基于作物物候特征的水稻种植面积提取研究
发布时间：2018-12-24 16:00
【摘要】：“民以食为天”这句话突出了粮食对于人类生存的重要性。作为一个拥有近14亿人口的发展中大国,粮食安全是维系社会稳定的前提。为确保我国粮食安全,国家和政府要坚守18亿亩耕地红线。水稻作为我国最主要的粮食之一,对保障粮食安全有着至关重要的战略意义。因此,及时准确地掌握水稻种植面积、种植制度和生产管理方式等信息显得尤为重要。随着我国人口的不断增长,城镇化进程加快,在这过程中不可避免的会出现占用耕地的情形,因而建立一个可靠的水稻面积检测体系是十分必要的。农业中原始获取水稻种植面积的方法费时费力,而且有很大的局限性,而遥感手段在农业中的运用使得获取作物种植面积的方法有了新的突破。大尺度水稻种植面积估算多以NOAA/AVHRR数据为基础,但其空间分辨率对于监测有很大的限制。EOS/MODIS的出现相对于NOAA/AVHRR在空间分辨率方面有很大的提高,提高了在大尺度上进行水稻种植面积估算的精度。本文以MODIS为数据源,在决策树分类方法的基础上结合水稻生长过程中EVI值的变化情况和水稻物候期(移栽期、抽穗期、成熟期)对水稻的种植面积进行识别,实现大尺度水稻种植面积的遥感提取。具体研究结果如下：(1)利用TIMESAT软件中的Savitzky-Golay(S-G)滤波、非对称高斯函数(AG)拟合、双逻辑曲线(D-L)拟合对时间序列EVI植被指数进行时间序列重构,减少数据中的异常值,提高数据可利用率。对经过三种方法重构后的曲线进行对比分析、定量统计分析,最终选择AG滤波作为EVI时间序列的重构函数。(2)利用经过AG滤波处理后的EVI时间序列提取水稻关键物候期(移栽期、抽穗期、成熟期)。将水稻生长期过程中开始的EVI最小值(EVImin)对应的日期作为水稻的移栽期,将EVI曲线中的最大值(EVImax)出现的日期定为抽穗期,成熟期是通过相对阈值法确定的,EVI=EVImin+ΔEVI*0.6时对应的日期定为成熟期。将提取的关键物候期的日期与气象台站观测的日期进行对比分析,发现大部分样本值落在误差为+16天的边界内。(3)将水稻生长过程中的EVI值、水稻的关键物候期和一些统计量(生长季长度,EVI曲线振幅)作为决策树中的分类条件,通过设置不同的阈值将整个研究区地物类型分为了水田、水体、林地、居民地和其他五类,并利用地表真实感兴趣区获得混淆矩阵,总体精度(Overall Accuracy)为94.37%,Kappa系数为0.9127,分类结果精度较高。将提取的水稻种植面积以市为单位进行统计与统计年鉴记录的当年水稻播种面积对比分析,在36个市中,有27个市的精度在72.22%以上,但也有8个市的精度低于60%。从总体看,利用16天合成的MODIS-E VI数据对大尺度水稻种植面积进行提取是可行的。
[Abstract]:The saying "the people take food as the sky" highlights the importance of food for the survival of mankind. As a developing country with a population of nearly 1.4 billion, food security is a prerequisite for maintaining social stability. In order to ensure our food security, the state and the government to adhere to 1.8 billion acres of arable land red line. As one of the most important grain in China, rice plays a very important strategic role in ensuring food security. Therefore, it is very important to grasp the information of rice planting area, planting system and production management. With the continuous growth of population and the acceleration of urbanization, it is inevitable to occupy cultivated land in the process, so it is necessary to establish a reliable rice area detection system. The original method of obtaining rice planting area in agriculture is time-consuming and laborious, and has great limitations. However, the application of remote sensing in agriculture has made a new breakthrough in the method of obtaining crop planting area. The estimation of large scale rice planting area is based on NOAA/AVHRR data, but its spatial resolution is very limited to monitoring. The appearance of EOS/MODIS is much higher than that of NOAA/AVHRR in spatial resolution. The precision of rice planting area estimation on large scale is improved. In this paper, MODIS is used as data source, based on the classification method of decision tree, the change of EVI value during rice growth and the phenological stage of rice (transplanting, heading and maturing) are combined to identify the planting area of rice. The large scale rice planting area can be extracted by remote sensing. The results are as follows: (1) using Savitzky-Golay (S-G) filter in TIMESAT software, asymmetric Gao Si function (AG) fitting, double logic curve (D-L) fitting to reconstruct time series EVI vegetation index. Reduce outliers in data and improve data availability. The curves reconstructed by three methods were compared and analyzed, and the quantitative statistical analysis was carried out. Finally, AG filter was selected as the reconstruction function of EVI time series. (2) the key phenology of rice (transplanting, heading and maturing) was extracted by EVI time series treated with AG filter. The corresponding date of EVI minimum (EVImin) in rice growing period was taken as the transplanting date of rice, the date of maximum (EVImax) in EVI curve was determined as heading date, and the mature period was determined by relative threshold method. The corresponding date of EVI=EVImin 螖 EVI*0.6 is set as maturity period. By comparing the date of the extracted key phenological period with the date observed by meteorological station, it is found that most of the sample values fall within the boundary of 16 days' error. (3) the EVI value in the rice growing process is changed. The key phenological period of rice and some statistics (length of growing season, amplitude of EVI curve) are used as the classification conditions in the decision tree. By setting different thresholds, the whole study area is divided into paddy field, water body and woodland. The confusion matrix is obtained by using the real area of interest of the earth surface and the other five categories. The overall accuracy of the confusion matrix is 94.37 kappa coefficient (0.9127), and the classification accuracy is high. The rice planting area extracted was compared with that recorded in the city and statistical yearbook. In the 36 cities, the precision of 27 cities was above 72.22%, but the precision of 8 cities was less than 60%. In general, it is feasible to extract large-scale rice planting area by using 16-day synthetic MODIS-E VI data.
【学位授予单位】：东北师范大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S511;S127
本文编号：2390804
[Abstract]:The saying "the people take food as the sky" highlights the importance of food for the survival of mankind. As a developing country with a population of nearly 1.4 billion, food security is a prerequisite for maintaining social stability. In order to ensure our food security, the state and the government to adhere to 1.8 billion acres of arable land red line. As one of the most important grain in China, rice plays a very important strategic role in ensuring food security. Therefore, it is very important to grasp the information of rice planting area, planting system and production management. With the continuous growth of population and the acceleration of urbanization, it is inevitable to occupy cultivated land in the process, so it is necessary to establish a reliable rice area detection system. The original method of obtaining rice planting area in agriculture is time-consuming and laborious, and has great limitations. However, the application of remote sensing in agriculture has made a new breakthrough in the method of obtaining crop planting area. The estimation of large scale rice planting area is based on NOAA/AVHRR data, but its spatial resolution is very limited to monitoring. The appearance of EOS/MODIS is much higher than that of NOAA/AVHRR in spatial resolution. The precision of rice planting area estimation on large scale is improved. In this paper, MODIS is used as data source, based on the classification method of decision tree, the change of EVI value during rice growth and the phenological stage of rice (transplanting, heading and maturing) are combined to identify the planting area of rice. The large scale rice planting area can be extracted by remote sensing. The results are as follows: (1) using Savitzky-Golay (S-G) filter in TIMESAT software, asymmetric Gao Si function (AG) fitting, double logic curve (D-L) fitting to reconstruct time series EVI vegetation index. Reduce outliers in data and improve data availability. The curves reconstructed by three methods were compared and analyzed, and the quantitative statistical analysis was carried out. Finally, AG filter was selected as the reconstruction function of EVI time series. (2) the key phenology of rice (transplanting, heading and maturing) was extracted by EVI time series treated with AG filter. The corresponding date of EVI minimum (EVImin) in rice growing period was taken as the transplanting date of rice, the date of maximum (EVImax) in EVI curve was determined as heading date, and the mature period was determined by relative threshold method. The corresponding date of EVI=EVImin 螖 EVI*0.6 is set as maturity period. By comparing the date of the extracted key phenological period with the date observed by meteorological station, it is found that most of the sample values fall within the boundary of 16 days' error. (3) the EVI value in the rice growing process is changed. The key phenological period of rice and some statistics (length of growing season, amplitude of EVI curve) are used as the classification conditions in the decision tree. By setting different thresholds, the whole study area is divided into paddy field, water body and woodland. The confusion matrix is obtained by using the real area of interest of the earth surface and the other five categories. The overall accuracy of the confusion matrix is 94.37 kappa coefficient (0.9127), and the classification accuracy is high. The rice planting area extracted was compared with that recorded in the city and statistical yearbook. In the 36 cities, the precision of 27 cities was above 72.22%, but the precision of 8 cities was less than 60%. In general, it is feasible to extract large-scale rice planting area by using 16-day synthetic MODIS-E VI data.
【学位授予单位】：东北师范大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S511;S127
【参考文献】
相关期刊论文 前8条
1 孙强;张三元;张俊国;杨春刚;;东北水稻生产现状及对策[J];北方水稻;2010年02期
2 DING MingJun;ZHANG YiLi;SUN XiaoMin;LIU LinShan;WANG ZhaoFeng;BAI WanQi;;Spatiotemporal variation in alpine grassland phenology in the Qinghai-Tibetan Plateau from 1999 to 2009[J];Chinese Science Bulletin;2013年03期
3 顾娟;李新;黄春林;;NDVI时间序列数据集重建方法述评[J];遥感技术与应用;2006年04期
4 阎静,王汶,李湘阁;利用神经网络方法提取水稻种植面积——以湖北省双季早稻为例[J];遥感学报;2001年03期
5 李儒;张霞;刘波;张兵;;遥感时间序列数据滤波重建算法发展综述[J];遥感学报;2009年02期
6 赖格英,杨星卫;南方丘陵地区水稻种植面积遥感信息提取的试验[J];应用气象学报;2000年01期
7 胡英敏;高琼;兰玉芳;金东艳;徐霞;;太仆寺旗2000—2008年EVI对气候及土地利用变化的响应[J];自然资源学报;2012年07期
8 杨小唤,张香平,江东;基于MODIS时序NDVI特征值提取多作物播种面积的方法[J];资源科学;2004年06期
，本文编号：2390804
本文链接：https://www.wllwen.com/kejilunwen/nykj/2390804.html
