18th Conference on Weather and Forecasting, 14th Conference on Numerical Weather Prediction, and Ninth Conference on Mesoscale Processes

Monday, 30 July 2001
A quantitative comparison of MM5 cloud forecasts and GOES cloud analyses
Michael A. Kelly, Litton-TASC, Chantilly, VA; and R. J. Alliss, M. E. Loftus, and J. C. Lefever
Poster PDF (102.2 kB)
Cloud forecasts produced by the Penn State-NCAR mesoscale model (MM5) are compared with cloud analyses derived from Geostationary Operational Environmental Satellite (GOES) 10-bit imagery. The objective of this study is to determine the capability of a high-resolution mesoscale model to produce accurate cloud forecasts for a variety of synoptic and mesoscale conditions. This approach contrasts with the usual practice of limiting in-depth analysis to specific case studies. The approach recognizes that a forecaster needs to know how the model performs on a day-to-day basis over a variety of situations. This work defines the baseline for comparison to simulations, which include assimilation of remotely sensed clouds.

The model was initialized twice per day and run for 24 hours during the month of February 2000. The inner grid was run at 4-km resolution over a 360-km by 360-km region. Convective heating for the inner grid is resolved explicitly by a microphysical parameterization which includes cloud water and ice. Vertically integrated cloud liquid water / ice for each model grid point were compared to the corresponding GOES 4-km pixel cloud assessment. Threat scores and percent correct ratios were used to measure accuracy for each forecast hour during the month-long experiment. In general, simulated values of vertically integrated cloud liquid water and ice paths are reasonable. However, the accuracy of individual cloud forecasts strongly depends on a pre-defined cloud optical depth threshold. Time of day was also an important factor that effected model performance. The model tended to generate clouds at the right time of day, but then failed to dissipate them as quickly as observed. In the absence of synoptic-scale forcing, model runs initiated after the period of maximum diurnal heating often failed to generate realistic cloud fields. Cloud forecasts produced with four different cloud microphysical parameterizations (Reisner, Schultz, NASA Goddard, simple ice) were compared and contrasted. On average, neither the simplest nor the most complex schemes produced the most accurate cloud fields.

Supplementary URL: