Add training with similar data to paper

[jacobbieker]

Detailed Description

MetNet was trained on GOES and MRMS data, and MetNet-2 on that plus some NWP initialization parameters. All that data is publicly accessible with no restrictions from NOAA, so adding that to HF datasets to make it easy to ideally, reproduce the paper would be great.

Context

This would add GOES-16/17 and MRMS precipitation data to HF Datasets, along with probably the HRRR NWP model for high-resolution data over the US. The data for GOES and HRRR is all available through Amazon open datasets, while the MRMS data has been archived and is fairly easily accessible.

Possible Implementation

[tarunluthra]

Hi jacob, what's the best place to source MRMS data?
I am extracting precipitation rate files from here, https://mrms.ncep.noaa.gov/data/2D/PrecipRate/
is there a better source for getting similar data?

[jacobbieker]

The best is probably the archive here: https://mrms.agron.iastate.edu which is a mix of on AWS and on a university's servers. The one you link to is only the near-realtime archive, so older things are deleted fairly quickly (I think a week?)

[tarunluthra]

Thanks, Jacob. For an initial baseline test of the model, how far back in time do I need to retrieve radar and GOES data? I also noticed that the model takes elevation as an input, so does the DEM need to be patched for the continental US region as some of the precipitation data isn't region specific.