Modeling alpine grassland cover based on MODIS data and support vector machine regression in the headwater region of the Huanghe River, China

Monitoring changes in grassland cover is essential in assessment of grassland health as well as the effects of anthropogenic interventions and global climate change on grassland ecosystems. Remote sensing is an effective approach for providing rapid and dynamic monitoring of vegetation cover over large grassland areas. In this study, four types of remote sensing retrieval models (i.e., pixel dichotomy models, univariate vegetation index (VI) regression models, multivariate regression models, and a support vector machine (SVM) model) are built to derive grassland cover based on moderate resolution imaging spectroradiometer (MODIS) data and the measured grassland cover data collected by unmanned aerial vehicle during the grassland peak growing season from 2014 to 2016. The optimal model is then used to map the spatial distribution of grassland cover and its dynamic change in the headwater region of the Huanghe River (Yellow River) (HRHR) of the northeastern Tibetan Plateau over the 16 years period (2001 to 2016). The results show that (1) the pixel dichotomy models based on MODIS VI data are inappropriate for estimating grassland cover in the HRHR when their endmembers (VIsoil and VIveg) are determined based only on the MODIS data; (2) the multivariate regression models present better performance than the univariate VI (normalized difference vegetation index (NDVI) or enhanced vegetation index (EVI)) models; (3) MODIS NDVI outperforms MODIS EVI for modeling grassland cover in the study area; (4) the SVM model based on nine factors is the optimal model (R2: 0. 75 and RMSE: 6.85%) for monitoring alpine grassland cover in the study area; and (5) majority of the grassland area (59.9%) of the HRHR showed increase in yearly maximum grassland cover from 2001 to 2016, while the average yearly maximum grassland cover for the 16 years exhibited a generally increasing trend from west to east and from north to south. This study provides a more suitable remote sensing inversion model to greatly improve the accuracy of modeling alpine grassland cover in the HRHR, and to better assess grassland health status and the impacts of warming climate to grasslands in regions of remote and harsh environments. •UAV is a valuable tool to improve monitoring/modeling accuracy of grassland cover.•The SVM model outperformed the linear and nonlinear regression models.•NDVI explained more variability than EVI in grassland cover inversion models.•Pixel dichotomy model was the least accurate