A diagnostic approach for verification of nowcasts

Over the past several years, a number of efforts have worked toward development of verification approaches that provide meaningful information about the quality of spatial forecasts, including high resolution NWP forecasts and convective nowcasts. These approaches focus on the evaluation of meaningful attributes describing forecast performance, rather than the single summary scores that are typically provided by traditional verification approaches. The measurements thus can provide useful information for improving forecasts or for decision making on the basis of the forecasts.

This paper describes an application of the so-called “Object-based” verification approach to the NCAR AutoNowcaster. The AutoNowcaster has been providing one-hour convective forecasts for the 2005 Dallas/Fort Worth AutoNowcaster demonstration project since March 2005, with “forecasters-in-the-loop” (i.e., forecasters at the DFW Weather Forecast Office) entering boundaries and shapes in regions where they expect new convective storms to initiate. The boundaries are then incorporated into the automated algorithm forecasts to create initiation fields and growth and decay fields. One goal of this evaluation is to investigate the benefits of the forecaster-in-the-loop. The object-based approach is well-suited for this evaluation because it allows measurement of a variety of attributes of the forecasts that are of interest to users and forecast developers. For example, the system can separately consider the growth and decay field and the initiation field and can evaluate a variety of basic attributes, including area size, location, intensity, and orientation. The approach allows detection of subtle differences in performance that may not be apparent in standard verification analyses. Results of this evaluation will be presented.

Benefits of the object-based approach for evaluating nowcasts will be discussed, in general, along with the benefits associated with other diagnostic approaches for evaluating nowcasts. Extensions to other variables (e.g., winds) and probabilistic forecasts will also be considered.