基于PROSPECT模型的植物叶片干物质估测建模研究

Plant Leaf Dry Matter Estimating and Modeling Based on PROSPECT Model

为了快速、准确估测植物叶片干物质含量,为作物长势及健康状况监控提供数据支撑,利用光谱分析技术探讨了干物质含量敏感光谱波段提取方法及其估测建模方法。试验数据由叶片辐射传输模型PROSPECT在干物质含量(0.001~0.02)g·cm^(-2)范围内进行模拟,随机产生1000条400~2500nm的光谱曲线,其中600条光谱曲线用于建模研究、400条光谱曲线作为模型验证数据,同时应用叶片光学特性数据库LOPEX93中325条叶片光谱-干物质含量数据进行进一步验证。首先应用试验数据进行局部敏感性分析,初步得到叶片干物质敏感波段范围,再运用改进Sobol算法进行全局敏感性分析,提取了干物质含量敏感的光谱波段范围,在此敏感波段范围运用波段组合算法计算归一化植被指数NDVI与叶片干物质含量相关系数,优选了4组相关性大的波段组合建立归一化干物质指数NDMI_((1644,1719))、NDMI_((1871,2294))、NDMI_((2150,2271))、NDMI_((1496,2282))用于干物质含量估测建模。结果表明:NDMI_((1644,1719))和NDMI_((1871,2294))模型中三次多项式形式(cubic)效果最佳、NDMI_((1496,2282))模型中幂指数形式(power)效果最佳,三者中NDMI_((1871,2294))的三次多项式模型最优,决定系数R^2为0.837,对叶片干物质含量具有较好的估测能力。

In order to estimate the dry matter of plant leaves quickly and accurately and provide data support for crop growth and health status monitoring,we used spectral analysis technique to explore the extraction method of dry matter sensitive spectral bands and its estimation modeling method.The experimental data were simulated by leaf optical properties spectra(PROSPECT)in direct mode,when dry matter in the range of 0.001-0.02 g·cm-2.From the randomly generated 1000 spectral curves between 400 nm and 2500 nm,600 spectral curves were used for modeling studies,and 400 spectral curves were used as model validation data.The models were further validated by the 325 leaves spectral-dry matter data from the Leaf Optical Properties Experiment 93(LOPEX93).Firstly,local sensitivities of leaf dry matter were preliminarily obtained by using the experimental data,and then global sensitivity was analyzed by using the improved Sobol algorithm.The range of spectral bands sensitive to dry matter was extracted,and the spectral combination algorithm was used to calculate the correlation coefficient between the normalized vegetation index NDVI and the dry matter of plant leaves.Four groups of correlation bands NDMI(1644,1719),NDMI(1871,2224),NDMI(2150,2271)and NDMI(1496,2282)were used as characteristic bands for dry matter estimation modeling.The results showed that the cubic polynomials in NDMI(1644,1719)and NDMI(1871,2294)models were the best,and that the power exponent in NDMI(1496,2282)model was the best.Among the three models,the third-order polynomial model of NDMI(1871,2294)was the best,and the coefficient of determination R2 was 0.837,which had good estimation ability.