Understanding Errors in Cloud Liquid Water Path Retrievals Derived from CloudSat Path-Integrated Attenuation

An algorithm that derives the nonprecipitating cloud liquid water path W-cld from CloudSat using a surface reference technique (SRT) is presented. The uncertainty characteristics of the SRT are evaluated. It is demonstrated that an accurate analytical formulation for the pixel-scale precision can be derived. The average precision of the SRT is estimated to be 34 g m(-2) at the individual pixel scale; however, precision systematically decreases from around 30 to 40 g m(-2) as cloud fraction varies from 0% to 100%. The retrievals of clear-sky W-cld have a mean bias of 0.9 g m(-2). Output from a large-eddy simulation coupled to a radar simulator shows that an additional bias of -8% may result from nonuniformity within the footprint of cloudy pixels. The retrieval yield for the SRT, measured relative to all warm clouds over ocean between 60 degrees N and 60 degrees S latitude is 43%. The SRT W-cld is compared with one estimate of W-cld from the Moderate Resolution Imaging Spectroradiometer (MODIS) using an adiabatic cloud profile and an effective radius derived from 3.7-mu m reflectance. A strong correlation between the mean MODIS W-cld and SRT W-cld is found across diverse cloud regimes, but with biases in the mean W-cld that are cloud-regime dependent. Overall, the mean bias of the SRT relative to MODIS is -13.1 g m(-2). Systematic underestimates of W-cld by the SRT resulting from nonuniform beamfilling cannot be ruled out as an explanation for the retrieval bias.