AROMA: Python code and data to reproduce results from Aichinger-Rosenberger & Sjoberg (2025)

This repository contains the model code and scaling routines as well as test data sets for validation of the AROMA model, a data-driven framework to retrieve thermodynamic profiles from GNSS-RO observations. 

Provided are:

• AROMA_final.pt: The AROMA model
• scaler_AROMA.sav / scaler_AROMA_targets.sav : scaling routines for input data and targets 
• internal_validation.py : Python script to evaluate performance on the test data set (y_aroma.pkl, y_test.pkl)
• cosmic2/spire/planetiq_era5.zip: NETCDF files already containing wetPf2, AROMA and ERA5 profiles of pressure, temperature and humidity for the performance evaluation using ERA5 as presented in the paper. These are called "wetPf3" here. Extract those folders in your working directory to work with the provided code for validation.
• validation_ERA5.py: Python script to evaluate performance against ERA5 (and wetPf2/CDAAC)
• RO_averages_test.csv: Average profiles from test period to calculate realtive errors
• raob.zip: Radiosonde data and text files with co-location pairs
• colloc_stats.csv: csv-file with the RO-RAOB pairs used 
• validation_raob.py: Python script to evaluate performance against radiosondes (and wetPf2/CDAAC)
• functions.py: different functions needed for the validation scripts

Place all these items in one directory. The zip-folders can just be extracted into the directory, then the code should be able to find the data. Exception is the path to the RO profiles (get them from CDAAC, https://doi.org/10.5065/T353-C093) for the radiosonde validation, which can be set in validation_raob.py