A new look at NICMOS transmission spectroscopy of HD189733, GJ-436 and XO-1: no conclusive evidence for molecular features

We present a re-analysis of archival HST/NICMOS transmission spectroscopy of three exoplanet systems; HD 189733, GJ-436 and XO-1. Detections of several molecules, including H20, CH4 and CO2, have been claimed for HD 189733 and XO-1, but similarly sized features are attributed to systematic noise for GJ-436. The data consist of time-series grism spectra covering a planetary transit. After extracting light curves in independent wavelength channels, we use a linear decorrelation technique account for instrumental systematics (which is becoming standard in the field), and measure the planet-to-star radius ratio as a function of wavelength. For HD 189733, the uncertainties in the transmission spectrum are significantly larger than those previously reported. We also find the transmission spectrum is considerably altered when using different out-of-transit orbits to remove the systematics, when some parameters are left out of the decorrelation procedure, or when we perform the decorrelation with quadratic functions rather than linear functions. Given that there is no physical reason to believe the baseline flux should be modelled as a linear function of any particular set of parameters, we interpret this as evidence that the linear decorrelation technique is not a robust method to remove systematic effects from the light curves for each wavelength channel. For XO-1, the parameters measured to decorrelate the light curves would require extrapolation to the in-transit orbit to remove the systematics, and we cannot reproduce the previously reported results. We conclude that the resulting NICMOS transmission spectra are too dependent on the method used to remove systematics to be considered robust detections of molecular species in planetary atmospheres, although the presence of these molecules is not ruled out.