Reducing Uncertainties in Satellite-derived Forest Aboveground Biomass Estimates using a High Resolution Forest Cover Map

Abstract:

Several studies to date have provided an extensive knowledge base for estimating forest aboveground biomass (AGB) and recent advances in space-based modeling of the 3-D canopy structure, combined with canopy reflectance measured by passive optical sensors and radar backscatter, are providing improved satellite-derived AGB density mapping for large scale carbon monitoring applications. A key limitation in forest AGB estimation from remote sensing, however, is the large uncertainty in forest cover estimates from the coarse-to-medium resolution satellite-derived land cover maps (present resolution is limited to 30-m of the USGS NLCD Program). The uncertainties in forest cover estimates at the Landsat scale result in high uncertainties for AGB estimation, predominantly in heterogeneous forest and urban landscapes. We have successfully developed an approach using a machine learning algorithm and High-Performance-Computing with NAIP air-borne imagery data for mapping tree cover at 1-m over California and Maryland. In a comparison with high resolution LiDAR data available over selected regions in the two states, we found our results to be promising both in terms of accuracy as well as our ability to scale nationally. The generated 1-m forest cover map will be aggregated to the Landsat spatial grid to demonstrate differences in AGB estimates (pixel-level AGB density, total AGB at aggregated scales like ecoregions and counties) when using a native 30-m forest cover map versus a 30-m map derived from a higher resolution dataset. The process will also be complemented with a LiDAR derived AGB estimate at the 30-m scale to aid in true validation.