Evaluating multitemporal vegetation indices from Zhuhai-1 hyperspectral images for detecting a rapidly spreading invasive species - Spartina alterniflora

Monitoring the spatiotemporal changes of Spartina alterniflora (SA) is essential in effectively managing coastal ecology since it is one of the most harmful invasive weeds worldwide. However, it remains challenging to accurately identify SA invasion, especially in regions subject to periodic tidal flooding. Recent studies have shown that utilizing traditional multitemporal vegetation indices (VIs), such as NDVI and EVI derived from multispectral image features, can improve the accuracy of identifying SA. Still, the application potential of multitemporal hyperspectral images with rich derived VIs has not yet been explored. The Zhuhai-1 hyperspectral satellite offers high spectral, spatial, and temporal resolution, providing crucial multitemporal features for accurately identifying SA. This study examined multitemporal VIs from nine months using hyperspectral images and common machine learning methods (i.e., K-nearest neighbor, support vector machine, random forest) to compare a variety of VIs' performance in identifying SA invasion in the Guangxi Zhuang Autonomous Region. Results showed that multitemporal VIs are more effective in identifying SA in periodic tidal flooding areas than individual hyperspectral parameters (spectral features, VIs, and spatial texture features). Significantly, the unique multitemporal VIs derived from red-edge bands of hyperspectral images constantly demonstrated higher accuracies (exceeding 91.6 %) than traditional NDVI (91.47 %) and EVI (84.78 %). Our results consistently identified June, February, and November as the most critical months for identifying SA invasion, as observed across all three algorithms and VIs. These months are connected to SA phenology's greening, yellowing, and withering. Results and findings from this study provided insight into the overwhelming potential of multitemporal hyperspectral image analyses to improve the monitoring and management of invasive species for sustainable coastal ecosystems. The same procedure, algorithms, indices, and features can be employed to effectively identify any other specific species or detailed land cover types.