Hello,
I just started to get into “nowcasting” and I am trying to do some machine learning stuff in this field. I am located in Germany and am using precipitation data from the DWD open data server. Also I am using python with the wradlib library.
I want to use the RADOLAN-YW product but am struggling to find historical data and recent data in the same format.
Right now I am using the historical data from Index of /climate_environment/CDC/grids_germany/5_minutes/radolan/reproc/2017_002/ (dwd.de) and the recent data from Index of /weather/radar/radolan/yw/ (dwd.de).
The issue is that the historcial data is in the RADOLAN version 2.18.3 with a grid size of 1100km x 900km and the recent data in the version 2.29.1 with a grid size of 900km x 900km. Also the radar station radius is different with 150km (historical) to 128km (recent).
But if I plot the data with the respective radolan_grids from the wradlib library, the data circles from the historical data are smaller than the recent ones:
(Since I am a new user, I can only uplad one picture)
Are there historical datasets with the RADOLAN version 2.29.1? Or ist there maybe something I am doing wrong?
Best
Max
Unfortunately i did not find both recent and historical datasets with the same format. The files from
Index of /climate_environment/CDC/grids_germany/5_minutes/radolan/recent/ (dwd.de)
also seem to be 900 x 900 while the historical data are 1100 x 900.
Historical data are reprocessed before publishing. I only found a document in German:
251_pdf.pdf;jsessionid=A9C29FFDAE4B1F845DFEE05DB5857D3B.live21061 (dwd.de)
The reprocessing is only done once per year for the year prior and not done continuously for recent data.
I still have 900x900 radolan bin files from /weather/radar/radolan/yw/ for a few days from August 2023 from an old project analysis. If you are interested, i can upload those and you can try to compare / overlay these datasets with the reprocessed historical data. However, there are the differences in both spatial/grid aspect as well as the reprocessing, so i am not sure if you will get a sufficiently good setup to do machine learning.
Thanks for the information. Somehow its hard to keep track of the different pieces and versions of information available from the DWD 
I also contacted the DWD directly and got a reply which confirmed my observation, that the historical (reprocessed) data is different format and based on the “DX” product, with 128 km radius around the radar stations on a 1100 km x 900 km grid. The new processing pipeline has 150 km radius with a 900 km x 900 km radius.
Maybe they can provid the raw data of the “new” measurements, so I possibly could create the precipitation grids myself.
For now I will stick to the older format for training some initial models.
