Hydrodynamic modeling of large remote forested floodplains, such as the Amazon, is hindered by the vegetation signal contained within Digital Elevation Models (DEMs) such as the Shuttle Radar Topography Mission (SRTM). Not removing the vegetation signal causes DEMs to be overelevated preventing the correct simulation of overbank inundation. Previous efforts to remove this vegetation signal have either not accounted for its spatial variability or relied upon single assumed error values. As a possible solution, a systematic approach to removing the vegetation signal which accounts for spatial variability using recently published estimates of global vegetation heights is proposed. The proposed approach is applied to a well-studied reach of the Amazon floodplain where previous hydrodynamic model applications were affected by the SRTM vegetation signal. Greatest improvements to hydrodynamic model accuracy were obtained by subtracting 50–60% of the vegetation height from the SRTM. The vegetation signal removal procedure improved the RMSE (Root-Mean-Square Error) accuracy of the hydrodynamic model than when using the original SRTM in three ways: (1) seasonal floodplain water elevation predictions against TOPEX/Poseidon observations improved from 6.61 to 1.84 m; (2) high water inundation extent prediction accuracy improved from 0.52 to 0.07 against a JERS (Japanese Earth Resources Satellite) observation; (3) low water inundation extent accuracy against a JERS observation improved from 0.22 to 0.12. The simple data requirements of this vegetation removal method enable it to be applied to any remote floodplain for which hydrodynamic model accuracy is hindered by vegetation present in the DEM.