基于遥感数据的新疆开-孔河流域农业区种植结构提取

Crop planting structure extraction based on remote sensing data in Kai-Kong River Basin,Xinjiang

农作物种植结构是农作物空间格局的重要组成部分,是区域土地资源和水资源优化配置的基础。以新疆开―孔河流域农业区为研究区域,综合利用作物物候信息和2016年的MODIS NDVI时序曲线,获得不同作物生长差异明显的关键期,选择关键期的Landsat 8 OLI影像,构建主要作物提取知识规则,基于决策树方法开展农作物的分类识别。开―孔河农业区2016年主要作物种植面积为5.07×10^5hm^2,其中棉花种植面积最大,为1.97×10^5hm^2,玉米、小麦次之。博斯腾湖和开都河农业区以辣椒、玉米和小麦为主要作物,种植结构比较零散;孔雀河农业区种植结构比较单一,以棉花和香梨为主要作物。与仅利用时间序列的MODIS数据进行作物分类识别的结果对比表明,综合利用MODIS和Landsat数据的作物识别精度有显著提高,总体分类精度从62.58%提高到88.37%,kappa系数从0.53提高到0.86。该方法综合利用了MODIS数据的时序特征和Landsat数据较高的空间分辨率特征,有效地避免了MODIS数据空间分辨率不足而导致分类精度较差的情况,又避免了Landsat数据时间分辨率不足所引起的时相选择盲目性或数据冗余,在干旱区农业种植结构的提取领域具有一定的应用价值。

Crop planting structure plays an important role in crop spatial pattern,and it is the foundation for the optimal allocation of regional land and water resources.In this paper,taking the Kai-Kong River Basin agricultural area in Xinjiang as the study area,a method for crop classification that comprehensively utilizes high temporal resolution MODIS images and high spatial resolution Landsat images was proposed.Due to large scope of study area,it is difficult to obtain crop sample points uniformly throughout whole research area with traffic and time constraints.MODIS and Landsat data for 2016,combined with crop phenology data,were used to construct experimental sample points;thereby providing a better solution for crop extraction in this area wherein sample access is difficult.NDVI time-series curves for different crops were constructed based on experimental sample points.Based on the NDVI time-series curves,the critical period of crops during growing season were obtained.For these key periods,Landsat 8 OLI images were selected.Next,extraction knowledge rules for the main crops were constructed,and identification and classification of crops were performed based on decision tree.In 2016,main crop planting area was 5.07 × 10^5 hm^2 of the Kai-Kong River Basin agricultural region,with largest planting area found for cotton (1.97 × 10^5 hm^2),followed by those for corn and wheat.Bosten Lake and Kaidu River agricultural area was dominated by pepper,corn,and wheat,and the planting structure was relatively scattered.Planting structure of the Peacock River agricultural area was relatively simple,with cotton and pear as main crops.A comparative experiment based on time-series MODIS images for crop recognition and classification was also conducted.Results were verified and compared with sample points of field survey.The accuracy of crop classification using MODIS and Landsat data was obviously improved as compared with the accuracy of crop classification using only time-series MODIS data.Overall classification accuracy increased from 62.58% to 88.37%,and kappa coefficient increased from 0.53 to 0.86.The use of high temporal resolution MODIS data and high spatial resolution Landsat data can improve the accuracy of crop extraction to a certain extent;this avoided (1) poor classification accuracy caused by the insufficient spatial resolution of MODIS data and (2) phase selection blindness or data redundancy caused by the insufficient resolution of Landsat data.Therefore,this approach has high potential application value in the extraction of crop planting structure in arid areas.