Biomass Mapping of US forests using synergy of SAR and optical Remote Sensing

Josef Kellndorfer, Woods Hole Research Center, josefk@whrc.org (Presenter)
Alessandro Baccini, Woods Hole Research Center, abaccini@whrc.org
Jesse B. Bishop, Woods Hole Research Center, jbishop@whrc.org
Oliver Cartus, Woods Hole Research Center, ocartus@whrc.org (Presenter)
Tina Cormier, Woods Hole Research Center, tcormier@whrc.org
Wayne S. Walker, Woods Hole Research Center, wwalker@whrc.org
Maurizio Santoro, GAMMA RS, santoro@gamma_rs.ch

The availability of several national EO datasets for ~2000, i.e. SRTM, USGS National Elevation Dataset (NED), National Land Cover Dataset (NLCD) and MRLC Landsat ETM+, represented a unique opportunity to produce baseline canopy height, CH, and aboveground biomass, AGB, maps for the US, i.e. the National Biomass and Carbon Dataset, NBCD 2000. Differentiation of the SRTM Elevations and NED allowed the estimation of the SRTM phase scattering center heights within the canopies, a key predictor for CH and AGB. The phase scattering heights, DEMs, NLCD and Landsat datasets were used as predictor layers in RandomForest for predicting CH and AGB. FIA survey data were used for model development and validation. CH and AGB maps were produced for the conterminous US at 30 m pixel size. Validation resulted in an RMSE for AGB of 55 (plot-), 19 (hexagon-) and 14 t/ha (county-level). For CH, the RMSE was 3.8 m (plot-level). To update NBCD, the feasibility of biomass retrieval with ALOS PALSAR was investigated. For the biomass retrieval with dual-polarization L-band, a retrieval algorithm was adopted (cf. Santoro et al., 2011), in which a semi-empirical model, relating backscatter to AGB, is trained using the NLCD canopy density map. The model, calibrated for each intensity image, was inverted to estimate the biomass. Where possible, a multi-temporal combination of single-image estimates was performed. A comparison with NBCD indicated the feasibility of the approach. While characterized by pronounced scatter at full resolution, the agreement of the ALOS and NBCD AGB maps improved substantially when aggregating the maps to coarser pixel sizes with no saturation noticeable in the ALOS maps up to ~200 t/ha. The feasibility of the L-band-based AGB retrieval, at least at aggregated scales, was further confirmed when comparing the ALOS maps to FIA county total AGB estimates (RMSE=5.2 Mio t, R2=0.95).

Presentation Type:  Poster

Session:  Science in Support of Decision Making   (Wed 10:00 AM)

Associated Project(s): 

• Kellndorfer, Josef: The National Biomass and Carbon Dataset 2000: A High Spatial Resolution Baseline to Reduce Uncertainty in Carbon Accounting and Flux Modeling ...details

Poster Location ID: 202