Convective Weather Forecasting: Benefits of a "Human over the Loop"
The Aviation Weather Group of the National Weather Service has funded NCAR during the past several years to install and run the Thunderstorm Nowcasting System (Auto–nowcaster or ANC) at the NWS Dallas/Ft.Worth (DFW) Forecast Office (FO). In 2010, a similar installation was completed in Melbourne FL that provides regional coverage overlap to three Air Route Traffic Control Centers (ARTCCs). Two primary objectives of the Forecaster–Over–the–Loop demonstration are to assess 1) the role of the NWS forecaster in providing value–added enhancements to gridded, automated, nowcast products produced by the Auto–Nowcaster, and 2) the usefulness of these products as guidance for the forecaster in producing short–term forecasts, area warning updates and CWSU outlooks for terminal/enroute aviation traffic. The overarching goal is to improve the consistency, reliability, and accuracy of 0–2 hour convective forecast products for automated aviation weather digital products (4–D grids) for the Next Generation Air Transportation System (NextGen).
Figure 1. Comparison of CSI performance of Auto–nowcaster thunderstorm nowcasts with (blue curves) and without (magenta curves) forecaster–input into the system computed for the boxed (green) regions shown. The white polygon is the 60 min nowcast of new storm initiation produced by the Auto–nowcaster. See text for more details.
Previous demonstrations of the ANC system for FAA–funded activities have shown that forecaster input into the ANC process, particularly in producing storm initiation nowcasts, added consistency, reliability and accuracy to the 0–1 hr short term, time and location specific thunderstorm nowcasts. The primary objective this year has been the evaluation of performance of the Auto–nowcaster thunderstorm initiation nowcasts with and without forecaster input into the process. Using standard statistical calculations of probability of detection (POD), false alarm rate (FAR) and critical skill index (CSI) over the whole NWS County Warning Area (CWA) domain, it is difficult to separate out the CSI skill associated with the storm initiation nowcasts, as nowcasts of extrapolated storms generally dominate the statistics over the very large domain where many storms are occurring simultaneously. For this reason, a new approach was taken by Rita Roberts and her team to perform the statistical evaluation of storm initiation nowcasts. The approach was simply to compute the statistics over smaller domain sizes that are more relevant to the scale on which the many different discrete areas of convection are occurring (Figure 1). We have found that this approach provides a more detailed and informative look at nowcast performance and a better scientific understanding of the factors that increase or decrease the accuracy of the nowcast.
Overlaid onto Fig. 1 are the CSI skill scores of ANC performance with (blue) and without (magenta) forecaster–input during the complete event for five of the subset boxes. The CSI scores in the top two plots (Boxes 1,2 and 4,3) show minor differences in skill with and without forecaster input. Within these boxed regions, very little initiation is occurring and most nowcasts are based on the extrapolation of existing storms. The benefit of forecaster involvement is much clearer in the statistics for the other three subset boxes (Boxes 1,1; 2,2; and 3,2). Forecaster–entered convergence boundaries clearly had an impact in the increased accuracy of the nowcasts. Significant new convection initiated in these boxed regions during the 10 hr period spanning this event and forecaster–entered boundaries aided in the timely nowcasts of new convection triggered by these boundaries. Increases of CSI scores ranged from 0.3 – 0.5, a substantial increase in accuracy.
Expanding upon these results an additional step was taken to combine the statistics from the individual sub–grid boxes over the duration of the event to evaluate the overall performance in storm initiation nowcasting when forecaster–entered boundaries are used in determining storm initiation. Results indicate that the use of forecaster–entered boundaries in the forecast process leads to increased POD of storm initiation and increase in overall accuracy of CSI.
Encouraged by these results, the NWS Meteorological Development Laboratory (MDL) has been collaborating with Rita Roberts,Dave Albo and Dan Megenhardt to transfer the components of the NCAR ANC forecaster interactive tools to MDL for inclusion in a prototype AWIPS system running at MDL. Surface convergence boundaries can now be entered by the forecasters on AWIPS. This has 1) increased the ease of forecasters for entering boundaries because of their familiarity with the AWIPS displays and 2) facilitated incorporating this task into the everyday routine of the NWS Short Term Forecaster.