三种回归分析方法在Hyperion影像LAI反演中的比较

借助GPS进行地面精确定位,利用LAI-2000冠层分析仅在攸县黄丰桥林场开展130个样地(60m×60m)的叶面积指数(Leaf Area Index,LAI)测量.采用FLAASH模块对Hyperion数据进行大气校正并与地面同步冠层观测数据进行拟合,通过研究地面实测LAI与Hyperion影像波段及其衍生的系列植被指数(NDVI、RVI等)的相关性,筛选出估算叶面积指数的植被指数因子.应用曲线估计、逐步回归及偏最小二乘三种回归分析技术分别建立叶面积指数的最优估算模型.结果表明:参与建模的因子中,比值植被指数(RVI)与LAI的相关性最大,敏感性最高,其次是SARVI0.1,NDVI705,NDVI,SARVI0.1,SARVI0.25;曲线估计、逐步回归分析和偏最小二乘回归三种分析方法所建的6个回归模型中,偏最小二乘回归的拟合效果最好,预测值与实测值的决定系数R2为0.84、曲线估计的拟合效果最低,预测值与实测值的决定系数R2为0.64;建模精度分析表明,选用5-6个自变量因子进行LAI建模是可靠的,以6个植被因子建立的偏最小二乘回归模型预测精度最高.

Comparison of three regression analysis methods for application to LAI inversion using Hyperion data

This paper focuses on Leaf Area index (LAI) inversion, using EO-1 Hyperion data for Huangfengqiao forest farm, YouXian County, Hunan Province. First, LAI was acquired using a LAI-2000 canopy analyzer at 130 sample plots (60 m × 60 m), with a Global Positioning System (Trimble GPS Geo XT). Second, atmospheric correction was applied to Hyperion data using the ground-synchronous canopy observation data. Third, effective vegetation indexes were selected to estimate LAI, according to research on correlation between LAI, bands and vegetation indexes derived from Hyperion imagery. Finally, an optimal estimation model of LAI was built by curve estimation, stepwise regression, and a partial least-squares regression algorithm. Results show that sensitivity of ratio vegetation index (RVI) was highest among all model factors, followed by SARVI_0.1, NDVI₇₀₅, NDVI, SARVI_0.1, and SARVI_0.25. Among all fit models, the effect of the partial least-squares regression was best, with R² coefficient 0.84, whereas the curve estimation effect was worst, with R² coefficient 0.64. Model precision analysis shows that it is reliable to build the model using 5 to 6 independent variables, and prediction accuracy of the partial least-square regression was the greatest.