Speaker
Description
Current remote sensing based approaches for forest biomass estimation evolve around the use of allometric relationships at different spatial and temporal scales. Three main input parameters are used for the biomass estimation: the tree height, either at the individual level, or at the respective resolution cell size, the allometric factor and the allometric exponent. This formula for the biomass estimation is widely known as the simplified power law function. The allometric factor alpha accounts for density variations in the forest stand. The allometric exponent on the other hand represents species composition and variations in the growing conditions, thus involves a more profound knowledge of the forests and the implementation of forest field inventory data.
Actual implementations of such an approach by means of TanDEM-X / GEDI measurements rely on forest height estimates obtained from the combination of interferometric TanDEM-X and GEDI waveform measurements and allometric parameters derived from the existing GEDI biomass product. This way, a forest height map at 25m resolution scale, matching the GEDI footprint of equally 25m radius, is generated. Future implementations must focus on the BIOMASS L2B FH product scale of 200m resolution. From this, a considerable scale difference becomes apparent.
As a preliminary step, this scale difference between Tandem-X/GEDI Products and the BIOMASS L2B FH scale is analyzed with existing waveform LiDAR measurements of the NASA LVIS instrument. Tropical forests in Gabon and Costa Rica were analyzed. By creating a baseline scenario with the provided LVIS biomass product, a multitude of different forest height map scales was analyzed with their respective effect on the outcoming biomass map. It became apparent that scale differences between different sensors cannot be neglected and induce – at least in the case of the analyzed parameter forest height – a considerable error. By the experiment with the mentioned LVIS dataset, biomass increases of up to 80t / ha were detected between a 25 m scale (e.g. Tandem-X, LVIS or GEDI) and a 200m scale (e.g. BIOMASS).
Throughout working with the LVIS dataset, extreme variations of the resulting biomass values for the individual canopy height classes became apparent. Now, in addition to the mentioned datasets and methods, simulated data from the UFZ FORMIND forest growth model was implemented, rendering a perfect sandbox for further analysis. The high variance in biomass values was also confirmed in the FORMIND biomass product. Different species compositions and growing conditions (which would directly attribute to changes in wood density) can be ruled out, as these variances are already prominent at rather small scales (e.g. 50 ha datasets of LVIS). Thus, density variations in the forests are suspected to be the underlying driver. As of right now, forest density can not be directly measured with remote sensing methods, thus, the complexity, or rather forest structure is applied as a proxy.
As of this moment, two approaches are being tested to estimate the forest structure at different spatial scales. One consists of a spatially adapted horizontal forest structure index, calculated from the phase centers of TanDEM-X measurements. The other revolves around using wavelet decompositions to likewise create a structure index and thus segment the canopy height to biomass point cloud into smaller areas, subsequently enabling the use of spatially adapted allometries.
Both approaches show promising results and both the performance and applicability will be further discussed by means of simulated and real experiment data.
