Diagnosing Cloudiness from Global Numerical Weather Prediction Model Forecasts.
Final rept. Oct 91-Sep 93,
PHILLIPS LAB HANSCOM AFB MA
Pagination or Media Count:
We investigated the utility of any information derivable from noncloud numerical weather prediction NWP model forecasts in inferring layer cloud amount distributions. This effort involved identifying and preparing a suitable source of the predictand cloud amount, generating and preparing a suitable source of the predictors NWP variables and geographic information, and combining them to form diagnostic relationships in a model output statistics approach. Both AFGWC RTNEPH cloud analyses and Phillips Laboratory Global Spectral Model PL GSM NWP forecasts were rendered on a 125 km equal-area grid in three cloud deck regimes high, middle, low. Two statistical methods CLOUD CURVE ALGORITHM CCA, a univariate method, and multiple linear regression MLR were used to relate the cloud amount to relative humidity CCA and to relative humidity and a large number of other NWP variables MLR. We found that the CCA method preserves the sharpness of the cloud distribution while sacrificing skill, while MLR produced cloud diagnoses that were more skillful but less sharp. The methods fall short of the error level standards established by Air Force requirements, but show potential for useful cloud forecast skill upon further refinement.