Comparing vegetation indices from Sentinel-2 and Landsat 8 under different vegetation gradients based on a controlled grazing experiment

Abstract

Grasslands contribute considerably to the global carbon cycle and livestock production. However, many of the world grasslands suffer from degradation caused mainly by overgrazing. Remote sensing methods are effective tools for monitoring and estimating grassland vegetation parameters. In this study, we compared the performance of vegetation indices (VIs) obtained from two different sensors to estimate grassland vegetation parameters under different vegetation and soil conditions based on typical grassland biomass gradients formed by long-term controlled grazing experiments. Sentinel-2 and Landsat 8 were selected as data sources to estimate two vegetation parameters, fresh aboveground biomass (AGB) and leaf area index (LAI). Field-measured fresh AGB and LAI data were collected from experimental grasslands with different grazing intensities (GI) in Hulunber, Inner Mongolia, China in 2019. Univariate linear mixed models were established between VIs and field measurements, and grazing intensities were considered as random factors. The results confirmed that vegetation parameters (AGB and LAI) and VIs decreased with increasing GI; however, the decreasing trend was insignificant when the GI exceeded 0.69 Au/ha. VIs derived from Sentinel-2 and Landsat 8 estimated fresh AGB and LAI at 80% accuracy. Sentinel-2 derived VIs yielded higher predictive accuracy than Landsat 8 for both fresh AGB and LAI. Comparing with the other VI inversion models, the normalised difference phenology index derived from Sentinel-2 images estimated the vegetation parameters the most effectively and accurately, with a coefficient of determination (R2) of 0.625 and relative root mean square error (RMSE%) of 18.105% for fresh AGB estimation and R2 of 0.559 and RMSE% of 14.953% for LAI estimation.