1、匿名评阅论文 论文编号 中国林科院硕士学位论文题目：基于多源数据的树种（组）空间分布信息提取方法English Title: Retriavel Method of Spatial Distribution Information of Forest Tree Species (Group) based on Multi-Sources Data专业： 地图学与地理信息系统学位论文答辩预定日期： 年 月 日I摘 要获取森林树种（组）空间分布信息是我国森林资源调查的重要内容之一，不仅可为国家森林资源管理宏观决策提供信息支撑，也是深入开展森林生态系统碳循环模型研究的重要输入数据之一。遥感技术为提取
2、树种（组）空间分布信息提供了一种有效的手段。中空间分辨率高重访周期遥感数据提取的时间序列 NDVI 数据能够较完整的反映植被的季相变化和物候差异，已被众多学者应用于大区域植被信息提取研究。同时，区域树种（组）的空间分布也受到诸多环境因素的影响，气温、降水量和地形等多类型数据的综合应用有利于树种（组）空间分布信息的提取。建立和检验树种（组）空间分布信息提取模型离不开大量地面实况调查数据的支撑，因此目前国外相关研究报道无一例外都采用了国家森林资源调查固定样地数据，但国内尚没有基于多源数据综合提取树种（组）空间分布信息的研究报道。为此本文在国外该方向最新研究进展基础上，开展了综合多源数据的树种（组）
3、空间分布信息提取方法研究。本文发展了一种以 MODIS NDVI 8 天合成时间序列数据（空间分辨率为 250m250m）和国家森林资源连续清查固定样地数据为主要数据源，综合利用气象观测数据和地形数据，基于梯度最近邻（GNN）方法的省级树种（组）单位面积胸高断面积（可理解为胸高断面积密度，单位为 m2/hm2，后文简称胸高断面积）和树种（组）成数估测方法。该方法首先利用典型对应分析（CCA）对特征变量进行特征变换，然后采用 k-最近邻（k-NN）方法对树种（组）胸高断面积和成数进行分层估测，其估测结果可用于间接表达主要树种（组）的空间分布。以黑龙江省和吉林省为试验区开展了实验研究，验证了所发展
4、方法的有效性，进而制作了黑龙江省 9 个树种（组）的胸高断面积分布图和吉林省 7 个树种（组）的成数分布图。本文主要研究工作和结论如下：（1 ） 研究了 k-NN 参数优选对估测精度的影响规律，摸索出了最佳 k-值的确定方法。本文在对黑龙江省的 9 个树种（组）的胸高断面积和吉林省的 7 个树种（组）成数进行估测时， 分别对 k 值进行了优选实验，分析了 k-NN 估测精度随 k 值的变化规律，总结出了最佳 k-值的确定方法。（2 ）发展了一种有效的分层估测方法。采用分层估测的方法对黑龙江省树种（组）的胸高断面积和吉林省树种（组）成数进行了估测，与不分层的直接估测法进行了精度对比，结果表明：对
5、于黑龙江省树种（组）胸高断面积估测，分层估测法的决定系数（R 2）比直接估测法平均高 0.07；对II于吉林省树种（组）成数的估测，分层估测法的均方根误差（RMSE）比直接估测平均低 0.1。本文发展的分层估测方法可以提高树种（组）空间分布信息提取精度。（3 ）采用两种精度评价方法对本文所发展的树种（组）空间分布信息提取方法进行了有效性检验。基于网格的精度评价结果显示：随着评价尺度（网格大小）的变化，黑龙江省 9 个树种（组）的胸高断面积的 RMSE 平均值在 0.44-1.68 之间；吉林省 7 个树种（组）成数的 RMSE 平均值在 0.35-0.65 之间。在吉林省实验区，以县为统计单元
6、采用森林资源二类调查加密样地进行了精度检验，结果表明：树种（组）成数估测 R2 为 0.83，RMSE 为 0.35。检验结果说明了本文所发展方法是有效的，所提取的主要树种（组）空间分布图对我国森林资源管理宏观决策、森林生态系统碳循环研究等具有潜在的重要应用价值。关键词：多源数据，GNN，CCA，k-NN，MODIS NDVI，树种成数，胸高断面积密度，制图IIIAbstract Access to forest tree species (group) of spatial distribution information is one of the important contents o
7、f Nation Forest Inventory, not only can provide information support for the national forest resources management decision-making, but also can be used as one of the important input data to the research of forest ecosystem carbon cycle models. Remote sensing technique provides a highly effective mean
8、s for extracting tree species (group) spatial distribution information. NDVI time-series data with medium spatial resolution and high revisit frequence incorporate intra-annual vegetation phenology and seasonal change and have been widely used in extracting vegetation information in large area. Furt
9、hermore, the spatial distribution of tree species (group) is influenced by environmental factors. We can extracte tree species (group) information more accurate by adding temperature, precipitation, topography and other types of data. The ground survey data is a support for establishing and testing
10、species (Group) spatial distribution information extraction model. Nation Forest Inventory plot data are widely used by scholars at home and abroad. But there is still no study on extracting tree species (group) spatial distribution information based on multi-sources data in China. Therefore, the pa
11、per carries on a method research of tree species (group) spatial distribution information retrival using mutli-sources data with the latest research developments abroad as basis. We developed a Gradient Nearest Neighbor (GNN) based approach for estimating provincial forest tree species (group) compo
12、sition and forest tree species (group) basal area per unit area (m2/hm2) distribution information with time series MODIS NDVI product of 250m pixel size and 8 days cloudy free composite and the permanent forest plot data collected by the National Forest Inventory (NFI) as the key data sources, and w
13、ith integrated utilization of weather observation data and topography data. The GNN method firstly applies Canonical Correspondence Analysis (CCA) to extract effective composited features from the original dependent and independent dataset, then it applies the k Nearest Neighbors (k-NN) method in th
14、e extracted feature space to estimate forest tree species (group) composition number and tree species (group) basal area using one two-layers stratification scheme, the result from which can be used to indicate the spatial distribution of forest tree species (group). The method described above has I
15、Vbeen studied with the whole Hei Longjiang Province and Jilin Province as test sites, and the basal area spatial density distribution map of 9 tree species (group) of Hei Longjiang Province and tree species composition spatial distribution map of Jilin Province of 7 tree species were produced. The r
16、esearch results are as follows:(1) We studied the effect of k-NN parameter optimization on the estimation accuracy. When we estimate basal area spatial density distribution map of 9 tree species (group) of Hei Longjiang Province and 7 ree species composition spatial distribution of Jilin Province, t
17、he value of k needs to be optimized respectively. We summed up the method to determine the optimal k- valueby analyzing the changing trend of k-NN estimation accuracy with the k values.(2) This paper developed a two-layers stratification estimation method. We used a two-layers stratification method
18、to estimate forest tree species (group) composition and forest tree species (group) basal area. The accuracy of direct estimation method and two-layers stratification estimation method was comapared. The results show that: the average R2 of estimating forest tree species (group) basal area of Hei Lo
19、ngjiang Province using two-layers stratification estimation method is 0.07 higher than that using direct estimation method; the average RMSE of estimating tree species (group) composition of Jilin Province using two-layers stratification estimation method is 0.1 less than that using direct estimatio
20、n method. So, It can improve the estimation accuracy using two-layers stratification estimation method.(3) We adopted two methods for validating the forest tree species (group) spatial distribution mapping results. In the case of the first validation method, the accuracy is computed by dividing the
21、whole coverage of the province into grids of several different size, taking the forest plot data collected by the NFI as reference and the grid as statistic unit. The average RMSE of estimating 9 tree species (group) basal area of Hei Longjiang Province is 0.44-1.68 and the 7 tree species (group) tr
22、ee species (group) composition of Jilin Province is 0.35-0.65. In the case of the second accuracy validation method, the accuracy for each tree species (group) is computed with the forest plot data of the 9 counties collected by the forest resources inventory in second level as reference data and ta
23、king county as statistic unit. The coefficient of determination (R2) of 0.83 and RMSE of 0.34 were achieved for the tree species (group) tree species (group) composition map of Jilin Province. The validation results show the effectiveness of the developed method of this paper, and the main tree spec
24、ies spatial distrition maps thuse produeed have great potential Vapplication value on supporting macro-scale decision-making for forest resources management and for regional forest ecological system carbon cycle researchs in China.Key words:Multi-data sources, GNN, CCA,k-NN, MODIS NDVI, Tree species
25、 composition, Basel area, MappingVI目 录摘 要 .IAbstract.III第一章 绪论 .11.1 研究背景与意义 .11.2 国内外研究现状及发展趋势 .31.2.1 高光谱分辨率遥感应用于树种（组）空间分布研究现状 .31.2.2 高空间分辨率遥感应用于树种（组）空间分布研究现状 .51.2.3 植被物候信息应用于树种（组）空间分布研究现状 .61.2.4 多源数据应用于树种（组）空间分布研究现状 .71.2.5 基于森林资源调查数据和高时间分辨率遥感数据的森林类型制图研究 .81.3 研究目标和研究内容 .121.3.1 研究目标 .121.3.2
26、研究内容 .121.4 技术路线 .131.5 论文结构 .15第二章 试验区和数据 .162.1 试验区概况 .162.1.1 黑龙江省概况 .162.1.2 吉林省概况 .172.2 数据情况 .182.2.1 样地数据 .182.2.2 时间序列 NDVI 数据 .222.2.3 时间序列气象数据 .222.2.4 地形及地理位置数据 .232.2.5 基于 Landsat TM/ETM+数据的植被覆盖度提取 .23VII2.3 本章小结 .24第三章 数据预处理与 GNN 估测方法 .253.1 时间序列特征变量处理 .253.1.1 MODIS NDVI 时间序列数据滤波处理 .25
27、3.1.2 MODIS NDVI 时间序列数据降维处理 .273.1.3 月平均气温和月总降水量数据的插值与降维处理 .283.1.4 树种（组）胸高断面积估测因变量和自变量 .293.2 GNN 估测方法 .293.2.1 典范对应分析（Canonical Correspondence Analysis, CCA） .303.2.2 k-NN 算法原理 .333.3 k-NN 分层估测方法 .343.4 估测精度评价方法 .353.4.1 基于多尺度网格的精度评价方法 .353.4.2 县级尺度精度评价方法 .383.5 本章小结 .39第四章 黑龙江省树种（组）胸高断面积估测结果与分析 .
28、404.1 样地树种（组）胸高断面积与自变量的定量关系分析 .404.2 k 值的优选 .414.3 k-NN 分层估测对估测精度的影响 .424.4 树种（组）胸高断面积的估测结果 .434.5 精度检验 .504.6 本章小结 .53第五章 吉林省树种（组）成数估测结果与分析 .545.1 样地树种（组）成数与自变量的定量关系分析 .545.2 k 值的优选 .545.3 k-NN 分层估测对估测精度的影响 .555.4 树种（组）成数的估测结果 .565.5 精度检验 .62VIII5.5.1 多尺度下的精度检验 .625.5.2 县级尺度下的精度验证 .635.6 本章小结 .64第六章 结论与展望 .666.1 结论 .666.2 展望 .67参考文献 .69在读期间的学术研究 .76致谢 .77IX图目录图 1-1 技术路线图 .14图 2-1 黑龙江省空间分布图 .17图 2-2 吉林省空间分布图 .18Fig.2-2 The spatial country distributi
