The use of cloud radar observations for model evaluation: A probabilistic approach

Abstract

Cloud radar observations provide detailed information about the vertical structure of cloud fields. They thereby constitute a highly desirable data source in the evaluation of cloud models and parameterizations. A major difficulty in using the data for that purpose lies in the inherent mismatch in the spatial scales of the observations with that of the model results. A probabilistic approach to overcome the scale mismatch is proposed. It is based on the reinterpretation of a model's spatial cloud distribution as a probabilistic forecast for the occurrence of cloud at any location within the model domain. Standard probabilistic forecast verification measures, such as the Brier score, Relative Operating Characteristics and reliability diagrams are used to evaluate model performance. The approach is first compared to other techniques in an idealized framework and is then applied to the evaluation of cloud predictions made by a Cloud Resolving Model for the ARM Southern Great Plains site in June/July 1997.