Combination of UAV Hyperspectral and LiDAR Data for Classifying Karst Wetland Vegetation Species Using the AEL-Stacking and LIME

Karst wetlands play a crucial role in global biodiversity conservation, water regulation, and carbon sequestration. Accurate classification of wetland vegetation species is vital for the effective conservation and restoration of these ecosystems. However, the issue of the challenges arises from the diversity of species and the spectral similarity of their canopies. This study addresses these challenges by integrating horizontal structural features from hyperspectral imagery (HSI), including vegetation indices and spectral and texture features, with vertical structural features derived from light detection and ranging (LiDAR) data, such as height variables, intensity variables, and canopy characteristics. To achieve precise vegetation species classification, we constructed adaptive ensemble learning stacking (AEL-Stacking) and deep learning models while exploring the impact of different feature datasets and classifiers on vegetation species mapping. The LIME (local interpretable model-agnostic explanations) method was utilized to assess the contribution of individual features to classification performance. Our findings reveal that (a) integrating HSI and LiDAR features achieved the highest overall accuracy (87.91% to 92.77%), surpassing their single feature datasets by 4.44% to 9.51%; (b) the AEL-Stacking outperformed the other models, with the accuracy improvements of 0.96% to 7.58% over the Swin Transformer; (c) there are more significant differences in classification results between the 4 classifiers based on HSI (the classification accuracy of Lotus is most affected by the classifier and the dataset); and (d) LiDAR features played a pivotal role in karst wetland classification, with most of vegetation species indicating the high sensitivity to DSM (digital surface model)-derived features. Our works highlight the critical role of HSI and LiDAR in improving karst wetland vegetation species classification.