基于PLS算法的棉花黄萎病高空间分辨率遥感监测

棉花黄萎病危害程度大，发生范围广，已成为中国乃至世界上棉花主要病害之一。论文基于野外定位调查数据及高空间分辨率遥感影像，利用变量投影重要性（VIP）准则筛选最优变量，用偏最小二乘回归（PLS）方法建立棉花黄萎病病情严重度的定量估测模型，并利用已建立的估测模型和高分辨率IKONOS影像获取了不同病情严重度的空间分布图。研究结果表明：在所分析的13个遥感因子中，增强植被指数（EVI）、再归一化植被指数（RDVI）、全球环境监测指数（GEMI）、差值植被指数（DVI）、修改型土壤调整植被指数（MSAVI）、归一化植被指数（NDVI）为棉花黄萎病病情严重度遥感估测的敏感因子，能够有效估测棉花黄萎病病情严重度，其模型预测值与实测值间的R2、RMSE和RE分别为0.78、0.45、9.2%。论文利用PLS算法和高分辨率卫星影像实现了棉花黄萎病病情严重的遥感监测，研究结果对实现大范围农作物病虫害的遥感监测具有重要的参考价值。

Remote sensing monitoring severity level of cotton verticillium wilt based on partial least squares regressive analysis

The objective of this study was to estimate the severity level of cotton verticillium wilt using high spatial resolution satellite data and partial least-squares regressive analysis. Firstly, remote sensing factors for monitoring cotton verticillium wilt were picked out by the pre-processed IKONOS image. Then the variable factors were selected according to an approach of Variable Importance in Projection, and used to establish the severity estimating model of cotton verticillium wilt using partial least squares (PLS) regression analysis. Finally the model was applied to calculate the severity level of each pixel in the region of cotton verticillium wilt. The results show that these vegetation indices, i.e. enhanced vegetation index (EVI), renormalized difference vegetation index (RDVI), global environment monitoring index (GEMI), difference vegetation index (DVI), modified soil adjusted vegetation index (MSAVI) and normalized difference vegetation index (NDVI), are sensitive factors for monitoring severity level of cotton verticillium wilt. The model based on those variables achieves better accuracy since precision assessment indices such as determination coefficient (R2), root mean square error (RMSE) and relative error (RE) which are 0.78, 0.45 and 9.2% respectively. Severity level of cotton verticillium wilt can be effectively estimated utilizing high spatial resolution image and partial least squares regression analysis and the result presents an important reference approach for further monitoring crop pests and diseases at large scale using airborne and airspace remote sensing data.