基于多源遥感数据的草地生物量反演

Retrieval of Grassland Biomass Based on Multi-source Remote Sensing Data

面向青海省黄南藏族自治州河南蒙古族自治县(河南县)开展实验。基于2018年5—8月的Landsat8、GF-1、GF-4遥感数据建立时空连续性的NDVI时间序列;将NDVI时间序列与人工采样草地生物量进行时空匹配,用于构建NDVI时间序列与草地生物量的经验模型;基于地理学第一定律,将基于点拟合的经验模型推广到面,实现在稀疏地面观测样本条件下的大区域、高精度草原生物量反演。提出的方法高效地构建了时空匹配的星基NDVI与时空分布稀疏的人工草地样方生物量的时间序列对,解决了当前遥感反演方法过度依赖稀疏的地面观测采样数据的问题,提升了反演的成功率和模型的泛化能力。基于提出的方法在试验区展开实验,决定系数(R^(2))平均为0.75,优于植被指数法(R^(2)取值0.3~0.5);均方根误差(RMSE)为1.10 kg/km^(2),优于植被指数法(RMSE取值10~60 kg/km^(2))。

The study is carried out in Henan Mongolian Autonomous County(Henan County),Huangnan Tibetan Autonomous Prefecture,Qinghai Province.First,a time series of the Normalized Difference Vegetation Index(NDVI)is constructed from the Landsat8,GaoFen-1(GF-1)and GaoFen-4(GF-4)remote sensing data from May of 2018 to August of 2018;Second,a spatiotemporal matching process is used for a good consistency of the NDVI time series and manually sampled grassland biomass,which will be helpful for constructing an accurate empirical model for detailing the relationship between the NDVI time series and the manually sampled grassland biomass;lastly,the point-based empirical models are generalized to the whole studying area based on the First Law of Geography.Overall,the study achieves large-scale and high-precision estimation of grassland biomass with the sparse and insufficient ground observation samples.The proposed method introduces a spatiotemporal model to spatiotemporally match satellite-based NDVI time series and the spatiotemporally sparse grassland biomass sampled,which solves the problem of excessively relying on sparse ground observation data in the course of the retrieval of grassland biomass,improves the efficiency and accuracy of the retrieval of grassland biomass,and promotes the generalization ability of the constructed empirical model.Based on this study,an experiment is carried out in the studying area,and the results show a rational precision.The average coefficient of determination(R^(2))of the proposed method is 0.75,which is higher than that of vegetation index methods(R^(2)ranging from 0.3 to 0.5).The Root Mean Square Error(RMSE)of the proposed method is 1.10 kg/km^(2),which is lower than that of vegetation index methods(RMSE ranging from 10 to 60 kg/km^(2)).