Non closure of the interannual water balance is a major issue in hydrological modelling. Because the potential causes (e.g. forcing biases, groundwater exchanges, anthropogenic influences, streamflow uncertainties) are so inextricably intertwined, many modelers resort to the calibration of 'sweep' parameters, leading to <em>ad hoc</em> corrections of water balance components to improve performance metrics. Such practices can be beneficial when applied 'for the right reasons', although this is seldom possible to verify. In this paper, we look at the untangling of water balance closure through the lens of scale: we hypothesize that the spatial variability of the water balance anomaly signal can be used to separate the different causes of non-closure and, in particular, to discriminate between those which have local determinants (such as groundwater leaks) and those involving non-local or systematic drivers (such as forcing biases). To test the idea, we compute long-term water balance anomalies for 3641 near-natural gauged catchments across Europe, using the CERRA-Land climate reanalyses (1984–2024). We then apply Thin Plate Splines to extract the <em>mesoscale</em> imbalance, a continuous field of low-frequency variability. We show that the extracted signal is most likely driven by the forcing biases of the CERRA-Land reanalysis. The approach is a step forward towards extracting the contribution of climate forcing biases across gauged catchments and beyond. The hypothesis and the associated methodology provide a tool to contain the uncertainties of 'sweeping', paving the way for hydrologically-grounded mesoscale corrections of the forcing variables in large-scale distributed hydrological models.