study. It takes the biomass model as an element in continuous function space. A group bases in this space were found to express as a linear combination for tree biomass unified model. Rigorous mathematical derivation was in model proposed process, and parameters were calculated using partial least squares method. Model results were compared with existing biomass model results and ground-measurement data. The whole modeling process is scientific and rigor. The model is a general type which is a new approach for above-ground biomass modeling. The model was applied to forest of Xiaoxing’an Mountain with the establishment of unified biomass model of same independent variable DBH for 16 major tree species. The average accuracy of unified biomass model is 5% higher than conventional statistical models.4、Forest biomass in inventory plot was calculated according to the tree, shrub and grass biomass models of different regions, as the basis for the establishment of remote sensing data model.5、Forest biomass estimation models in various regions were established by stepwise regression, BP neutral network and Erf-Bp neural network methods. The results show that the stepwise regression model was less precise, about 75%, which was difficult to achieve accuracy; Erf-BP neural network has high precision, about 80%, but is difficult to promote in large region for its own algorithm characteristics.6、Partial least squares model is a new method to estimate forest biomass. Compared with the regression model, the accuracy of this method to estimate forest biomass is higher, above 80%. In particular, nonlinear partial least squares model is better. However, the algorithm is complex and costs long running time. Because of the uncertainty of nonlinear model form, it is a problem to estimate by remote sensing in large area, so the method only stops at a small area experiment.7、Joint equations and measurement error model to estimate the forest biomass is a new statistical method to extract spatial parameters in remote sense. The results of the researches show that this method is more than reasonable than conventional statistical methods. Joint equations and measurement error model has good application in forest stand growth model, but have not yet applied to remote sensing information extraction project. In this study, joint equations and measurement error model is introduced into the remote sensing estimation of forest biomass, on one hand to explore a new remote sensing estimation model, on the other hand to provide a new method to estimate information of joint multi-sensor. Multi-angle remote sense was used in the joint equations of biomass and leaf area index. And the biomass estimation was combined with physical model and statistical model, with the average test accuracy of 83.3%, RMSR of 17.72 in needle model, and the average test accuracy of 83.0%, RMSE of 20.28 in broad-leaves. Laser radar was used in the joint equations of biomass and tree heights, with the average test accuracy of 81.0%, RMSR of 15.19. Microwave remote sense was used in the joint equations of biomass and backscattering coefficients. The joint method of active and passive remote sense for forest biomass estimation was tried, with the average test accuracy of 83.9%, RMSR of 20.36. To some extent, these methods all improved the estimation accuracy of forest biomass. Finally, considered the implementability of biomass estimation, the joint equations of biomass and crown density were chose in this study, with the average test accuracy of 83.1%, RMSR of 20.01.8、Analysis of spatial and temporal changes in forest biomass. Under the foundation of the GIS spatial analysis and geo-statistical analysis, a computer program for analyzing change-driven factors of regional forest biomass was developed for calculating importance value of each factor to biomass change quantitatively, using canonical correlation analysis, principal component analysis and partial least squares algorithm based on extensive collections of regional meteorological data, business activity data and socio-economic data for several years.According to the conclusions of the study, it is found that the stepwise regression model can not meet the required precision, the neural network model and partial least squares algorithm with high precision are complex and only used in small experimental area, the joint equations and the measurement error model is the best with high precision, simple algorithm and suitability for remote sensing estimation in large area. On the analysis of spatial and temporal changes in forest biomass, it showed that management measures are the main driving factors of forest biomass changes in the 70-80 years; three categories of factors all have played important roles in the 80-90 years from the view of importance values of various factors affecting biomass change; from the lat 90s to the present, the significance of management measures reduced, but natural factors and socio-economic factors increased, which indicated the initial success of natural forest protection project.更多还原