基于高光谱数据的互花米草叶片功能性状反演

互花米草成功入侵的关键是其生长繁殖能力以及对环境的适应能力，叶片含水率、相对叶绿素含量、碳氮比、总氮、总磷以及比叶面积等叶片功能性状反应的是互花米草对资源的利用能力以及环境的适应能力。以江苏盐城滨海湿地为研究对象，进行互花米草叶片功能性状与高光谱数据的关系研究。通过对原始光谱数据以及一阶微分转换光谱数据进行主成分分析提取新的主成分变量作为自变量分别建立不同性状的逐步回归、BP神经网络、支持向量机、随机森林4种预测模型，通过比较构建模型的R²以及RMSE选择最优模型，进而基于相关性分析得到的敏感波段构建最优模型，验证其准确性和适用性。研究结果发现：（1）一阶微分数据的建模效果优于原始光谱数据；（2）通过对不同功能性状的预测建模，发现4种模型的预测效果排序为：随机森林>支持向量机>BP神经网络>逐步回归，其中随机森林模型的准确性高、稳定性强，明显优于其他3种模型，而逐步回归模型的效果最差，不适用于互花米草叶片功能性状的高光谱建模；（3）通过对相关性分析得到的敏感波段建立随机森林模型，建模R²均大于0.90，验证R²介于0.73-0.95之间，进一步证实了随机森林模型的准确性和稳定性。研究结果表明，高光谱数据可以作为快速监测互花米草生长状况的有力手段，而随机森林模型可以作为高精度模型实现对互花米草不同叶片功能性状的估测。

Inversion of functional traits of Spartina alterniflora leaves based on hyperspectral data

The key to the successful invasion of Spartina alterniflora is its ability to grow and reproduce and its ability to adapt to the environment. Leaf water content, relative chlorophyll content, carbon-to-nitrogen ratio, total nitrogen, total phosphorus, and specific leaf area, and other leaf functional traits reflect S. alterniflora''s ability to utilize resources and adapt to the environment. This study was carried out in the coastal wetland of Yancheng, Jiangsu Province, to study the relationship between leaf functional characteristics of S. alterniflora and hyperspectral data. Principal component analysis was performed on the original spectral data and first-order differentially transformed spectral data to extract new principal component variables as new independent variables. Then stepwise regression, BP neural network, support vector machines, and random forest regression models for different leaf functional traits were established. The optimal model is selected by comparing the decision coefficient R² and Root mean square error (RMSE) of the constructed model, and then the optimal model is constructed based on the sensitivity band obtained by the correlation analysis to verify its accuracy and applicability. The results indicated that (1) the accuracy of the first derivative transformation of the hyperspectral data was better than that of the original spectral data. (2) Through predictive modeling of different leaf functional traits, it was found that the prediction effects of the four models are ranked as follows: random forest > support vector machines > BP neural network > stepwise regression. The random forest model had high accuracy and stability, which was obviously better than the other three models. The stepwise regression model had the worst effect and was not suitable for modeling and predicting leaf functional characteristics of S. alterniflora based on hyperspectral data. (3) Using the sensitive bands obtained by correlation analysis as independent variables, we established random forest models with different leaf functional traits. The R² of the constructed models were all greater than 0.90, and the R² of the verification model was between 0.73 and 0.95, which further confirmed the accuracy and stability of the random forest model. These results showed that the hyperspectral data can be used as a powerful means to quickly monitor the growth status of S. alterniflora, and the random forest model can be used as a high-precision model to estimate the different leaf functional characteristics of S. alterniflora.