基于Sentinel-1雷达影像的玉米倒伏监测模型

在玉米发生倒伏灾害后,为定量监测区域尺度下的玉米倒伏程度,该研究以2017年8月8日因强风和强雨造成大面积玉米倒伏的小汤山国家精准农业研究示范基地作为研究区,提取倒伏前后Sentinel-1A雷达影像的多种强度信息,与实测倒伏样本关联分析,筛选出玉米倒伏前后最佳敏感后向散射系数。采用自然高与植株高的比值作为倒伏程度评价指标并构建比值公式,最终得到倒伏监测模型。结果表明,倒伏前后玉米植株高度的最优敏感后向散射系数分别为σVH和σVV+VH。32个建模点的实测差值结果与模拟差值结果的R~2为0.896(P0.01)。15个检验样本点和总样本点的倒伏程度分类准确度均达到100%。模型求解的自然高与植株高的比值与实测的比值总体相关性达到0.899。其中,中度倒伏类型的相关性最好,严重倒伏次之,轻度倒伏最差。该研究结果表明,在倒伏发生后,基于Sentinel-1A雷达后向散射系数构建的倒伏监测模型能在区域尺度下有效的实现玉米倒伏程度的分级监测。

Monitoring model of maize lodging based on Sentinel-1 radar image

Abstract: In order to monitor the degree of maize lodging quickly and accurately at the regional scale, the study was conducted in Xiaotangshan National Experiment Station for Precision Agriculture, where serious maize lodging occurred on August 8, 2017, due to strong wind and heavy rain. Firstly, different backscatter indicators before and after lodging were obtained from 2 Sentinel-1A radar images. Forty-seven ground sample points were obtained. The correlation coefficients between the natural height and plant height of 32 modeling samples and their backscatter indicators were investigated to choose the best backscatter indicator of maize lodging. When the backscatter indicators were used to simulate height of samples before and after lodging directly, there was a large error between the backscatter indicators of radar and those of lower or higher plant. Therefore, to establish more accurate lodging monitoring model, firstly backscatter coefficients after lodging respectively at the channel VH and the channel VV+VH were used to build difference formula in this study. We found that backscatter coefficient after lodging at the channel VH had the best correlation with the height before lodging, and that at the channel VV+VH had the best correlation with the height after lodging. The formula''s values were obtained by the simulated plant height before lodging subtracting the simulated natural height of the plant after lodging. Then, the ratio of the natural height to the plant height was used as a standard for distinguishing the degree of lodging. The results of the difference were brought into the ratio formula of natural height and plant height. By setting a reasonable ratio interval, the intervals with different degrees of lodging were separated. Finally, the lodging monitoring model was obtained. The results were verified by using the remaining 15 ground samples. The results show that lodging difference obtained by the difference formula is 100. Measured difference and simulated difference have an extremely significant relationship level (P<0.01). The classification accuracy for lodging degree is very high by using the 15 test samples and the total samples, both reaching 100%. The overall relativity between the ratio of simulated natural height to plant height and the ratio of measured natural height to plant height is 0.899. Among them, the relativity of moderate lodging is the best, followed by serious lodging, and mild lodging is the worst. The reason is that the surface vegetation in the areas of moderate lodging is not destroyed, and the surface information of the farmland is not exposed to the radar''s channel. Therefore, the simplex backscatter information for plant structure makes it possible to obtain the highest accuracy. There is a large difference in plant structure before and after lodging in serious lodging areas, and the water on the ground and weeds are mixed into the information of the backscatter indicators of radar. So the accuracy will be reduced. There is only a little difference in plant structure before and after lodging in mild lodging areas, and the reset rate of mild lodging areas is better than the former 2 types of lodging. Meanwhile, the error caused by non-structural information is more. Therefore, the accuracy is the worst. The final remote sensing mapping with a high accuracy is the same as the ground field lodging condition basically. This study shows that the lodging monitoring model based on dual polarized Sentinel-1A radar image can effectively monitor the degree of maize lodging at the regional scale. It should be noted that there still are some deficiencies in this study that will be improved in the future.