基于地块单元的冬小麦遥感估产方法研究

APPROACH FOR WINTER WHEAT YIELD ESTIMATION WITH REMOTE SENSING IMAGE AT FIELD SCALE

关键词:  土地承包经营权  地块单元  冬小麦  归一化植被指数  比值植被指数

Key words:  land contractual management right  filed scale  winter wheat  normalized difference vegetation index (NDVI)  ratio vegetation index (RVI)

摘要:

目的 以高分时序遥感影像为基础数据源，结合土地承包经营权地块数据，对冬小麦遥感估产方法及其精度进行研究和分析。方法 文章以河南省兰考县为研究区，采用2019年4―5月份的GF-1C和GF-6 2 m PMS遥感影像提取了研究区冬小麦种植空间分布，并在地块单元控制下对冬小麦种植面积进行了修正和精度验证。其最优提取结果的修正阈值为0.93，地块单元内冬小麦总体分类精度为95.66%，Kappa系数为0.89。利用3月7日至5月20日6期GF-1 WFV遥感影像序列NDVI和RVI与冬小麦种植地块单元数据进行空间统计，得出各冬小麦种植地块单元内NDVI和RVI均值，通过分析冬小麦测产地块单元内均值植被指数与产量间的敏感性，提出一种组合均值植被指数的冬小麦遥感估产模型构建方法，通过交叉验证法对不同的估产线性回归模型进行精度评价。结果 由4个均值植被指数组合变量的多元线性回归模型为最佳，决定系数为0.922 0，预测误差为40.96 g/m²，预测精度为93.13%。通过该模型得出兰考县冬小麦平均产量为6 047.25 kg/hm²，较2017年河南省统计年鉴研究区冬小麦平均单产6 001 kg/hm²有所提高，土地承包经营权地块内和地块外冬小麦总产量分别为2.76亿 kg和4 650万 kg。结论 该方法实现了冬小麦估产结果以像元为单位向以地块单元为单位的转变，解决了模型构建时光谱信息与实测产量间对应问题，为利用国产高分卫星进行县域地块尺度遥感单产精准化估算提供了方法支撑。

Abstract:

The field-scale data, such as land contractual management right, provides accurate and basic information for precision agriculture. The remote sensing approach and its accuracy for winter wheat estimation are proposed and analyzed based on high-resolution time series images and field-scale data of land contractual management right. Taking Lankao county, Henan province as the study area, the spatial distribution of winter wheat area was extracted by using GF-1C and GF-6 PMS remote sensing images with 2 m resolution on April 3 and May 1, 2019. Based on this, the winter wheat area was amended and its overall accuracy was verified, which depended on the filed-scale data. Consequently, the correction threshold value for the optimal extraction result of winter wheat area was 0.93 with the overall accuracy and Kappa coefficient were 95.66% and 0.89, respectively. Then, the field-based mean values for normalized difference vegetation index (NDVI) and ratio vegetation index (RVI) were calculated by time series of GF1-WFV images (6 scenes) which were selected from 7 March to 20 May, 2019. By analyzing the sensitivity between the mean vegetation indices and winter wheat yield at filed-scale, a remote sensing yield estimation model of winter wheat combining mean vegetation indices was proposed, and the accuracy of different linear regression models were evaluated by cross validation. The result showed that the linear model built with four mean vegetation indices (3 NDVI and 1 RVI) was the optimal model among the four-different yield estimation models, and its coefficient of determination R², root mean square error (RMSE) and predication accuracy were 0.922 0, 40.96 g/m², 93.13%, respectively. With the optimal estimated yield model, the average yield of winter wheat in study area was 6 047.25 kg/hm², higher than that of 6 001 kg/hm² of Henan Statistical Yearbook in 2017. The total yield of winter wheat was 2.76×10⁸ kg and 4.65×10⁷ kg within and outside of the filed-scale data, respectively. In summary, this approach achieves the transformation of winter wheat yield estimation from pixel to filed-scale, which can provide a solution to deal with the correspondence between spectral information and measured yield during modeling, suggesting a possibility that it can also provide an effectively method for utilizations of GF remote sensing images in precision agriculture at filed scale.