不同机器学习算法在草原草地生物量估算上的适应性研究

准确估算草地地上生物量(Aboveground biomass，AGB)对于科学调整草畜关系、保护生态环境和实现草地资源的可持续发展具有重要意义。本文以锡林郭勒盟不同草地类型为研究对象，基于遥感数据、气象数据和数字高程模型数据，利用支持向量机(Support vector machines,SVM)、BP神经网络(BP neural networks,BP)和随机森林(Random forest,RF)三种机器学习算法建立AGB估算模型，评估三种机器学习算法模型估算AGB的潜力。精度验证结果表明，在研究区内不区分草地类型整体建立估算模型时RF算法的回归精度最高(R=0.88，RMSE=0.10，MSE=0.01，MAE=0.07)。SVM算法建立的模型在草甸草原和荒漠草原回归精度较高，而RF算法回归能力在典型草原具有相对优势。不同特征变量对估算AGB的贡献分析结果表明，植被覆盖度(Fractional vegetation cover,FVC)、归一化植被指数(Normalized difference vegetation Index,NDVI)、增强植被指数(Enhanced vegetation index,EVI)和降水量(Precipitation,PRCP)四个变量对AGB估算结果的影响较大。本文研究结果为干旱/半干旱区草地地上生物量估算精度的提高和方法的选择提供科学建议。

Adaptation of Different Machine Learning Algorithms for Grassland Biomass Estimation

Accurate estimating of grassland aboveground biomass (AGB) is important for scientific adjustment of the grass-livestock relationship,ecological environment protection and sustainable development of grassland resources. Based on remote sensing,meteorological and digital elevation model data,we paper used three machine learning algorithms,including the Support vector machines (SVM),BP Neural Networks (BP) and Random forest (RF) to establish a grassland AGB estimation model in Xilin Gol League and evaluate the estimation potential of the three models. The accuracy validation results showed that the RF algorithm had the highest regression accuracy in the study area (R=0.88,RMSE=0.10,MSE=0.01,MAE=0.07). The model established by the SVM algorithm had higher regression accuracy in the meadow steppe and desert steppe,while the regression ability of the RF algorithm had a relative advantage in the typical steppe. The analyses results of the contribution of different characteristic variables to the estimated AGB showed that the fractional vegetation cover (FVC),Normalized Difference Vegetation Index (NDVI),Enhanced vegetation index (EVI) and Precipitation (PRCP) greatly influence the AGB. The results of this study provide scientific suggestions to improve the accuracy of above-ground biomass estimation and selecting methods in arid/semi-arid grassland areas.