叶冠尺度野鸭湖湿地植物群落含水量的高光谱估算模型

利用高光谱遥感技术定量估测野鸭湖湿地植被含水量,对于监测和诊断野鸭湖湿地植被的生理状况及生长趋势具有重要意义,也能够为高光谱遥感影像在野鸭湖湿地植被含水量诊断中的实际应用提供理论依据和技术支持.采用Field Spec 3野外高光谱辐射仪,获取了野鸭湖典型湿地植被冠层和叶片的光谱,并测定了对应的含水量.以上述实测数据为基础,首先以芦苇为例初步探明了不同含水量水平下典型湿地植被冠层和叶片光谱反射率的响应模式,然后采用相关性及单变量线性与非线性拟合分析技术,从冠层和叶片两种层次,对不同尺度下的含水量与“三边”参数及高光谱植被指数进行了分析拟合,并采用交叉检验中的3K-CV方法对估算模型进行了测试和检验,确立了不同尺度下野鸭湖湿地植被含水量的定量监测模型.结果表明:(1)随着含水量水平的增加,芦苇冠层与叶片光谱在可见光波段(350-760 nm)和红外波段(760-2500 nm)的反射率均呈逐渐降低趋势.(2)不同尺度含水量与选取的光谱特征参数整体上相关性较强,与“三边”参数基本上都呈极显著相关,相关系数最大达到0.906;与高光谱指数全部呈极显著相关,相关系数最小为0.455,最大达到0.919,并通过选取不同尺度上相关性最佳的光谱特征参数,分别基于“三边”参数和高光谱植被指数构建了不同尺度下的含水量估算模型.其中,冠层尺度下,黄边面积(SDy)与SRWI( Simple Ratio Water Index)的估算效果最好,估算模型分别为y=-9.462x2 -2.671x+0.608和y=0.219e1.010x；叶片尺度下,红边面积(SDr)与WI( Water Index)的估算效果最好,估算模型分别为y=0.562x+0.376和y=2.028x2 -0.476x-1.009.通过3K-CV的交叉验证,不同尺度下的含水量估算模型均取得了较为理想的预测精度,预测精度的最小值为94.92％,最大值为97.06％,表明估测模型具有较高的可靠性与普适性.(3)高光谱植被指数与含水量拟合方程的拟合度相对高于“三边”参数与之建立方程的拟合度,说明多波段组合的光谱特征参数更适合含水量的判别.

Hyperspectral estimation models for plant community water content at both leaf and canopy levels in Wild Duck Lake wetland

Quantitative estimation of vegetation water content with hyperspectral remote sensing technique is of great significance for vegetation physiological status and growth trend monitoring. It also provides a theoretical foundation for actual application of vegetation water content diagnosis using hyperspectral remote sensing images in Wild Duck Lake wetland. The hyperspectral reflectance and corresponding water content of canopy and leaf of typical wetland vegetation were measured by Field-Spec 3 wild high-spectrum radiometer. We used reed as an example to prove the response mode of the spectral reflectance in different water content levels. Then the correlation among water content, trilateral parameters,and hyperspectral vegetation index and hyperspectral estimation models were obtained by using regression and correlation analysis in canopy and leaf levels. In additiona, we made use of Three K-fold Cross Validation to test and inspect the hyperspectral estimation models. The results show: a)The canopy and leaf spectral reflectance of reed in visible bands (350-760nm) and infrared bands (760-2500nm) reflectivity tends to reduce gradually. b) We found strong correlation between the selected spectrum characteristic parameters and different water content scales. For trilateral parameters, they show significant correlation with a maximum correlation coefficient of 0.906. For the hyperspectral vegetation index, they all show significant correlation, with a minimum correlation coefficient of 0.455 and a maximum of 0.919. Based on the trilateral parameters and hyperspectral vegetation index, we choose the spectrum characteristic parameters that have the best correlation in different scales to construct the water content estimation models. SDy and simple ratio water index have the best effect at canopy level; the best models are evaluated and validated as y=-9.462x²-2.671x+0.608 and y=0.219e^1.010x, respectively. SDr and water index have the best effect at leaf level; the best models are evaluated and validated as y=0.562x+0.376 and y=2.028x²-0.476x-1.009, respectively. According to Three K-fold Cross Validation examination, these estimation models have the satisfactory prediction accuracy, with a minimum prediction accuracy of 94.92% and a maximum of 97.06%. Those results indicate that the parameters adopt in this paper have great reliability and applicability for water content estimation. c) We make use of the hyperspectral remote sensing to analyzee and estimate wetland vegetation water content, and this method is a new nondestructive testing technology. Compared to the traditional physiological and biochemical analysis method in the laboratory, the remote sensing method not only gets non-point source information, but also reduces the analysis cost greatly, and greatly improves the efficiency of the analysis. The hyperspectral vegetation index has a higher degree of fitting than the trilateral parameters, so the spectrum characteristic parameters of band combination are more suitable for water content discrimination. d) For future research, we need to analyze further and systematically the relationship between spectral indexes and water content, and between the spectral indexes and any single band or combination of two bands in ratio (SD) and the normalized (ND) from 350 to 2500 nm. We plan to use the special handheld blade spectrum detectors to determine the leaf spectral reflectance, so that the reliability and accuracy of estimation model will be further enhanced.