Potential of hyperspectral data and machine learning algorithms to estimate the forage carbon-nitrogen ratio in an alpine grassland ecosystem of the Tibetan Plateau

Abstract The carbon-nitrogen (C:N) ratio plays a crucial role in regulation of the nutrient utilization efficiency and growth rate of plants. However, challenges are faced when hyperspectral data are directly used to estimate the C:N ratio due to the lack of corresponding sensitive feature bands. This study aims to explore the feasibility of using important bands (IBs) determined for the C:N ratio and known absorption bands (KBs) of protein, chlorophyll, N, and carbon-containing compounds from hyperspectral measurements to estimate the forage C:N ratio. Random forest (RF) and support vector machine (SVM) algorithms are employed to establish a model for the estimation of the forage C:N ratio. The results show that the KBs exhibit good performance in estimating the forage C:N ratio (V-R2 of 0.70–0.80, with a mean of 0.77), and the IBs derived from the red and red-edge regions significantly contribute to the forage C:N ratio estimation, with V-R2 of 0.77–0.80. This study also demonstrates that the models based on combined bands (CBs) (the combination of KBs and IBs) slightly improve the accuracy of the forage C:N ratio estimation. Moreover, further optimization of the CBs produces satisfactory estimation of the forage C:N ratio (V-R2 = 0.82, V-RMSE = 5.53), explaining 85–92% of the variation in the forage C:N ratio at the different growth stages (May to November). Overall, the direct estimation of the forage C:N ratio in alpine grassland using hyperspectral feature bands is promising.