We present a novel physics-based retrieval method to infer total column mixing ratios of methane (XCH4) and carbon dioxide (XCO2) from space-borne short-wavelength infrared (SWIR) Earth radiance observations over the cloud-covered ocean. In nadir observing geometry in the SWIR spectral range, backscattering at the ocean surface is negligible. Hence, space-borne radiance measurements of ocean scenes generally do not provide sufficient level to retrieve XCO2 and XCH4. Our approach specifically targets cloudy GOSAT ocean soundings to provide sufficient radiance signal in nadir soundings in ocean areas. Currently, exploiting space-borne SWIR soundings over oceans relies on soundings in Sun glint geometry, observing the specular solar reflection at the ocean surface. The glint observation mode requires cloud-free conditions and a suitable observation geometry, severely limiting their number and geographical coverage. The proposed method is based on the existing RemoTeC algorithm that is extensively used to retrieve CH4 and CO2 columns from GOSAT SWIR measurements over land. For ocean pixels, we describe light scattering by clouds and aerosols by a single-layer water cloud with Gaussian height distribution. We infer the height and the geometrical thickness of the cloud layer jointly with the droplet size and the number density and the column abundances of CO2, CH4, and H2O. The CO2 and CH4 column product is validated with ground-based total column measurements performed at eight stations from the TCCON network that are geographically close to an ocean coastline. For the TCCON site with the most robust statistics (Lauder, New Zealand), we find a retrieval bias of 0.36% for XCH4 combined with a standard deviation of retrieval errors of 1.12%. For XCO2, the bias is 0.51% combined with a standard deviation of 1.03%. Averaged over all TCCON sites, our retrievals are biased -0.01% for XCO2 and -0.32% for XCH4. The standard deviation of station biases amounts to 0.45% for XCO2 and 0.35% for XCH4.
