基于Sentinel-2的绿洲-荒漠过渡带植被地上生物量估算

Estimation of aboveground vegetation biomass in oasis-desert transition zone based on Sentinel-2

开展绿洲-荒漠过渡带植被地上生物量监测是植被生长状况评价与荒漠化监测的重要手段。文中利用Sentinel-2影像数据构建了地上生物量估算模型,比较了统计模型和两种机器学习算法模型的性能,并对渭干河-库车河绿洲的绿洲-荒漠过渡带的植被地上生物量进行了估算。结果显示,在统计模型中,红边三角植被指数(RTVI)与地上生物量的非线性模型拟合效果最好,且相关最显著。在机器学习算法中,随机森林模型优于支持向量机回归模型。通过验证发现,RTVI非线性估测模型和随机森林模型具有较好的外推能力。在绿洲-荒漠过渡带植被地上生物量的反演中,随机森林模型表现出较高的精度,验证集R^(2)为0.65,RMSE和MAE分别为255.08g·m^(-2)和192.93g·m^(-2)。相较其他模型,随机森林模型可以在小样本情况下更精确,对科学监测绿洲-荒漠过渡带植被地上生物量和维护绿洲的稳定发展提供依据。

Monitoring the above-ground biomass of vegetation in the oasis-desert transition zone is an important means to evaluate vegetation growth and monitor the desertification.In this study,Sentinel-2 image data was used to construct an above-ground biomass estimation model.The performance of the statistical model and two machine learning algorithm models were compared,and the above-ground biomass of vegetation in the oasis-desert transition zone of the Weigan-Kuqa River oasis was estimated.The results showed that,among the statistical models,RTVI has the best fitting effect and the most significant correlation with the nonlinear model of above-ground biomass.In machine learning algorithms,the random forest model is superior to the support vector machine regression model.The results show that the RTVI nonlinear estimation model and the random forest model have better extrapolation abilities.In the inversion of above-ground biomass of the oasis-desert transition zone,the random forest model achieves higher accuracy,the verification set R2 is O.65,RMSE and MAE are 255.08g·m^(-2)and 192.93g·m^(-^(2)),respectively.Compared with other models,the random forest model can be more accurate in the case of small samples,and provide a basis for scientific monitoring of the aboveground biomass of vegetation in the oasis-desert transition zone and maintaining the stable development of the oasis.