search

google-deepmind / graphcast

Apache License 2.0
4.53k stars 572 forks source link

Data downloaded from ECMWF is different from the sample data #36

Open NightWalkerqqq opened 9 months ago

NightWalkerqqq commented 9 months ago

I have downloaded ERA5 data following instructions on ECMWF. To check the accuracy of data, I downloaded data on 2022-01-01, just the same as the example data provided in Google Cloud Bucket, but actually they were slightly different. Would I ask which API should be used to download these data, or could you publish a script or instructions for constructing input data?

Thanks a lot !

Tlmnk commented 9 months ago

I am not sure, but maybe this could explain it: ERA5 is on 0.25 degree resolution while the final model is run using data from the ECMWF high resolution model that has a 0.1 degree resolution and is then downsampled to 0.25 degrees before being used as input. Maybe this explains the difference?

HisKim1 commented 2 months ago

Same issue. I tried the same process on 2022-01-01 to generate the other dataset but got a slightly different dataset especially the polar region's total_precipitation_6hr. I guess other variables are directly downloaded from ECMWF and no need to pre-process so the difference would be just a noise but the pattern on the figure of the difference btw mine and google bucket's total_precipitation_6hr doesn't seem like a noise. I really appreciate it if you share how to solve this issue