基于随机森林的高寒草地地上生物量高光谱估算

草地地上生物量（Aboveground biomass，AGB）是衡量草地生产力的关键因素，准确测定草地AGB具有重要意义。高光谱因具有时效性强、不破坏草地等特点被广泛用于草地生理生态指标的测定。本研究提取和计算了海北试验站高寒草地冠层的原始光谱（Original spectrum，OR）反射率、一阶微分光谱（First derivative spectrum，FD）反射率、光谱位置面积参数（Spectral parameters of spectral position and area，PA）和植被指数（Vegetation indices，VI）4种不同类型的特征变量，使用连续投影算法（Successive projections algorithm，SPA）和递归特征消除算法（Recursive feature elimination，RFE）进行特征选择，采用随机森林算法（Random forest，RF）构建草地AGB估测模型。结果表明：在由4种特征变量分别构建的草地AGB估测模型中，基于VI的RF模型精度最高（测试集R²=0.70，RMSE=557.87 kg·ha^-1），实测AGB与估测AGB的线性R²达到0.72；不同类型特征变量组合构建的草地AGB估测模型中，PA+VI组合的RF模型精度最高（R²=0.71，RMSE=548.97 kg·ha^-1），实测AGB和估测AGB的线性R²达到0.73。

Hyperspectral Estimation of Aboveground Biomass of Alpine Grassland based on Random Forest Algorithm

Aboveground biomass (AGB) is a key indicator of grassland productivity and it is important to measure grassland AGB accurately in grassland resource survey. Hyperspectrum is an effective method to measure the physiological and ecological indexes of grassland without physical damage to the grassland. In this study, the original spectrum (OR), the first derivative spectrum (FD), spectral parameters of spectral position and area (PA) and the vegetation index (VI) of alpine grassland canopy were calculated near Haibei National Field Research Station for Alpine Grassland Ecosystem (Haibei Station). Based on the above variables, feature selection was performed with successive projections algorithm (SPA) and recursive feature elimination algorithm (RFE), and the model was constructed with random forest algorithm (RF). The results showed that among the grassland AGB estimation models constructed by different variables, the accuracy of RF model based on VI was the highest (R²=0.70, RMSE=557.87 kg·ha^-1), and R² of measured AGB and predicted AGB was 0.72. Among the grassland AGB estimation models constructed by different combination of variables, the accuracy of RF model of PA + VI combination was the highest (R²=0.71, RMSE=548.97 kg·ha^-1), and R² of measured AGB and predicted AGB was 0.73.