Closed JackKelly closed 2 years ago
my guess is that, sometimes, the variables are in the wrong order before appending to the Zarr (maybe dataset.to_array()
puts the channels into a non-deterministic order?)
rough plan of action:
if appending to existing zarr then:
@jacobbieker this issue might also affect the EUMETSAT native-to-Zarr conversion.... but let me do a bit more digging into the NWP to Zarr conversion...
Okay! I believe Satip reorders the dims before writing to disk, but I'd have to check tomorrow.
OK, I've forced the order of the channels in the NWP zarr, and am writing again to /mnt/storage_ssd_8tb/data/ocf/solar_pv_nowcasting/nowcasting_dataset_pipeline/NWP/UK_Met_Office/UKV/zarr/UKV_intermediate_version_3.zarr
PR to follow...
In Satip, I use transpose
instead of reindex here: https://github.com/openclimatefix/Satip/blob/9091a92fc1ef05d84821d2e41dfd42734797c8fc/satip/utils.py#L214 which looking at the docs for transpose and reindex look like they do similar things, but not the same. So that could be the issue with the satellite Zarrs
The problem is in the intermediate Zarr: Sometimes the channel labelled "si10" (10 meter wind speed) has data from the
hcct
channel in it!