干旱胁迫下冬小麦叶面积指数的高光谱监测

Hyperspectral Monitoring of LAI in Winter Wheat under Drought Stress

为实现干旱条件下冬小麦叶面积指数(LAI)的快速精准监测,为我国干旱半干旱地区的冬小麦安全生产提供理论和技术支持,为干旱条件下冬小麦LAI的快速精准监测提供参考,以不同干旱条件下的冬小麦为研究对象,测定关键生育时期冬小麦的LAI和冠层高光谱反射率,探索冬小麦LAI和高光谱反射率间的响应趋势,分别采用一阶微分(1ST)、标准正态变换(SNV)和对数变换(Log)3种方法对原始光谱反射率(R)进行预处理;对比分析不同预处理后光谱反射率与LAI的相关关系,采用偏最小二乘回归(PLSR)算法构建不同预处理下冬小麦LAI的监测模型。结果表明,在可见光区域和1100~1800 nm光谱区域,随着LAI的升高,光谱反射率呈现出逐渐增大的趋势;3种预处理方法均改善了光谱反射率与冬小麦LAI的相关性,其中1ST预处理后的光谱反射率与LAI的相关性最高,在1209 nm处相关系数达到了0.550。基于1ST预处理后光谱反射率,采用PLSR构建的冬小麦LAI监测模型表现最佳,相较R-PLSR模型,1ST-PLSR校正模型和验证模型的决定系数分别提高了42.974%和8.842%,均方根误差分别降低了33.710%和8.111%,相对分析误差分别提高了50.838%和8.813%。

To enable the rapid and accurate monitoring of leaf area index(LAI)in winter wheat under drought conditions,for providing theoretical and technical support for the safe production of winter wheat in arid and semi-arid regions of China and providing reference for rapid and accurate monitoring of winter wheat LAI under drought conditions.In this study,taking winter wheat under different drought conditions as the research object,the LAI and canopy hyperspectral reflectance of winter wheat in key reproductive period was measured the response trend between LAI and hyperspectral reflectance was explored,and the raw spectral reflectance(R)was pre-processed using three methods includingfirst-order differentiation(1ST),standard normal variation(SNV),and logarithmic transformation(Log).The correlation between spectral reflectance and LAI after different preprocesses was comparatively analyzed,and the partial least squares regression(PLSR)algorithm was employed to construct models for monitoring LAI in winter wheat under different pretreatments.The results indicated that spectral reflectance exhibited a gradual increase in tandem with rising LAI in the visible region and the 1100–1800 nm spectral region.All three preprocessing methods improved the correlation between spectral reflectance and LAI of winter wheat,with the highest correlation following 1ST pretreatment,which reached 0.550 at 1209 nm.The LAI monitoring model for winter wheat constructed using PLSR based on 1ST preprocessed spectral reflectance performed the best.Compared to the R-PLSR model,the r2 of the 1ST-PLSR correction model and validation model increased by 42.974%and 8.842%,while the root mean square error decreased by 33.710%and 8.111%,and the ratio of performance to deviation increased by 50.838%and 8.813%.