Separating Shrub Cover From Green Vegetation in Grasslands Using Hyperspectral Vegetation Indices

Abstract

Shrub cover is a key parameter for monitoring woody plant encroachment (WPE) in the background of global grassland degradation. At the critical early encroachment stage, many shrub species are small, and it is impossible to be detected through height parameters. Hyperspectral vegetation indices (HVIs) are valuable tools to estimate vegetation cover, while the effect of shrubs of different cover levels (e.g., low: <35%, moderate: 36–55%, high: >56%) on the spectrum of mixed grassland is still unclear. In this study, narrowband HVIs were created for shrub cover estimation using field hyperspectral data collected in Cypress Hills Interprovincial Park, SK. Published HVIs and newly developed HVIs were examined for the relationship between green shrub cover and total green vegetation (GV) cover. Averaging normalized difference vegetation index (NDVI) was also used to minimize in the variation of shrub cover and GV in each quadrat. Results showed that HVIs SR [600, 800] and NDVI [705, 750] had the highest R2 with green shrub and GV cover (R2 = 0.34 and R2 = 0.30, p-value < 0.05). Blue bands (449 and 489 nm) and SWIR bands (1619 and 1739 nm) combination exhibited the highest accuracy for shrub and GV covers estimation (R2 = 0.60 and 0.40, p-value < 0.05). Averaging NDVI positively correlated with green shrub cover below 30% or above 70%. Three separability metrics (JM, TD, and M) results also showed the red and NIR region may promise to distinguish different shrub cover levels.