基于STARFM的草地地上生物量遥感估测研究——以甘肃省夏河县桑科草原为例

遥感数据具有实时、动态、大范围等特点，在草地资源监测与管理研究中获得了广泛应用。然而，单一的遥感植被指数无法同时满足草地地上生物量观测中时空分辨率的需求。因此，本研究基于时间序列Landsat NDVI和MODIS NDVI数据，结合时空融合算法（spatial and temporal adaptive reflectance fusion model， STARFM），生成了2000-2016年高时空分辨率的植被指数数据集（NDVI_STARFM，时间分辨率为16 d，空间分辨率为30 m），并基于2013-2016年地面实测草地地上生物量数据，构建了夏河县桑科草原高寒草地地上生物量遥感反演模型，分析了2000-2016年研究区草地地上生物量生长状况和变化趋势。结果表明：1）基于NDVI_STARFM的最优估测模型为乘幂模型，其R²为0.58，均方根误差（root mean square error， RMSE）为795.62 kg·hm^-2，模型的表现能力次于Landsat NDVI最优估测模型（R²=0.76，RMSE=634.83 kg·hm^-2），而优于MODIS NDVI最优估测模型（R²=0.24，RMSE=937.79 kg·hm^-2）；2）基于NDVI_STARFM最优估测模型对各样区草地地上生物量总产的估测精度优于MODIS NDVI而次于Landsat NDVI，总体精度达84.05%；3）2000-2016年来，夏河县研究区草地地上生物量总体呈现增加趋势，其中90%左右的区域年增量大于30 kg·hm^-2，草地地上生物量呈现减少趋势的区域仅占2.30%。

Estimating grassland above ground biomass based on the STARFM algorithm and remote sensing data——A case study in the Sangke grassland in Xiahe County,Gansu Province

Characteristics of remote sensing data include that they are real-time， dynamic and large-scale， so such data have been widely used in grassland resource monitoring and management research. However， a single remote sensing vegetation index can not meet the needs of temporal and spatial resolution in grassland above ground biomass （AGB） monitoring. Therefore， this study generated a high spatial and temporal resolution vegetation index data set based on a time series of Landsat NDVI and MODIS NDVI data， combined with the spatial and temporal adaptive reflectance fusion model （STARFM）. The data set so generated （NDVI_STARFM） had a temporal resolution 16 d and a spatial resolution 30 m. The optimum grassland above ground biomass inversion model was constructed based on measured grassland above ground biomass and NDVI_STARFM during the grass growth seasons of 2013-2016_. Finally， the spatiotemporal dynamic variation trends of grassland above ground biomass in the study area were analyzed for the period from 2000-2016. It was found that： 1） the optimal estimation model based on NDVI_STARFM was a power model， with an R² of 0.58 and an RMSE 795.62 kg·ha^-1. The performance of this model was lower than that of the Landsat NDVI optimal estimation model （R² =0.76， RMSE=634.83 kg·ha^-1）， but better than that of the MODIS NDVI optimal estimation model （R² =0.24， RMSE=937.79 kg·ha^-1）. 2） The overall accuracy of the optimal estimation model was 84.05%， it was higher than that of MODIS NDVI but lower than that of Landsat NDVI. 3） The grassland above ground biomass showed an increasing trend in most areas from 2000-2016. About 90% of the study area showed an increasing trend with annual increment more than 30 kg·ha^-1， while only 2.3% of the study area showed a decreasing trend.