森林生物量遥感降尺度研究

森林生物量是评价全球碳氧平衡、气候变化的重要指标。目前已有基于星载激光雷达数据的全球森林生物量产品,但空间分辨率较低,不能很好地满足小区域森林调查和动态监测的需要。针对这一现状,以美国马里兰州两个森林分布状况不同的区域为研究区,基于CMS(Carbon Monitoring System)30 m分辨率和GEOCARBON 1 km分辨率森林地上生物量产品以及TM等数据源,通过升尺度模拟低分辨率生物量数据和直接使用低分辨率产品两种方式,分别尝试建立了多光谱地表参数和低分辨率森林地上生物量之间的统计关系,以此作为降尺度模型实现了森林地上生物量空间分辨率从1 km到30 m的转换,并对降尺度结果进行精度评价和误差分析。结果表明:模拟数据降尺度后的30 m分辨率森林地上生物量空间分布和CMS森林地上生物量分布状况大致相同,RMSE=59.2—65.5 Mg/hm~2,相关系数约为0.7;其降尺度结果优于GEOCARBON产品直接降尺度结果RMSE=75.3—79.9 Mg/hm~2;相较于线性模型,非线性模型能更好地呈现森林地上生物量和地表参数间的关系;总体上,降尺度生物量呈现高值区低估,低值区高估的现象。

Spatial downscaling of forest biomass based on remote sensing

Forest biomass is an important indicator that can be used to evaluate global carbon-oxygen balance and climate change. At present, the spatial resolution of existing global forest biomass products based on spaceborne large footprint LiDAR data is too coarse to meet the needs of local forest investigation and dynamic monitoring. Therefore, it is necessary to determine a downscaling method to produce high spatial resolution forest biomass products from coarse resolution products. Two areas with different forest distribution patterns in Maryland, USA, were selected in this study to establish statistical relationships between low resolution multispectral data upscaled from TM data and forest aboveground biomass (AGB), which were upscaled from CMS (Carbon Monitoring System) forest AGB products or directly from GEOCARBON AGB products. The statistical relationships were then used as a downscaling model to downscale the forest AGB products from a spatial resolution of 1 km to 30 m. Results showed that the spatial distribution of downscaled 30 m-resolution biomass from simulated forest AGB was roughly the same as that of the CMS biomass. The RMSE was between 59.2 Mg/hm² and 65.5 Mg/hm². The correlation coefficient reached 0.7. Downscaled 30 m-resolution biomass from GEOCARBON forest AGB had a higher RMSE, which was between 75.3 Mg/hm² and 79.9 Mg/hm². Compared with the linear model, the non-linear model showed the relationship between AGB and multispectral data more effectively. In general, there was an AGB underestimation at high values and overestimation at low values.