Hyperspectral database prediction of ecological characteristics for grass species of alpine grasslands

Abstract

Alpine grasslands are being degraded because of human activities and associated global climate change. Mapping the spatial distributions and ecological characteristics of grass species is essential for scientific management of grasslands. However, traditional field-survey methods are costly or even impossible owing to poor accessibility. Hyperspectral remote sensing provides solutions for the purpose. This study was conducted in Shenzha County of the Qiangtang Plateau, north-western Qinghai–Tibet Plateau, to examine the potential of using hyperspectral data for identifying the grass species and predicting their ecological characteristics in the alpine grasslands dominated by Stipa purpurea with co-existing species Leontopodium nanum and Oxytropis microphylla. Hyperspectral data were collected in 106 sample quadrats and the ecological characteristics of each quadrat (number and height of plants, vegetation cover, etc.) were measured. The results of spectral data analysis and regression modelling showed the following. (i) The near- and middle-infrared region was more appropriate than the visible region for discriminating the grass species. (ii) The enhanced spectral variables had much higher correlations with the ecological characteristics than the original bands. (iii) Most of the 23 derived enhanced spectral variables were significantly correlated with the number and height of the dominant species plants within the quadrats. (iv) The vegetation cover could be accurately predicted by using the models based on the enhanced spectral variables of the field-collected hyperspectral data with the relative RMSE values <28%. (v) The ecological characteristics of the dominant species could be more accurately estimated than of co-existing species. Overall, this study suggests that the hyperspectral database method provided great potential to predict the ecological characteristics of grass species in alpine grasslands.