The unprecedented hydraulic visibility of rivers surfaces deformation with SWOT satellite is tremendous information for refined hydrological-hydraulic modeling. But the estimation of uncertain or unknown discharge and bathymetry-friction in a spatialized hydrodynamic model from water surface elevation (WSE) and width (WSW) observations is a difficult high-dimensional inverse problem faced with equifinality. This article newly studies variational data assimilation (VDA) of WSE into a 1D Saint-Venant differentiable river network model fed by a semi-distributed hydrological model. A pre-processing chain enables (i) building effective hydraulic model geometry from WSE altimetry (Sentinel 3, drifting ICESat2) and WSW (Sentinel 1 images), and (ii) filtering noisy SWOT level 2 WSE before assimilation. The simultaneous inference of spatially distributed inflow hydrographs, bathymetry-friction at network scale, on the large poorly gauged Maroni basin (French Guiana), is done by VDA of nadir and in situ WSE or SWOT 1-day WSE only. A systematic improvement obtained for the fit to assimilated WSE and in validation of discharge at 5 gauges inside the network: 70% of data-model misfit in [−0.25 ; 0.25m], NRMSE on discharge between 0.11 and 0.26 for SWOT only on a large flood given unfavourable hydrological prior. SWOT WSE density enables to infer detailed spatial variability on channel bottom elevation given width from images and detailed temporal variabilities of hydrological inflow hydrographs. The approach is transposable to other rivers networks worldwide in view to tackle the double regionalization problem of hydrological and hydraulic parameters from sparser but increasingly massive and informative data.