芒果叶片水分含量估算光谱指数模型的建立

Establishment of hyperspectral index model for water content estimation of mango leaf

【目的】建立芒果叶片水分含量高光谱指数模型,实现芒果叶片水分含量快速无损监测,为芒果田间水分管理、日灼病防控、产量提升提供参考依据。【方法】于广西右江河谷百色国家农业科技园区生态遥感试验站采用离体测量法,同步测定芒果叶片反射率(R_(λ))、水分含量、干物质含量、叶绿素含量和类胡萝卜素含量等生化参数。基于实测数据集校正PROSPECT模型,并生成模拟数据库。从实测数据集(n=200)和模拟数据库(n=1000)中各自随机选取85%样本作为建模集(n=1020),建立最优芒果叶片水分含量估算高光谱指数模型,并利用剩余15%样本作为验证集对指数模型进行精度验证。【结果】以实测数据校正PROSPECT模型后,PROSPECT模型模拟精度整体提升。比较实测数据与模拟生化组分数据分布特征,得出模拟数据库中光谱数据与实测光谱数据分布相近。对芒果叶片水分光谱吸收波段进行分析,结果表明,1400和1900 nm波段附近出现水分强吸收波谷。对芒果叶片水分含量进行高光谱指数开发,发现二重差值指数[DDn(1255,25)]与叶片水分含量间的相关性最佳、均方根误差最小(相关系数为0.86,均方根误差为2.58×10^(-3) g/cm~2),因而[DDn(1255,25)]为芒果叶片水分含量最优高光谱指数。利用多项式拟合[DDn(1255,25)]与芒果叶片水分含量,结果表明,基于[DDn(1255,25)]的一次多项式回归模型的决定系数为0.73,F检验值为2757.18,在验证集中表现最优。【结论】[DDn(1255,25)]指数与芒果叶片水分含量的相关性最高,基于该指数建立的一次多项式回归模型能快速无损监测芒果叶片水分含量,可用于芒果叶片水分含量无损监测,服务芒果田间水分管理、日灼病防控等生产活动。

【Objective】Targeting for non-destructive monitoring of leaf water content in mango trees,hyperspectral index model for leaf water content in mango was established to lay a basis for fields water management,prevention of sunscald and production increase.【Method】In this study,measurements were conducted in an Integrated Remote Sensing Experimental Site for mango trees located in the Baise National Agricultural Sci-tech Zone,Guangxi,China.Leaf samples were collected using the detached branch method.Leaf reflectance(R_λ) and biochemical parameters(leaf water content,dry matter mass content,chlorophyll content,carotenoid content) were measured for each leaf sample.Based on the measured data,the PROSPECT model was calibrated and used for data simulation.85% of leaf samples were randomly selected from the measured dataset(n=200) and the simulated database(n=1000) served as the calibration subset(n=1020).Hyperspectral index models for estimating the leaf water content of mango were calibrated from the calibration subset,and validated with the remaining 15% of leaf samples to access the accuracy of the model.【Result】After correcting the PROSPECT model with measured data,the overall accuracy of the model simulation was improved.By comparing the distribution characteristics of the measured and simulated biochemical components data,it was concluded that the spectral data in the simulated database was similar to the measured spectral data.The spectral absorption bands of leaves were analyzed,and the results showed that there were strong absorption troughs near 1400 and 1900 nm.The spectral index of mango leaf water content was developed,and it was found that the correlation between the double difference index[DDn(1255,25)]and the leaf water content was the best,and the root-mean-square error was the smallest(correlation coefficient was 0.86,root mean square error was 2.58×10~(-3) g/cm~2).[DDn(1255,25)]was the optimal hyperspectral index of water content in mango leaves.Using polynomial fitting[DDn(1255,25)]and the water content of mango leaves,the results showed that the coefficient of determination(R~2) of the linear polynomial regression model based on[DDn(1255,25)]was 0.73,and the F test value was 2757.18,which had the best performance in the verification set.【Conclusion】The correlation between[DDn(1255,25)]index and water content of mango leaves is the highest.The linear polynomial regression model based on the[DDn(1255,25)]can estimate the leaf water content of mango trees quickly and non-destructively which can be applied to fields of water management,prevention of sunscald and other production activities.