欢迎访问《麦类作物学报》编辑部网站

为确定适用于冬小麦植株水分诊断的最佳高光谱指数及其在植株水分处于适宜状态时的阈值，设置4个水分处理（灌溉定额分别为0、60、120和180 mm），获取了小麦关键生育时期（返青期、拔节期和灌浆期）的冠层高光谱反射率、植株含水率、土壤含水率和产量等数据。依据高光谱指数与冬小麦植株含水率之间的相关性对高光谱指数进行筛选，以筛选的高光谱指数为输入变量，分别构建一元回归、偏最小二乘回归、随机森林回归和支持向量回归的冬小麦植株含水率估测模型。考虑到土壤含水率对冬小麦植株含水率的影响，进一步量化了当日植株含水率与不同时间土壤含水率的关系，通过产量比较法分别确定了冬小麦植株水分、土壤水分的阈值。结果表明：（1）在返青期、拔节期和灌浆期，一元回归模型的精度（r²=0.673，RMSE=3.144%，RE=5.489%）较好，能确定高光谱指数阈值，可以较精准、快捷地实现冬小麦水分诊断。机器学习算法中随机森林回归的模型精度（r²=0.904，RMSE=1.701%，RE=3.606%）最高，但模型参数较多，无法给出高光谱指数阈值。（2）当日植株含水率与其前一天0~50 cm土层的含水率之间具有较强的正相关关系（r²为0.708，RMSE为2.436%，RE为7.755%）。（3）返青期、拔节期和灌浆期估测冬小麦植株水分最佳高光谱指数分别为 MCARI/0SAVI、PRI3和VEG，其相应的阈值分别为0.765 1~1.130 1、0.155 2~0.225 7、1.633 9~1.668 5。因此，可根据植株含水率与土壤含水率之间的关系确定冬小麦在关键生育期内所处水分状态，从而采取相应对策。

英文摘要:

In order to select a reliable hyperspectral index for winter wheat plant water diagnosis and determine the threshold of the selected hyperspectral index for an appropriate water state of plant, in this study, four winter wheat water treatments(irrigation amount of 0, 60, 120, and 180 mm) were set up. The data of canopy hyperspectral reflectance, plant water content, soil water content, and yield at key growth stages(regreening stage, jointing stage and filling stage) were obtained. The hyperspectral index was screened according to the correlation between the hyperspectral index and the water content of winter wheat plants. The selected hyperspectral index was used as the input variable to construct the water content estimation model of winter wheat plants by simple regression, partial least squares regression, random forest regression, and support vector regression. Considering the influence of soil moisture content on winter wheat plant moisture content, the relationship between plant moisture content and soil moisture content at different stages was further quantified. The hyperspectral index thresholds of winter wheat plantmoisture and soil moisture were determined according to the yield comparison method. The results showed that at regreening stage, jointing stage and filling stage, the accuracy of the univariate regression model(r²=0.673，RMSE=3.144%，RE=5.489%) was better. It could determine the threshold of hyperspectral index and realize the moisture diagnosis of winter wheat accurately and quickly. Among the machine learning algorithms, the accuracy of random forest regression model(r²=0.904, RMSE=1.701%, RE=3.606%) was the highest, but there were too many model parameters to give the hyperspectral index threshold. There was a strong positive correlation between the water content of plant on the day and the water content of the 0-50 cm soil layer the day before(r²=0.708, RMSE=2.436%, RE=7.755%). MCARI/OSAVI, PRI3, and VEG were the best hyperspectral indices for estimating the water content of winter wheat plants at the regreening stage, jointing stage and filling stage, respectively. When the plant was in a suitable water state, the corresponding hyperspectral index thresholds were 0.765 1-1.130 1,0.155 2-0.225 7, and 1.633 9-1.668 5, respectively. Therefore, the water status of winter wheat during the critical growth period can be determined based on the relationship between plant moisture content and soil moisture content, and corresponding countermeasures can be taken.

查看全文查看/发表评论下载PDF阅读器

关闭