青藏高原草地地上生物量估算

Estimation of grassland aboveground biomass on the Qinghai⁃Tibet Plateau

及时准确评估草地产草量对草地资源的科学管理和可持续发展具有重要意义。青藏高原自然环境特殊,气候差异显著,地形复杂,仅依靠遥感信息准确监测草地地上生物量(Aboveground Biomass,AGB)变化有较大限制。基于青藏高原草地AGB野外实测数据与Landsat遥感影像,探索了植被指数表征草地AGB信息的有效性,评估了气象和地形信息对准确估算草地AGB的影响,综合利用气象、地形和遥感信息,在新一代地球科学数据和分析应用平台(Google Earth Engine)上构建了梯度增强回归树草地AGB估算模型,绘制了青藏高原多年草地AGB空间分布图。结果表明:(1)基于单因素遥感因子的线性回归模型仅能解释8%—40%的草地AGB变化情况,其中绿色归一化植被指数(Green Normalized Difference Vegetation Index,GNDVI)对草地AGB解释能力较强(40%)。(2)基于遥感因子构建的梯度增强回归树模型测试集R~2为0.57。分别添加气象、地形信息,模型对草地AGB的估测准确性有所提升,测试R~2为0.62和0.63。(3)基于气象、地形和遥感因子的多因素估测模型能够提高草地AGB估测精度,经递归特征消除法优选后,基于13个特征变量的梯度增强回归树模型拟合效果最好(训练数据集R~2=0.79,RMSE=43.42 g/m^(2),P<0.01;测试数据集R~2=0.66,RMSE=53.64 g/m^(2),P<0.01),可以解释66%草地AGB变化情况。(4)2010年青藏高原平均AGB为94.58 g/m^(2),2015年93.63 g/m^(2),2020年100.78 g/m^(2)。青藏高原西北部草地AGB较低,东南部草地AGB较高,整体呈现自西北向东南逐渐增加的分布格局。研究结果为准确估算青藏高原草地产草量和碳储量等研究提供重要参考。

Timely and accurate assessment of grass production is of great significance to the scientific management and sustainable development of grassland resources.The Qinghai⁃Tibet Plateau has specially natural environment with significant climate differences and complex topography.There are major limitations in relying only on remote sensing information to accurately monitor changes in aboveground biomass(AGB)in grasslands.In this study,based on the grassland AGB field measurement data and Landsat remote sensing images on the Qinghai⁃Tibet Plateau,the validity of vegetation indices in characterizing grassland AGB information was explored,and the influence of meteorological information and topographic information on accurate estimation of grassland AGB on the Qinghai⁃Tibet Plateau was assessed.Based on the meteorological,topographic,and remote sensing image data,gradient boosting regression tree models for estimating grassland AGB were constructed on a new generation of earth science data and analysis application platform(Google Earth Engine),and the spatial distribution of multi-year grassland AGB on the Qinghai⁃Tibet Plateau was mapped.The results show that:(1)The linear regression model based on single remote sensing factor could only explain 8%—40%of the changes in grassland AGB,among which the green normalized difference vegetation index(GNDVI)was more capable of explaining grassland AGB(40%).(2)The gradient boosting regression tree model constructed with remote sensing factors had an R2 of 0.57 for the test dataset.With the addition of meteorological information and topographic information,respectively,the accuracy of the model for estimating the grassland AGB was improved,with the test dataset R2 of 0.62 and 0.63.(3)The multi-factor estimation model coupled with meteorological factors,topographic factors,and remote sensing factors could effectively improve the accuracy of grassland AGB estimation.After optimization by the recursive feature elimination method,the gradient boosting regression tree model based on 13 feature variables had the best fitting effect(training dataset:R2=0.79,RMSE=43.42 g/m^(2),P<0.01;testing dataset:R2=0.66,RMSE=53.64 g/m^(2),P<0.01),which could explain 66%of grassland AGB variation.(4)The average grassland AGB on the Qinghai⁃Tibet Plateau was 94.58 g/m^(2)in 2010,93.63 g/m^(2)in 2015,and 100.78 g/m^(2)in 2020.The grassland AGB in the northwestern part of the Qinghai⁃Tibet Plateau was lower,and the grassland AGB in the southeastern part was higher.The overall distribution pattern showed a gradual increase from northwest to southeast.The research results provide important references for studies such as the accurate estimation of grassland yield and carbon storage on the Qinghai⁃Tibet Plateau.