Examining Verification Methods for Evaluation of Cloud Predictions

Cloud predictions are essential for many applications, yet verification of cloud forecasts remains challenging. In particular, evaluation of the ability of gridded cloud predictions to replicate spatial features is not straightforward due to the complex nature of cloud fields (e.g., multiple layers) and the limitations associated with cloud observations (e.g., based on surface stations, active sensors, and . Some of the new spatial verification methods (e.g., object-based, neighborhood, scale separation, image warping) – which previously have primarily been applied to numerical weather prediction (NWP) model forecasts of precipitation – show promise as alternative methods for evaluating cloud predictions. Although these approaches are suitable for application to any variable characterized by coherent spatial features, they have not been widely applied to other fields. In a project collaboration between NCAR and the U.S. Air Force’s 557th Weather Wing, several standard approaches and a number of spatial methods have been examined for regarding their usefulness for evaluating gridded cloud predictions from advection and NWP models. The verification approaches examined include the Method for Object-based Diagnostic Evaluation (MODE); several distance measures; the Structure, Amplitude, and Location (SAL) approach; and image warping. Results of application of several of these approaches to a set of advection and NWP-based cloud amount predictions will be presented and compared; positive and negative aspects associated with application of each method will be discussed.