Abstract: Estimating aboveground biomass (AGB) in grassland helps determine the theoretical stocking capacity. A functional relationship between remote sensing vegetation index and climate variables and AGB observations based on artificial neural networks (ANNs) was constructed,which was then used to estimate the AGB of grassland on the Tibetan Plateau. The influence of each climatic factor on the change of AGB was analyzed based on ridge regression. Results showed that the R² between the simulated and measured values was 0.92 (0.88),and the RMSE was 18.48 (23.62) g·m^-2 in the training (test) period. AGB of the grassland types from the steppe to the meadow to the grass-forb community decreased in turn. AGB increased first and then decreased with the increase in altitude. AGB was the highest in the areas with elevations between 3 400 and 3 800 m. It was found that the areas with under-or over-estimated values from ANNs accounted for 1% and 10% of the total area,respectively,compared with the five mechanism models. It mainly resulted from the deviations between the mean value of the training data and that of the simulated values from the mechanism models in the corresponding regions. The influencing factors,sorted by importance from high to low,are atmospheric CO₂ concentration,saturated water vapor pressure deficit,previous-year precipitation,mean wind speed,and mean air temperature,respectively.

Key words: Grassland, Aboveground biomass, Artificial neural networks

摘要： 草地地上生物量(Aboveground biomass，AGB)的估算有助于理论载畜量的确定。基于人工神经网络(Artificial neural networks，ANNs)，利用遥感植被指数和气候变量与AGB观测值构建函数关系，进行了青藏高原草地AGB的模拟，并基于岭回归分析了每个气候因子对AGB变化的影响强弱。结果表明，在训练期(测试期)，ANNs的模拟值与实测值之间的R²为0.92(0.88)，RMSE为18.48(23.62)g·m^-2。草地类型从草丛到草甸再到草原，AGB依次减少。AGB随海拔的升高先增加后减少。海拔3 400～3 800 m的区域AGB最高。ANNs与5个机理模型对比，发现ANNs模拟值偏低和偏高的面积分别占总面积的1%和10%，主要原因是训练资料的均值与相应地区中机理模型模拟值的偏差所致。影响因子按重要性从高到低的排序分别为大气CO₂浓度、饱和水汽压差、前一年降雨量、平均风速和平均气温。

关键词: 草地, 地上生物量, 人工神经网络