In an article in Hydrological Processes in 2004(bates, 2004), I argued that the use of remote sensing data had allowed a significant breakthrough to be made in flood dinundation modelling. Over the preceding 5 years, flood inundation research had moved rapidly from being a 'data-poor' to a 'data-rich' science (see also Di Baldassarre and Uhlenbrook, 2011), with attendant possibilites for model development and process insight. Until the late 1990s, the only data available to build, parameterize, calibrate and validate hydraulic models were from limited ground topographic surveys and sparse ground gauging stations with spacings of between 10 and 60 km. Occasionally, air photos of flooding were avaliable, but these were not used to test model performance in a systematic way because the available terrain data were insufficiently detailed to provide confidence in distributed model predictions. The errors in the then-available data and ther limited information content meant that it was very difficult to discriminate between different model physics and parameters, such that many different models could fit the available validation data equally well yet lead to different future predictions or process inferences. This equifinal behaviour also retarded model development and testing, and in particular prevented the more widespread adoption of two-dimensional (2D) hydraulic models in two main ways. First, the ground survey terrain data that were available were typically captured as a series of cross sections perpendicular to the channel and floodplain that were more easily integrated with one-dimensional (1D) models. Second, the use of widely spaced point-gauge data for calibration and validation meant that the only aspect of model performance that could be tested effectively was the ability of a model to route a wave in 1D along a river network. Under the circumstances, 2D models added considerable additional complexity for no discernible net benefit in prediction (see Bates et al., 1998 for an example of an overspecified 2D model tested in terms of its flow routing performance only).