SSOF model of order 97 for HD 26965, which separates the two dominant modes of telluric variability. Top: the telluric and stellar templates are shown above in gray and blue while the two telluric feature vectors are shown below in orange and green. For comparison, we also show portions of simulated H₂O (turquoise) and O₂ (pink) spectra. The simulated lines are those whose line depths are >0.1% in typical conditions at Kitt Peak. The line locations and emmissivities are from HITRAN2020 (R. Tolchenov et al. 2005; I. E. Gordon et al. 2011, 2022; S. Yu et al. 2014; T. Furtenbacher et al. 2020). The translucent portions of the feature vectors show the true values, while the solid portion is a version that has been smoothed by the instrumental LSF used during fitting. The orange feature vector captures the time variability of O₂ lines while the green basis vector captures the time variability of H₂O lines. Bottom: The telluric feature scores as a function of time (in modified Julian date, MJD). The scores for the O₂ feature vector (orange) are highly correlated with observation air mass (black) while the scores for the H₂O feature vector (green) are more erratic but clearly identified in multiple orders (see Figure 9). Δχ² is one minus the ratio of the final model χ² to the model χ² if you set all of this feature’s scores to zero. Higher values mean the feature vector is more important and captures more variance. The residuals between v_⋆,97 and ﹩{v}_{\star }^{{\mathtt{SSOF}}}﹩ are shown in Figure 11.