Estimating total aboveground, stem and branch biomass using multi-frequency SAR

We have investigated the retrieval of total aboveground, stem and branch biomass using multi-temporal SAR data acquired at C-, L-, and P-band by Radarsat-2, ERS-2, ALOS PALSAR, and SETHI for a forest site in Sweden. A semi-empirical model was used to model C- and L-band backscatter as function of total aboveground, stem and branch biomass; a linear regression model was used for P-band. Calibrated with the aid of inventory plots, the models were used to estimate total aboveground, stem, and branch biomass for each individual scene in the multi-temporal and multi-frequency stack of observations. Individual estimates were then combined using weights reflecting each scenes' sensitivity to the biomass component in question. At the plot scale, the retrieval errors were 37, 40, and 33 % for total aboveground, stem, and branch biomass in terms of the relative RMSE; the error further reduced to 33, 38, and 27 % when calibrating models per species. We find that i) the retrieval of total aboveground, stem and branch biomass benefits from the availability of multi-temporal/-frequency data as the lowest retrieval errors with the best single image (P-band) were about 6 to 8 % higher, ii) there is no benefit in estimating total aboveground biomass via independent estimates of stem and branch biomass, iii) consideration of species specific differences associated with different allocations of biomass to stems and branches allows for improving the retrieval.