Use time series to create forcing files

[shartgring]

Kind of request

Possible enhancement of setup_precip_forcing, or alternatively a new feature. I am working on this in a branch as part of making things easier for stress-testing and model coupling

Enhancement Description

Currently, only gridded data can be used to create forcing files. In some cases it can be convenient to create forcing files from timeseries (csv), either uniformly or using interpolation methods

Use case

• When gridded data does not have the desired coverage or temporal resolution, which can be supplied by multiple local stations
• When working with uniform design rainfall events (e.g. to compute boundary conditions, stress-testing, climate analysis)

Additional Context

Might this be something that would also fit in HydroMT core?

Would this be something to use as an enhancement of the current setup_precip_forcing (and let it detect if precip_fn refers to gridded data or timeseries), or add it as a new method?

And, which interpolations would be good to include? Thiessen polygons, Inverse Distance Weighting, ... ? Literature might be good source for methods, see for example https://cdn.knmi.nl/knmi/pdf/bibliotheek/knmipubIR/IR2009-04.pdf

[hboisgon]

It's been on the wishlist for very long and there is an issue open in hydromt core: Deltares/hydromt#101

I would prefer to create a different method, the arguments will be completely different so it would be better to separate I think.
We can start with one interpolation but would be good to have several available even if we implement them later.

[shartgring]

Interpolation docs shared by FEWS team: https://publicwiki.deltares.nl/display/FEWSDOC/Interpolation+Spatial+Transformations

[hboisgon]

Create a new hydromt-wflow method and workflow: setup_precip_from_point_timeseries

Steps in the workflow: using metpy

• Read input data as geodataset - geometry type is point only (we could decide to support vector later)
• Time resampling
• Filter the stations with nodata for a specific timestep to leave them out of the interpolation (any other missing treatment eg fill with zeros should be done beforehand)
• Call the interpolation method from metpy: interpolate_to_grid
• The method should be wrapped for xarray (ufunc) - (metpy can only be used timestep per timestep)
• Fill any remaining missings with zero or with other gridded data (nice to have)?
• Export - write

Note: for later if we decide to support polygon timeseries, we can use xugrid regridding methods to go from mesh/polygon to regular grid

[shartgring]

Alternatively, there is a more precipitation focused package https://docs.wradlib.org/en/latest/ipol.html (thanks @RubenImhoff!)