Short-Range Cloud Amount Forecasting With Model Output Statistics

摘要

TASC conducted a feasibility study of the utility of short-range (3-, 6-, and 9-hour) forecasts of total cloud amount and probability of cloud cover categories using the Model Output Statistics (MOS) approach. The MOS models were built using output from the Air Force Global Spectral Model (GSM) and the Real-Time NEPHanalysis (RTNEPH) model. The RTNEPH total cloud amount fields were also used as ground truth. Models were developed on the eighth-mesh RTNEPH grid (approximately 40 km resolution) for 71 data sets covering a range of locations (tropics and midlatitudes), seasons, and times of day. Results were compared with persistence forecasts using the Brier score, 20/20 score, and sharpness measure. The MOS-based forecasts outperformed persistence with respect to the overall reduction in mean square error, but forecast sharpness was sacrificed. Estimates for future model performance for total cloud cover and probabilities of cloud cover categories were determined for two sample cases. Considering the short data records used to develop the equations, the MOS-based forecast models were remarkably robust in most cases. Recommendations for model improvements and additional performance assessment included: using higher spatial resolution model data that better match the resolution of the eighth-mesh forecasts, using cloud layer variables in model development, developing regression coefficients using additional seasons of data, using non-linear models, and continuing model evaluation using independent cloud data. Model output statistics, Cloud forecasting, Statistical forecasting.