基于PLSR的典型沼泽湿地植物叶片性状与光谱模型构建——以三江国家级自然保护区为例

通过植物光学特性测量叶片性状是一种非破坏性的、长期的湿地动态监测方法。选择三江国家级自然保护区多种典型湿地植物为研究对象，探究植物性状与叶片光谱之间的联系。研究表明：叶片氮含量与光谱的模型构建效果最好，模型R²为0.61，均方根误差（RMSE）为2.3862；叶片含水量、叶片磷、可溶性糖、纤维素和木质素含量之和的模型一般，R²在0.38~0.55范围内，RMSE在0.0004~10.7019范围内；淀粉含量拟合效果较差， R²为0.29，RMSE为0.0106。光谱预测重要性的结果表明，可见光与近红外边缘范围内的光谱信息对于叶片含水量、叶片氮含量、叶片磷含量、单位叶面积质量、纤维素和木质素之和、可溶性糖和淀粉的预测具有最高的重要性。

Models Building Between Leaf Spectral Reflectance and Traits for Typical Marshland Wetland Plants Based on PLSR: A Case of Sanjiang National Nature Reserve

In this article, the importance of each band for predicting leaf character data was obtained by calculating variable importance of prediction (VIP) values. In the summer of 2020, the full-wavelength spectral information (350-2500 nm) of 12 species of typical wetland plants in Sanjiang National Nature Reserve was collected by ASD Lab spectrometer. Seven functional traits of plant leaves were tested in the laboratory. Through principal component analysis, the optimal principal component number of the spectrum required for modeling was obtained. The traits and spectrum models of 12 wetland plants were established by Partial Least Squares Regression (PLSR). The R² and RMSE were used to test the fitting effect of the models. The VIP value is used to determine the most important spectral region. Finally, two main conclusions were drawn. Firstly, compared with studies on single species in other ecosystems, the traits-spectral general models for 12 typical wetland plants were satisfied except for starch. Among the seven traits, spectral model of nitrogen had the highest model accuracy. The fitting results of LMA, water content, leaf phosphorus, sugar, cellulose, and lignin were good, with R² ranging from 0.38 to 0.55. The fitting result of starch content model was not ideal which R² was only 0.29. Secondly, the VIP value showed the importance of each band in model construction. The first and second peaks of the VIP value are in the range of 400-405 nm and 708-719 nm. Therefore, for 7 traits, visible spectrum and near infrared edge are the most important for model construction. Inversion of plant traits through hyperspectral data has become one of the important means of ecological environment monitoring. But due to the complexity of wetland environment, spectral inversion of leaf traits of wetland plants had rarely studied. Based on previous modeling methods of grassland and forest ecosystems, this study established general models for 7 leaf functional traits for 12 typical wetland plants, it is feasible to monitor wetland plant traits by using spectrum. However, the fitting effect of starch content model was not ideal and needs to further improvement.