Handle dry point

[saeed-moghimi-noaa]

No description provided.

[saeed-moghimi-noaa]

@WPringle Please document the problem and your current solution. Thanks.

[ghost]

currently, the issue is that a large amount of mesh nodes are dry (NaN water elevation) throughout the ensemble space. These nodes are not all colocated; one particular mesh node might be dry in one ensemble member and wet in another. This is problematic for the polynomial fitting, as it expects to be fitting to a continuous function of water elevation.

The initial solution was to assign all dry nodes in the study area a minimum (near-zero) water elevation value. This allowed the polynomial to run, but it is not an ideal solution because the minimum value acts as a totally flat plateau of water elevation, and the accuracy of the resulting surrogate reflected that.

The new solution is to assign dry nodes a "real" water elevation with linear falloff away from the nearest true wet node. This allows the polynomial to fit to a gradient curve instead of to a discontinuous plateau.

[WPringle]