This presentation details our approach to the Ariel Data Challenge 2024, focusing on extracting atmospheric spectra from simulated ARIEL data of exoplanet transit observations, and estimating the level of uncertainty.

Our approach consisted of three main stages after data preprocessing.

First, we estimated the atmospheric opacity for each wavelength by leveraging polynomial fitting to detrend light curves and determine transit depths. This process was applied to an average over neighboring wavelengths for noise reduction.

Next, a differentiated postprocessing strategy was employed, classifying each spectrum as low or high dynamics based on its correlation with known spectra labels. This allowed us to tailor smoothing, clipping, and sigma estimation strategies to each category, improving accuracy.

Finally, we computed the Principal Component Analysis of all resulting spectra. We kept only the most significant components, reducing the global dimensionality. This effectively reduced residual noise and artifacts for a final accuracy boost.

The presentation will delve into the rationale and implementation of these stages, illustrating our journey towards a competitive solution in this challenging competition.