Mapping of Native Plant Species and Noxious Weeds in Typical Area of the Three-River Headwaters Region by Using Worldview-2 Imagery

Mapping of native plant species and noxious weeds is critical to monitor grassland degradation in the Three-River Headwaters Region. The grass species in the study area were divided into native plant species and noxious weeds based on the applicability of grazing. Field data of the two kinds of grass species were collected and divided into ten coverage grades with an interval of 10%. The eight original bands were used to derive 37 features by Random Forest (RF) algorithm, including first derivative (FD), vegetation indexes (VI), biochemical indexes (BI), hat transform (KT) and gray level co-occurrence matrix (GLCM). The importance of each feature was calculated and 17 of them were selected by RF, reflecting their superiority in identifying native plant species and noxious weeds. The random forest algorithm was also used in classification of the native plant species with an overall accuracy (OA) of 43.2% (8 bands), 45.9% (37 features) and 51.3% (17 selected features) and an 10% grade expansion accuracy (GEA) of 59.4%, 64.8% and 70.2%, respectively. The noxious weeds with a higher overall accuracy (OA) of 62.1% (8 bands), 64.8% (37 features) and 67.5% (17 selected features) and an 10% grade expansion accuracy (GEA) of 86.4%, 83.7% and 89.1%, respectively. Therefore, the classification of native plant species and noxious weeds coverage grades with the interval of 10% demonstrated the potential of the WorldView-2 data for mapping native plant species and noxious weeds in the typical area of Three-River Headwaters Region.