基于无人机遥感的夏玉米水分胁迫指数改进方法

冠层温度（canopy temperature，Tc）是作物水分胁迫计算的基础。准确地剔除热红外图像中的土壤背景，可以提高作物水分的监测精度。该研究以4种水分处理的拔节期夏玉米为研究对象，借助无人机可见光和热红外图像，采用红绿比值指数（red-green ratio index，RGRI）法提取研究区域的面状玉米冠层温度的空间分布信息，并分析每幅热红外图像上冠层温度的累积频率。该并提出了两种改进作物水分胁迫指数（crop water stress index，CWSI）性能的方法，一是使用基于正态分布的不同统计分位数分割冠层温度，并基于不同统计分位数上的平均冠层温度计算CWSI（记为CWSITcF%）。二是基于冠层温度方差（canopy temperature variance，Var），将玉米冠层数据分为4个区间：区间Ⅰ，Tc≤40，Var≤10；区间Ⅱ，Tc≤40，10< Var≤20；区间Ⅲ，35< Tc<45，Var>20；区间Ⅳ，40< Tc<50，0< Var≤20，并在各自区间上选择最敏感的统计分位数计算CWSI（记为CWSIn）。研究结果表明：1）利用2020年和2021年两年数据计算的CWSIn与作物生理指标（气孔导度Gs、净光合速率Pn、蒸腾速率Tr）间的决定系数R2分别为0.72、0.52、0.62 ，nRMSE分别为23.96%、24.06%、25.60%，模型拟合精度高于原始CWSI（R2分别为0.73、0.34、0.46，nRMSE分别为23.69%、28.27%、30.21%），但与CWSITcF%差别不大（R2分别为0.74、0.54、0.61，nRMSE分别为22.87%、23.74%、25.61%）；2）虽然CWSITcF%能提高诊断作物水分胁迫的精度，但最敏感的冠层温度区间在年际间相差较大（2020，61.17%；2021，49.38%；两年数据，83.51%），而CWSIn稳定性更高（与生理指标间的nRMSE分别为：2020年16.60%、27.37%、28.49%；2021年21.60%、18.95%、22.64%）。因此，综合来看 CWSIn可以更加精确地监测作物水分胁迫，利用该改进方法可为无人机遥感精准监测作物水分胁迫状况提供参考。

Improved method of crop water stress index based on UAV remote sensing

Abstract: Canopy temperature is one of the most important parameters to calculate the crop water stress index (CWSI). Among them, the unmanned aerial vehicles (UAVs) with thermal infrared remote sensing function have been widely used to extract the crop canopy temperature in recent years. The monitoring accuracy of crop water can be improved to accurately remove the soil background in the thermal infrared images. In this study, an improved method was proposed to monitor the CWSI using UVA remote sensing. The research object was firstly selected as the summer maize at the jointing stage. Four water treatments were then carried out: T1 (50% of the field capacity), T2 (65% of the field capacity), T3 (80% of the field capacity), and T4 (95%-100% of the field capacity). The spatial distribution of surface maize canopy temperature in the study area was extracted by the red green ratio index (RGRI) with the visible and thermal infrared images from UAVs. Specifically, the UAV images were selected as 5 days in 2020 and 7 days in 2021. The cumulative frequency was then analyzed for the canopy temperature on each thermal infrared image. Soil water content data at a depth of 40cm was also collected at the same time. Subsequently, two methods were proposed to improve the performance of CWSI. The accuracy was evaluated using the correlation between the CWSI and crop physiological indexes (Stomatal conductance, Gs; Net photosynthetic rate, Pn; and Transpiration rate, Tr) with the soil moisture content. Method 1, the canopy temperature was segmented using different statistical quantiles in the normal distribution. Some unreasonable values were then eliminated for the low and high temperature in the histogram of canopy temperature frequency. The commonly-used statistical quantiles were 93.32%, 84.13%, 69.15%, 61.79%, 50%, 38.21%, 30.85%, and 15.87%, respectively. The intervals 0.62% and 2.28% were used to eliminate the high temperature values on the right side of the frequency histogram, and then the CWSI (denoted as CWSITcF%) was calculated from the average canopy temperature over different statistical quantiles. The statistical quantiles with the best correlation with the crop physiological indexes were selected as the most sensitive canopy temperature interval. In Method 2, maize canopy data was divided into four parts: Interval Ⅰ, Tc <40, Var<10; Interval Ⅱ, Tc <40, 10< Var<20; Interval Ⅲ, 35< Tc <45, Var >20; and Interval Ⅳ, 40< Tc<50, 0< Var<20, according to the variance of canopy temperature. The most sensitive statistical quantile on the respective interval was then selected to calculate the CWSI (denoted as CWSIn). The results showed that: 1) The coefficients R2 of CWSIn and crop physiological indexes (Gs, Pn, Tr) calculated from 2020 and 2021 were 0.72, 0.52 and 0.62, and normalized root mean square error (nRMSE) were 23.96%, 24.06% and 25.60%. The accuracy of model fitting was higher than that of original CWSI (R2 were 0.73, 0.34, 0.46, nRMSE were 23.69%, 28.27%, 30.21%), but little different from CWSITcF% (R2 were 0.74, 0.54, 0.61, nRMSE were 22.87%, 23.74% and 25.61%) 2) The most sensitive canopy temperature range varied greatly between years (2020, 61.17%; 2021, 49.38%; and Two-year data, 83.51%), although the CWSITcF% was provided a higher accuracy of crop water stress. There was also some influence on the application of CWSITcF% in practice. The CWSIn presented a higher stability for the crop physiological indexes (The nRMSE between physiological indicators were 16.60%, 27.37% and 28.49% in 2020 and 21.60%, 18.95%, 22.64% in 2021). Therefore, the improved CWSIn can be expected to more accurately monitor the crop water stress under UAVs remote sensing, compared with the original.