【作者】 孔令寅；

【导师】 鲍艳松； 延昊；

【作者基本信息】 南京信息工程大学 ， 大气遥感科学与技术， 2012， 硕士

【摘要】 农业正在由传统农业向由遥感技术等高科技为特征的高科技农业发展,高科技农业技术成果也正在迅速的向农业的各个领域渗透。随着遥感技术的发展,利用遥感技术对农作物的发育期、长势、产量等进行识别与监测已成为可能；遥感技术为现代化农业的发展提供了实时快捷和准确的大范围资料。本文使用MODIS-EVI数据与冬小麦长势和产量数据,研究冬小麦生长变化与EVI变化的关系。找到利用EVI数据识别冬小麦关键发育期的方法。基于识别发育期,建立冬小麦遥感长势监测模型和估产模型。使用相同发育期EVI数据对冬小麦进行长势监测和产量估算。在研究冬小麦关键发育期遥感识别方法时,首先,选择合适的Savitzky-Golay滤波方法对EVI时间序列进行平滑处理,获取整个发育期冬小麦冠层的EVI；然后,分别使用最大变化斜率法、窗口极值法与转折点法识别冬小麦返青期、抽穗期、成熟期。研究结果表明：基于2029个冬小麦观测数据进行遥感识别发育期的平均绝对偏差为11.2天,均方根误差为13.94天。其中返青期有548个数据,平均绝对偏差为10.59天,均方根误差为12.84天；抽穗期数据766个,平均绝对偏差11.05天,均方根误差14.07天；成熟期715个数据,平均绝对偏差9.26天,均方根误差14.57天。验证结果表明使用遥感数据识别冬小麦发育期是可行的,识别数据是可以用于研究中的。基于识别发育期数据,使用2006-2010年遥感数据监测冬小麦长势,并且利用研究区遥感图像和地面站点数据比较基于识别发育期的长势监测方法和未考虑发育期长势监测方法。基于发育期长势监测方法进行长势监测的遥感图像结果显示2010年大部分地区长势好于2009年,与实际情况相同。而未考虑发育期的遥感长势监测方法显示：2010年大部分地区冬小麦长势差于2009年,与实际情况不符合。在站点数据对比中,随机选择了7个站点,基于识别发育期长势监测结果的相关系数从0.51至0.86,呈明显的正相关性。而传统长势监测方法监测结果相关系数为-0.96到0.61。在区域和站点尺度上的比较都可以发现：与未考虑发育期的遥感长势监测方法相比,基于识别发育期的长势监测方法监测效果更好。分别基于三个主要发育期和全发育期建立了冬小麦估产模型,发现全发育期产量估产模型精度较高。通过相同方法的不同发育期对比发现：使用基于发育期EVI的遥感估产模型估产误差较小,返青期估产误差从-6.96%到5.84%,抽穗期估产误差从-6.73%到8.69%,成熟期估产精度误差为-7.87到4.7%,全发育期估产精度为-1.04%到1.57%。而使用未考虑发育期方法(即使用相同时间EVI进行遥感估产)估产,在儒略历70天时估产精度从-9.1%到10.74%,儒略历115天时估产精度从-7.17%到11.51%,儒略历165天时估产精度从-8.44%到9.14%,同时使用三个时间段估产精度从-11.34到6.27。通过比较可以发现,基于相同发育期EVI建立的遥感估产模型精度要好于未考虑发育期模型。通过不同方法的相同发育期对比发现：全发育期估产精度误差从-1.04%到1.57%,是误差最小的估产模型。冬小麦返青期、抽穗期、成熟期的发育情况好坏预示着冬小麦产量的高低,做好这三个主要发育期的管理工作对提高冬小麦产量有着重要的作用。更多还原

【Abstract】 Agriculture has changed from traditional agriculture to high-tech agriculture. Remote sensing technology as one high-tech has become more important to agriculture. With the development of remote sensing technology to provide fast, accurate, large area, and high precision data, remote sensing based growing stage monitoring, growing status, yield estimation have become more and more important to fine agriculture, especially in China with a large pressure of population. This study researched the relationship of winter wheat growth and EVI data by using the MODIS-EVI data and the field data of winter wheat. Based on the detected growing stages, this study built the model of growth monitoring and yield prediction of winter wheat, which differs from traditional methods in that the EVI data in same stages was applied to monitor growth and estimate yield of winter wheat,.Savitzky-Golay Filter as an effective method was adopted to build high-quality EVI time series data. Then, the maximum variation of slope method, the window of extreme value method, and the turning point method were used to monitor planting time, heading time, and harvest time. The result shows that mean absolute deviation (ADE) for three growing stages was11.2day and RMSE was13.94days for all samples of2029winter wheat data. In addition, the error statistics in planting time had an ADE of10.59days and an RMSE of12.84days with a sample number of548. An ADE of11.05days and an RMSE of14.07days were obtained in heading time with a sample number of766. An error of ADE-9.26days and RMSE-14.57days was obtained in harvest time with a sample number of715data. The evaluation demonstrated that it was feasible to monitor growing stages by using remote sensing data.Based on the detected crop stage data, crop growth monitoring was carried out by using remote sensing data from2006to2010. Further evaluation by using field leaf erea index (LAI) shows that the new monitoring method by using the same stage’EVI data was better than traditional method by using same Julian day’EVI data. The new method gave an improved statistics with correlation coefficients ranging from0.51to0.86compared with field LAI data. However, traditional method had an wide variation of correlation coefficient of-0.96～0.61. Further monitoring result given by the new method showed that winter wheat in2010grew better than that in2009, which agrees well with field LAI data. But the traditional method showed that winter wheat grew in2010worse than that in2009, which contradicts with field LAI data. It can be concluded that compared with the traditional method, the new growth monitoring method, based on same growing stage, was better.As growth of winter wheat in planting time, heading time, and harvest time was important to the yield forming of winter wheat, this study built yield prediction models of winter wheat based on three key stages and all stages. The evaluation for six years shows that the yield prediction error changed from-6.96%to5.84%for planting time model, from-6.73%to6.89%for heading time model, and from-7.87%to4.7%for harvest time model, respectively. The error of yield prediction for all stage model varied from-1.04to1.57for six years, which indicates that the all stage model was the best. However, the traditional prediction model based on julian day of70,115, and165had an error of-9.1～10.74%,-7.17%to10.51%, and-11.34%to6.27%, respectively. The three time model based on julian date had an error of-11.34%6.27%for six years. Above comparison indicates that yield prediction models of winter wheat based on the same growing stage had a better performance than that based on the same julian day.更多还原