Coupled surface water - groundwater interaction models are ideally the best choice of models to simulate complex hydrological processes in regional river basins. Such fully integrated models are less tenable with comprehensive parameter sensitivity and non-linear prediction uncertainty analysis workflows that require numerous runs of the simulation model. This limits the application of these models in practical decision-making contexts where reliability of the predictions is important. We proposed to overcome this limitation by the paired use of an approximate and fast-running surrogate model constrained by the quantitative and qualitative knowledge of process details from the complex model. While the complex model focuses on process dynamics as much possible as underpinned by available data, the surrogate model is purpose-built for the predictions of interest with approximation of the system dynamics and scale supported by parameterisation schemes that are suitable for predictions of interest. We used a low-fidelity MODFLOW model together with a complex MIKE SHE model built for the North West Bangladesh region to assess district scale water balance and climate change impacts on future water balance. The MIKE SHE model is underpinned by detailed conceptualization of the alluvial aquifer and channel bathymetry, simulation of the catchment processes and flow routing by the MIKE 11 routine and calibrated to observed water levels and flow data. Each model run takes several hours to complete one simulation. The MODFLOW model approximates the process details with appropriate upscaling of parameters and was subjected to calibration and uncertainty using the PEST suite to simultaneously improve the match to observed water levels and to maintain district-scale water balance comparable with the complex model. The fast running (less than 2 minutes) surrogate MODFLOW model was then used in the scenario analysis for climate change impacts. The results indicated that the surrogate MODFLOW model simulation is able to achieve matches to the observed water levels that are comparable to the complex model simulation in majority of the locations across the region. The simplified and up scaled parameterization scheme of the MODFLOW model enabled computationally efficient uncertainty analysis of climate change impacts on district scale water balance. The study indicated that low-fidelity surrogate models developed for specific objectives can be useful when used in conjunction with complex models that can simulate complex SW - GW interaction processes. This study was undertaken as part of an Australian Government initiative to increase water and food security in Asia.