Hyperspectral imaging data have been rarely focused on studies of mangrove pests and diseases. With leaf hyperspectral imaging data, this study aims to extract the sensitive spectral and textural features related to information of mangrove pest and disease using successive projection algorithm (SPA), and to model and visualize leaf traits in response to different pest and disease severity using random forest (RF). The results showed that multiple repetitions of SPA and RF modeling operations could provide a robust set of sensitive features and reliable accuracies of vegetation parameter estimation. Among the five types of features (450 bands of original and first derivative reflectance, 52 vegetation indices, 112 texture features, and all coupling features), the RF models with 33 sensitive features chosen from the coupling of all the 1064 features, 13 sensitive wavelengths with first derivative reflectance, and 30 sensitive wavelengths with first derivative reflectance reported the optimal validation performance (mean R²_Val = 0.752, 0.671, and 0.658) in estimating pest and disease severity, leaf SPAD-502, and leaf NBI values, respectively. Moreover, the two leaf trait values increased with decreasing severity of pest and disease based on the leaf trait visualization map using the optimal SPA-RF model. We conclude that the combination of SPA-RF model and hyperspectral imaging had great potential in detecting the spatial distribution of leaf traits under different pest and disease severity. The leaf-level study could lay foundation for early warning and monitoring of mangrove pests and diseases at the landscape or region level.

高光谱成像数据很少关注红树林病虫害的研究。本研究旨在利用叶片高光谱成像数据，使用连续投影算法（SPA）提取与红树林病虫害信息相关的敏感光谱和纹理特征，并使用随机算法对响应不同病虫害严重程度的叶片性状进行建模和可视化。森林（RF）。结果表明，多次重复 SPA 和 RF 建模操作可以提供一组稳健的敏感特征和可靠的植被参数估计精度。在5类特征（450个原始和一阶导数反射率、52个植被指数、112个纹理特征和所有耦合特征）中，从所有1064个特征的耦合中选择了33个敏感特征的RF模型，² _Val  = 0.752、0.671 和 0.658）分别用于估计病虫害严重程度、叶片 SPAD-502 和叶片 NBI 值。此外，基于使用最佳 SPA-RF 模型的叶性状可视化图，两个叶性状值随着病虫害严重程度的降低而增加。我们得出结论，SPA-RF 模型和高光谱成像的结合在检测不同病虫害严重程度下叶片性状的空间分布方面具有很大的潜力。叶级研究可为红树林病虫害在景观或区域层面的预警和监测奠定基础。