Feature: Interpolation of 3D parameters to be used as field parameters in ERT

[oddvarlia]

With the Drogon case and with APS plugin for RMS continuous 3D parameters like perm, poro or Gaussian Random Fields (GRF) used by APS need to fill the ERTBOX grid that is used in the ERT config file.
The wanted feature is escribed as follows:

• Perm or Poro fields are modelled in RMS in the geomodel/simulation grid, but these need to be extrapolated/interpolated to fill in the whole ERTBOX grid. The same can be said for user defined trends for GRF's made in RMS for a zone in a modelling grid.

About the ERTBOX grid:

• The ERTBOX grid is a 'shoebox' grid, a box with a regular grid. Its purpose is to be container for the field parameters that are updated by ERT. It has dimensions equal to the dimensions for the modelling grid for NX, NY and the same orientation and x,y location as the modelling grid. The number of layers is defined to be large enough that this container grid has space enough for all grid values in the modelling grid for one arbitrary zone. Hence it must at least be as large as the number of layers for the zone with largest number of layers. But if the FMU setup also includes structural uncertainty, the number of layers must be large enough to contain the realisation with most layers for the largest zone.
• For each field parameter included in the ERT config file, it must fill the ERTBOX grid. Since there can be many inactive and collapsed grid cells in the modelling grid, the ERTBOX grid will contain many grid cells with missing value after copying the values from the geomodel/simulation grid into the ERTBOX grid. A problem when working with structural uncertainties is that the number of active grid cells in the geomodel grid may vary from realisation to realisation. ERT Ensemble smoother however requires that the state vector has fixed length for all realisations. This can be achieved by using the ERTBOX grid which has fixed size so that all field parameters regardless of which zone and realisation they belong to have the same size. The problem is that Ensemble smoother will then create linear combination of state vectors where some model parameters such as grid cell values for field parameters may be active (have physical meaning) while in other realisation may be missing (non-physical meaning). It is not a good idea to mix physical realistic values with dummy values that are far from physically realistic in linearcombinations of state vectors.
• One proposal to mitigate the effect described above, it is best to fill also grid cell values that belongs to inactive cells with physical meaningfull values since they might be used when calculating linear combinations of state vectors.. To be able to do this we need methods to fill in values that are not physical (not sampled from the geomodel grid/simulation grid) in the ERTBOX grid.

Questions: This task - interpolate/extrapolate values in 3D grid to fill the whole ERTBOX grid with sensible values can be a part of fmu-tools, but can also belong to another repository for FMU or ERT. Maybe it should belong to the RMS part of the FMU workflow? Interpolation / extrapolation to fill in the ERTBOX will be asked for as soon as some users adopt the new ERTBOX way of working with fields and at the same time has structural uncertainty. It is also possible that the users don't see the need for this and forget the consequences this can have on updated field parameters (with the effect that can have on history matching and predictions). NOTE: The old way of treating resampling and field parameters have the same weakness.