Support for single source binary forcing files

[donaldwj]

Short description explaining the high-level reason for the new issue.

Current behavior

Currently formulations need a csv forcing file for each catchment. We need to integrate the capability to drive all catchment using a single forcing file for all catchments.

Expected behavior

Forcing are regrided to provide per-catchment forcings

Proposed Solution

1. Solution 1

• Each MPI rank reads part of a complete forcing file
• Each rank broadcast its part of the forcing grid to all other ranks
• Individual ranks then load pre stored regridding weights for each assigned catchment
• Forcing are then regridded and/or aggregated for each catchment as needed
• This approach requires a full forcing grid for the overall domain to be loaded in each mpi rank
• This approach minimizes forcing I/O time as parallels NetCDF reads will never be for overlapping data segments

2. Solution 2

• Each MPI rank reads the bounding box for each assigned catchment from a forcing file
• Each rank then load precomputed weights for regridding. These weight are different than those in solution 1 as they will be for different matrix sizes.
• Forcing are then regridded as needed
• This approach minimized I/O time as forcing are not shared between MPI ranks
• This approach requires more NetCDF reads 1 per catchment instead of 1 per MPI rank
• NetCDF I/O time will be higher as parrallel reads will be in overlapping regions for at least some catchments

[mattw-nws]

Do we want to call this done after #406 (and #417)? Or should this be used as the Epic for generalized gridded forcing processing to include use of a spatial coupler etc.?

[mattw-nws]

Closing as completed in #417.