Publications

The FAA requires short-term forecasts of the development and motion of high reflectivity regions to plan for weather avoidance in the en route and terminal areas. Specific needs include choice of air routes and anticipating when to open or close approach/departure gates, descent corridors, and runways. This report compares storm-tracking algorithms for making short-term (0-30 minute) forecasts of high reflectivity areas, to serve these air traffic control needs. The area forecasts are made by moving the key features of the current reflectivity map according to the velocities derived from the storm trackers. The NEXRAD centroid, correlation, and Crane peak-cell trackers are compared against themselves, persistence, and a best-fit extrapolation. Two performance measures are used: (a) overlap of predicted versus actual areas (b) accuracy in flight-path choice. The second method is a new way of scoring the predictor performance and is particularly suited to aviation needs. Five storms are considered, three in Massachusetts and two in Oklahoma. The correlation and peak-cell trackers generally performed well in the Massachusetts storms, close to a best correlation fit extrapolator. The centroid tracker behaves erratically, due to contour merging and splitting. The centroid tracker performed well on compact, Oklahoma storms where the correlation and peak-cell trackers were misled by storm propagation, an effect to be expected when there is high vertical shear of the horizontal wind. It is recommended that either the correlation or centroid tracker be used, depending on the type of storm expected. The centroid tracker would be used on compact storms; the correlation tracker would be used on storms without substantial propagation. The forecasts appear to be skillful in predicting high-reflectivity areas; however, they are less skillful in anticipating flight-paths which do not intersect these areas. Inclusion of forecasts of storm growth and decay will probably be required to improve the performance; anticipating growth and decay will also be important for forecasts of greater than 30 minutes.

Airspace utilization and safety could benefit significantly from accurate, real-time, short-term predictions of hazardous weather regions (e.g., 5-30 minutes). For some hazards, such as heavy turbulence, the detection process itself is in an immature stage. No universally accepted algorithm exists for indicating the regions of current turbulence - let alone predicting it. For other hazards, such as hail and more particularly for heavy rain, the detection process is in a more mature state. In fact heavy rain may be unambiguously associated with high dBZ (reflectivity), if no ice phases are present. Hail is also associated with high reflectivities. We have therefore chosen to place our initial emphasis on the prediction of reflectivity contours in the context of ATC (air traffic control) operations. For all or our prediction techniques, we begin by collecting fixed dBZ-level contours on a fixed-elevation scan by fixed-elevation scan basis, and then combining these elevation cell slices into volume cells as is done in the algorithm of Bjerkaas and Forsyth (1980). To these volume cells we attach translations vectors to make the desired prediction: at this time no provision is made for the growth or decay of reflectivity cells. We generate our translation vectors using each of several algorithms which have already been described elsewhere. Firstly, we use the centroid-tracking approach of Bjerkaas and Forsyth (1980). This is the current tracker of choice in the NEXRAD (Next Generation Weather Radar) program. Secondly, we use tracking vectors of clusters of volume cells, as described ny Crane (1979): much of this work was performed under the sponsorship of the Federal Aviation Administration (FAA). Thirdly, we generate translation vectors by cross-correlating low-altitude (0-4 cm) CAPPIs (constant-altitude plan position indicators): this correlation is done either for the entire storm, or for 30 km by 30 km segments of the storm. This approach has been motivated by the work of Rinehart and Garvey (1978), although we generally use a CAPPI of liquid water content. Fourthly, we use as a prediction the current, composite reflectivity map - our so-called status-quo prediction.