基于赤池信息量准则的冬小麦叶面积指数高光谱估测

摘要

冬小麦叶面积指数（leaf area index，LAI）是描述冠层结构的重要参数之一，对评价其长势和预测产量具有重要意义。该文利用灰色关联分析（grey relational analysis，GRA）对植被指数进行排序，用偏最小二乘法（partial least squares regression，PLS）选择不同的植被指数个数作为自变量进行回归建模，通过赤池信息量准则（Akaike’s information criterion， AIC）选择AIC值最小的模型作为冬小麦LAI最优估算模型，即GRA、PLS和AIC 3种方法整合建立冬小麦LAI最优估算模型。使用2008－2009年在中国北京通州区和顺义区获取的整个生育期冬小麦LAI和配套的光谱数据进行建模，利用2009－2010相关数据进行验证。研究表明：采用GRA评价标准与冬小麦LAI关联度最大的植被指数是VOG1，关联度最小的植被指数是SR；通过AIC建立的以8个植被指数作为自变量的冬小麦LAI模型效果最优，建模集的决定系数R2和标准误SE分别为0.76和0.009，验证集的R2和相对均方根误差RRMSE分别为0.63和0.004，预测模型和验证模型均具有较高的精度和可靠性。结果表明采用GRA-PLS-AIC方法进行冬小麦LAI反演是可行的，为提高冬小麦LAI遥感预测精度提供了一种有效的方法。%Winter wheat leaf area index (LAI) is one of important parameters in describing the canopy structure, which is particularly significant in the analysis of winter wheat growth and the yield prediction. The objective of the study was to demonstrate the feasibility of remote sensing monitoring on winter wheat LAI and its expansibility in spatial and temporal scale. Canopy LAI variables from remote sensing data were investigated using empirical statistics inversion model. This study focused on analyzing the correlation between vegetation index and LAI. After sorting the vegetation index using grey relational analysis (GRA), the number of independent variables of different vegetation indices was chosen to participate in the regression using the partial least squares regression (PLS). Based on these LAI models, the Akaike’s information criterion (AIC) values were calculated, and the model with the smallest AIC value was chosen as the optimal winter wheat LAI estimation model, i.e. the optimal winter wheat LAI estimation model was established by integrating the methods of GRA, PLS and AIC. Spectral reflectance of leaves and concurrent LAI parameters of samples were acquired in Tongzhou and Shunyi District, Beijing City, China during 2008-2009, which were for model establishment. Fourteen vegetation indices related to LAI were chosen to evaluate the model of LAI. Firstly, the correlation coefficient was analyzed, and it was found that there was significant negative correlation between transformed chlorophyll absorption in reflectance index / optimized soil adjusted vegetation index(TCARI/OSAVI), structure insensitive pigment index (SIPI), photon radiance index (PRI) and LAI, and significant positive correlation between normalized difference vegetation index(NDVI), simple ratio index(SR), OSAVI, normalized difference vegetation index 705 (NDVI705), modified red edge simple ratio index(MSR705), modified red edge normalized difference vegetation index (mNDVI705), Vogelmann index 1 (VOG1), modified chlorophyll absorption in reflectance index (MCARI2), modified soil adjusted vegetation index (MSAVI), modified simple ratio (MSR) and LAI. Secondly, the related degree order between 14 vegetation indices and winter wheat LAI could be drawn as follows: VOG1 > SIPI > MCARI2 > NDVI > MSR > mNDVI705 > OSAVI > NDVI705 > TCARI / OSAVI > TCARI > PRI > MSAVI > MSR705 > SR. Among them, the biggest GRA correlation degree was VOG1, whose value was 0.9211 and the smallest was SR, whose value was 0.6178. Thirdly, in accordance with the arrangement size of GRA, we used the PLS algorithm to increase the number of independent variables in turn to build 9 winter wheat LAI inversion models. Based on the AIC, we filtered and optimized the 9 winter wheat LAI models. Then, the optimal winter wheat LAI model was constructed by 8 independent variables, which were VOG1, SIPI, MCARI2, NDVI, MSR, mNDVI705, OSAVI and NDVI705. The decision coefficient (R2) and standard error (SE) of the GRA-PLS-AIC method modeling were respectively 0.76 and 0.009, which had a higher ability to predict winter wheat LAI. Considering the temporal characteristics of winter wheat LAI, we incorporated the relevant data from 2009 to 2010 to the model and evaluated its ability of estimating the winter wheat LAI in different years. TheR2, relative root mean standard error (RRMSE) and the slope of the fitted line between measured and predicted LAI value in validation set by GRA-PLS-AIC method were respectively 0.63, 0.004 and 0.68. It shows that the model has a higher predictive ability, which lays an important foundation for improving the precision of forecasting winter wheat LAI using remote sensing method.