We present an adapted woody biomass retrieval approach for tropical savanna areas appropriate for use with satellite acquired L-band SAR imagery. We use the semiempirical water cloud model to describe the interaction between the SAR signal and vegetation and re-arrange the model to predict biomass. Estimations are made using dual polarization SAR imagery collected by ALOS PALSAR during 2008 in combination with community woodland inventory data from pine savanna areas in Belize. Estimation accuracy is assessed internally by the fit of the model to the ground training data, and then validated against an independent external dataset, quality controlled using Worldview II imagery. The internal validation shows a biomass estimation with an RMSE of 25 t/ha and a coefficient of determination ${{bf R}^{bf 2}}$ of 0.70, while the external validation indicates an RMSE of 13 t/ha with ${{bf R}^{bf 2}}$ of 0.53. This approach to biomass estimation appears to be most influenced by the plots with higher tree numbers and where the trees were more homogeneous. The existence of many similar sized individuals in those plots influence the SAR backscatter and is predicted to be the cause the elevated level of saturation found in this study (${>} {bf 100}{bf t}/{bf ha}$) with complete saturation predicted as a result of number density increases, and concurrently increasing basal area, both not exclusively dependent on biomass.