Estimation of Stipa krylovii biomass based on hyperspectral remote sensing

Hyperspectral remote sensing technology can provide an estimate of the the above-ground biomass (AGB) of grasslands efficiently, non-destructively, and rapidly. This is of great significance for the dynamic monitoring and efficient management of grassland animal husbandry and the balance between supply and demand of forage and livestock. In this study, we obtained spectral reflectance and AGB data from Xianghuang Banner pasture in Xilin Gol League, Inner Mongolia, to explore the most appropriate spectral variables and vegetation index for estimating the AGB of the Stipa krylovii community during the growing season. Among other variables, the original hyperspectral data, first-order differential (trilateral optical parameters), vegetation index, green peak and red valley values were used. A model for estimation of AGB was established using regression models. The accuracy of different models was compared using R-square (R²), root mean square error (RMSE), and mean relative error (MRE) as loss functions. The results show that the Stipa community exhibited dynamic changes throughout the growth of forage, with the AGB increasing gradually in July-August and decreasing significantly in August-September. Differential processing of the original hyperspectrum could improve the correlation between the sensitive band and AGB. Among the hyperspectral variables, the linear model constructed using the red-edge slope (Dr) had the best accuracy (R² = 0.94, RMSE = 1.97 g·m⁻², and MRE = 1.97 g·m⁻²). For the vegetation indexes, the polynomial model constructed using soil-adjusted vegetation index had the best accuracy (R² = 0.92, RMSE = 1.15 g·m⁻², and MRE = 1.39 g·m⁻²). This study provides a scientific basis for estimation of the AGB of Stipa communities with different hyperspectral variables and offers a method of and technical support for rapid and accurate remote sensing monitoring of natural grassland in pastoral areas.