Evaluating the Cloud Cover Forecast of NCEP Global Forecast System with Satellite Observation

To assess the quality of daily cloud cover forecast generated by the operational global numeric model, the NCEP Global Forecast System (GFS), we compose a large sample with outputs from GFS model and satellite observations from the International Satellite Cloud Climatology Project (ISCCP) in the period of July 2004 to June 2008, to conduct a quantitative and systematic assessment of the performance of a cloud model that covers a relatively long range of time, basic cloud types, and in a global view. The evaluation has revealed the goodness of the model forecast, which further illustrates our completeness on understanding cloud generation mechanism. To quantity the result, we found a remarkably high correlation between the model forecasts and the satellite observations over the entire globe, with mean forecast error less than 15% in most areas. Considering a forecast within 30% difference to the observation to be a one, we find that the probability for the GFS model to make good forecasts varies between 58-73%, which is much better or at least not worse than the forecasts from comparison climatology and random models in most cases. We found that the model has bias tendency for total (~10% too low) and high clouds (5-10% too high), and is about right on low and mid clouds. Additionally, we propose some potential linkages between strong positive/negative biases and background characteristics, including forecast biases and El Nino - Southern Oscillation (ENSO) events, negative bias on total and low clouds versus cold oceanic currents in mid latitude belts, positive bias on high clouds versus presentence of monsoon in equator belts, and incoherent behavior of vertical cloud structure in West Wind Drift belt. Further investigations on these matters are worthy to uncover details of these potential linkages to update our understanding on cloud generation mechanism and help refinement on