small inconcistency in time dimension for time-invariant data arrays? about mikeio HOT 6 CLOSED

We are considering ways to improve consistency with xarray, currently experimenting with a dfs engine for dfs{2,3].

It could be as simple as:

import xarray as xr
import mikeio

ds = xr.open_dataset("some_meteo.dfs2", engine="mikeio")

That would be great for structured data, but I am still searching for the optimal solution for unstructured meshes, let me know if you have any suggestion.

from mikeio.

You likely know that this topic is also realized by xarray: https://docs.xarray.dev/en/stable/roadmap.html#specific-needs
Maybe also worth to look at xarray extension approach by Deltares: https://github.com/Deltares/xugrid

In general, I find your geometry approach in mikeio much tidier than having additional dimensions like x_nodes etc. to store the mesh in the same ds. But keeping it seperate will require to also modify the geometry for all the .sel and aggregation methods. So, unfortunately no, no really good ideas right now from my end. But I am happy to test and contribute once your experiments mature :-)

from mikeio.

@Hendrik1987 how did you select in time?
I get no time dimension.

>>> ds = mikeio.read("waves.dfs2")
>>> ds
<mikeio.Dataset>
dims: (time:3, y:31, x:31)
time: 2004-01-01 00:00:00 - 2004-01-03 00:00:00 (3 records)
geometry: Grid2D (ny=31, nx=31)
items:
  0:  Sign. Wave Height <Significant wave height> (meter)
  1:  Peak Wave Period <Wave period> (second)
  2:  Mean Wave Direction <Mean Wave Direction> (degree)
>>> ds.isel(time=0)
<mikeio.Dataset>
dims: (y:31, x:31)
time: 2004-01-01 00:00:00 (time-invariant)
geometry: Grid2D (ny=31, nx=31)
items:
  0:  Sign. Wave Height <Significant wave height> (meter)
  1:  Peak Wave Period <Wave period> (second)
  2:  Mean Wave Direction <Mean Wave Direction> (degree)
>>> ds = mikeio.read("waves.dfs2",time=0)
>>> ds
<mikeio.Dataset>
dims: (y:31, x:31)
time: 2004-01-01 00:00:00 (time-invariant)
geometry: Grid2D (ny=31, nx=31)
items:
  0:  Sign. Wave Height <Significant wave height> (meter)
  1:  Peak Wave Period <Wave period> (second)
  2:  _Mean_ Wave Direction <Mean Wave Direction> (degree)
ds.sel(time="2004-01-01")
<mikeio.Dataset>
dims: (y:31, x:31)
time: 2004-01-01 00:00:00 (time-invariant)
geometry: Grid2D (ny=31, nx=31)
items:
  0:  Sign. Wave Height <Significant wave height> (meter)
  1:  Peak Wave Period <Wave period> (second)
  2:  Mean Wave Direction <Mean Wave Direction> (degree)
>>>

from mikeio.

Sorry for the incomplete / incorrect report @ecomodeller
It actually does not happen during time select or aggregation, but when reading a time-invariant dfsu:

from mikeio.

Sorry for the incomplete / incorrect report @ecomodeller It actually does not happen during time select or aggregation, but when reading a time-invariant dfsu:

That is by design. If you know you are expect a time-invariant dfs file, you can do mikeio.read("time-invariant.dfsu", time=0).

In a NetCdf file you can be explicit about the dimensions, e.g. time can be a dimension with a single timestep, or have no time dimension, but in a dfs file, there will always be a start_time even if we want to interpret the file as time invariant.

It is a not easy to come up with a good design for singleton dimensions, but I hope this is good enough.

from mikeio.

Your suggestion mikeio.read("time-invariant.dfsu", time=0) works.

Both, ds and ds2 above carry the information on the first time step, or not? Only the array shape is different.

I personaly would expect a behaviour where after write and read ds2 is the same as ds. But the current design is definetely good enough!

from mikeio.