Identification of Stellera chamaejasme distribution based on multi-temporal Sentinel-2A images

Stellera chamaejasme has been one of the main invasive noxious weeds in the alpine grasslands of the Qinghai-Tibet Plateau in recent years. Timely and efficient investigation and monitoring can provide important technical support for the integrated control of S. chamaejasme and the restoration of degraded grasslands. In this study, Sentinel-2A multi-spectral images of before flowering and full flowering were selected, and the Google Earth Engine platform cloud removal, environmental factor masking, feature selection, and random forest classification were combined to explore a regional scale remote sensing identification method for S. chamaejasme. We found that seven features of S. chamaejasme classification were extracted and four feature combinations were constructed through the calculation of the S. chamaejasme sensitivity index and the secondary dimensionality reduction, combined with Spearman rank correlation analysis and random forest importance ranking. Compared with the combination of single temporal features, the combination of multi-temporal features effectively improved the recognition accuracy of S. chamaejasme. Among them, the total classification accuracy of the six features combination scheme based on the random forest model was 84.62%, and the classification accuracy of S. chamaejasme was all more than 80%, showing the best recognition effect. This study shows that the image cloud removal and mask preprocessing can effectively reduce the classification interference information, and the combination of multi-temporal features extracted before flowering and in full bloom enhances the spectral difference between S. chamaejasme community and other plant communities, indicating good application potential in the regional scale remote sensing recognition of S. chamaejasme.