Neil I. Fox*, David M. Jankowski, Elizabeth Hatter and Elizabeth Heiberg University of Missouri - Columbia, Columbia, Missouri, USA

Transcription

1 5.10 FORECASTING STORM DURATION Neil I. Fox*, David M. Jankowski, Elizabeth Hatter and Elizabeth Heiberg University of Missouri - Columbia, Columbia, Missouri, USA 1. INTRODUCTION The approach to severe weather warning using radar-based algorithms generally involves locating the center of a storm cell on successive images and deriving a propagation vector to indicate the predicted motion of the storm. This methodology has been successful in forecasting the onset of severe weather, but for flash flood forecasting the duration of heavy precipitation is of interest, and to forecast this it is the motion of the trailing edge of the storm that appears critical. Collier and Fox (2003) and subsequently Hatter (2004), investigated the meteorological and hydrological components of flash flood events. In each case the findings indicated that, of the many factors required to identify threatening situations, storm velocity and duration were the least well understood and quantified. There have been several studies examining the relationship between storm motion and storm propagation from the development of new cells within the storm system, but none have looked explicitly at the issue of forecasting of storm duration, which could be expected to be well correlated with total storm precipitation. Further to this, event organizers and emergency services are often concerned with the time that precipitation will end. This paper addresses two issues: the first is an exploration of the difference in velocity between the centroid track usually calculated and used for warning purposes and the velocity of the trailing edge. The second part of the paper is an investigation of the effectiveness of a measurement of the velocity of the trailing edge of the storm as a flash flood potential indicator by examining correlations of storm velocity measures and precipitation totals. * Corresponding author address: Neil I. Fox, Univ. of Missouri - Columbia, Dept. of Soil, Environmental and Atmospheric Sciences, 332 ABNR Building Columbia, MO 65211; foxn@missouri.edu. The current storm-cell identification and tracking algorithm (SCIT) in the WSR-88D radar systems tracks the storm centroid. A linear least-squares method is used to determine movement using current and past mass-weighted centroid location (Johnson et al. 1998). In contrast to work that deals with the motion of MCCs (e.g., Corfidi et al. 1996; Chappell 1986; Collier and Fox 2003), the SCIT is concerned with the motion of the individual storm cells that make up the cluster. This is because severe weather is associated with individual cells, and the aim is to forecast the motion of these smaller areas where the severe weather threat is found. While the SCIT algorithm is very useful in determining the motion of severe weather events associated with individual cells (tornadoes, large hail) it is less useful when trying to forecast flash floods associated with an entire cluster of cells. Because the cells can be moving relatively quickly in comparison to the storm propagation as a whole, the cell motion predicted by the storm tracking algorithm can be misleading if one is trying to determine how long it will rain over a particular area. It would be more beneficial to take into account storm motion and propagation as a whole, rather than individual storm cell motions. Other short-period forecast systems (e.g. TITAN: Dixon and Wiener 1993) have similar approaches to storm tracking. 2. STORM DURATION PREDICTION A storm center traveling with a velocity v c will reach a location at a distance, D, in a time, t c, given by D/v c. Similarly the trailing edge will reach the same location at a time (D + D)/v r, where v r is the velocity of the rear edge of the storm and D is the distance between the centre and the rear edge of the storm. Making the crude assumption that D >> D, then the predicted duration (T) of the storm over the point at a distance, D, from the center of the storm is given by:

2 1 T = D vr 1 vc vr D v = c vcvr Here we define (v c v r)/v cv r as the storm duration factor (SDF) and this is expected to be a clear measure of the expected duration of heavy precipitation over a point in the storm s future track. It is easy to see that this parameter tends to 0 when the two velocities are both large and similar (fast moving storm), but is not quite so small when they are small and similar (slow moving storm). A storm which has redevelopment on the trailing edge will have a reasonable v c, but a small v r, leading to a large value of the parameter and this is when we expect the storm to persist over a given area. In (1) we also consider only the magnitudes of the velocities. In reality the propagation of the center and rear edge of the storm will most often proceed in different directions. For convenience the SDF ignores this. Practically, this is consistent with the concept of an objective measure of flash flood threat rather than an explicit precipitation forecast. Physically one would not expect the directions of these two vectors to diverge greatly, and given that the storm does have width, and will not move in a linear fashion, it seems reasonable to take the SDF as a good measure. How good a measure is what this work set out to establish. The SCIT analyzed centroid velocities are used for v c. This algorithm often re-detects storm cells as new entities, leading to jumps in diagnosed cell velocity or two velocities for what would be considered a single storm for the purposes of precipitation forecasting. By using the fastest velocity of those that are available for a cluster of cells (that appear as a continuous rainfall area), forward propagation of the storm by development on the leading edge of the storm is accounted for, and one determines the earliest time that rain will fall at the location of interest. v r is found by locating the trailing edge on subsequent scans. In order to reduce errors in diagnosed velocities the difference in position of the rear edge was found at 12-minute (for the 6-minute radar scan cycle) or 15-minute (for the 5-minute scan cycle) intervals. The trailing edge is found by tracing the diagnosed centroid velocity vector backwards until the observed reflectivity falls below a threshold value of 30 dbz. Again, by ignoring storm-cell continuity this (1) methodology implicitly accounts for redevelopment on the rear flank and possible training of storms. Even negative values of v r are permitted. Such values would lead to negative values of the SDF which could equally result from v r > v c. This latter condition may occur if the cell is dying out, or if the velocities are similar but have slight errors. 3. DATA For this study, radar data was examined from a number of events of different types and from a variety of areas of the US. Level II Nexrad data was acquired from the NCDC archive and processed using WATADS. Using the available NSSL algorithm suite SCIT tracks and cell tables were used to assess the centroid velocity at each time step. One-hour rainfall accumulations for the hour subsequent to the scan analyzed were also available and used to correlate the SDF for a point 25km ahead of the storm center. Other distances were also considered, but the results are not presented here. The first case occurred on October 1998, near Kansas City and St. Joseph, Missouri. Thunderstorms produced extremely heavy rain, which led to severe flash floods over west-central Missouri during this event. The thunderstorms were relatively slow moving and convection continued to redevelop over the same areas. These factors allowed for heavy precipitation to fall for an extended period of time. Radar data from the Kansas City, MO radar (KEAX) was used. Precipitation totals varied widely over the area. Gauges at Kansas City, MO and St. Joseph, MO, recorded totals of mm and mm, respectively. In the storm event reports, however, much heavier amounts were recorded in the area. Some areas received as much as 76.2 mm to 127 mm in 2 hours and the highest 24-hour rainfall totals reported were mm. Data was used from the St. Louis, MO radar (KLSX) from the event of 07 May 2000 over eastcentral Missouri. Thunderstorms produced extreme flash flooding over east central Missouri during the overnight and early morning hours of 6-7 May Glass et al. (2001) described the conditions before the event as seemingly benign, however, continual convective development led to a nearly stationary system for several hours. Over 305 mm of rain fell

3 over a fairly broad span of Franklin County Missouri, with unofficial reports of up to 406 mm west of Union, MO (Glass et al. 2001). According to the event record for the storm, the rain fell at a rate of 76.2 mm per hour from about 0800 to 0900UTC. An example image from this case is shown in Figure 1. The storms were moving from SW to NE and the magenta line illustrates the diagnosed rear edge of the storm of interest. This shows the redevelopment at the rear edge and potential training in this case. they did contribute to the overall persistence of the flood threat by maintaining soil moisture levels and river levels. Figure 3 is an example of a typical image from July 1 st, 2002 from the San Antonio radar. Although there is structure in the distribution of cells, as the motion was from SE to NW each cell is isolated and does not show signs of training. Figure 2: Example image of storm structure from San Antonio radar on July 1 st, RESULTS 4.1 Rear edge vs. centroid velocity Figure 1: Example image of storm structure from St. Louis radar on May 7 th, Another case included in this study used data from the Huntsville, AL (KHTX) radar on 6 May This was also a severe flooding event enacted by repeated redevelopment and propagation of cells over similar tracks. In the case of training storms the rear edge identification will often produce very slow motion vectors in comparison with the centroid motion vector, especially if there is no real gap between the cell as it moves away and the developing cell at the rear. Data was taken from the San Antonio / Austin, TX (KEWX) radar for an extended period of rainfall events from 28 June 9 July This comprised a whole series of discrete rainfall episodes which resulted from a variety of mechanisms and storm types. Some storms propagated onshore from the Gulf of Mexico, while others developed over the elevated topography west of the area and moved eastward over the same region. The result was a series of heavy precipitation events over a number of days that produced a number of severe floods and flash floods. However, many of the storms did not, of themselves, produce floods, Comparisons of rear edge velocities and those of the centroid found by the SCIT algorithm are shown in the scatterplot in Figure 3. This is composed of 356 points, the majority of which (240) are from the extended San Antonio case. While there are many points that lie close to the line of equality, there are a significant number of points showing the rear edge propagating significantly more slowly than the centroid. As was seen in the reflectivity imagery, and illustrated by the scattergram, the cases from the San Antonio (blue crosses) had storms that were mostly isolated cells that propagated without significant development. In these cases the cells had a fairly constant shape and the two velocities were similar. The data from the Missouri cases (green crosses) show a larger difference with a large number of the cells having much slower rear edge velocities. The two cases represented by these were both significant, damaging flash floods caused by repeated redevelopment and training.

4 The data from the Huntsville case (red crosses) show a mixture of cells with slower rear edges and those that move at approximately the same speed as the centroid. Although the cells were similar to those seen in the Missouri cases there was sufficient separation between existing cells and those that developed to their rear that redevelopment resulted in cells that were identified as new rather than rearward extensions of the existing cell. not others. Therefore, the intention is to continue analyzing data and separate data by some objective assessment of storm type. Figure 4: One-hour precipitation accumulation at 25km versus 1/v c. Figure 3: Comparison between centroid and rear edge velocities. Points depicted in blue are from the San Antonio case study, those in green are from the St. Louis and Kansas City cases, and those in red from the Huntsville radar. The solid line is that of equality. 4.2 Comparisons with precipitation totals In order to determine what measure of storm velocity provided the best indicator of future precipitation graphs were prepared that plotted 1/v c, 1/v r, and the SDF against one hour precipitation totals from the forecast time T = 0 to T + 60 minutes at a distance of 25km ahead of the centroid location. These are shown in figures 4, 5 and 6. None of these graphs provide a clear indication that there is a reliable relationship between the parameters. Despite the clustering of points close to the zero precipitation axis there is some sign in figure 6 that the precipitation accumulation increases with increasing SDF. The number of points clustered close together, and close to both axes, makes interpretation difficult. There is some sign that the precipitation accumulations increase with the values of each velocity measure. Some of the problems in determining relationships may be due to the range of storm types incorporated into each graph. It may be that the analysis would be useful for some types of storms and Figure 5: One-hour precipitation accumulation at 25km versus 1/v c. Figure 6: One-hour precipitation accumulation at 25km versus the SDF. 5. DISCUSSION AND NEXT STEPS Although there clearly appears to be a difference between the diagnosed centroid motion and that of the rear edge, there is little evidence that the simple use of the differential velocity provides a good

5 prognostic indication of imminent rainfall totals. Two types of cases have been observed. There is one where genuine back-building occurs and the storm has a very slow rear edge velocity. In these cases there is some evidence that this approach has some value, but more of these cases need to be examined to ascertain accurate results. The next steps in the investigation are to redo the analyses accounting for cell size. This moves away from the range independent measure of SDF presented in equation (1), as it necessitates the use of the following formulation: 1+ D = D 1 T D v r v c This now depends on the relative distance of the area in question to the size of the storm cell. It therefore also makes sense to study this without reference to a fixed value of D, but to a fixed surface location (such as a rain gauge site). These modifications will form the basis of the next phase of these studies. The use of rain gauge data will also permit the comparison of rear edge velocity and precipitation end time. (2) Glass, F.H., J.P. Gagan, and J.T. Moore, 2001:The extreme east-central Missouri flash flood of 6-7 May Preprints, Precipitation Extremes: Prediction, Impacts, and Reponses, Albuquerque, NM. Amer. Meteor. Soc., Hatter, E.A., 2004: Using Radar and Hydrologic Data to Improve Forecasts of Flash Floods in Missouri. MS Thesis, University of Missouri. Johnson, J.T., P.L. MacKeen, A. Witt, E.D. Mitchell, G.J. Stumpf, M.D. Eilts, and K.W. Thomas, 1998: The storm cell identification and tracking algorithm: An enhanced WSR-88D algorithm. Wea. Forecasting, 13, ACKNOWLEDGEMENTS This work was funded partly under a COMET Partners Project no S and by the University of Missouri Research Council. REFERENCES Chappell, D. F., 1986: Quasi-stationary convective events. Mesoscale Meteorology and Forecasting, P. S. Ray, Ed. Amer. Meteor. Soc., Collier, C.G. and N.I. Fox, 2003: Assessing the flooding susceptibility of river catchments to extreme rainfall in the United Kingdom. Int. J. River Basin Man. 1 (3). Corfidi, S.F., J.H. Merritt, and J.M. Fritsch, 1996: Predicting the movement of mesoscale convective complexes. Wea. Forecasting, 11, Dixon, M. and G. Weiner, 1993: TITAN: Thunderstorm Identification, Tracking Analysis and Nowcasting A radar based methodology. J. Atmos. Oceanic. Technol., 10,