42Model Replacement Benchmark

To Do

Write first draft of the paper @Stephan R

Have everyone review the paper outline

Figure out where to submit

Figure out where to upload the data (Azure or Pangeo Google Cloud) @Stephan R

Also DOI data repository for ESSD

Download and process the data for all variables @Stephan R

Write script to extract single levels/time slices from full data

Upload

sudo mount -v -t cifs -o username=ge35maj,password=rsc2016mwrWCB //nas.ads.mwn.de/tuze/bib/FD_Share/m1524895 mediaTUM/

Create benchmark evaluations

New figure: Straight line must be some sort of error.

Method	Z500 RMSE (3 days/5 days) [m^2/s^2]	850 hPa Temperature (3/5/ days) [K]
Persistence	936/1033	4.29/4.56
Climatology	1075	5.51
Weekly climatology	816	3.50
Linear Regression (direct)	714/814	3.19/3.52
Linear Regression (iterative)	719/812	3.17/3.48
Fully convolutional network (direct)	626/757	2.87/3.37
Fully convolutional network (iterative)	1114/1559	4.48/9.69
T42 IFS	489/743	3.09/3.83
Operational IFS	154/334	1.36/2.03

Persistence/Climatology @Stephan R

Simple Z500 Dense NN (let’s not do this one because CNNs are better anyway)

Simple Z500 CNN @Sebastian S @Stephan R

Use this as example script for data handling

This is a 1-hidden-layer dense NN trained on every 6th time step (sample)

Operational ECWMF from TIGGE @Stephan R

IFS run at T21 @Peter D

Create figures and tables for paper @Stephan R

RMSE vs lead time for different methods

3/5 day table

Figure out licence

For code repository @Stephan R

For data @Peter D

Add documentation to the repository @Stephan R

Quick start guide for downloading and reading data (in docs and notebook, potentially binder)

Add download API documentation (so people can download their own high-res data)

Please read this and comment on the open questions. I also tried to divide the work for the next steps. Please indicate whether you will have the time in the next two months or so.

Here is the paper draft: https://www.overleaf.com/3526418756wqbjngbqgtfr

I created a github repository for the project: https://github.com/raspstephan/weather-benchmark

This repository should contain (all with documentation):

Scripts to download the ERA5 data

Scripts to regrid the data

​​To Do

To Do