Multiyear Evaluations of a Cloud Model Using ARM Data

This work uses long-term lidar and radar retrievals of the vertical structure of cloud at the Atmospheric Radiation Measurement (ARM) program’s Southern Great Plains site to evaluate cloud occurrence in multiyear runs of a cloud system–resolving model in three configurations of varying resolution and sophistication. The model is nudged to remain near the observed thermodynamic state and model fields are processed to mimic the operation of the observing system. The model’s skill in predicting cloud occurrence is evaluated using both traditional performance measures that assume ergodicity and probabilistic measures that do not require temporal averaging of the observations.

The model shows considerable skill in predicting cloud occurrence when its thermodynamic state is close to that observed. The overall bias in modeled cloud occurrence is relatively small in all model runs, suggesting that this field is relatively well calibrated. The Brier scores attained by all configurations also suggest considerable model skill. Greater differences in performance are found between seasons than between model configurations during the same season, despite substantial differences between the computational costs of the configurations. Several significant seasonal dependencies are identified, most notably greater conditional bias, but better timing, of boundary layer cloud in winter, and substantially less conditional bias in high cloud during summer.