Abstract

A conceptual model is presented for developing a new tool for nowcasting severe thunderstorms using existing operational data. Selected output from two operational, automated, weather detection and forecasting systems have been combined together within a fuzzy logic–based, data fusion system to test the concept and produce 15-min nowcasts of severe weather. The NCAR Auto-Nowcast System provides information and nowcasts on the evolving boundary layer and storm initiation, growth, and decay. The National Severe Storms Laboratory Warning Decision Support System (WDSS) identifies severe weather attributes within storms and provides storm-centric and specific detections of strong winds, mesocyclones, tornadoes, and probabilities of hail and severe hail. A modified version of the Auto-Nowcast System is employed as the engine for combining the Auto-Nowcast gridded output with the object-based WDSS output. Severe thunderstorm nowcasts are compared with available spotter reports for a multicellular, hail-producing squall-line event and a tornadic supercell event. Proof of concept is demonstrated and the results are encouraging as some skill is observed with the 15-min nowcasts. Many challenges still exist in producing a robust tool and these challenges are discussed.

1. Introduction

Two of the strategic goals of the National Weather Service (NWS) are to 1) improve short-term probabilistic and severe thunderstorm forecasting and 2) to increase the lead time on warning the public of the potential for severe weather (Smith et al. 1998a). The challenges for forecasters are to identify the storms that will become severe and determine the communities and rural areas that will be impacted by severe storms within the next hour. Severe storms include those that produce large hail (>1.91 cm in diameter), tornadoes or strong surface wind (>25 m s−1). In this paper, we present and test a concept for merging the capabilities of two operational forecast guidance systems to produce a new guidance tool for the prediction and warning of storm intensification and severity. Specific attributes of the National Severe Storms Laboratory (NSSL) Warning Decision Support System (WDSS; Johnson et al. 2000) and the National Center for Atmospheric Research (NCAR) Auto-Nowcast System (hereafter referred to as the ANC system; Mueller et al. 2003) are combined in a fuzzy logic–based, data fusion system to produce short-term, time- and location-specific nowcasts of severe storms.

To facilitate the achievement of the NWS goals, the System for Convection Analysis and Nowcasting (SCAN) project (Smith et al. 1998b), a collaborative effort between the NWS, NSSL, and NCAR, was instituted to design a path for implementing state-of-the-art detection and forecast guidance tools into the NWS modernized Automated Weather Interactive Processing System (AWIPS). SCAN ran during the summers of 1997–2000 at the Washington, D.C.–Baltimore, Maryland, Weather Forecast Office (WFO) in Sterling, Virginia. Both the ANC system, which provides short-term, time- and place-specific nowcasts (i.e., 0–1-h forecasts) of thunderstorms, and the WDSS, which provides automatic detection of severe storm attributes, ran at the Sterling WFO office. Guidance products from both systems were displayed on workstations located next to the AWIPS displays at the severe weather forecast desk. Neither system produced storm severity nowcasts.

The WDSS and ANC systems also ran at the Bureau of Meteorology (BOM) weather forecast office in Sydney, New South Wales, Australia, as part of the Sydney 2000 Forecast Demonstration Project (FDP; Keenan et al. 2003) that was sanctioned by the World Weather Research Program. The goal of this FDP was to demonstrate state-of-the-art nowcasting systems from around the world in an operational setting and to quantify the benefits of real-time nowcasting systems. All of the demonstration systems ran independent of each other and products were displayed separately on a common Web page (Keenan et al. 2003). The FDP was designed to allow a small set of products from all of these systems to be combined into a forecaster warning guidance tool (Bally 2004). It was this tool that the forecasters monitored for guidance on nowcasting severe weather. Sills et al. (2004), in their study of a severe weather event during the Sydney FDP, iterated the need to improve capabilities for detecting and forecasting low-level convergence boundary development and propagation in order to improve prediction of severe weather. Assessments of forecaster interaction with the Sydney 2000 FDP systems provided by Anderson-Berry et al. (2004) and Keenan et al. (2003) document the need of the forecasters to have accurate, timely, and clearly rendered information on storms that have the potential to become severe.

In section 2 we describe the specific attributes and products from the WDSS and ANC systems that lend themselves to storm severity nowcasting and demonstrate the concept of combining these features to produce storm severity nowcasts. We test the concept using data from Sterling, Virginia, and Sydney, Australia, and present nowcasting results in section 3. Implications of this study for future endeavors are discussed in section 4.

2. Existing operational tools

a. WDSS

The WDSS is composed of several severe weather [mesocyclone, tornadic vortex signature (TVS), hail, downburst] detection algorithms that ingest primarily Weather Surveillance Radar-1988 Doppler data, along with Geostationary Operational Environmental Satellite imagery and numerical model output (Johnson et al. 2000). The output from the different WDSS algorithms (listed in Table 1) is generally in the form of a single point of information per storm characterizing a particular storm severity attribute, for example, the storm-relative location and intensity of mesocyclonic shear. Two of the WDSS algorithms provide probabilities that storms contain hail or severe hail. These probabilities represent the uncertainty of occurrence for hail within storms at the current volume time; they are not a prognostic outlook for hail-producing storms. Output fields from the WDSS algorithms are of particular interest for inclusion when developing a system that nowcasts the intensification and severity of storms. The WDSS also runs a Storm Cell Identification and Tracking algorithm (SCIT; Johnson et al. 1998) that predicts future storm location using a centroid-based, storm extrapolation technique. WDSS also makes use of version 2 of the Rapid Update Cycle (RUC-2) numerical model output to characterize the near-storm environment (NSE) parameters, for example, the storm-relative helicity.

Table 1.

A list of WDSS severe weather detection and prediction algorithms [from Johnson et al. (1998) unless otherwise noted].

A list of WDSS severe weather detection and prediction algorithms [from Johnson et al. (1998) unless otherwise noted].
A list of WDSS severe weather detection and prediction algorithms [from Johnson et al. (1998) unless otherwise noted].

Forecasters can view the output from these algorithms as either a symbolic graphical overlay onto radar data, in time–height profiles, or in the form of a cell-attributes table that ranks storms by severity, that is, by the presence and intensity of severe weather attributes within cells. The WDSS display shown in Fig. 1 illustrates all of these attributes. The WDSS has run at selected NWS WFOs and the cell-attribute table has been included as part of the AWIPS 5.0/SCAN 2.0 build.

Fig. 1.

Graphical display of 2256 UTC 2 Jun 2000 data from Sterling, VA, on the NSSL WDSS. Components of the display include (a) a storm cell attribute table; (b) radar reflectivity imagery with multiple overlay capabilities including color-coded, storm cell ID numbers (boxed), storm tracks (magenta), and county locations (white); and (c) time series plots for a selection of algorithm output fields.

Fig. 1.

Graphical display of 2256 UTC 2 Jun 2000 data from Sterling, VA, on the NSSL WDSS. Components of the display include (a) a storm cell attribute table; (b) radar reflectivity imagery with multiple overlay capabilities including color-coded, storm cell ID numbers (boxed), storm tracks (magenta), and county locations (white); and (c) time series plots for a selection of algorithm output fields.

b. ANC system

The ANC system is a data fusion system composed of several feature detection algorithms that ingest all available operational datasets and provide automated detection of features and precursor signatures relevant to thunderstorm initiation, growth, and decay. A key aspect of the ANC system is its ability to identify (automatically or via human insertion) low-level convergence boundaries, such as thunderstorm outflows, synoptic fronts, and sea-breeze fronts that trigger storm initiation and impact storm evolution and longevity (Wilson and Megenhardt 1997). Predictor fields for nowcasting convective storms include 1) location and extrapolation of surface convergence features and boundary layer characteristics, 2) identification of cloud type and location of cloud growth aloft as an indirect measure of atmospheric instability, and 3) storm characterization fields that include the location, track, and growth rate of storms. The predictor fields are transformed into thunderstorm likelihood fields ranging from 1 (high likelihood) to −1 (low likelihood) using a fuzzy logic algorithm that applies user-defined membership functions (McNeill and Freiberger 1993) to the predictor fields. Weighted likelihood fields are combined to produce 30- and 60-min nowcasts (black polygons in Fig. 2) of storm initiation, growth, and decay. The complete set of ANC system membership functions and weighted likelihood fields is provided by Mueller et al. (2003) and Roberts and Rutledge (2003). While the ANC system was not originally designed to produce nowcasts of storm severity, Table 2 lists a subset of feature detection fields that have applicability toward this objective.

Fig. 2.

Example of ANC system output from Sterling, VA, on 2 Jun 2000. (a) ANC system products overlaid onto the reflectivity field (grayscale) at a forecast time (2213 UTC). The location of the surface convergence line is shown by the thin solid black line and its 30-min extrapolated position by the thick dashed black curve. The 30-min nowcasts (curved polygons) are overlaid onto the reflectivity field at forecast time and (b) the reflectivity field at verification time (2243 UTC).

Fig. 2.

Example of ANC system output from Sterling, VA, on 2 Jun 2000. (a) ANC system products overlaid onto the reflectivity field (grayscale) at a forecast time (2213 UTC). The location of the surface convergence line is shown by the thin solid black line and its 30-min extrapolated position by the thick dashed black curve. The 30-min nowcasts (curved polygons) are overlaid onto the reflectivity field at forecast time and (b) the reflectivity field at verification time (2243 UTC).

Table 2.

A list of predictor fields produced by algorithms running in the NCAR ANC system that are applicable for storm severity prediction.

A list of predictor fields produced by algorithms running in the NCAR ANC system that are applicable for storm severity prediction.
A list of predictor fields produced by algorithms running in the NCAR ANC system that are applicable for storm severity prediction.

Figure 3 provides an example of the steps involved in creating one of the thunderstorm likelihood fields used in combination with other likelihood fields to generate a final thunderstorm likelihood field and storm initiation nowcast. Data from Sterling, Virginia, on 2 June 2000 are used to illustrate the process of creating a likelihood field of the strength of vertical velocities along a convergence boundary (see Table 2 under “boundary characterization”). The presence of strong vertical motions above converging boundary layer winds and under conditions of favorable atmospheric instability can lead to thunderstorm formation (Wilson et al. 1992). Identifying the existence of a convergence boundary is the first step in the process. On 2 June a convergence boundary visible on radar in the form of a thin reflectivity line with maximum intensities on the order of 15–25 dBZ was detected by a boundary detection algorithm. The algorithm output showing the position of this convergence boundary (solid line) is overlaid onto the reflectivity image in Fig. 3a. The algorithm also computes the movement and future (extrapolated) position of the boundary at nowcast time, as indicated by the dashed line in Fig. 3d. An elliptical zone surrounding the extrapolated boundary, termed the lifting zone, is created that represents the region where storms are expected to occur during the nowcast period (Wilson and Mueller 1993), based on statistical studies of storm initiation locations relative to surface convergence boundary locations (Wilson and Schreiber 1986).

Fig. 3.

Process for transforming VDRAS vertical velocity and surface convergence boundary information into one likelihood field within the ANC system. (a) Location of VDRAS horizontal winds relative to the surface convergence boundary (solid black line) on 2 Jun 2000. (b) VDRAS vertical velocity field (shaded) with VDRAS wind field overlaid. Darker shades represent upward velocities. (c) Membership function for transforming vertical velocities into 0–1 likelihood or interest values. (d) Convergence boundary lifting zone, based on the 15-min extrapolated boundary position (dashed line), containing the vertical velocity likelihood values.

Fig. 3.

Process for transforming VDRAS vertical velocity and surface convergence boundary information into one likelihood field within the ANC system. (a) Location of VDRAS horizontal winds relative to the surface convergence boundary (solid black line) on 2 Jun 2000. (b) VDRAS vertical velocity field (shaded) with VDRAS wind field overlaid. Darker shades represent upward velocities. (c) Membership function for transforming vertical velocities into 0–1 likelihood or interest values. (d) Convergence boundary lifting zone, based on the 15-min extrapolated boundary position (dashed line), containing the vertical velocity likelihood values.

Through the use of the four-dimensional Variational Doppler Radar Assimilation System (VDRAS; Sun and Crook 2001) running in the ANC system, high-resolution three-dimensional boundary layer winds and thermodynamic analysis of the boundary layer are routinely available every 12 min. The VDRAS horizontal wind field, shown in Fig. 3a, provides information on the location and strength of converging winds and vertical motions (Fig. 3b). Of primary interest are the vertical velocity values in the vicinity of the boundary. These vertical velocity values are assigned to discrete 10-km segments of the boundary using the maximum value of the VDRAS vertical velocity at ∼1 km above ground level along the segment (see Mueller et al. 2003). The grid within the elliptical, lifting zone is then populated with these maximum values of vertical velocity. Next, a membership function (Fig. 3c) is applied to the data to convert the vertical velocity values into likelihood values for thunderstorm initiation. The range of velocity values associated with thunderstorm development and the corresponding likelihood values assigned in the membership function were initially determined based on case study analysis and then refined based on additional analysis of numerous thunderstorms. Figure 3d shows the resulting thunderstorm likelihood field produced; one of several likelihood fields that will subsequently be weighted and combined together to produce a final thunderstorm likelihood field. In Fig. 3d, the lighter-shaded sections along the lifting zone represent areas of stronger upward velocities and higher likelihood that thunderstorms will form there.

While the likelihood fields, such as the one displayed in Fig. 3d, are available for display in the ANC system, forecasters generally prefer to view only the final thunderstorm likelihood field and the thunderstorm nowcast products that are provided in symbolic form as polygons and boundaries. The nowcast polygons (see Fig. 2) represent areas in the final thunderstorm likelihood field that exceed a preselected threshold value. The threshold value is initially defined based on trial and error and later refined based on verification studies of hundreds of events. More details on the final nowcast fields are presented in the following sections (see also Fig. 12).

Fig. 12.

Likelihood fields contributing to the severe storm nowcast produced at 0452 UTC. Likelihood values range from 0 to 1. Weights and threshold are shown in the lower left-hand corner. The black polygon [white polygon in (b)] is the predicted location of the severe storm and hazardous weather. (a) The vertical velocity, (b) boundary collision, (c) TVS low-level shear, (d) mesocyclone, (e) POSH, and (f) final nowcast likelihood field.

Fig. 12.

Likelihood fields contributing to the severe storm nowcast produced at 0452 UTC. Likelihood values range from 0 to 1. Weights and threshold are shown in the lower left-hand corner. The black polygon [white polygon in (b)] is the predicted location of the severe storm and hazardous weather. (a) The vertical velocity, (b) boundary collision, (c) TVS low-level shear, (d) mesocyclone, (e) POSH, and (f) final nowcast likelihood field.

In addition to being run at the NWS WFO in Sterling, Virginia, and the BOM weather office in Sydney, Australia, this system has also been run continuously at the WFO at the White Sands Missile Range in New Mexico since 1997 (Roberts et al. 1999; Saxen et al. 1999) and is currently running at the NWS WFO in Fort Worth, Texas (Roberts et al. 2005).

c. Conceptual idea: Merging capabilities

As the atmosphere becomes more unstable, it is beneficial for forecasters to know which storms are going to become severe within the next hour. In this section we demonstrate the concept of merging WDSS and ANC system capabilities to produce very short-term (<30 min) nowcasts of storm severity as the first step in providing automated guidance to the forecaster.

The concept for combining data from the ANC system and WDSS to obtain predictions for locations of severe weather is as follows. Selected output parameters (Pi; refers to parameters listed in Tables 1 and 2) from each system are converted to storm severity likelihood values. Each parameter is converted to a storm severity likelihood value by a membership function. This function may be based on statistical studies, theoretical studies, and/or forecaster insight. The individual likelihood values Li are given (multiplied by) a weight Wi and summed ΣWiLi. The greater the summed value, the greater the likelihood of a severe storm. A threshold is typically set on the summed field that marks the likelihood level for thunderstorms to occur. This whole procedure is often referred to as fuzzy logic.

The use of membership functions illustrates the strength of using fuzzy logic in a nowcasting system. By specifying the range of values observed during analysis of numerous severe storm events for any one predictor field (see, e.g., Fig. 3c) and assigning a range of interest or likelihood values to the predictor values, more information is retained about the predictor features than would have been possible using a binary yes–no type of nowcast system. This fuzzy logic approach reflects what a forecaster typically does as he or she builds a conceptual model in their mind of how a particular weather event will evolve. Once the predictor field has been converted to a likelihood field, as shown in Fig. 3, each grid point in the likelihood field is multiplied by a weighting factor. The weighting factor can be viewed as a measure of uncertainty in the importance of a particular predictor feature in the overall production or nowcast of a severe storm. The ability to specify a range of values in a membership function allows one to account for the variability in values from one geographical region to the next. Thus, there is generally less need to change the membership functions when running the system in more than one geographical region and a tendency to change the weights applied to each likelihood field instead. When substantial differences do exist in the characteristics of storms or the environment from one location to the next, it is necessary to modify the membership functions, as was done for the Sydney storms in section 3b.

Combining ANC system and WDSS data together necessitates rendering the WDSS object-based, point data into a form similar to the gridded ANC system fields. Figure 4 illustrates this process using data from 2 June 2000. First, the point data from the WDSS probability of hail (POH) algorithm (rendered as cross hairs in Fig. 4a) are expanded to an area representing a typical storm size; that is, the point value is assigned to all grid points within 5-km radius of the point to represent a typical mature storm size of ∼100 km2. The circular, shaded areas in Fig. 4a are the graphical expansions of the POH output onto the storm scale and are in reasonable agreement spatially with the dimensions of the >30 dBZ storms (black-contoured regions). Next, a membership function for converting POH values to likelihood values of storm severity (Fig. 4b) is applied to the gridded POH field. The membership function in this case is based on 1) general trends in the data observed with several hail events; 2) consistent, significant threshold levels associated with the precursor signatures; and 3) the level of confidence in the precursor signature for predicting storm severity. Here, a simple membership function is applied to the POH field such that higher likelihood values are assigned to POH values >50%. The resulting POH likelihood field is shown in Fig. 4c. Darker colors represent a higher interest or likelihood for severe weather.

Fig. 4.

Rendering WDSS POH objects to gridded values of hail probability and then to the likelihood of a severe storm. (a) Contours of 35 dBZ are overlaid onto the rendered WDSS POH objects. The crosses represent the point WDSS POH locations. See the text for the rendering process used to represent the gridded values of hail probability (shaded circles). (b) Membership function for converting POH values to the likelihood of a severe storm. (c) POH likelihood of a severe storm after applying the membership function in (b).

Fig. 4.

Rendering WDSS POH objects to gridded values of hail probability and then to the likelihood of a severe storm. (a) Contours of 35 dBZ are overlaid onto the rendered WDSS POH objects. The crosses represent the point WDSS POH locations. See the text for the rendering process used to represent the gridded values of hail probability (shaded circles). (b) Membership function for converting POH values to the likelihood of a severe storm. (c) POH likelihood of a severe storm after applying the membership function in (b).

The following discussion documents conceptually how the WDSS and ANC system likelihood fields are used to produce storm severity nowcasts. The fields that are ingested into a severe storm nowcasting system represented by this conceptual model are listed in Tables 1 and 2. First, in order to obtain a 15-, 30-, or 60-min nowcast of severe storm location, all detected features of interest including POH, probability of severe hail (POSH), and convergence boundary and storm detections are extrapolated into the future using either an automated boundary extrapolation algorithm or applying storm cell tracker algorithms like SCIT or the Thunderstorm Identification, Tracking, Analysis and Nowcasting (TITAN; Dixon and Wiener 1993) algorithm to the relevant storm data.

The next step is to determine the storm severity at nowcast time using the fuzzy logic approach. Only three likelihood fields are discussed here for simplicity: vertical velocity strength along the convergence boundary (from Fig. 3d), POH (from Fig. 4c), and POSH. Based on examination of numerous severe weather events and verification data in the form of spotter reports, each of the likelihood fields is multiplied by a weight factor representing the confidence and relative importance of a particular field toward the final nowcast. At each grid point, the weighted likelihood values from each field are summed to produce the final nowcast likelihood field for storm-specific, severe weather. Figure 5a shows conceptually what the final nowcast likelihood field might look like. Outside the lifting zone the potential for storms to be severe would have a low likelihood, while within the lifting zone likelihood values would be higher depending on the magnitude of the value at each grid point of the weighted, summed, likelihood fields. In Fig. 5a, the highest likelihood (lighter colors) for severe weather would be located where storms overlap with the lifting zone and have the most favorable POH, POSH, and vertical velocity summed likelihood values. The final storm severity nowcast is obtained by applying a threshold limit to the final likelihood field in Fig. 5a. Values exceeding this threshold limit represent the area in which severe weather is expected. The threshold is user selectable and would typically be based upon an examination of the storm severity data from a specific geographical region and optimized by the validation of nowcast performance for several test case studies. Using a 0.75 likelihood threshold value, arbitrarily selected here for illustration purposes, only two regions of severe storm nowcasts are produced as defined by the black polygons in Fig. 5b.

Fig. 5.

Conceptual model for production of storm severity nowcasts using ANC system and WDSS likelihood fields. (a) Final likelihood field composed of weighted likelihood fields (POH, POSH, and boundary-relative vertical velocity) that have been summed together at each grid point. Summed likelihood values for three specific grid points are shown in figure text. (b) Storm severity nowcast polygons that are produced after a 0.75 threshold limit was applied to the final likelihood field in (a).

Fig. 5.

Conceptual model for production of storm severity nowcasts using ANC system and WDSS likelihood fields. (a) Final likelihood field composed of weighted likelihood fields (POH, POSH, and boundary-relative vertical velocity) that have been summed together at each grid point. Summed likelihood values for three specific grid points are shown in figure text. (b) Storm severity nowcast polygons that are produced after a 0.75 threshold limit was applied to the final likelihood field in (a).

3. Test of concept

To test the conceptual model presented in section 2, two severe weather events have been processed using a modified ANC system that ingests WDSS data and produces 15-min storm severity nowcasts. The modified ANC system, or ANC–WDSS system, is also capable of producing 30-min nowcasts. However, 15 min was chosen as a first step to better understand the data and system and their relation to severe weather reports. Longer intervals will be tried at some point in the future. It would likely be more meaningful to represent the nowcasts in probabilistic form but the Auto-Nowcast fuzzy logic system is not currently designed to provide this type of output.

The severe weather events in this section include a squall line that produced large hail near Washington, D.C., on 2 June 2000 and a tornadic hailstorm that occurred over Sydney, Australia, on 3 November 2000. A brief description of the events and an examination of the storm severity nowcasts follow.

a. Washington, D.C.—2 June 2000

On 2 June a cold front located in Pennsylvania was approaching Virginia and was expected to arrive in the Washington, D.C., area at around 2100 UTC. The forecasters’ attention was focused on the arrival of the cold front and not on a semistationary, mesoscale convergence boundary located 40 km northwest of Washington, D.C., and evident on both radar and satellite images (S. Zubrick, scientific operations officer, Washington, D.C.–Baltimore, Maryland, WFO, 2000, personal communication). Had the forecasters been monitoring this particular area, they would have noticed a strengthening of the surface convergence boundary with time as it slowly moved southeastward toward them. Growth of cumulus clouds above this boundary in the visible and infrared imagery was clear as well. In the radar image an intense, an almost-continuous line of storms (see Fig. 2b) developed on the convergence boundary, well ahead of the expected cold front, which had stalled in Pennsylvania. This squall line produced heavy rain, large hail, and strong winds throughout the area.

The 15-min ANC–WDSS storm severity nowcasts for this event were produced using output from all of the algorithms listed in Tables 1 and 2 with the exception of the Downburst Detection And Prediction Algorithm (DDPDA) and Tornado Detection Algorithm (TDA), which were not available for 2 June. Figure 6 shows the membership functions and weights applied to each of the ANC–WDSS predictor fields. The membership functions for the boundary-related predictor fields (Figs. 6a–e) are based on the examination of years of thunderstorm-related data collected in conjunction with operations of the ANC system at various places in the United States. The reader is referred to Mueller et al. (2003) for substantial detail on the physical basis for the predictor fields and the processes used to define the limits for the membership functions, which are the same as is shown in Figs. 6a–e.

Fig. 6.

Membership functions and weights used in the modified ANC–WDSS severe storm nowcasting system for the 2 Jun 2000 Sterling, VA, event: (a)–(e) boundary-related fields and (f)–(k) storm characterization fields.

Fig. 6.

Membership functions and weights used in the modified ANC–WDSS severe storm nowcasting system for the 2 Jun 2000 Sterling, VA, event: (a)–(e) boundary-related fields and (f)–(k) storm characterization fields.

Figure 6 also shows the membership functions associated with the storm characterization fields. The TITAN algorithm was used to detect and analyze existing storms and extrapolate their position 15 min into the future. Both 35- and 50-dBZ storm (radar) echoes were tracked with time. The TITAN vectors used to extrapolate the storms were used to extrapolate the WDSS predictor features 15 min into the future also. While the forecast position of the 50-dBZ storm echoes was extremely important for the accuracy of the location of the severe storm nowcasts, other storm attributes were also important in determining whether a storm would become severe. Increasing storm area and storm growth rate (Figs. 6f and 6g), while not specific predictors of storm severity, are necessary features for monitoring storm persistence and longevity. The membership functions for these storm predictor fields are shown in Figs. 6f–l. Given that this is the first attempt at producing automated severe storm nowcasts, a first guess was made of the weights to use based on careful examination of the evolution of the two case studies presented in the paper. These weights are shown in Fig. 6 (W values) for each field. Similarly, a first guess at a threshold value (0.7) was made and applied to the final summed likelihood field. The likelihood fields that contributed most significantly to the nowcasts for severe storms are shown in Table 3, in order from top to bottom.

Table 3.

Predictor features that contributed to the automated, 15-min, ANC–WDSS severe storm nowcasts produced for both the Washington, D.C., and Sydney, Australia, events. The features are listed in order of importance (from top to bottom) by their level of contribution to the automated nowcast.

Predictor features that contributed to the automated, 15-min, ANC–WDSS severe storm nowcasts produced for both the Washington, D.C., and Sydney, Australia, events. The features are listed in order of importance (from top to bottom) by their level of contribution to the automated nowcast.
Predictor features that contributed to the automated, 15-min, ANC–WDSS severe storm nowcasts produced for both the Washington, D.C., and Sydney, Australia, events. The features are listed in order of importance (from top to bottom) by their level of contribution to the automated nowcast.

The 15-min storm severity nowcasts from the ANC–WDSS system are shown in Figs. 7 and 8 in the form of polygons that bound the locations where severe storms are expected. Because the TDA was not available for 2 June, this test of the ANC–WDSS does not produce nowcasts of storm severity type, although ideally this is what one would like a future nowcasting guidance tool to provide. The surface convergence boundary that triggered the initial convection moved to the SE with time and continued to force new storm development along the leading edge of the squall line as it approached the Washington, D.C., area and its surrounding suburbs (see Figs. 7a,c and 8a,c). Storms moved at a speed and direction comparable to the motion of the surface convergence boundary and therefore the boundary-relative steering flow likelihood field (see Table 3) indicated a broad, favorable region for continued convection along the boundary. The other fields listed in Table 3 provided more discrete, focused information on which storms along the boundary had the potential to become or stay severe. For example, the strongest vertical motions were located along the central sections of the boundary and the WDSS algorithm output indicated high probabilities for hail along the SW portion of the squall line, but very low probabilities for the storms along the NE portion of the line. All of the storms having reflectivity values >50 dBZ were extrapolated forward in time, but not all of these storms had severity nowcasts associated with them. This is evident along the NE section of the squall line (see Figs. 8a and 8c in particular) where no 15-min nowcast polygons were produced. The lack of nowcasts for this region is because the summed likelihood values were not high enough to meet the final threshold for severe storm occurrence.

Fig. 7.

Fifteen-minute nowcasts (black polygons) predicting the location of severe storms in VA, Washington, D.C., and MD, on 2 Jun 2000. Nowcast polygons are overlaid onto the Sterling, VA, radar reflectivity field at two nowcast issue times: (a) 2253 and (c) 2258 UTC. The corresponding verification of the nowcasts based on radar reflectivity data ∼15 min later and spotter reports are overlaid in (b) and (d), respectively.

Fig. 7.

Fifteen-minute nowcasts (black polygons) predicting the location of severe storms in VA, Washington, D.C., and MD, on 2 Jun 2000. Nowcast polygons are overlaid onto the Sterling, VA, radar reflectivity field at two nowcast issue times: (a) 2253 and (c) 2258 UTC. The corresponding verification of the nowcasts based on radar reflectivity data ∼15 min later and spotter reports are overlaid in (b) and (d), respectively.

Fig. 8.

Same as in Fig. 7, but for two later nowcast periods at (a) 2304 and (c) 2309 UTC 2 Jun 2000.

Fig. 8.

Same as in Fig. 7, but for two later nowcast periods at (a) 2304 and (c) 2309 UTC 2 Jun 2000.

Verification of the nowcasts in the form of radar reflectivity at nowcast validation times and from spotter reports of severe weather are shown in Figs. 7b,d and 8b,d. Storm severity nowcasts for storms along the SW portion of the squall line validated quite well in the sense that each of those storms did produce ≥0.75 in. (1.91 cm) hail. It is difficult to provide a more rigorous estimate or magnitude of nowcast performance given the uncertainty in spotter reports as a result of latency in reporting and temporal and spatial inaccuracies inherent in the reports. Given the density of the spotter network in Virginia and the Washington, D.C., area, it is interesting to note that no hail >1.91 cm was reported by spotters in Washington, D.C., with the very large storm (see Figs. 7d and 8b,d) that passed over the city, validating the “null” or no nowcast issued in this area. While the output from the WDSS system did indicate low to moderate probabilities for hail, there were no additional substantiating signatures to support a severe storm nowcast. It appears that this methodology may have some skill in differentiating between the different levels of potential storm severity, as the technique was able to discriminate which part of the line would produce severe weather. The boundary-relative vertical velocity and POSH likelihood were key fields in targeting the central and SW portions of the squall line as having the highest likelihood for severe storms. However, this is only one event; examination of additional cases would be required to determine the representativeness of this result.

b. Sydney, Australia—3 November 2000

On 3 November, during the Sydney FDP, severe weather hit the Sydney metropolitan area in the form of heavy rain, flash flooding, giant (7 cm) hail, and weak tornadoes. The interaction of three, low-level convergence boundaries played an important role in enhancing storm organization and the development of rotation within the severe storms (Sills et al. 2004). Using the ANC–WDSS system, 15-min nowcasts of severe storms were produced every 5 min, corresponding to the radar update rate, during the 2.0-h period of severe weather. Output from all of the algorithms listed in Tables 1 and 2, with the exception of the DDPDA, which was not available for 3 November, were used in the ANC–WDSS system. Similar to the Sterling case, the TITAN algorithm was use to track storms and extrapolate the WDSS predictor fields. The membership functions and weights for each of the fields are shown in Figs. 6 and 9. The boundary characterization membership functions and weights for Sydney are the same as for Sterling (Figs. 6a–d) because the range of predictor values in the membership functions covered the important ranges for both geographical regions. In the Sydney area, storm structure and evolution were typical of maritime-type storms, while continental storms were observed in Sterling. Differences in the storm area, growth rate, and magnitude of the vertical motions were evident and thus a separate set of storm-related membership functions was required for Sydney (see Fig. 9). Differences in the microphysics of the storms in Sydney, particularly in the intensities of radar-based features, also required a different membership function for the likelihood of small hail (see the POH function in Fig. 9). A membership function for maximum TVS shear was created also, because of the availability of the WDSS TDA output from Sydney.

Fig. 9.

Membership functions and weights used in the ANC–WDSS severe storm nowcasting system for the Sydney, Australia, 3 Nov 2000 event. The membership functions for the other predictor fields are the same as in Fig. 6.

Fig. 9.

Membership functions and weights used in the ANC–WDSS severe storm nowcasting system for the Sydney, Australia, 3 Nov 2000 event. The membership functions for the other predictor fields are the same as in Fig. 6.

Figures 10 and 11 show the nowcasts issued at four time periods spaced 30 min apart. Included in these figures are the corresponding validation plots with all available spotter reports overlaid. The features that contributed significantly toward the storm severity nowcasts are listed in the second column of Table 3. The nowcasts represent regions where severe weather is likely to occur in the next 15 min, but do not provide information on the type of severe weather expected.

Fig. 10.

Fifteen-minute nowcasts (black polygons) at (a) 0332 and (c) 0402 UTC for severe storms in the vicinity of Sydney, Australia, on 3 Nov 2000. Convergence boundary locations are represented by the black lines. Corresponding verification reflectivity data and spotter reports are overlaid in (b) and (d), respectively.

Fig. 10.

Fifteen-minute nowcasts (black polygons) at (a) 0332 and (c) 0402 UTC for severe storms in the vicinity of Sydney, Australia, on 3 Nov 2000. Convergence boundary locations are represented by the black lines. Corresponding verification reflectivity data and spotter reports are overlaid in (b) and (d), respectively.

Fig. 11.

Same as in Fig. 10, but at two later nowcast time periods: (a) 0427 and (c) 0452 UTC.

Fig. 11.

Same as in Fig. 10, but at two later nowcast time periods: (a) 0427 and (c) 0452 UTC.

The spotter reports provide evidence for the type of weather that occurred. Unfortunately, great uncertainty exists in the temporal accuracy of the reports called in to the Sydney BOM forecast office. This uncertainty is reflected in the time intervals listed for spotter reports of a specific severe weather event; for example, the spotter report of 2–4 cm hail on the ground is listed as occurring some time between 0415 and 0445 UTC and F1 tornadoes occurred between 0430 and 0530 UTC. A damage survey and research conducted by Sills et al. (2004) immediately after the tornadic event was able to narrow down the time interval of the spotter reports of tornado occurrence to 0505–0525 UTC. This shows how difficult it is to obtain a quantitative estimate of performance, based on spotter reports, for all nowcast time periods in that interval. Given the approximate locations of the spotter reports, it can only be said that the 15-min nowcasts do a reasonably good job of earmarking the specific regions for severe storms.

The nowcasts issued at 0452 UTC are of particular interest. This nowcast time is about 15 min prior to the start of an F1 tornado and giant hail (7 cm at 0510 UTC) occurring over Sydney. Some of the ANC–WDSS features listed in Table 3 and present in the likelihood fields at this time period are shown in Fig. 12 with the nowcast polygon overlaid for reference (same nowcast as in Figs. 11c and 11d). It is clear that all of these features are contributing to the final likelihood field produced (Fig. 12f), but at different levels of likelihood values. Upward vertical velocities associated with the three boundaries are detected and nowcast to be present in the next 15 min with varying levels of likelihood (Fig. 12a) for severe weather expected to occur over the broad lifting zone regions. The strongest likelihood values (0.5–0.75) coincide with the strongest vertical velocities anticipated from the collision of these boundaries. The region of these high values also corresponds to the higher (1.0) likelihood values associated with the anticipated boundary collisions (Fig. 12b); the collision of surface convergence boundaries is considered a significant factor in the rapid intensification of storms and for tornadogenesis (see Sills et al. 2004 and their references).

The WDSS-based fields provide crucial information on the likelihood for specific types of severe weather to occur. At this time period, the likelihood values for tornadoes (Fig. 12c) and large hail (Fig. 12e) are very high, ranging from 0.6 to 1.0 and 1.0, respectively. And although the mesocyclonic shear within the storms was not particularly strong, as reflected by the lower likelihood values for this field (Fig. 12d), the combination of all the WDSS-based fields plays an important role in narrowing down the regions for potential severe storms into much more focused areas. Figure 12f shows the results once the likelihood fields have been summed together and a threshold value (0.7) applied. Not all of the likelihood features line up in the same place but their combined information has defined the area of most concern for severe weather in general, which is very important guidance for forecasters wanting to alert specific communities rather than issuing a county-wide alert. This is confirmed by the spotter reports (overlaid onto Fig. 12f) that show the type of weather that was either ongoing at a time period close to 0450 UTC or that occurred near the nowcast time.

Figure 12 also illustrates an important point in assigning the appropriate weights to predictor likelihood fields. For Sydney, the first guess at a weight for the TVS shear field (see Figs. 9 and 12) is 0.1. However, since tornadoes present a greater threat to life and property, it likely would have been more appropriate to set this weight to a value similar to that for the mesocyclonic shear, that is, W = 0.25. Had this been done, the area just north of the polygon in the final likelihood field (Fig. 12f) would have increased in likelihood value well above the threshold limit and resulted in an extension of the severe storm nowcast area to the north. A comparison of this revised nowcast polygon with the spotter’s reports would have shown very good agreement and validation. We anticipate that the inspection of likelihood fields from several additional cases of severe weather events will illuminate the best way to modify the ANC–WDSS nowcasting system in the future for prediction of a specific hazardous weather event, like tornadoes, in addition to predicting the timing and location of severe storms.

4. Summary and discussion

Feature detection and nowcast output from two distinctly different systems, the WDSS and ANC systems, have been combined to produce 15-min, storm severity nowcasts. This is a first exploratory step toward identifying the type of system needed to produce guidance tools for nowcasting severe storms. A conceptual idea for combining these datasets was presented and tested on a hail-producing squall line and a tornadic, supercell storm. In this test, reasonable 15-min nowcasts were produced that showed some discriminate skill in identifying which storm cells in a multicellular line would produce damaging hail and the potential to outline a specific region of severe weather associated with a supercell.

While only 15-min nowcasts were shown for the two case studies presented, it is conceivable that 30–60-min nowcasts could also have been produced. For short-term (15 min) nowcasts, the inclusion of the ANC system boundary layer information, particularly the location of maximum vertical velocities along the boundary and the anticipated collision zone of two or more convergence boundaries, provides 1) heightened areas of interest, 2) anticipated storm longevity, and 3) added weight to severe weather detection features identified by the WDSS system of algorithms. The WDSS parameters, such as TVS, POH, POSH, and mesocyclonic shear, provide important forecast information for the short-period nowcasts that possibly could be extended to be useful in 30-min nowcasts. The storm-specific nature of these data helps to focus the nowcast regions into much smaller areas. At longer nowcast periods (from 30 to 60 min), boundary layer convergence features, particularly boundary collisions and atmospheric instability conditions, are likely the more significant factors for nowcasting the location and intensity of severe weather over a broader region.

This study has illustrated some of the challenges and issues associated with designing tools for the production of short-term, storm severity nowcasts. These include 1) finding the best methodology for combining object-based, storm features with larger-scale gridded features, 2) the importance of having temporally and spatially accurate spotter reports to verify storm severity nowcasts, and 3) determining the relative benefits of producing location-specific, storm severity nowcast regions versus hazard-specific nowcast regions. The ANC–WDSS combined nowcasting system used data from two case studies to test the concept and skill in predicting the timing and location of severe storms. However, the information exists in the system to be able to produce hazard-specific nowcast regions. This can be achieved by increasing the weights of specific predictor fields associated with a particular weather hazard. Results from this study suggest that the ideal guidance tool should provide comprehensive products that include location, timing, and hazard-specific nowcasts. Developing a comprehensive severe storm nowcasting system is possible using the fuzzy logic approach demonstrated in this study.

The above challenges will need to be met in a robust nowcasting system that allows for the fusion of all different types of data including other severe weather detection fields such as 1) the total lightning field for predicting the onset of supercell tornadoes (Goodman 2003; Goodman et al. 2003), 2) NWP model and satellite stability fields, and 3) polarization radar data for the detection of hail. The system needs to make use of both the object-based and gridded capabilities of systems like the new WDSS II (Stumpf et al. 2003) that pull in information from several radars over a multiple-WFO domain and also incorporates forecaster input into the algorithms in real time, such as the latest version of the regional ANC system (Roberts et al. 2003).

Recent advances in the assimilation of Doppler radar and mesonet data into mesoscale and cloud models (Verlinde and Cotton 1993; Sun and Crook 1997, 1998; Xu et al. 2001; Weygandt et al. 2002a, b; Synder and Zhang 2003; Zhang et al. 2004) have led to dramatic improvements in very short-range, numerical model forecasts of convection on high spatial and temporal scales. Systems such as the Weather Research and Forecasting (WRF) model (Michalakes et al. 2001) and the Center for Analysis and Prediction of Storms Advanced Regional Prediction System (Xue et al. 2003) model are assimilating radar data into their numerical models and providing significantly improved prediction of large-scale storm systems and realistic prediction of storm structure (e.g., formation of bow echoes). The WRF model forecasts using explicit treatment of convection on a 4-km grid are found to be far superior to 10-km grid forecasts in their ability to forecast mesoscale convective systems (Done et al. 2004). The four-dimensional VDRAS that runs in a real-time mode routinely produces forecasts, updated every 12 min, of wind shifts and surface convergence that are precursors to the development of convective storms (Crook and Sun 2002; Caya et al. 2003). Use of these high-resolution forecasts will become important not only as stand-alone output but also for integration into fuzzy logic systems that incorporate both model and observational datasets (Saxen et al. 2004).

Designing tools to ease the forecaster’s workload by integrating many datasets together will ensure that the forecaster will have more time to direct his or her energy toward improving severe weather forecasts and increasing the lead time on warning the public of the potential for severe weather in their area. As an example, 30- and 60-min nowcasts provided by the ANC system on 13 June 1998 were used by Washington, D.C.–Baltimore, Maryland, WFO forecasters as guidance in providing a special 45-min advisory to concert officials at Robert F. Kennedy Memorial Stadium in the District of Columbia of when a severe thunderstorm and gust front would reach the stadium (Roberts et al. 1999). Currently, the NWS is sponsoring a test of a human-interactive ANC system at the Fort Worth, Texas, WFO from March 2005 to March 2007 and will be evaluating the forecasters’ use of this tool in their daily operational routine. Based on a year of experience with the ANC system, the forecasters find the final, fuzzy likelihood field (similar to Fig. 12f) more informative and useful than a display of nowcast polygons (Roberts et al. 2005). While the ANC system is currently stand alone, parallel NWS efforts are simultaneously under way to incorporate ANC system algorithms into the AWIPS infrastructure. A natural follow-on to the ANC system integration into AWIPS, in the near future, could be an ANC–WDSS type of guidance system for severe storm nowcasting.

Ultimately, tools designed for storm severity nowcasting should be compatible with the NWS AWIPS Graphical Forecast Editor so that a forecaster can interactively edit and modify nowcast guidance fields to produce meaningful, automated, graphical products for the community. During the Sydney 2000 Olympics games, Olympic forecasters had access to information from the FDP advanced technological systems (including WDSS and ANC system products) via a state-of-the-art Thunderstorm Interactive Forecast System (TIFS; Bally 2004) to interactively produce severe weather warnings tailored to the needs of several Olympic-specific end users. The human impacts study conducted following the FDP (Anderson-Berry et al. 2004) indicated that the TIFS was well received by experienced forecasters who felt that their warnings were superior to what they would have issued using routine storm tracking techniques. Although the forecasters had high confidence in their ability to identify any threatening weather in the data, they felt they could produce even better forecasts with access to more advanced technology and the automation of some products, allowing them more time to study the weather situation and local conditions.

Acknowledgments

The authors thank Kevin Thomas for providing the WDSS output files and Arthur Witt for providing Fig. 1. James Wilson and Thomas Saxen provided insightful comments on an earlier version of the paper. The NEXRAD Operational Support Facility supported the running of the Auto-Nowcast system and WDSS at the Washington, D.C., WFO during the SCAN demonstration. The USWRP provided support for running the Auto-Nowcast during the Sydney 2000 FDP. The Australian BOM provided support for running WDSS during the Sydney 2000 FDP.

REFERENCES

REFERENCES
Anderson-Berry
,
L.
,
T.
Keenan
,
J.
Bally
,
R.
Pielke
Jr.
,
R.
Leigh
, and
D.
King
,
2004
:
The societal, social, and economic impacts of the World Weather Research Program Sydney 2000 Forecast Demonstration Project (WWRP S2000 FDP).
Wea. Forecasting
,
19
,
168
178
.
Bally
,
J.
,
2004
:
Generating severe weather warnings from TITAN and SCIT thunderstorm tracks.
Wea. Forecasting
,
19
,
64
72
.
Caya
,
A.
,
N. A.
Crook
, and
J.
Sun
,
2003
:
The use of an evaporation scheme in a boundary-layer model for real-time 4D-VAR radar data assimilation and forecasting of convergence lines. Preprints, 31st Conf. on Radar Meteorology, Seattle, WA, Amer. Meteor. Soc., 110–113
.
Crook
,
N. A.
, and
J.
Sun
,
2002
:
Assimilating radar, surface, and profiler data for the Sydney 2000 Forecast Demonstration Project.
J. Atmos. Oceanic Technol.
,
19
,
888
898
.
Dixon
,
M.
, and
G.
Wiener
,
1993
:
TITAN: Thunderstorm Identification, Tracking, Analysis, and Nowcasting—A radar-based methodology.
J. Atmos. Oceanic Technol.
,
10
,
785
797
.
Done
,
J.
,
C.
Davis
, and
M.
Weisman
,
2004
:
The next generation of NWP: Explicit forecasts of convection using the Weather Research and Forecasting (WRF) model.
Atmos. Sci. Lett.
,
5
,
110
117
.
Goodman
,
S.
,
2003
:
Atmospheric electrical activity and the prospects for improving short-term, weather forecasting. Preprints, 12th Int. Conf. on Atmospheric Electricity, Versailles, France, International Assembly of Meteorology and Atmospheric Science of the International Union of Geodesy and Geophysics, 1–4
.
Goodman
,
S.
, and
Coauthors
,
2003
:
The north Alabama lightning mapping array: Recent severe storm observations and future prospects. Preprints, 12th Int. Conf. on Atmospheric Electricity, Versailles, France, International Assembly of Meteorology and Atmospheric Science of the International Union of Geodesy and Geophysics, 423–437
.
Johnson
,
J. T.
,
P.
MacKeen
,
A.
Witt
,
E. D.
Mitchell
,
G.
Stumpf
,
M.
Eilts
, and
K.
Thomas
,
1998
:
The Storm Cell Identification and Tracking algorithm: An enhanced WSR-88D algorithm.
Wea. Forecasting
,
13
,
263
276
.
Johnson
,
J. T.
,
M. D.
Eilts
,
D.
Ruth
,
W.
Goodman
, and
L.
Rothfusz
,
2000
:
Warning operations in support of the 1996 Centennial Olympic games.
Bull. Amer. Meteor. Soc.
,
81
,
543
554
.
Keenan
,
T.
, and
Coauthors
,
2003
:
The Sydney 2000 World Weather Research Programme Forecast Demonstration Project: Overview and current status.
Bull. Amer. Meteor. Soc.
,
84
,
1041
1054
.
McNeill
,
D.
, and
P.
Freiberger
,
1993
:
Fuzzy Logic—The Revolutionary Computer Technology That Is Changing Our World.
Simon and Schuster, 319 pp
.
Michalakes
,
J.
,
S.
Chen
,
J.
Dudhia
,
L.
Hart
,
J.
Klemp
,
J.
Middlecoff
, and
W.
Skamarock
,
2001
:
Development of a next generation regional weather research forecast model.
Developments in Teracomputing: Proceedings of the Ninth ECMWF Workshop on the Use of High Performance Computing in Meteorology, W. Zwieflhofer and N. Kreitz, Eds., World Scientific, 269–276
.
Mueller
,
C. K.
,
T.
Saxen
,
R.
Roberts
,
J.
Wilson
,
T.
Betancourt
,
S.
Dettling
,
N.
Oien
, and
J.
Yee
,
2003
:
NCAR Auto-Nowcast system.
Wea. Forecasting
,
18
,
545
561
.
Roberts
,
R. D.
, and
S.
Rutledge
,
2003
:
Nowcasting storm initiation and growth using GOES-8 and WSR-88D data.
Wea. Forecasting
,
18
,
562
584
.
Roberts
,
R. D.
,
T.
Saxen
,
C.
Mueller
,
J.
Wilson
,
A.
Crook
,
J.
Sun
, and
S.
Henry
,
1999
:
Operational application and use of NCAR’s thunderstorm nowcasting system. Preprints, Int. Conf. on Interactive Information and Processing Systems, Dallas, TX, Amer. Meteor. Soc., 158–161
.
Roberts
,
R. D.
,
T.
Saxen
,
C.
Mueller
, and
D.
Albo
,
2003
:
A forecaster–computer interactive capability of the NCAR Auto-Nowcaster system for improved storm initiation nowcasts. Preprints, 31st Conf. on Radar Meteorology, Seattle, WA, Amer. Meteor. Soc., J20–J23
.
Roberts
,
R. D.
, and
Coauthors
,
2005
:
The NWS/NCAR Man-In-The-Loop (MITL) nowcasting demonstration: Forecaster input into a thunderstorm nowcasting system. Preprints, Symp. on Nowcasting and Very Short Range Forecasting, Toulouse, France, WMO WWRP and Amer. Meteor. Soc., CD-ROM
.
Rotunno
,
R.
,
J. B.
Klemp
, and
M. L.
Weisman
,
1988
:
A theory for strong, long-lived squall lines.
J. Atmos. Sci.
,
45
,
463
485
.
Saxen
,
T.
,
C.
Mueller
,
T.
Jameson
, and
E.
Hatfield
,
1999
:
Determining key parameters for forecasting thunderstorms at White Sands Missile Range. Preprints, 29th Int. Conf. on Radar Meteorology, Montreal, QC, Canada, Amer. Meteor. Soc., 9–12
.
Saxen
,
T.
,
C.
Mueller
,
J.
Wilson
,
R.
Roberts
,
E.
Nelson
,
D.
Ahijevych
, and
S.
Trier
,
2004
:
Updates to the Auto-nowcaster for the 2004 convective weather season. Preprints, 11th Conf. on Aviation, Range, and Aerospace Meteorology, Hyannis, MA, Amer. Meteor. Soc., CD-ROM, P5.1
.
Sills
,
D. M. L.
,
J. W.
Wilson
,
P. I.
Joe
,
D. W.
Burgess
,
R. M.
Webb
, and
N. I.
Fox
,
2004
:
The 3 November tornadic event during Sydney 2000: Storm evolution and the role of low-level boundaries.
Wea. Forecasting
,
19
,
22
42
.
Smith
,
S. B.
, and
Coauthors
,
1998a
:
The System for Convection Analysis and Nowcasting (SCAN). Preprints, 16th Conf. on Weather Analysis and Forecasting, Phoenix, AZ, Amer. Meteor. Soc., J22–J24
.
Smith
,
S. B.
,
J. T.
Johnson
,
R. D.
Roberts
,
S. M.
Zubrick
, and
S. J.
Weiss
,
1998b
:
The system for convection analysis and nowcasting (SCAN): 1997–1998 field test. Preprints, 19th Conf. on Severe Local Storms, Minneapolis, MN, Amer. Meteor. Soc., 790–793
.
Smith
,
T. K.
,
K.
Elmore
, and
S. A.
Dulin
,
2004
:
A damaging downburst prediction and detection algorithm for the WSR-88D.
Wea. Forcasting
,
19
,
240
250
.
Snyder
,
C.
, and
F.
Zhang
,
2003
:
Asimilation of simulated Doppler radar observations with an ensemble Kalman filter.
Mon. Wea. Rev.
,
131
,
1663
1667
.
Steiner
,
M.
,
R. A.
Houze
Jr.
, and
S. E.
Yuter
,
1995
:
Climatological characterization of three-dimensional storm structure from operational radar and rain gauge data.
J. Appl. Meteor.
,
34
,
1978
2007
.
Stumpf
,
G. J.
,
T. M.
Smith
,
K. L.
Manross
,
V.
Lakshmanan
, and
K. D.
Hondi
,
2003
:
Severe weather warning and application development at NSSL using multiple radars and multiple sensors. Preprints, 31st Conf. on Radar Meteorology, Seattle, WA, Amer. Meteor. Soc., 555–558
.
Sun
,
J.
, and
N. A.
Crook
,
1997
:
Dynamical and microphysical retrieval from Doppler radar observations using a cloud model and its adjoint. Part I: Model development and simulated data experiments.
J. Atmos. Sci.
,
54
,
1642
1661
.
Sun
,
J.
, and
N. A.
Crook
,
1998
:
Dynamical and microphysical retrieval from Doppler radar observations using a cloud model and its adjoint. Part II: Retrieval experiments of an observed Florida convective storm.
J. Atmos. Sci.
,
55
,
835
852
.
Sun
,
J.
, and
N. A.
Crook
,
2001
:
Real-time low-level wind and temperature analysis using single WSR-88D data.
Wea. Forecasting
,
16
,
117
132
.
Verlinde
,
J.
, and
W. R.
Cotton
,
1993
:
Fitting microphysical observations of non-steady convective clouds to a numerical model: An application of the adjoint technique of data assimilation to a kinematic model.
Mon. Wea. Rev.
,
121
,
2776
2793
.
Weisman
,
M. L.
, and
J. B.
Klemp
,
1986
:
Characteristics of isolated convective storms.
Mesoscale Meteorology and Forecasting, P. S. Ray, Ed., Amer. Meteor. Soc., 331–358
.
Weygandt
,
S.
,
A.
Shapiro
, and
K.
Droegemier
,
2002a
:
Retrieval of model initial fields from single-Doppler observations of a supercell thunderstorm. Part I: Single-Doppler velocity retrieval.
Mon. Wea. Rev.
,
130
,
433
453
.
Weygandt
,
S.
,
A.
Shapiro
, and
K.
Droegemier
,
2002b
:
Retrieval of model initial fields from single-Doppler observations of a supercell thunderstorm. Part II: Thermodynamic retrieval and numerical prediction.
Mon. Wea. Rev.
,
130
,
454
476
.
Wilson
,
J. W.
, and
W. E.
Schreiber
,
1986
:
Initiation of convective storms at radar-observed boundary-layer convergence lines.
Mon. Wea. Rev.
,
114
,
2516
2536
.
Wilson
,
J. W.
, and
C. K.
Mueller
,
1993
:
Nowcasts of thunderstorm initiation and evolution.
Wea. Forecasting
,
8
,
113
131
.
Wilson
,
J. W.
, and
D. L.
Megenhardt
,
1997
:
Thunderstorm initiation, organization, and lifetime associated with Florida boundary layer convergence lines.
Mon. Wea. Rev.
,
125
,
1507
1525
.
Wilson
,
J. W.
,
G. B.
Foote
,
N. A.
Crook
,
J. C.
Fankhauser
,
C. G.
Wade
,
J. D.
Tuttle
,
C. K.
Mueller
, and
S. K.
Krueger
,
1992
:
The role of boundary-layer convergence zones and horizontal roles in the initiation of thunderstorms: A case study.
Mon. Wea. Rev.
,
120
,
1785
1815
.
Xu
,
M.
,
D. J.
Stensrud
,
J-W.
Bao
, and
T. T.
Warner
,
2001
:
Applications of the adjoint technique to short-range ensemble forecasting of mesoscale convective systems.
Mon. Wea. Rev.
,
129
,
1395
1418
.
Xue
,
M.
,
D. H.
Wang
,
J. D.
Gao
,
K.
Brewster
, and
K. K.
Droegemeier
,
2003
:
The Advanced Regional Prediction System (ARPS), storm-scale numerical weather prediction and data assimilation.
Meteor. Atmos. Phys.
,
82
,
139
170
.
Zhang
,
F.
,
C.
Snyder
, and
J.
Sun
,
2004
:
Impacts of initial estimate and observation availability on convective-scale data assimilation with an ensemble Kalman filter.
Mon. Wea. Rev.
,
132
,
1238
1253
.

Footnotes

* The National Center for Atmospheric Research is sponsored by the National Science Foundation

Corresponding author address: Rita D. Roberts, NCAR Research Application Program, P.O. Box 3000, Boulder, CO 80307. Email: rroberts@ucar.edu