东北主要绿化树种叶面积指数(LAI)高光谱估算模型研究

A Study for Estimating the Main Tree Species Leaf Area Index in Northeast Based on Hyperspectral Data

以东北主要绿化树种为研究对象,分别在长春市南湖公园和长春公园获取了共240组树冠高光谱反射率及相应的LAI数据。对数据进行相关分析,以确定反演LAI的敏感波段,而后分别运用6种植被指数、神经网络以及小波分析等3种方法进行估算。研究结果表明,3种方法估算树冠LAI都取得了较好的效果:①与RVI、NDVI相比,由DVI、RDVI、MSAVI、TVI等植被指数建立的估算模型可以提高LAI的估算精度;②神经网络在拟合光谱反射率与树冠LAI关系时明显优于植被指数法(R2达0.850);③小波能量系数与LAI相关性较好,单变量回归分析R2可达0.683,部分小波能量系数估算LAI的精度优于植被指数法,并且验证R2也较高,说明其稳定性较好,多元变量回归分析能够实现各小波能量系数间的优势互补,R2可达0.794。

LAI is a key biophysical variable,used in most global models of climate,ecosystem productivity,biogeochemistry,hydrology and ecology.Hyperspectral remote sensing provides an effective method to monitor the physiological and biochemical parameters of vegetation canopy.Two experiments were carried out in South Lake Park and Changchun Park respectively with the major greening tree species in the Northeast as the study object.Totally 240 groups of hyperspectral reflectance and corresponding LAI were obtained.After analyzing the correlations of reflectance and derivative reflectance with LAI,an evaluation of canopy LAI retrieval methods was conducted using 6 vegetation indices,neural network method and spectrum analysis of wavelet energy coefficient and the estimate effects of three methods were compared.The results indicated that all these methods had an ideal effect on the estimation of LAI： ① Compared with RVI and NDVI,Four vegetation indices（DVI,RDVI,MSAVI,TVI） may enhance the precision to estimate LAI;② The estimations were further improved when neural network method was used（R2 reached 0.850）;③ Spectral wavelet energy coefficients showed a better correlation with canopy LAI.R2 of single variable regression analysis may reach 0.683 and the accuracy of individual wavelet coefficients to estimate LAI was superior to vegetation index method with higher R2 in validated model.When a subset of wavelet coefficients was analyzed,it was found that multi-variable regression analysis had reduced the error in the retrieved parameter with R2 of measured value and predicted value of the LAI being 0.794.When the LAI was small,vegetation indices were obvious for removing the noise from soil and atmospheric effect and obtained good evaluation result.Neural network method and wavelet analysis could weaken the effect by saturation at high LAI levels and showed better effect for all LAI.