中高分辨率遥感影像森林类型精细分类与森林资源变化监测技术研究

发布时间：2018-05-09 03:07

本文选题：森林类型 + 精细分类　；参考：《中国林业科学研究院》2016年博士论文

【摘要】：近年来,随着遥感技术的发展和遥感影像林业应用的深入,应用中高空间分辨率遥感影像开展森林类型精细识别和森林资源动态变化监测成为目前研究的热点之一。虽然遥感影像分类技术取得了长足的发展,但是已有研究表明,森林类型信息获取中仍存在精度不高、详细程度和可信度差等技术难点,以及森林资源动态变化监测尚未完全克服信息获取周期长、变化信息提取困难、新方法应用少、自动化程度低、成果质量和精度欠佳等突出问题。本文面向国家森林资源监测调查的行业应用需求,重点研究高精度森林类型精细识别方法和森林资源动态变化监测技术,为及时、准确掌握森林资源现状和动态变化趋势提供可靠支撑,为森林资源空间配置、优化调整与辅助决策提供技术支持。本文主要内容和结论如下:(1)以嘉陵江上游甘肃省小陇山林业实验局百花林场为例,探讨复杂中山区域、多源数据支持下,高空间分辨率遥感影像森林类型层次化精细分类方法。以SPOT5和高分一号(GF-1)遥感影像为主要数据源,综合利用影像光谱特征、植被指数特征、纹理特征与时相特征、地形特征、森林资源“二类调查”成果数据与林相图等辅助信息,及典型地类与主要森林类型外业调查样本数据,发展了针对暖温带典型天然次生林区、复杂山区地形条件下高空间分辨率遥感影像森林类型多层次信息提取与森林类型精细识别的有效方法。采用分层随机抽样的独立检验样本对分类结果中7类林地与森林类型进行精度验证,并对5类主要森林类型精细识别结果进行面积统计,与“二类调查”及影像解译结果各类型面积统计值进行对比分析。研究结果表明,本文所发展的分类方法对森林类型信息提取精度较高,有林地、其他林地、苗圃地等7类林地和森林类型总体分类精度达92.28%,总Kappa系数为0.899 6;油松林、华山松林、日本落叶松林、栎类落叶阔叶林、其他落叶阔叶混交林等5类主要森林类型面积统计结果的平均相对精度为92.4%。本文发展的多源数据支持下的多层次森林类型精细分类方法是一种有效的森林类型信息精准监测方法,具有精度高和可信度高的优势,且森林类型精细识别详细程度达到优势树种(组)级别,是解决复杂山区森林类型信息提取与精细识别的一种有效手段,可满足森林资源调查、变化监测、数字更新等林业应用需求。(2)以甘肃省天水市为例,以1990年~2015年五期冬夏时相landsattm/oli遥感影像为主要数据源,结合辅助数据和外业实地样本点,在对光谱特征、指数特征、时相特征等分类特征综合分析的基础上,选取ndvi、ndwi、ndi和mtvi等四个指数作为特征变量,发展了基于两种非参数分类器(随机森林(rf)和参数优化支持向量机(posvm))分类后比较法的森林资源变化监测技术。研究结果表明:引入多元特征和稳健、优化的非参数分类器,可显著提高分类精度和分类结果的可信度,降低类别混淆和结果的不确定性。两种分类方法均取得了较好的分类效果,具有较高的空间一致性,且时序分类结果及逐期变化分析结果可准确、客观地反映该区域近30年来森林资源时空动态变化。随机森林(rf)分类方法在分类精度、效率、计算量和稳定性方面明显优于参数优化支持向量机(posvm)分类方法,随机森林(rf)方法对于复杂地形、破碎地貌区域和典型植被(森林-灌草-草地)交错过渡区具有的较强的适应性,可应用于大区域、复杂地形、过渡区域的植被/森林制图和动态变化监测。监测结果表明:1990年~1996年林地转化为非林地为3.764%,非林地转化为林地为3.024%。林地面积净减少0.74%。1996年~2002年林地转化为非林地为5.648%,非林地转化为林地为2.914%。林地面积净减少2.734%,林地减少呈现加剧趋势。2002年~2008年林地转化为非林地为5.574%,非林地转化为林地为6.631%。林地面积净增加1.057%。2008年~2015年林地转化为非林地为6.563%,非林地转化为林地为15.446%。林地面积净增加为8.883%。该区域近30年来森林资源变化的总体趋势:以2002年(2002期影像)为界,林地面积为先减少后增加,2002年后林地面积增加显著。林地面积增加主要原因为其它类型向有林地转化,由于1999年以后,随着天然林保护工程、退耕还林工程等林业重点工程的实施,使得该区域森林覆盖率上升趋势明显,林地面积显著扩大。林地面积明显减少的区域在空间上主要集中在武山县和张家川回族自治县,在林地与其他地类交错过渡地带、林缘区域表现尤为明显,尤其在2002年之后。林地面积减少的可能原因为自然因素及人类活动影响使该区域原有的森林、灌木群落等植被遭到破坏,林地转变为耕地、草地和建设用地等,局部生态环境状况有进一步恶化的风险和可能。(3)在对地类、林地-非林地信息提取与变化监测结果分析的基础上,进一步对“常绿针叶林”、“落叶阔叶林”、“其他林地”三类林地内类型的动态变化(转化)进行深入分析,对四个时间段内三类林地内类型的“增加”(转入)、“减少”(转出)变化区域、空间分布和变化程度状况进行过程分析。研究结果表明:(1)1990年~1996年间,“常绿针叶林”类型增加(转入)、减少(转出)均较为明显。增加区域分布于秦岭南坡小陇山林区的各个林场;减少区域位于秦岭北坡、沿渭河流域的带状区域,以及嘉陵江上游低海拔区域的河流及山谷两侧。1996年~2002年间、2002年~2008年间,在整体上“常绿针叶林”类型增加、减少均不明显。2008年~2015年间,“常绿针叶林”类型增加十分显著。增加区域分布于小陇山林区、西秦岭林区和关山林区的所有林场,仅在龙门林场中部有小片减少区域。(2)1990年~1996年间,“落叶阔叶林”类型增加(转入)、减少(转出)均较为明显。增加区域主要分布于太碌、立远、东岔林场,以及党川、观音林场、龙门林场南部;减少区域分布于秦岭南坡小陇山林区的各个林场,尤其在小陇山林区北部边缘区域和林区中部表现尤为明显。1996年~2002年间、2002年~2008年间,“落叶阔叶林”类型增加区域主要分布于小陇山林区的麦积林场、东岔林场、立远林场等。“落叶阔叶林”类型减少主要分布于东岔、立远林场及小陇山林区北部边缘区域。2008年~2015年间,“落叶阔叶林”类型增加区域主要分布在秦岭北坡渭河南岸区域、麦积林场,以及天水市辖温泉、尖山林场。“落叶阔叶林”类型减少主要分布在麦积林场和滩歌林场。(3)1990年~2015年间,“其他林地”类型增加(转入)、减少(转出)区域均主要分布于林区边缘,增减变化幅度不大。基于两种非参数分类器分类后比较法的遥感影像变化监测技术,探讨了典型黄土高原丘陵沟壑与陇山-西秦岭山地交接过渡区域近30年来森林资源空间分布规律、时间变化趋势及变化影响因素,以期为该区域森林动态变化定量分析及综合评价、森林资源空间配置与优化调整、经营管理与辅助决策、林业工程进展监测、生态环境评价以及森林保护措施制定等提供一定的参考。
[Abstract]:In recent years, with the development of remote sensing technology and the in-depth application of remote sensing image forestry, the application of high spatial resolution remote sensing images to carry out fine recognition of forest types and monitoring the dynamic change of forest resources has become one of the hotspots of current research. Although the remote sensing image classification technology has made great progress, the existing research has shown that forest type In the acquisition of type information, there are still some technical difficulties, such as low precision, detailed degree and poor reliability, and the dynamic change monitoring of forest resources has not completely overcome the long period of information acquisition, the difficulty in extracting the change information, the low application of new methods, the low degree of automation, the poor quality of the results and the poor precision. In order to provide a reliable support for the timely and accurate grasp of the current situation and dynamic change trend of forest resources and provide technical support for the spatial allocation of forest resources, and the technical support for the optimization adjustment and auxiliary decision, the main content and the main content of this paper are the main content and the main content of this paper. The conclusions are as follows: (1) taking the Baihua forest farm of the Xiaolong Forestry Experiment Bureau of the upper reaches of the Jialingjiang River in the upper reaches of Gansu Province as an example, the detailed classification method of the high spatial resolution remote sensing image forest types was studied under the support of the multi source data, and the remote sensing image of SPOT5 and GF-1 was used as the main data source, and the image spectral characteristics were used synthetically. The index features, texture features and temporal features, terrain features, forest resources "two types of survey" data and forest phase map and other auxiliary information, as well as typical and main forest types of survey sample data, developed a typical natural secondary forest area of warm temperate zone, high spatial resolution remote sensing image under the complex mountainous terrain conditions. The effective method of multi level information extraction and fine recognition of forest types was used. The independent test samples of stratified random sampling were used to verify the accuracy of 7 types of forest and forest types in the classification results, and the area statistics of the fine recognition results of the 5 types of main types of forest types were carried out, and each type of "two types of investigation" and the results of image interpretation were used. The results show that the classification method developed in this paper has high precision for extracting forest type information. The overall classification accuracy of 7 types of woodland and forest types, including woodland, other woodlands and nursery fields, is 92.28%, the total Kappa coefficient is 0.8996, oil pine forest, Huashan pine forest, Japanese Larix forest and oak deciduous broad-leaved forest, The average relative accuracy of the statistical results of 5 major forest types, such as other deciduous broad-leaved mixed forests, is the precise classification method of multi level forest types supported by the multi source data supported by 92.4%., which is an effective method for accurate monitoring of forest type information, with high accuracy and high credibility, and the fine recognition of forest types. It is an effective means to solve the information extraction and fine recognition of forest types in complex mountainous areas. It can meet the needs of forest resources survey, change monitoring and digital updating. (2) taking Tianshui, Gansu as an example, the landsattm/oli remote sensing image of winter and summer in 1990 ~2015 year is taken as an example. On the basis of comprehensive analysis of spectral features, exponential features and temporal features, the main data sources, based on the comprehensive analysis of spectral features, exponential features and temporal features, are based on the combined analysis of four indices, such as NDVI, NDWI, NDI and mtvi, and develop two nonparametric classifiers (random forest (RF) and parameter optimization support vector machine (posvm)). The research results show that the introduction of multiple features and robust and optimized non parametric classifiers can significantly improve the classification accuracy and the reliability of the classification results, reduce the confusion of categories and the uncertainty of the results. The two classification methods have achieved a better classification effect and have a higher space. Consistency, time series classification results and phase by phase analysis results can be accurate, objectively reflecting the temporal and spatial dynamic changes of forest resources in the region during the last 30 years. The stochastic forest (RF) classification method is obviously superior to the parameter optimization support vector machine (posvm) classification method in the classification accuracy, efficiency, calculation and stability, and the random forest (RF) method is used. Complex terrain, broken landform area and typical vegetation (forest - grasses and grassland) interlaced transitional zone have strong adaptability. It can be applied to large area, complex terrain, vegetation / forest mapping and dynamic change monitoring in transition region. The results of monitoring show that in 1990, 3.764% of forest land was converted to non woodland in ~1996 and 3. of non woodland to Forestland in 1990. 024%. woodland area net reduction in 0.74%.1996 year ~2002 forest land conversion to non woodland 5.648%, non woodland conversion to woodland to 2.914%. woodland area net reduction of 2.734%, woodland decrease trend in.2002 year ~2008 year ~2008 forest conversion to non woodland is 5.574%, non woodland to woodland to 6.631%. woodland area net increase 1.057%.2008 year ~2015 1.057%.2008 year ~2015 The conversion of woodland to non woodland was 6.563% in the year, and the net area of non woodland converted to woodland was 15.446%. forest area net increase in the total trend of forest resources change in the area of 8.883%. in the last 30 years. In 2002 (2002 period images), the area of woodland was reduced first and then increased, and the forest land accumulation increased significantly after 2002. The main reason for the increase of forest land area was the others. Since 1999, since 1999, with the implementation of the key forestry projects such as natural forest protection project, returning farmland to forest engineering, the forest coverage rate of the region is rising obviously, the area of forest land is greatly enlarged. The area of the forest area is obviously reduced in the space mainly concentrated in Wushan County and Zhangjiachuan Hui Autonomous County, in the forest area. After 2002, the possible reasons for the reduction of forest land area are the natural factors and the influence of human activities, which cause the destruction of the original forest, shrub community and other vegetation, the transformation of the woodland into arable land, the grassland and the construction land, and the local ecological environment. The risk and possibility of one step worsening. (3) on the basis of the analysis of the information extraction and change monitoring results of the land class, woodland and non woodland, the dynamic changes (transformation) of the three types of woodland types in the "evergreen coniferous forest", "deciduous broad-leaved forest" and "other woodland" were further analyzed, and the three types of woodland types within the four time periods were analyzed. "Increase" (transfer), "reduce" (turn out) change area, space distribution and change degree state of the process analysis. The results show: (1) in 1990 ~1996, the "evergreen coniferous forest" type increased (transfer), reduce (transfer) is more obvious. Increase the area distribution in the small Longshan Forest Area of the south slope of Qinling Mountains; reduce the area On the northern slope of Qinling Mountains, the zonal region along the Weihe River Basin and the rivers and valleys on the upper reaches of the Jialing River in the upper reaches of the Jialing River in ~2002 years of.1996 years, the "evergreen coniferous forest" type increased in the period of ~2008 2002, and the decrease of the "evergreen coniferous forest" type is very significant in the ~2015 years of.2008. The small Longshan Forest Area, the western Qinling Mountains forest area and the Guan Shan Forest area all forest farms, only in the middle of the Longmen forest farm, there are small areas. (2) in 1990 ~1996, the type of "deciduous broad-leaved forest" increased (transfer), and the decrease (transfer) was more obvious. The region is distributed in every forest farm in the small Longshan Forest Area of the southern slope of Qinling Mountains, especially in the northern edge of the Xiaolong forest area and the middle of the forest area, especially in the middle of.1996 year ~2002. In 2002, the increasing area of "deciduous broad-leaved forest" was mainly distributed in the Maiji forest farm in the Xiaolong forest area, the East Fork forest farm and the Li Yuan forest farm, etc. " The type of deciduous broad-leaved forest is mainly distributed in Dong Cha, Li Yuan forest farm and the northern edge of Xiaolong mountain forest area in.2008 year ~2015, the increasing area of "deciduous broad-leaved forest" is mainly distributed in the South Bank of Weihe area of Qinling Mountains north slope, Maiji forest farm, Tianshui jurisdiction hot spring, Jianshan forest farm and the type of "deciduous broadleaf forest". In the Maiji forest farm and the tan song forest farm. (3) during the period of ~2015 in 1990, the types of "other woodlands" were increased (transferred), and the regions were mainly distributed on the edge of the forest area, and the changes were not significant. Based on the remote sensing image changing monitoring techniques of two non parametric classifiers, the typical Loess Plateau hilly and gully and longs were discussed. The spatial distribution law of forest resources in the transition region of mountain and West Qinling Mountains mountain area over the past 30 years, the time change trend and the influence factors, in order to be the quantitative analysis and comprehensive evaluation of the forest dynamic changes in this region, the spatial allocation and optimization of forest resources, the management and auxiliary decision, the progress monitoring of forestry engineering, and the ecological environment evaluation And provide some reference for the formulation of forest protection measures.

【学位授予单位】：中国林业科学研究院
【学位级别】：博士
【学位授予年份】：2016
【分类号】：S757;S771.8

【参考文献】