Interpolation of input fields in ww3_prnc

[ukmo-juan-castillo]

While testing a development branch (see #285) we discovered some problems when reading from input an air density field. We checked the input file and it is OK, but in the model output some zeroes appear in the middle of the domain, and strange interpolation is made along the land/sea boundary. Two main problems were found:

1. For points in the input field adjacent to land, interpolation still takes place with land values, which are set to zero by default. In consequence, the grid interpolation that takes place in ww3_prnc generates non-physical values. This can be fixed by checking if there are adjacent land points and recalculating the normalization factors ignoring these land points. An improvement is possible by calculating weights in land points adjacent to ocean instead of setting them to zero, so that interpolation takes place with valid values. This could be useful for atmosphere fields where in principle there are valid values in the whole grid domain, but the modifications required are large.

2. When the boundary cells read from the nesting file are wrong and are not ocean grid points (as it is the case in the ww3_tp2.15 regtests), the weights are still used and are all zero, so the final interpolated value is zero. This can be avoided by just not interpolating this type of points and keeping the original value. It is not ideal but it will avoid having spurious values at the land/sea boundaries.

It is quite possible that this fix will change results of some tests, as it affects how the forcing fields are read.