美国Corn Belt地区农作物遥感估产模型及NAO对该地区农作物产量的影响研究

CORC > 成都山地灾害与环境研究所 > 中国科学院水利部成都山地灾害与环境研究所 > 成都山地所知识仓储（2009年以前）

题名	美国Corn Belt地区农作物遥感估产模型及NAO对该地区农作物产量的影响研究
作者	李爱农
学位类别	博士
答辩日期	2007
授予单位	中国科学院水利部成都山地灾害与环境研究所
授予地点	成都
导师	周万村 ; 王昂生
关键词	遥感估产模型人工神经网络 SCE-UA算法 NAO 遥相关植被响应
学位专业	自然地理学
中文摘要	“民以食为天”，粮食生产是关系到维护世界和平，保持社会稳定和人民安居乐业的重大问题。对粮食作物产量进行及时、准确的预测，对于国家及时了解农作物产量，制定粮食进出口政策、宏观调控粮食价格以及粮食贸易都具有重要意义。同时，气候变化对农业及粮食保障的影响也已经成为当今人类社会普遍关注的问题。研究气候变化对植被生长的影响能够指示农业生态系统对气候变化的反应能力，能够指示出全球植被受大尺度气候变化影响的典型区域及其敏感性程度。本文以美国Corn Belt地区为研究区域，选择多年(22年)、多时相(24、23/年)、多平台(TM、NOAA AVHRR、MODIS Terra)遥感数据、多年农作物单产测量统计数据为数据源，运用遥感估产原理及人工神经网络方法开展了农作物单产估测遥感模型研究；为了探索气候涛动对全球范围内植被生长的宏观影响，本文还使用了对北半球气候有广泛影响的典型气候涛动-北大西洋涛动(NAO)指数，以及多年(22年)时序(12个月)AVHRR NDVI遥感数据，采用经验正交法(EOF)和“时滞”相关法研究了全球植被的涛动显著区域与NAO的空间相关及其时空传输模式；文章最后还探索了NAO对Corn Belt地区农作物单产的可能影响。论文获得了如下一些主要的结论：(1) 人工神经网络模型能够很好地表达作物光谱辐射值与作物单产之间复杂的非线性映射关系，使用ANN方法发展的作物单产模型与线性模型相比有效地提高了单产估算的精度，其被评估的估产精度达到了85%左右。这个精度能满足县域精确农业单产估测与制图的需要。模型集是基于县域的，有别于美国农业部“年内方法”建模过程中采用加权的方法计算州级作物单产，这为发展大面积县域运行化遥感估产提供了借鉴。(2) 在建模过程中发展的基于SCE-UA的全局优化算法的模型训练程序，具有全局搜索、避免局部收敛的优点，训练更高效、灵活。使用“泛化”控制函数可以有效防止ANN模型训练中的“过学习”，多折交叉验证评估结果显示模型具有很好的稳定性和“泛化”能力。GIS与ANN技术相结合发展遥感单产估测运行化系统，能使用GIS的数据管理和可视化能力来调用和管理ANN单产估测模型集，具有进一步发展成农作物估产与农业决策支持系统的潜力。(3) 在全球尺度上NDVI变化的时-空异质性较强，5-9月份NDVI变化的空间异质性最大。全球NDVI年际变化趋势最明显的区域主要集中在北半球中高维度地带以及南北回归线以内的局部地区。NDVI在时-空尺度上涛动区域与其变化趋势在空间上相关联。(4) 月NAO 指数与全球NDVI在特定区域、一定时间段内表现出了一定的空间相关关系，伴随着不同的时间延迟，并表现出了明显的时-空分布特征和四种空间传输模式。研究还发现，NDVI动态变化显著的区域和植被受NAO影响较敏感的区域在空间上存在关联。本研究对NAO与NDVI时-空相关关系的传输模式1和模式2都有较好的动力学解释。这种延迟相关特性可以为粮食估产和农业水资源管理提供可参考指标。通过研究NAO对Corn Belt地区农作物单产可能的影响发现，NAO与该地区玉米单产相关性较弱。在区域的东南角出现了可信度较高的弱相关区域，分析其原因与NAO影响该地区的降水有直接的关系。本研究获得了两个方面的学术成果：(1) 解决了农作物冠层光谱辐射值与单产间复杂非线性映射关系的表达问题；(2) 归纳了NAO对全球植被生长影响区域可能的四个“时滞”相关动态传输模式；论文还提出了植被对NAO响应敏感的区域与其本身涛动变化区域在时空上具有关联关系的观点。论文的创新工作有：(1) 本研究在县域基础上，创新性地开展了基于ANN模型技术的遥感单产估算模型研究，选择作物生长期内的所有遥感时序数据代替通常使用的几个特定时期的遥感辐射值，在ANN模型建模过程中，发展了新的样本训练程序，创新性地使用了SCE-UA优化算法，这些都是有别于其它类似研究的创新之处；(2) 使用长序列(1982-2003)高分辨率(8km×8km) NDVI对全球范围内的植被生长状况进行了分析，在时空尺度上比前人研究有所提高，研究了NAO对全球植被生长状态的影响，并从动力学角度解释了几种典型的“遥相关”时空传输模式，这有别于过去静态的研究问题。
英文摘要	Choosing Corn Belt of U.S. as a study area, selecting RS data of multi-annual(22 years), multi-temporal (24 or 23 per year), multi-sensor(Landsat TM、NOAA AVHRR、MODIS Terra) as well as data of multi-annual crop yield, and applying crop estimation principle and artificial neural network technologies, this dissertation carried out the research on crop yield estimation models. To explore the influence of climate oscillation on global vegetation growth, it also applied NAO (North Atlantic Oscillation) index together with AVHRR NDVI and empirical orthogonal functions (EOF) and time-lag methods to the study of spatial correlations and spatio-temporal transfer patter between global oscillation typical areas and NAO. Finally, it explored the possibility of NAO impact on crop yield in Corn Belt area. This dissertation drew such main conclusions as follows:(1) Artificial Neural Network (ANN) models reflect very well the complex non-linear relation between spectral radiation and crop yield, which compared with linear models effectively improves the precision of crop yield estimation to about 85%, meeting the precision needs of crop yield estimating and mapping within a county scale. (2) The developed model training program based on SCE-UA optimal algorithm is characteristic of global searching, making training more flexible and robust. Using generalization control-function can prevent over-learning in the training. Multi-fold cross-validation result shows the model possesses good stability and generalization capability. Combining GIS with ANN technologies, crop yield estimation system will be developed to manage ANN yield estimation model, with the potential to further develop into crop estimation and agricultural decision supporting system. (3) There is a strong spatio-temporal heterogeneity for NDVI changes on global scale, especially from May to September. The most obvious area is concentrated in the high-latitude belts of North Hemisphere as well as part of Tropic of Cancer and Capricorn areas. The oscillation areas and its change tendency are closely related at tempo-spatial scale.(4) Monthly NAO index and global NDVI present certain spatial correlations within specified areas and time phase, along with apparent tempo-spatial distribution features and four transmitting patterns with different time lags. The study also finds out that there are kinds of relations between NDVI dynamic-change typical areas and NAO affection-sensitive areas. Moreover, it provides good dynamics explanation for transfer patter 1 and 2. This lag can offer reference index for crop estimation and agricultural water resources management, benefiting the improvement of global carbon cycling module research. Through the analysis of NAO impact on crop yield in Corn Belt area, it is found that there is weak relativity with corn yield, in accordance with the conclusion of global vegetation’s response to NAO. At the same time, there appear high-reliable weak-relativity districts in southeastern area, resulting directly from the fact that NAO affects the rainfall in this region.This study acquires two academic results:(1) It solves the expression problem of complex non-linear relation between crop spectrum radiation values and crop yield.(2) It concludes the four time-lag dynamic transfer patterns. It also explores the correlation between sensitive areas of vegetation’s response to NAO and the oscillation changing areas, which has not been discussed before.The innovation work includes:(1) On county scales, it innovatively develops RS yield estimation model based on ANN technology, selecting all the RS time-series data within crop growth cycle rather than several radiation values within some specified periods. A new training program is developed and SCE-UA optimal calculation method was applied, all this differing from other similar researches.(2) By the application of long-term and high-resolution RS image, NDVI is implemented to analyze global vegetation growth, improved than past study at tempo-spatial scale. It studies the impact of NAO on vegetation growth and interpreted several typical connection transfer patterns from the angle of dynamics, different from previous static researches.
语种	中文
学科主题	摄影测量与遥感技术
公开日期	2010-10-21
分类号	TP7;S5
内容类型	学位论文
源URL	[http://ir.imde.ac.cn/handle/131551/2273]
专题	成都山地灾害与环境研究所_成都山地所知识仓储（2009年以前）成都山地灾害与环境研究所_数字山地与遥感应用中心
推荐引用方式 GB/T 7714	李爱农. 美国Corn Belt地区农作物遥感估产模型及NAO对该地区农作物产量的影响研究[D]. 成都. 中国科学院水利部成都山地灾害与环境研究所. 2007.