Non-orographic gravity wave drag dataset

Public data repository for Chantry et al. (2020) on NOGWD emulation using machine learning

See paper for details on data generation.

To unzip:

tar -xzf 2015-01-01.h5.tgz

Files are formatted year-month-day.h5 based on forecast start date.
Files contain arrays x & y
Inputs: x(191,ntime,nspace)
Outputs: y(126,ntime,nspace)
ntime=36 (every 20 hours from 20:720 hours)
nspace=65440 (subsampling every 10th gridpoint of 654400 original points)
See paper for explanation of the 191 & 126.
Data has not been normalised.

Challenge is for a given 191 inputs, predict the 126 outputs.

Training data: used 12 30-day periods in 2015.
Testing: 3 provided dates in 2016.
Validation: 3 provided dates in 2017