AbstractStation Helgoland Roads in the south-eastern North Sea (German Bight) hosts one of the richest longterm time series of marine observations. Hydrodynamic transport simulations can help understand variability in the local data brought about by intermittent changes of water masses. The objective of our study is to estimate to which extent the outcome of such transport simulations depends on the choice of a specific hydrodynamic model. Our basic experiment consists of 3,377 Lagrangian simulations
in time-reversed mode initialized every 7 h within the period Feb 2002–Oct 2004. Fifty-day backward
simulations were performed based on hourly current fields from four different hydrodynamic models that are all well established but differ with regard to spatial resolution, dimensionality (2D or 3D), the origin of atmospheric forcing data, treatment of boundary conditions, presence or absence of baroclinic terms, and the numerical scheme.
The particle-tracking algorithm is 2D; fields from 3D models were averaged vertically. Drift simulations were evaluated quantitatively in terms of the fraction of released particles that crossed each cell of a network of receptor regions centred at the island of Helgoland. We found
substantial systematic differences between drift simulations based on each of the four hydrodynamic models. Sensitivity studies with regard to spatial resolution and the effects of baroclinic processes suggest that differences in model output cannot unambiguously be assigned to certain model properties or restrictions. Therefore, multi-model simulations are needed for a proper identification of uncertainties in long-term Lagrangian drift simulations.