【作者】 孙华生；

【导师】 黄敬峰；

【作者基本信息】 浙江大学 ， 农业资源利用， 2009， 博士

【摘要】 水稻是中国最重要的粮食作物之一。获取大范围的水稻种植空间分布、面积和产量信息对指导水稻生产、合理分配水资源，以及监测大气环境变化等具有重要的意义。由于农业生产具有覆盖面积大、季节性强、区域差异大、单位面积的经济效益低等特点，通过地面调查方法获取每年的农作物种植信息，不论在技术方面还是在经济可承受性方面都是一件非常困难的事情，而利用遥感技术则是解决这个问题的可行且有效的方法。与常规的统计方法相比，应用遥感方法获取作物种植信息具有它独特的优势。由于遥感具有覆盖面大、短时间内可重复观测，以及成本相对较低等特点，并与地理信息系统和全球定位系统相结合，不仅可以提取农作物种植面积，而且可实现空间分布的准确定位，全面地监测农作物的整个生长发育过程。研究利用EOS-MODIS数据空间覆盖面大和时间分辨率高的优势，选取覆盖中国范围的数据，实现对全国范围水稻种植和生长信息的提取。研究目标是解决水稻遥感估产中最关键的技术问题，为实施大面积水稻遥感估产提供理论与试验依据。研究的主要内容包括：中国水稻遥感信息获取区划、水稻关键生长发育期识别、水稻种植空间分布和面积信息提取，以及水稻生长发育状况分析等4个在大尺度水稻遥感估产中最关键的部分。具体内容如下：首先，研究对研究的选题、国内外的研究进展、拟采用的技术处理方法进行简单的介绍，主要阐述了研究的背景、意义和选题依据。研究对国内外在遥感估产方面已经开展过的课题项目，以及过去研究中采用的技术处理方法进行简要概括，总结了前人在相关研究中已经取得的研究成果和仍然存在的需要进一步解决的问题，为研究目标的实现和可能取得的创新提供帮助。然后，研究对各个具体部分分别采取的技术处理方法和技术路线，以及所需要的数据进行说明，并对研究目标实现的具体步骤进行了总体设计，为研究的开展提供了指导。研究详细地介绍了各个部分的具体内容、采用的具体处理方法和得出的主要结论。其中，在中国水稻遥感信息获取区划的研究中，研究以全国水稻种植区为对象，通过分析并选取对遥感技术信息获取有重要影响的因素，主要包括耕作制度、地形因素、种植结构和大气噪声等，采用恰当的区划指标，利用定性和定量相结合的分析方法进行区划。根据耕作制度的差异把全国分为4个水稻遥感信息获取区，再根据地形、种植结构和大气噪声等因素对遥感信息获取的影响分成19个亚区。区划结果对水稻遥感信息获取时选择合适的遥感获取方式、恰当的空间分辨率与时相的遥感图像，以及对遥感信息提取结果准确度的验证等提供参考。在水稻关键生长发育期识别的研究中，研究以2005年的多时相MODIS数据为例，提取全国范围内的水稻关键生长发育期。首先，利用傅立叶低通滤波和小波低通滤波平滑处理后的时间序列EVI(Enhanced Vegetation Index)，然后根据水稻在移栽期、分蘖初期、抽穗期和成熟期的EVI变化特征，分别对各个生长发育期进行识别。通过将利用MODIS数据识别的结果与当年气象台站的地面观测数据进行比较，各个生长发育期的提取结果的误差绝大部分在±16天以内，经过F检验表明利用MODIS提取的结果与地面观测数据在0.05水平下具有显著的一致性。研究中的提取方法可以被用于其他年份的水稻生长发育期识别，而且根据其他作物的生长发育特点，也可能被用于识别其他作物的生长发育期。在水稻种植空间分布和面积信息提取的研究中，探讨了提取中国水稻种植空间分布及其种植面积信息的方法。研究利用覆盖面积大、高时间分辨率、低成本的MODIS数据，实现了对整个中国范围的全面覆盖。研究识别稻田的依据是根据在灌水移栽期有水的特征来提取水稻。通过分析对陆地比较敏感的MODIS前7个波段反射率的特点，确定对植被和土壤湿度敏感波段，用来构建植被指数和土壤水敏感指数以扩大感兴趣地物与其背景的差异。在选取的典型试验样区内，根据稻田在灌水移栽期所表现出的特有的光谱特征，得出判断水稻的判别条件。根据识别水稻的判别条件，并按照单季稻、早稻和晚稻生长期的差异得出2000-2007年全国单季稻、早稻和晚稻的空间分布状况和面积统计数据，然后对利用MODIS数据识别的结果与各年的农业统计数据进行比较，分析利用MODIS数据提取水稻的面积精度。为了验证识别结果在空间位置上准确性，研究选取了4个具有代表性的试验样区，对利用MODIS数据提取的结果与利用中等空间分辨率的遥感图像的分类结果进行叠加分析，检验其在空间位置上的匹配性。结果表明：研究中利用MODIS数据提取水稻的算法是有效的；提取结果的精度取决于水稻与其他地物的混合程度，混合像元中水稻的纯度越高，那么提取结果的精度就越高；云覆盖对最终的分类结果的精度也会产生很大的影响，在多云的地区，云覆盖成为光学传感器应用的一个重要的限制因素。在水稻生长发育状况分析研究中，探讨了对水稻长势进行定量化分析的方法。研究突破了过去的研究仅得出的与往年同时期或者同一时期不同地区相比较的好、持平、差等定性分析结论。研究以2005年数据为例，通过在典型试验区建立水稻植被指数与其生物物理参数的关系模型，反演出水稻在不同时期的LAI和FPAR。试验结果表明EVI反演水稻生物物理参数的效果比NDVI更好。因此，研究最终选择EVI作为反演水稻生物物理参数的依据。根据对水稻生长发育期的识别结果，从而得出全国单季稻、早稻和晚稻的生长季的开始和结束日期，并在像素水平上识别水稻的生长季，通过时间序列EVI反演各个时期的LAI和FPAR，再进一步通过光能利用效率模型(LUE模型)得到各个时期的NPP，最后得出单季稻、早稻和晚稻在整个生长季内的生物量，实现对长势的定量化分析，并为进一步的单产分析提供参考依据。最后，研究总结了以上各个部分得出的主要结论和取得的创新，并展望了在将来的工作中仍需要进一步解决的问题。更多还原

【Abstract】 Paddy rice is one of the major staple crops in China. It is essential to get the spatial distribution, planting area and yield information of paddy rice at large scale for guiding rice production, regulating water use, and monitoring environmental changes. However, agricultural production has the characteristics of large coverage, frequent changes seasonally and regionally, and low benefit in per unit area, therefore, it is unfeasible to get the annual information through ground in situ measurement neither in the aspect of technical issues nor in the aspect of economic sustainability. Remote sensing technology is a feasible and efficient way to solve these problems. Compares with the traditional methods, remote sensing has its own advantages for acquiring planting information of crops. For remote sensing has the characteristics of large and frequent coverage, as well as its low costs, then it can get not only the area information, but also the accurate spatial position and the whole process of crop growth information associates with GIS and GPS. In the study, planting and growth information of paddy rice was extracted in all of China using EOS-MODIS data with large coverage and high spatial resolution. The goals of the study were to solve the most important technical problems in rice yield estimation using remote sensing technology, and provide theoretical and experimental bases for large-region rice yield estimation. The main contents of the study including regionalization of paddy rice information acquirement through remote sensing technology in China, detecting major growth stages of paddy rice, extracting the spatial distribution and area information, and getting growth information of paddy rice. The four above mentioned parts are the most important problems for rice yield estimation at large scale using remote sensing technology, so they were studied in detail in the study. Main content of each part is introduced as follows:Firstly, the dissertation gave a brief introduction of the topic, made a summarization of the precious studies, and described the methods in the study. The background, significance, and goals of the study were introduced, and a summary of the precious studies and the issues that need to be solved was made. It is helpful for the realization of the objectives and the potential innovation in the study. Then, detailed methods and the flowcharts were given for each part in the study; an introduction of the data used and an overall design for each step in the study were given as well.Subsequently, detailed introductions of the contents, methods and conclusions for each part in the study were given. In the study of regionalization of paddy rice information acquirement through remote sensing technology, major impacting factors, including rice farming systems, topography, planting structure and atmospheric noises, were analyzed and appropriate indices were chosen. Regionalization was executed by qualitative and quantitative analyses. Rice planting area in the whole country was divided into 4 regions by the difference of rice planting rotation, and subsequently divided 19 sub-regions by the differences of topography, land surface feature structure and atmospheric noise. It might provide useful information to the selection of remote sensing images with appropriate spatial resolution and date, as well as to the accuracy evaluation of the classification.In the study of major growth stages detection of paddy rice, phenological information of paddy rice in the whole country of China was extracted using multi-temporal MODIS data in 2005 as a case study. Firstly, time-series of MODIS-EVI (Enhanced Vegetation Index) was smoothed by Fourier and Wavelet low pass filtering, then the stages of transplanting, beginning of tillering, heading, and maturation were identified by their respective characteristics. The MODIS-derived results were compared with the statistics of meteorological observatories, most of the errors of the MODIS-derived results were within±16 days, and F test indicated that all of the results had significant consistency at the level of 0.05. The detecting methods could be used to extract rice growth stages in other years, and they might be able to extract the growth stage information of other crops according to their characteristics potentially.In the study of spatial distribution and area information extraction, spatial distribution and area information were extracted in all of China using MODIS data with large coverage, high temporal resolution, and low costs. The basis of rice classification in the study was relied on identifying the spectral characteristics of water in the "flooding and transplanting" period. Vegetation and soil moisture sensitive bands were selected by comparing the first 7 bands of MODIS that are sensitive to land surface, then they were used to calculate vegetation and water sensitive indices to amplify the differences between the regions of interest and the background. Spatial distribution and area information of single, early, and late rice in 2000-2007 were extracted by identifying the unique characteristic of rice fields in the "flooding and transplanting" period with the assistance of the growth period information. Areal accuracy of the MODIS-derived results was validated comparing with the agricultural statistics, and four typical test regions were selected to test the spatial matching by overlaying the results derived from MODIS data and the data with medium spatial resolution. Tests showed that the algorithms for rice identification in the study were effective; and the accuracy of the MODIS-derived results was relied on the purity of rice in the mixed pixels, the higher the purity is, the higher the accuracy of the results obtained; cloud contamination had big impacts on the results, therefore, optical sensors were seriously restricted in the regions with frequent clouds.In the study of growth status monitoring, methods of quantitative analyses were studied to break through the traditional methods, for the methods in the past only got the qualitative results of better, even, or worse than the former years or than other regions in the same period. Models were built to inverse the biophysical parameters of rice, including LAI and FPAR, by vegetation indices. Tests showed that EVI was superior to NDVI for getting the biophysical parameters. The growing seasons were identified at pixel level according to the results of the beginning date and end date for single, early and late rice, then LAI and FPAR were inversed by EVI for each period, and NPP was calculated by LUE (Light Use Efficiency) model, eventually, the results of biomass in the whole growing season for single, early, and late rice were calculated. The results could show the growth status of paddy rice, and might provide references for the analysis of the yield in per unit area.Eventually, a summary of the main conclusions and innovations in the study were made, and at the same time gave an expectation of the issues that need to be solved in the future.更多还原