The interpretation of HST transmission spectroscopy signals has recently been the subject of much debate, in particular the NIR NICMOS data of HD 189733. At optical wavelengths, a high-altitude haze has been confirmed with both STIS and ACS, whereas the presence of molecules has been claimed with NICMOS. However, this detection of molecules has been disputed based on the ad hoc model used to remove the systematics, the choice of which changes the interpretation of the transmission signal. Here, we introduce a powerful new technique, Gaussian Processes (GPs), to model the systematics and simultaneously extract the transmission spectrum, and demonstrate its application to the NICMOS data. GPs are a Bayesian technique widely used in the machine learning community, which allow us to define a distribution over functions. Rather than impose a strict, functional form of systematics correction, we marginalise over potentially infinite numbers of basis functions, effectively inferring the form of the systematics correction from the data itself. This results in a more robust interpretation of the signal. We also present similar analyses of HST/WFC3 observations of HD 189733, which bridge the gap between the current optical and NIR spectrum.
|Title of host publication||AAS Meeting #231|
|Publication status||Published - 01 Sep 2011|