Sometimes the user input to SWAN is such that SWAN produces unreliable and sometimes even
unrealistic results. This may be the case if the bathymetry or the wave field is not well resolved. Be aware
here that the grid on which the computations are performed interpolates from the grids on which the input
is provided; different resolutions for these grids (which are allowed) can therefore create unexpected
interpolation patterns on the computational grid. In such cases SWAN may invoke some **internal
scenarios** or **limiters** instead of terminating the computations. The reasons for this model policy is that

- SWAN needs to be robust, and
- the problem may be only very local, or
- the problem needs to be fully computed before it can be diagnosed.

Examples are:

- The user can request that
__refraction__over one spatial grid step is limited to at most the length of a directional sector associated with one sweep (which in the absence of a current is 90^{o}). See command`NUMERIC`. This may be relevant when the depth varies considerably over one spatial grid step (e.g. at the edge of oceans or near oceanic islands with only one or two grid steps to go from oceanic depths to a shallow coast). This implies inaccurate refraction computations in such grid steps. This may be acceptable when refraction has only local effects that can be ignored but, depending on the topography, the inaccurately computed effects may radiate far into the computational area. - SWAN cannot handle wave propagation on
__super-critical current flow__. If such flow is encountered during SWAN computations, the current is locally reduced to sub-critical flow. - If the
__water depth__is less than some user-provided limit, the depth is set at that limit (default is 0.05 m, see command`SET`). - The
__user-imposed wave boundary conditions__may not be reproduced by SWAN, as SWAN replaces the*imposed*waves at the boundaries that propagate into the computational area with the*computed*waves that move out of the computational area at the boundaries. - SWAN may have
__convergence__problems. There are three iteration processes in SWAN:- an iteration process for the spatial propagation of the waves,
- if ambient currents are present, an iteration process for spectral propagation (current-induced refraction and frequency shift) and
- if wave-induced set-up is requested by the user, an iteration process for solving the set-up equation.

- ad 1
- For spatial propagation the change of the wave field over one iteration is
limited to some realistic value (usually several iterations for stationary
conditions or one iteration or upgrade per time step for nonstationary
conditions; see command
`NUMERIC`). This is a common problem for all third-generation wave models (such as WAM, WAVEWATCH III and also SWAN). It does not seem to affect the result seriously in many cases but sometimes SWAN fails to converge properly.

For curvilinear grids, convergence problems may occur locally where in some points in the grid, the directions separating the 4 sweeping quadrants coincide with the given spectral directions. - ad 2
- For spectral propagation (but only current-induced refraction and frequency shift) SWAN may also not converge.
- ad 3
- For the wave-induced set-up SWAN may also not converge.

`PRINT`file.

Because of the issues described above, the results may look realistic, but they may (locally) not be accurate. Any change in these scenarios, limiters or shortcomings, in particular newly discovered coding bugs and their fixes, are published on the SWAN web site and implemented in new releases of SWAN.