Cloud predictions diagnosed from mesoscale weather model forecasts

Abstract

Observed cloud characteristics, such as cloud cover, type, and base and top altitude, are of interest to the U.S. Air Force operational community for mission support. Predictions of such cloud characteristics are useful in support of a variety of mission activities. In this paper, a model output statistics approach to diagnosing these cloud characteristics from a forecast field generated by a mesoscale numerical weather prediction model is presented. Cloud characteristics information obtained from the air force RTNEPH cloud analysis supplied the cloud predictands, and forecast fields from the MM5 mesoscale numerical weather prediction model provided the weather variable predictors. Multiple linear regression (MLR) and multiple discriminant analysis (MDA) were used to develop the predictand-predictor relationships using 10 days of twice-daily cloud analyses and corresponding forecasts over a theater-scale grid. The consequent relationships were then applied to subsequent gridded forecast fields to obtain estimates of the distribution of the cloud characteristics at the forecast times. The methods used the most recent 10 days of cloud analyses and weather forecasts to develop the relationship for each successive application day. The gridded cloud characteristics were diagnosed for 10 days in each of January and July of 1992 over a theater-scale region in southern Europe. The resulting diagnosed cloud predictions were verified against the RTNEPH analyses for forecast durations of 6-36 h at 6-h intervals. It is found that both the MLR and the MDA methods produced small mean errors in all the cloud variables. When compared with persistence, MLR showed skill in rmse in January, while MDA did not. On the other hand, MDA obtained a better score than MLR in percent diagnosed in the correct cloud amount category. Furthermore, the category selection method used with the MDA scheme effectively reproduced the cloud variables' category frequency distribution compared with that of the verification data, while MLR did not. In July, both methods showed skill with respect to persistence in cloud amount. Verification results for cloud type, base altitude, and thickness did not show appreciable skill with respect to persistence. Cloud-ceiling altitude diagnoses showed consistent skill compared to persistence for both methods in both months. Visual depictions of the total cloud amount distribution as diagnosed by the methods showed that the MDA algorithm is capable of generating useful cloud prediction products. The images produced by the MLR scheme had unrealistically flat gradients of total cloud amount and too many occurrences of partly cloudy skies. The multiple discriminant analysis scheme is considered to be a useful short-term solution to the U.S. Air Force need for predictions of cloud characteristics in theater-scale areas.