【作者】 张远；

【导师】 吴嘉平； 齐家国；

【作者基本信息】 浙江大学 ， 农业资源利用， 2009， 博士

【摘要】 水稻是世界上非常重要的农作物，也是拥有13亿人口的中国最主要的粮食作物，占全国粮食产量的40％以上。水稻产量无疑是今后农业生产和国家安全方面值得关注的重要问题。开展水稻遥感研究对农业与环境可持续发展、食品与水安全和温室气体减排等都具有重要意义。本研究选取浙江富阳和海宁市部分区域的水稻田作为研究区，目的是利用微波遥感技术与生态模型对水稻田遥感信息提取、水稻结构参数反演和水稻田对气候环境影响评价等方面进行综合探讨和研究。农业生产中常用的水稻面积的统计方法周期长，不仅需要花费大量的人力物力，而且获得面积的真实性也受到制约。遥感技术凭借实时、快速、大范围、客观准确等优点用于提取水稻种植面积是对传统方法的一个有力补充。另外，水稻结构参数的遥感定量反演一直是水稻遥感生物量估算和估产研究中的一个主要问题。因此，本研究尝试开发了一个用于模拟水稻散射特性和结构参数反演的一阶散射模型，为应用遥感技术进行水稻生物量估算提供了一个新的手段。目前，分析水稻田对气候环境的影响是生态领域的一个研究热点，本研究利用一个基于模拟生态过程的DNDC（Denitrification andDecomposition）模型对水稻生长期内的甲烷排放进行了模拟。通过情景分析，提出了有利于甲烷气体减排的耕作管理措施，并为今后减少和控制水田甲烷排放提供科学依据。具体研究内容和结果概述如下：（1）利用富阳市水稻生长期内获取的三景不同时期ALOS／PALSAR数据进行合成，生成多时相彩色复合图，利用支持向量机（SVM-Support Vector Machine）算法进行分类。目的是充分利用这三个水稻生长期内各地物的后向散射系数的时间变化差异进行感兴趣目标（水稻田）的提取。本研究主要是依据水田在不同时期的地物特征变化很大，而其他地物变化相对较小的独特特征，在时间维空间内进行向量分割，提取出水稻的种植面积分布。并利用高分辨率（2.5m）ALOS／PRISM数据、土地利用调查矢量图件及其实地踏查进行对比及分类精度验证。研究结果表明水稻类别的提取精度可以达到90％。（2）利用上述SVM自动分类方法，结合逐级分类、人工辅助纠正，对研究区浙江海宁市部分区域的水稻田进行提取。以水稻的分类专题图为掩模，得到只包含水田的雷达原始后向散射图层。利用与雷达卫星过境近同步的实测地面水稻结构参数作为模型输入变量，对广泛使用的用于模拟树木后向散射系数的辐射传输模型进行内部参数修改，得到适用于模拟水稻作物的冠层后向散射系数的模型。结果表明，改进模型的模拟值与地面实测点对应的雷达遥感数据后向散射系数值进行比较，两者具有很好的一致性。（3）以后向散射系数图像、雷达波束入射角为输入变量，利用改进的一阶辐射传输模型以及遗传算法优化工具GAOT，在Matlab环境下，对海宁水稻部分种植区内的两个重要结构参数（水稻高度和密度）进行反演。再通过对水稻测量数据进行统计分析，得到了的水稻生物量生长方程，进而对研究区内水稻田生物量进行了空间分布制图和定量估算。为进一步开展遥感估产以及农田生态系统碳平衡定量研究提供重要参数。（4）利用由覆盖全海宁市ALOS／AVNIR-2多光谱数据提取的水稻面积信息和实地调查获得的水稻耕作管理信息为输入变量，应用生物地球化学模型（DNDC模型）对研究区水稻田进行生态过程模拟。通过对不同耕作模式下稻田温室气体CH₄排放量进行情景分析与定量估算，提出了有利于稻田甲烷减排的耕作措施。综上所述，本研究在以下几方面取得了进展：验证了在水稻主要生长期内获得的多时相ALOS／PALSAR雷达数据提取种植面积的实用性；在原有辐射传输模型基础上进行改进，得到了适用于模拟水稻后向散射系数的一阶散射模型；利用改进的模型和遗传算法优化工具进行水稻结构参数反演，并对研究区开展水稻生物量空间分布制图和估算；利用DNDC模型模拟和定量估算稻田CH₄的排放量，进而提出了具体减排耕作措施。更多还原

【Abstract】 Paddy rice is one of the most important crops in the world, and the major staple food inChina. It accounts for more than 40% of food production for the population of 1.3 billions.Rice yield is one of the major concerns related to farm practice and national security. Remotesensing on rice plays important role in yield estimation, and provides key information foranalyzing and assessing the impact of paddy field on agriculture and environmentsustainability, food and water security and greenhouse gas mitigation. Fuyang City andpartition of Haining City of Zhejiang Province were selected as study area. The objective ofthe study is to integrate microwave remote sensing technology and ecology model forextracting paddy rice area, retrieving rice biophysical parameters and evaluating impacts ofpaddy fields on climate environment.Conventional methods in estimating rice area are not only time and labor consuming butalso questioned in accuracy. With its advantage of acquiring geo-information in timely,speedy and objective way over large areas, remote sensing is one powerful approach inextracting rice area to conventional methods. Otherwise, quantitative retrieval of rice structureparameters by remote sensing technology is a key topic in rice biomass and yield estimation.Attempt was made to develop one first-order radiation transfer model for simulating ricescattering character and retrieving physical parameters, which is a new means to apply remotesensing technology in rice biomass estimation. Analyzing the impacts of paddy field onclimate currently becomes a hot topic in ecology community. A process-based ecologicalmodel, DNDC （Denitrification and Decomposition） model was utilized to simulate methaneemissions from rice field during the growing season. With scenario analysis, feasible farmmanagement method was proposed for methane mitigation, which provides scientific basis toreduce and control methane emission from paddy field. Specific contents and results weresummarized below: （1） Pseudo-color composite imagery derived from three-date ALOS/PALSAR imagesacquired in key dates during rice growing season in Fuyang City was used to discriminatedifferent land cover with support vector machine algorithm （SVM-Support Vector Machine）.Temporal variation of backscattering coefficient during rice growing season was seriouslytaken into account in extracting the target object due to unique character of paddy rice, whichvaries greater than other land cover. Vector segmentation was taken in temporal dimensionspace to extract rice planting area. The derived map of rice planting area has high accuracy,which was validated by ALOS/PRISM imagery with a spatial resolution of 2.5 meter, routinesurvey data of land use/cover and in situ field investigation. The study results indicated thatpaddy rice could be extracted with accuracy of around 90%.（2） The SVM classifier used above and stepwise manual method were integrated toextract paddy rice in partition of Haining City, Zhejiang Province. And then, the rice thematicmap was taken as a mask to subset the raw PALSAR image for obtaining backscatter image.Some field measured structure parameters acquired synchronically with ALOS satelliteoverpassing the study area were used to modify the inter parameters of a 1^st-order forestcanopy scattering model for developing rice backscattering coefficients simulation model.The simulated results were validated by PALSAR imagery-based backscattering coefficients,good agreement was obtained.（3） Taken paddy rice imagery backscattering coefficients and radar signal beam incidenceangle as inputs, the modified 1^st-order radiation transfer model and genetic optimizationalgorithm were utilized to retrieve two-rice structure parameters, plant height and density, ofpaddy rice in this study area at the environment of Matlab. One empirical allometric equationwas derived from in situ measurements, which was used to estimate and map rice biomass atspatial scale in the study area. This result would also provide one key parameters for remotesensing estimation of rice yield and carbon balance in agricultural ecosystem quantification.（4） The rice planting area extracted from multispectral ALOS/AVNIR-2 imagery coveringHaining and farm management practice information were taken as inputs to run thebiogeochemical DNDC model for simulating rice paddy ecology process. Furthermore,scenario analysis and quantitative estimation were taken to estimate methane emission frompaddy fields given various farm management practice. Efficient method was also taken up with for controlling and mitigating methane emissions from paddy fields.In summary, this study showed the feasibility of using new multi-temporal microwaveRadar ALOS data acquired in key dates during rice growing season to extract paddy ricegrowing area was tested. A modified rice backscatter model was developed on the basis oforiginal radiation transfer model for simulating paddy rice backscattering characeteristics. Themodified model and Genetic Algorithm Optimization Tool were synergized to reversephysical parameters and map rice biomass at spatial scale in the study area. At last, on thebasis of paddy rice planting area and farming practice data, biogeochemical DNDC modelwas utilized to simulate quantitatively estimating methane （CH₄） emission from paddy fieldsand represented some specific mitigation scheme.更多还原