VLT/FORS2 comparative transmission spectroscopy II: confirmation of a cloud-deck and Rayleigh scattering in WASP-31b, but no potassium?

We present transmission spectroscopy of the hot-Jupiter WASP-31b using FORS2 on the VLT during two primary transits. The observations cover a wavelength range of $\approx$400-840nm. The light curves are corrupted by significant systematics, but these were to first order invariant with wavelength and could be removed using a common-mode correction derived from the white light curves. We reach a precision in the transit depth of $\approx$140 ppm in 15 nm bins, although the precision varies significantly over the wavelength range. Our FORS2 observations confirm the cloud-deck previously inferred using HST/STIS. We also re-analyse the HST/STIS data using a Gaussian process model, finding excellent agreement with earlier measurements. We reproduce the Rayleigh scattering signature at short wavelengths ($\lesssim$5300 $\AA$) and the cloud-deck at longer wavelengths. However, our FORS2 observations appear to rule out the large potassium feature previously detected using STIS, yet it is recovered from the HST/STIS data, although with reduced amplitude and significance ($\approx$2.5$\sigma$). The discrepancy between our results and the earlier STIS detection of potassium ($\approx$4.3$\sigma$) is either a result of telluric contamination of the ground-based observations, or an underestimate of the uncertainties for narrow-band features in HST/STIS when using linear basis models to account for the systematics. Our results further demonstrate the use of ground-based multi-object spectrographs for the study of exoplanet atmospheres, and highlight the need for caution in our interpretation of narrow-band features in low-resolution spectra of hot-Jupiters.