Abstract: In order to explore the information loss in UAV （Unmanned aerial vehicle） monitoring of grassland biomass， this study compared the ground hyperspectral and on-site measured data with the multispectral images of the desert grassland community of Seriphidium transiliense， and used the vegetation index combined with linear regression method to construct the inversion model of aboveground biomass of the plant community at different phenological periods. The error and rule of UAV acquisition of aboveground biomass were discussed and the modified model was established. The result shows that：（1）In April， June and September， the best UAV multi-spectral models were NDVI （Normalized difference vegetation index）， RVI （Ration vegetation index） and NDVI， with accuracy of 65.61%， 48.26% and 61.59%， respectively. Among all the ground hyperspectral data， NDVI is the most optimal， and the accuracy was 71.77%， 53.63% and 67.58%， respectively. （2）The ground hyperspectral retrieval ability was better than the UAV multispectral retrieval ability， the loss rates of multispectral data from UAV at each phenological stage were NDVI > RVI > DVI （Difference vegetation index）， and the loss rates of optimal model were 8.58%， 9.89% and 8.87%， respectively. （3）The UAV multi-spectral accuracy of all models was improved by 0.02%~3.74%， The modified NDVI mode in April had the highest accuracy of 66.36 %（y=403.431x+17.5936）. In summary， by screening the best inversion models， analyzing the information loss of different platforms， and modifying the UAV multi-spectral of the UAV can give full play to the advantages of different platforms and improve the estimation accuracy. It also proves the feasibility of using high-resolution remote sensing data to modify low-spatial-resolution data.

Key words: Unmanned aerial vehicle, Ground hyperspectral, Aboveground, Inversion, Information loss, Modification

摘要： 为探究无人机监测草地生物量的信息损失，本文以伊犁绢蒿（Seriphidium transiliense）荒漠草地多光谱影像对照地面高光谱及实测数据，以植被指数结合线性回归法构建不同物候期群落地上生物量反演模型，探讨无人机获取地上生物量的误差与规律并建立修正模型。结果表明：（1）4月、6月、9月无人机多光谱最佳模型分别为归一化植被指数（Normalized difference vegetation index，NDVI）、比值植被指数（Ration vegetation index，RVI）和NDVI，精度分别为65.61%，48.26%和61.59%；地面高光谱均以NDVI最优，精度分别为71.77%，53.63%和67.58%。（2）地面高光谱反演能力优于无人机多光谱，各物候期无人机多光谱损失率为NDVI>RVI>差值植被指数（Difference vegetation index，DVI），最佳模型损失率分别为8.58%，9.89%和8.87%。（3）无人机多光谱所有模型精度提高0.02%~3.74%，修正后4月NDVI精度最高为66.36%（y=403.431x+17.5936）。综上，通过筛选最佳反演模型，分析不同平台的信息损失，并对无人机多光谱修正能发挥不同平台优势，提高估算精度，证明了高分辨率遥感修正低空间分辨率数据的可行性。

关键词: 无人机, 地面高光谱, 地上生物量, 反演, 信息损失, 修正