Mapping the occurrence and spatial distribution of noxious weed species with multisource data in degraded grasslands in the three-river headwaters region, China.
The invasion of noxious weed species has long been associated with the degradation of alpine grasslands ecosystems . However, traditional in situ-based methods for surveying noxious weed species are generally time consuming and inefficient over large-scale areas. This paper investigates the possibility of applying multisource data to map the occurrence and spatial distribution of noxious weed species in degraded alpine grasslands in the Three-River Headwaters Region, China. Sentinel-2 image-related vegetation indices (VIs), field sample data and environmental variables were integrated to build a noxious weed species detection model based on the maximum entropy (MaxEnt) species modeling framework. The modeling results suggest that based on both training and testing AUC (area under the receiver operating characteristic (ROC) curve) values higher than 0.82, the VI-only variable model, the environmental-only variable model and the combined environmental and VI variables model, all yielded good simulation results. The spatial distributions of noxious weed species mapped by the VI-only variable model and the combined environmental and VI variable model were more concentrated, while the VI-only variable model yielded more scattered results. This analysis also explains why noxious weed species are mainly distributed in the low-elevation flat riverine zone in the study area. The model combining Sentinel-2 imagery-related VIs, environmental variables and in situ sample data proposed in this study can successfully map the occurrence and spatial distributions of noxious weed species. The method and results of this research can be used to help monitor noxious weed species invasions and better manage grassland ecosystems.