基于Geosail模型和SVR算法的叶面积指数遥感反演

Remote sensing inversion of leaf area index based on Geosail model and SVR algorithm

叶面积指数(LAI)控制着植物冠层的多种生理和生态过程,是陆地生态、水文模型中不可或缺的植被参数,因此准确反演区域LAI对研究植被与土壤侵蚀具有重要意义。本文以北京地区阔叶林为研究对象,利用Geosail模型模拟LAI和7种植被指数:比值植被指数(RVI)、归一化植被指数(NDVI)、绿波段植被指数(GNDVI)、重归一化植被指数(RDVI)、土壤调整植被指数(SAVI)、调整土壤亮度植被指数(OSAVI)和修正的土壤调整植被指数(MSAVI),并采用支持向量机回归(SVR)算法和4种统计回归方法(线性函数、二次函数、指数函数和对数函数)建立LAI反演模型,同时通过Landsat 8 OLI遥感数据和实测数据验证模型精度。研究表明:1) SVR算法相比其他统计回归方法可以提高LAI反演的模型精度和预测精度; 2) OSAVI指数在LAI反演方面的表现要优于NDVI等指数;3) NDVI指数的建模精度很高,但预测精度较低; 4) OSAVI和SVR算法构建的模型精度和稳定性更好,是LAI反演的优选模型,其预测结果最为精确。因此,基于Geosail模型和SVR算法的反演方法可提高LAI反演精度,为大区域LAI反演的应用提供了新的思路,扩展了Geosail模型、SVR算法和Landsat 8 OLI遥感数据在LAI反演方面的应用潜力。

[ Background ] As an indispensable vegetation parameter in land ecosystem and hydrological models, Leaf area index controls many physiological and ecological processes of plant canopies. Therefore, real-time and accurate acquisition of regional LAI is very important for studying vegetation and soil erosion. Combining the physical model and the statistical model to estimate physiological parameters of the plant is non-destructive, simple, and highly efficient, which is one of the major approaches to quantitative remote sensing.[ Methods ] In the paper, leaf area index of broad-leaved forest was studied in Beijing. Geosail model, a combination of a geometric model and a mixed medium model, was used to simulate the broad band reflectance of canopy. Prospect model, a kind of leaf optical physical model, was used to simulate the leaf hyperspectral reflectance of broad-leaved forest. The leaf hyperspectral reflectance was converted into leaf broad band reflectance by the spectral response function, and then Geosail model used leaf broad band reflectance to simulate reflectance of canopy of broad-leaved forest. LAI and 7 kinds of remote sensing vegetation indexes were generated by simulated canopy reflectance, including RVI, NDVI, GNDVI, RDVI, SAVI, OSAVI, and MSAVI. Then 4 types of statistical regression methods (Linear function, quadratic function, exponential function, logarithmic function) and SVR algorithm were used to establish LAI inversion models. The accuracy of LAI inversion models was verified by Landsat 8 OLI remote sensing data and measured data.[ Results ] The analysis showed: 1) SVR algorithm could improve accuracy and prediction accuracy of LAI inversion models than other statistical regression methods. 2) The prediction results of LAI inversion models showed that the performance of OSAVI was better than that of NDVI and other vegetation indices in the field of LAI inversion. This indicated that OSAVI could eliminate the most influence of atmospheric condition and soil background by using the correction factor of canopy background in computing formula, and had better anti-interference ability. 3)The LAI inversion modeling and models prediction showed that the modeling accuracy of NDVI index was very high, but in reality, the prediction accuracy of NDVI index was relatively low. 4) The accuracy and stability of the model constructed by the OSAVI and SVR algorithm were better, and it was the preference model for LAI inversion. Its prediction results were the most accurate, its value of coefficient of determination ( R 2 ) was 0.852 8, its value of root mean square error (RMSE) was 0.204 6, and its value of Slope was 0.988 1.[ Conclusions ] Therefore, the inversion method based on Geosail model and SVR algorithm was feasible, which could improve the accuracy of LAI inversion and provide new ideas and methods for the application of LAI inversion in large area. Through the LAI inversion model, the ground measured data could be converted to the remote sensing image data scale, which expanded the application potential of Geosail model, SVR algorithm and Landsat 8 OLI remote sensing data in LAI inversion.