基于Landsat 8 OLI和MODIS数据的高寒草地盖度升尺度效应研究——以夏河县桑科草原试验区为例

Scaling-up methodology for alpine grassland coverage monitoring based on Landsat 8 OLI and MODIS remote sensing data:A case study in XiaheSangke grassland

基于Landsat 8OLI和MODIS卫星遥感资料,结合2013-2014年甘南州夏河县桑科草原试验区野外实测盖度数据,对草地盖度敏感的OLI波段及其组合指数进行了筛选,构建并确立了试验区基于MODIS植被指数的草地盖度反演模型,探讨了不同空间升尺度方式对草地盖度的空间效应。结果表明,1)OLI比值植被指数r(Band7/Band5)对草地盖度的响应最敏感,基于该指数的草地盖度最优反演模型为yoli=-270.064xoli+115.987(R2=0.833,P<0.001);2)基于MODIS MEVI和Landsat 8OLI比值植被指数r(30m)反演的盖度重采样数据(250m)的对数模型为最优草地盖度评估模型(R^2=0.795,P<0.001),其决定系数较MODIS MEVI与基于农业多光谱照相机(agricultural digital camera,ADC)的盖度数据建立的回归模型高(R^2=0.706),绝对误差和相对误差较低;3)基于OLI和MODIS植被指数的3种草地盖度回归模型(方法1~3)精度均优于直接使用MODIS植被指数建立的回归模型(方法4);方法3利用OLI比值指数r反演盖度(30m),将其升尺度至250m,再反演盖度构建的对数模型的精度最高(R^2=0.795),其次依次为方法2构建的模型(R2=0.760)、方法1构建的模型(R^2=0.730)和方法4构建的模型(R2=0.706)。

This research used remote sensing data of MODIS and Landsat 8OLI,combined with ground observations during 2013 and 2014in XiaheSangke grassland,Gansu Province.Both individual bands and combinations of bands of Landsat 8OLI were tested,with a view to selecting band combinations sensitive to grassland coverage.Then,grassland coverage inversion models were established based on MODIS vegetation index data.At the same time,the spatial scale effect was analyzed with a 30 mresolution and up-scaled to 250 mresolution for spectral reflectance,vegetation index,and estimated grassland coverage.It was found that：1）the ratio of Band7/Band5 of the OLI data was the most sensitive combination for detecting grassland coverage,and the best grassland coverage inversion model was the linear function：yoli=-270.064xoli＋115.987,R2=0.833,P〈0.001;2）The best grassland coverage inversion model was the logarithmic model（y=64.160ln（xMEVI）＋136.927,R2=0.795,P〈0.001）,which was established by using MODIS MEVI and up-scaling ratio index of OLI.The coefficient of fit was higher than for the models based on MODIS MEVI and Agricultural Digital Camera pictures（R2=0.706）,and its average absolute error and average relative error were lower.3）The accuracy of the logarithmic model（R2=0.795）based on MODIS MEVI and up-scaling grassland coverage using a ratio index（Band7/Band5）of OLI was higher than other models.