Evaluation of the current Global Energy and Water Cycle Experiment (GEWEX) Continental-Scale International Project (GCIP)/Americas Prediction Project (GAPP) Surface Radiation Budget Product as produced by NOAA/NESDIS operationally, revealed that several aspects of the retrieval methodology related to snow, need to be revisited. These include: improved information on snow extent; ability to detect clouds over snow; and diurnal variability in snow. In this presentation we will review progress made on these issues. Specifically, a Cloud Detection and Snow Mapping Algorithm (CDSMA) has been developed in the framework of a reprocessing activity aimed at producing a homogeneous and improved Surface Radiation Budgets product from the beginning of the GCIP activity.

Improvements have been implemented in several stages. Initially, the NOAA/NESDIS operational daily snow-mapping algorithm has been applied and the resulting snow information was used as input to the CLAVR (Cloud from AVHRR) cloud detection algorithm. The results showed that this approach overestimated the cloud amount over snow. Subsequently, a Modified CLAVR (MCLAVR) algorithm was applied in which a test was added to better identify clouds over snow. Two types of Clear-sky Composite Maps (CCM), one for snow-free and one for snow-covered surfaces were used to re-allocate mixed pixels into clear and cloudy categories. Downward solar radiative fluxes at 0.5x0.5 degree latitude/longitude resolution have been calculated with these inputs and evaluated against SURFRAD stations. For January 1997 and 1998 snow situations over 3 of the SURFRAD stations, hourly root mean square error decreased from 176.30 and 126.75 in the operational version to 85.41 and 83.25 W/m**2 in the reprocessed version, respectively. The correlation coefficient increased from 0.56 and 0.59 to 0.81 and 0.81, respectively. To achieve additional improvemnts the CDSMA development was undertaken where cloud detection is produced in the form of a probability function for each pixel based on three-threshold tests using multi-spectral observations. The thresholds are derived for each pixel under snow-free and snow-covered situations. This approach eliminates the need to re-allocate mixed pixels into clear and cloudy categories using subjective criteria, as was necessary with the products of CLAVR. Snow mapping and cloud masking is combined into a single program for efficiency and is applied on an hourly time scale. Evaluation of this new approach will be done with one year of reprocessed data.