Accounting for the Effects of Unresolved Variability in the Estimation of Cloud Properties from Reflected Solar Radiation

Clouds are an important but poorly understood component of the climate system. Observations from passive meterological satellite imagers can provide insights into cloud processes and cloud-radiation interactions. Geostationary satellites fully resolve the diurnal cycle, and thus allow to monitor the development of clouds throughout the day. However, their spatial resolution lags significantly behind that of polar-orbiting satellites. To date, the most advanced geostationary imager is the SEVIRI sensor onboard the current Meteosat satellites, with 3x3km2 nadir resolution versus 1x1km2 or higher for the MODIS and AVHRR sensors. Current methods for estimating cloud optical thickness, effective radius and water path are highly non-linear and very sensitive to uncertainties in both forward model and observations, which can result in large systematic and random errors of retrieved cloud properties. A point of particular concern is the plane-parallel albedo bias, which introduces a significant dependence on sensor resolution. To address this point, we propose to include information on unresolved variability in retrievals by using the so-called unscented transform. This method is able to establish the link between the mean and covariance matrix of reflectances on the one side, and cloud properties on the other side. The accuracy of this method is quantified using MODIS data at its original resolution of 1x1km2 versus a reduced resolution of 3x3km2. It is shown that both scatter and biases in retrieved cloud properties can be substantially reduced. The influence of the correlation of cloud properties on cloud radiative effects is studied. Finally, options for estimating the required information on unresolved variability are discussed. A particularly promising option is the use of the broadband high-resolution visible channel of the SEVIRI sensor.