Context. Ground-based transit observations are affected by both telluric absorption and instrumental systematics, which can affect the final retrieved transmission spectrum of an exoplanet. To account for these effects, a better understanding of the impact of different data analyses is needed to improve the accuracy of the retrieved transmission spectra.
Aims: We propose validating ground-based low-resolution transmission spectroscopy using transiting white dwarfs. These targets are selected to have transit parameters comparable with typical transiting hot Jupiters but nondetectable transmission signals due to their extremely high surface gravities. The advantage here is that we know beforehand what the final transmission spectrum should be: a featureless flat spectrum.
Methods: We analyzed two transiting white dwarfs analogous to hot Jupiters, KIC 10657664B and KIC 9164561B. We used various noise models to account for the systematic noise in their spectroscopic light curves following common procedures of transmission spectroscopy analyses. We compared the derived transmission spectra with the broadband transit depth to determine whether there are any artificial offsets or spectral features arising from light-curve fitting.
Results: The results show a strong model dependence, and the transmission spectra exhibit considerable discrepancies when they are computed with different noise models, different reference stars, and different common-mode removal methods. Nonetheless, we can still derive relatively accurate transmission spectra based on a Bayesian model comparison.
Conclusions: With current ground-based instrumentation, the systematics in transit light curves can easily contaminate a transmission spectrum, introducing a general offset or some spurious spectral features and thus leading to a biased interpretation on the planetary atmosphere. Therefore, we suggest that any wiggle within the measurement errors in a transmission spectrum should be interpreted with caution. It is necessary to determine the dependence of results on the adopted noise model through model comparison. The model inferences should be examined through multiple observations and different instruments.