Extended Data Fig. 3: Illustration of the baseline determination and column density derivation methods.

a) The baseline derivation method for the 3 μm absorption feature of H₂O. Data are in black, the mean derived baseline in red, the 1σ baseline errors in dark red and the conditioning points for the Gaussian Process Regression model as blue crosses. b(i): The baseline determination method for F322W2 spectra with partial wavelength coverage of the 3 μm absorption feature of H₂O. On a flux scale, a template spectrum (red) is fitted to the data range in orange to derive the continuum (blue). b(ii): On the optical depth scale, the flux-space-fitted template spectrum (blue) is fitted to determine the column density. c) the column density derivation method for the CO₂ absorption feature at 4.27 μm. The wings of a saturated CO₂ band (red) are multiplied by the α ratio (derived from unsaturated bands) to determine its column density. The inset in panel C shows a typical unsaturated CO₂ feature, one of eight averaged to determine the ratio α. The total CO₂ area (purple) is divided by the area under the wings (red) to derive an average α = 7.62 ± 0.96. Note that in panels b(ii) and c the y axis scale is not inverted, as would be usual in the optical depth space. This choice visually highlights those panels containing data on an optical depth scale.