The bathymetry, current, water level, bottom friction and wind (if spatially variable) need to be provided to
SWAN on so-called input grids. It is best to make an input grid so large that it completely covers the
In the region outside the input grid SWAN assumes that the bottom level, the water level and bottom friction are identical to those at the nearest boundary of the input grid (lateral shift of that boundary). In the regions not covered by this lateral shift (i.e. in the outside quadrants of the corners of the input grid), a constant field equal to the value at the nearest corner point of the input grid is taken. For the current and wind velocity, SWAN takes 0 m/s for points outside the input grid.
In SWAN, the bathymetry, current, water level, wind and bottom friction may be time varying. In that case they need to be provided to SWAN in so-called input time windows (they need not be identical with the computational, output or other input windows). It is best to make an input window larger than the computational time window. SWAN assumes zero values at times before the earliest begin time of the input parameters (which may be the begin time of any input parameter such as wind). SWAN assumes constant values (the last values) at times after the end time of each input parameter. The input windows should start early enough so that the initial state of SWAN has propagated through the computational area before reliable output of SWAN is expected.
One should choose the spatial resolution for the input grids such that relevant spatial details in the bathymetry, currents, bottom friction and wind are properly resolved. Special care is required in cases with sharp and shallow ridges (sand bars, shoals) in the sea bottom and extremely steep bottom slopes. Very inaccurate bathymetry can result in very inaccurate refraction computations the results of which can propagate into areas where refraction as such is not significant (the results may appear to be unstable). For instance, waves skirting an island that is poorly resolved may propagate beyond the island with entirely wrong directions. In such a case it may even be better to deactivate the refraction computations (if refraction is irrelevant for the problem at hand e.g. because the refracted waves will run into the coast anyway and one is not interested in that part of the coast). In such cases the ridges are vitally important to obtain good SWAN results (at sea the waves are 'clipped' by depth-induced breaking over the ridges which must therefore represented in SWAN computation). This requires not only that these ridges should be well represented on the input grid but also after interpolation on the computational grid. This can be achieved by choosing the grid lines of the input grid along the ridges (even if this may require some slight "shifting" of the ridges) and choosing the computational grid to be identical to the input grid (otherwise the ridge may be "lost" in the interpolation from the bottom input grid to the computational grid).
Finally, one should use a time step that is small enough that time variations in the bathymetry, current, water level, wind and bottom friction are well resolved.