Training Set and Test Set for Fast Atmospheric Radiative Transfer Inversion - Clear sky

This dataset is organized into two ZIP folders: a Training Set and a Test Set.

1. Training Set

The Training Set contains two subfolders:

Input Files: This folder includes all input variables describing the atmospheric scenarios necessary for simulations.

Output Files: This folder contains the corresponding simulated spectra generated by the forward model.

2. Test Set

The Test Set consists of three subfolders:

Input Files: Similar to the training set, this folder contains the input variables for the atmospheric scenarios.

Output Files: This folder holds the simulated spectra for the test cases.

A Priori Files: This folder contains the a priori data generated by perturbing true profiles according to a precision matrix SaS_aSa, constructed based on error estimates from the UK Met Office for the routine assimilation of IASI products into their operational Numerical Weather Prediction (NWP) system.

Data Generation Process

The definition of our datasets requires input variables that describe atmospheric scenarios and output variables representing the corresponding FORUM simulated spectra. The atmospheric state was constructed using data from the ECMWF (European Center for Medium-range Weather Forecast) Re-Analysis ERA5 database. Concentrations of additional atmospheric gases, including CO2, CO, CH4, NO2, HNO3, and SO2, were obtained from the CAMS (Copernicus Atmosphere Monitoring Service) global Greenhouse Gas reanalysis (EGG4). The remaining atmospheric gases were sourced from the Initial Guess 2 (IG2) database.

Surface emissivity profiles were derived from the Huang database, utilizing a tailored procedure to associate the most suitable profile based on the time and geolocation. 

For spectral simulation, the radiative transfer problem was addressed using the CLouds and Atmospheric Inversion Module (CLAIM), an advanced version of the inversion code previously used, based on the LBLRTM (Line By Line Radiative Transfer Model) forward model. The inversion process employed by CLAIM integrates the Optimal Estimation (OE) approach, the Gauss-Newton minimization method, the Levenberg-Marquardt technique, and a final Tikhonov regularization step using the Iterative Variable Strength (IVS) method.

Instrument characteristics for the FORUM were taken from one of the two instrument concepts available during the phase A/B1 of the FORUM E2E project.