We evaluated a new approach to take clouds into account in ozone profile retrievals from backscattered ultraviolet radiance measurements as performed by the Global Ozone Monitoring Experiment (GOME). In this approach ozone profiles are retrieved using cloud fractions, cloud optical thicknesses and top pressures retrieved from oxygen A-band measurements combined with measurements between 350 nm and 390 nm. This approach (CUVO2), is compared with two commonly used approaches in ozone profile retrievals, namely to treat clouds as an effective ground surface albedo (CaA); and using effective cloud fractions and top pressures retrieved from the oxygen A-band by assuming a cloud optical thickness of 40 (Ceff). Using simulated GOME retrievals we show that the CaA and Ceff approaches lead to significant biases in the mean ozone concentrations of up to -85% and 18% near the surface, respectively. With the CUVO2 approach these errors are reduced to below 3%. Retrievals from 233 GOME measurements using the three approaches were validated with ozonesonde measurements at 5 different locations. For most cases the results are as expected from the simulations. For scenes with strong indications for the presence of inhomogeneous clouds, all studied approaches do not correct for cloud sufficiently. The standard deviation of the differences between retrieved profiles and sonde sonde profiles is about 2.5 times larger in the troposphere than expected from the simulations, which indicates that other error sources than clouds dominate the variation in retrieval and/or validation.
