Recent research modelling floodplain inundation processes has concentrated on issues surrounding the level of physical, topographical, and numerical solver complexity needed to represent floodplain flows adequately. However, during flooding episodes the channel typically still conveys the bulk of the flow. Despite this, the effect of channel physical processes and topographic complexity on model results has been largely unexplored. To address this, the impact of channel cross-section geometry, channel long-profile variability and the representation of hydraulic structures on floodplain inundation are explored using a coupled dynamic 1D-2D hydraulic model (ESTRY-TUFLOW) of the Carlisle floods of January 2005. These simulations are compared with those from a simplified 1D-2D model, LISFLOOD-FP. In this case, the simpler model is sufficient to simulate the far-field peak flood elevations. However, comparison of channel dynamics suggests that the full shallow water approximation used by ESTRY-TUFLOW gives a more robust performance when models calibrated on maximum floodplain water elevations are used to predict channel water levels. Examination of the response of ESTRY-TUFLOW to variations in channel geometric complexity shows that downstream variations in the channel long profile are more important than cross-section variability for obtaining a dataset-independent calibration. The results show, in general, that as model physical complexity is increased, calibrated parameters become less ‘effective’, and as a consequence, the values of performance measures reduce less rapidly away from the optimum value. This means that often more physically complex models are less likely to yield different optimum parameter values when calibrated on different datasets resulting in a more robust numerical model. Lastly, the inclusion of bridge structures can simulate substantial local backwatering effects, but the variability in observed water and wrack marks is such that it is not possible to discern the effect of the bridges at this site in the post-event observational dataset. Copyright © 2011 John Wiley & Sons, Ltd.