Abstract

The Sydney 2000 Olympic Games World Weather Research Programme Forecast Demonstration Project (WWRP FDP) aimed to demonstrate the utility and impact of modern nowcast systems. The project focused on the use of radar processing systems and products for nowcasting, including severe weather. The forecast problems facing the Australian Bureau of Meteorology (BoM) on these short timescales during the FDP are briefly described. The observing system is then discussed and enhancements to the network that supported the Olympic Games forecast requirements and the WWRP FDP project are outlined. In particular, issues related to radar calibration and quality control are discussed in some detail. The paper concludes with a brief discussion on the observing system requirements to meet such modern nowcast systems, areas of further development, and impacts that the FDP had on BoM nowcasting systems. The need for end-to-end design of systems from data gathering, to analysis and product generation is emphasized.

1. Introduction

The World Weather Research Programme (WWRP) sanctioned the Sydney 2000 Forecast Demonstration Project (FDP). The aim of this project was “To demonstrate the capability of modern forecast systems and to quantify the associated benefits in the delivery of a real-time nowcast service.” Here, nowcasts refer to forecast weather on the 0–6-h time frame with the emphasis on severe weather detection, forecasts of storm movement, precipitation forecasts, and wind shifts. These nowcasts are an important component of operational forecasts for aviation meteorology and public safety as well as having significant implications for weather-sensitive industries and sporting events. The FDP was formed to support the provision of weather services provided by the Bureau of Meteorology (BoM) for the Sydney 2000 Olympics, but was continued after the games in order to sample the beginning of the “storm season.” Given the potentially high impact of weather during the Olympics the aim of this project was to demonstrate how operationally tested state-of-the-art nowcast systems can provide an improved nowcast service. As the project developed, the main focus of activities centered on radar-based methodologies for detecting and forecasting significant weather on the 0–2-h time frame. Significant weather included wind shifts and nonsevere rain affecting sporting events or aircraft operations as well as severe weather.

This paper will describe the primary forecast problems relevant to the FDP, the observing system in the Sydney area, improvements to the network for the FDP, and the requirements for the successful implementation of the WWRP forecast systems. This latter section will include discussion of quality control issues vital for the successful implementation and operation of these systems. The paper will conclude with a discussion on how well the observing system supported the requirements for the FDP products and the forecast problems generally, as well as the impact on BoM observing system designs. Note that the individual FDP systems and specific events will be discussed in detail in other papers in this special issue.

2. Primary forecast problems and requirements

a. Climate

Sydney's climate is characterized by many different types of extreme weather, including severe convective storms and bushfires. For example, on 14 April 1999, a severe hailstorm struck Sydney's eastern suburbs with hailstones of a diameter greater than 9 cm causing widespread damage with costs estimated to be greater than $1 billion Australian. On 6 August 1986 more than 300 mm of rain fell on the city in 1 day. On 21 January 1991, a severe thunderstorm produced wind gusts estimated to be in excess of 230 km h−1 (BoM 1995). The city is surrounded on three sides by eucalypt forests that impinge into urban areas in many parts, bringing a very real bushfire risk. For example, the January 1994 fires totally destroyed more than 200 urban houses and three people were killed (NSW Rural Fire Service 1998).

The average rainfall throughout the year is quite uniform, but convective weather strongly peaks in the summer. The timing of the Olympic Games was chosen to minimize the possibility of severe weather, but severe thunderstorms have been observed during September. The FDP period was extended beyond the Olympics to November in order to improve the probability of sampling severe weather events.

Sydney, on Australia's east coast, is built in a natural basin with higher-elevation terrain to its north, south, and west where the Great Dividing Range reaches more than 1000 m in elevation (Fig. 1). This orography plays a major role in the weather experienced in Sydney, particularly on the amount and distributions of precipitation, the position and strength of wind changes, and the development of thunderstorms.

Fig. 1.

Observing network and topography in the Sydney area. The circles enclose the area with dual-Doppler radar coverage

Fig. 1.

Observing network and topography in the Sydney area. The circles enclose the area with dual-Doppler radar coverage

b. Operational responsibilities

The Bureau of Meteorology's New South Wales (NSW) Regional Office has the responsibility of providing forecasts for the Sydney area. The forecasting service is divided into different areas of responsibility with aviation forecasters, public weather forecasters, hydrologists, and severe weather forecasters working closely together in the NSW Regional Forecasting Centre in Sydney.

One of the major customers of the bureau is the aviation industry with its focus on operations at Sydney Airport. The BoM has a team of specialized aviation forecasters to provide a dedicated forecasting service at Sydney Airport. There are a wide range of weather phenomenon that can cause disruptions to the effective and efficient operation of the airport, including wind shifts, thunderstorms, below airport minima conditions (low cloud and/or visibility), fog, turbulence, and wind shear. Sydney is Australia's busiest airport and delays flow onto the whole country. Variations in surface wind velocity can have major safety and efficiency ramifications to operations.

Storms of almost any intensity can affect airport operations. For safety reasons, the risk of lightning associated with thunderstorms causes all ground operations to cease until the thunderstorms are well clear of the airport perimeter. During this time no aircraft are loaded, unloaded, or refueled, sometimes leading to long delays in flight schedules. Lightning alerts are issued for Sydney Airport when thunderstorms are within 10, then 5 n mi (∼18, 9 km) of the airport and are forecast to cross the airfield. A graphical lightning alert product depicting thunderstorm movement was developed for use by the airlines prior to the Olympics. It has the capability of using predicted thunderstorm tracks from any thunderstorm tracking system such as the Thunderstorm Identification, Tracking, Analysis, and Nowcasting algorithm (TITAN; Dixon and Wiener 1993) and is delivered via a fax and via a Web page (Bally 2004, in this issue). Systems such as the Auto-nowcaster (ANC; Mueller et al. 2003) and the BoM three-dimensional radar picture (3D-Rapic) display of weather radar data allowed meteorologists to analyze the location of convergence lines, which can directly affect storm growth, movement, and dissipation.

During the Olympic Games in September 2000, there was the additional requirement for the BoM to provide detailed site-specific forecasts for the Olympic venues. The Olympic nowcast and warning services consisted of advices1 of weather likely to impact on sporting events within a timescale of hours. These included severe and nonsevere thunderstorms and heavy rainfall events. For some sports, for example, softball, short-term information about heavy showers was also crucial, since there was a requirement to cover the pitching area when this occurred. Wind changes were also of crucial importance, especially for the sailing events, which were the most weather sensitive of all the Olympic events. For this reason a special weather forecast office was established at the sailing venue at Rushcutters Bay. Forecasts included a morning forecast of hourly wind speeds and directions expected in three different parts of the harbor as well as offshore. The details of the methods used to communicate WWRP information to forecasters and weather clients and their effectiveness is discussed in detail by Anderson-Berry et al. (2004) in this special issue.

The specific issues relating to the need for high-resolution Doppler radar systems include 1) the radar detection of existing storms in a complex mix of mountainous, coastal, and urban environments, and 2) coverage of the planetary boundary layer in order to observe mesoscale convergence associated with wind shifts and convective forcing. In order to address these issues, a composite strategy to integrate a variety of observations to take advantage of the strengths of each system is needed. This must be done in real time to ensure the timely delivery of products to the forecasters. The provision of a forecasting service depends heavily on identifying the needs of the users of this service and the identification of the systems and infrastructure required to provide the service.

c. Forecast challenges

In addition to threatening public safety and property, thunderstorms have a major impact on the aviation industry, electricity suppliers, and transport authorities. Forecasters rely heavily on a sophisticated observational network when predicting the development and intensity of thunderstorms. Storms are often seen to develop on the ranges to the west of Sydney and track toward the east in the predominantly westerly upper-level airflow (Potts et al. 2000). Forecasters are faced with the challenge of predicting how a thunderstorm will behave as it moves from the mountains, eastward toward the coast. Forecasters assess the potential movement of a thunderstorm using upper-level wind flights, atmospheric profilers, and model data. Once thunderstorms form, their intensity and structure are monitored using radar, and the prediction of potential development or decay is highly dependent on the environment into which they track. A volunteer team of about 1000 storm spotters telephone reports of severe weather to the forecasting center. A network of automatic weather stations (AWSs) report surface conditions.

Boundaries can play a major role in thunderstorm initiation and evolution. There are many different types of wind shifts or boundaries that affect the Sydney area. Sea breezes can develop along the coast and extend many tens of kilometers inland. Precipitation can produce outflows, and offshore and onshore winds can develop and decay with the diurnal cycle as pressure gradient winds dominate. Arguably the most significant of the boundaries in many cases are cold fronts. One significant form that affects Sydney is the “southerly burster” or “buster.” This is a particularly severe form of southerly change where the frontal behavior is modified by the topography of southeast Australia (e.g., Colquhoun et al. 1985; McInnes 1993). Forecasters face the challenge of detecting and accurately predicting the movement and intensity of the boundaries. Other major systems affecting the Sydney area are intense offshore low pressure systems (e.g., Holland et al. 1987), which can rapidly develop bringing severe winds, large swells, and flooding rains.

Observations and predictions of the wind direction and speed are critical to airport operations and bushfire fighting operations. A change in wind can dramatically change the behavior of a fire, sometimes turning the relatively long flank of a fire into an intense “head” fire. This was a particular issue for periods early in the main FDP period when there were several fires in the area surrounding Sydney.

The variety of the severe meteorological conditions affecting Sydney pose a number of challenges in designing an observational network capable of providing a diverse range of data at sufficient temporal and spatial resolution for forecasters to produce detailed forecasts and warnings. It is important to note that the storms or winds do not have to reach severe criteria to affect airport operations or sporting events.

3. FDP systems: Aims and requirements

As discussed by Keenan et al. (2003), the FDP systems had three main objectives: rain/cell position forecasts, severe weather, and product dissemination. The FDP system products and data requirements are listed on Table 1, and these are discussed in more detail in other papers in this special issue. With a focus on the use of radar-based guidance for 0–2-h forecasts, there are clearly several common themes. The rain forecasts required high-quality high-resolution radar reflectivity data and in some cases some additional input such as NWP wind fields. The severe weather algorithms also require high-resolution reflectivity data as well as velocity data. The latter is clearly important for damaging wind detection.

Table 1.

FDP forecast systems, products, data sources, and references

FDP forecast systems, products, data sources, and references
FDP forecast systems, products, data sources, and references

In many respects the two most experimental systems were the polarimetric hydrometeor classification scheme (that requires high-quality polarimetric data) (Vivekanandan et al. 1999; Keenan 1999) and the 4D data assimilation system, an adjoint model within the National Center for Atmospheric Research (NCAR) Auto-nowcaster (Mueller et al. 2003; Crook and Sun 2002). The former is at the leading edge of radar technology and is only just entering operations while the latter is attacking the problem of how to assimilate multiple–Doppler radar data, surface data, and sounding data on a very fine resolution grid for analysis and short-term forecasts of convective initiation.

Successful operation of all of these systems is highly dependent on obtaining high-quality weather radar data. Significant issues include calibration for quantitative precipitation estimation, clutter rejection in the urban environment around Sydney, radar sensitivity for clear-air coverage (for detecting convergence lines), and velocity data quality control. These will be discussed in detail below. Other data sources also had quality control issues. In particular the spatial distribution of the AWS network was less than ideal to support integrated observing approaches and will be discussed later. However, the operation of the FDP systems was critically dependent on the quality of the input data. This became a significant issue in the buildup phase of the FDP. Preexisting observational standards required considerable enhancements to meet FDP requirements.

The FDP systems were automatically producing products and guidance for the forecasters. However, the products require some interpretation and assessment of their reliability for a given situation, and the time for training on a number of diverse systems was very limited. During the FDP there were experienced operators for each of the systems present. Each day, a system “champion” was chosen to be the primary contact between the FDP team and the forecasters. A single point of contact was necessary as the forecasters were extremely busy during significant weather events. The forecasters used the output as guidance, and tools were developed to manipulate and edit the product output (Bally 2004). Overall, the interactions were valuable for all concerned. This is discussed in more detail by Anderson-Berry et al. (2004) in this special issue.

4. Weather events during the Sydney FDP

The demonstration project was conducted over the period September–November 2000. This period included the Sydney 2000 Olympic and Paralympic Games. The timing of the games was chosen to minimize the probability of severe weather. It was for this reason that the FDP continued for 2 months following the games to cover the onset of the “severe weather season.”

Table 2 shows a list of significant weather events during the games. The most significant convective weather occurred in the last month of the project (3 and 30 November). While there were several instances of intense convection, the number of severe events was quite small. There are sufficient cases for testing many of the individual components of the systems and several cases for the study of nonsevere convection. The focus for some of the systems was on nowcasting widespread rain events. Again there were a limited number of examples, but these are sufficient for some validation studies (Ebert et al. 2004), in this special issue.

Table 2.

Weather types observed during the FDP, Sep–Nov 2000. Frontal types associated with the convergence line verifications are also indicated. SBF indicates sea-breeze front, while the lowercase letters indicate the motion of the front (e.g., seF is a southeasterly front). An asterisk indicates that TIFS nowcasts were made during this period

Weather types observed during the FDP, Sep–Nov 2000. Frontal types associated with the convergence line verifications are also indicated. SBF indicates sea-breeze front, while the lowercase letters indicate the motion of the front (e.g., seF is a southeasterly front). An asterisk indicates that TIFS nowcasts were made during this period
Weather types observed during the FDP, Sep–Nov 2000. Frontal types associated with the convergence line verifications are also indicated. SBF indicates sea-breeze front, while the lowercase letters indicate the motion of the front (e.g., seF is a southeasterly front). An asterisk indicates that TIFS nowcasts were made during this period

The sea breeze occurred on most days and while this may seem a relatively minor component of nowcasting, it had implications for air quality, aviation weather, and fire weather with several instances of bushfires in the areas just surrounding Sydney, as well as convective initiation (and decay). For example, the sea breeze affected the motion of the tornadic storm of 3 November. Particularly clear and interesting examples of sea breezes are listed as “fair weather days” in Table 2. The discussion will now turn to the observational network that was in place to support the forecasters in addressing these issues and what was needed to produce radar data of sufficient quality for the objective forecast aids to supply useful products.

5. Operational network

The observing system in the Sydney region is shown in Fig. 1. Forecasters had access to wind and thermodynamic soundings from Sydney Airport twice a day (1130 and 2330 UTC) and an additional (radar tracked) pilot-balloon (pibal) observation. The wind measurements are obtained using radar tracking of the sondes and have a precision of better than 1 m s−1. A small number of additional sondes were launched on request when severe weather events were threatening.

There were a variety of satellite data available including National Oceanic and Atmospheric Administration (NOAA) polar orbiters, and hourly visible, IR, and water vapor images from the Japanese Geostationary Meteorological Satellite-5 (GMS-5). These data are used both manually for subjective interpretation to identify synoptic and subsynoptic weather features, and are assimilated into the numerical weather prediction systems. The most routinely used data, evaluated manually by the forecasters, are the IR and visible imagery from the GMS satellite. This provides 1-km resolution visible and 4-km resolution IR images as well as 6.7-μm water vapor images.

In the years leading up to the Olympics the AWS network was expanded significantly with an emphasis on providing data at the Olympic venues. This was a necessary step for the site-specific forecasts for the Olympics, both for establishing some climatology of the sites and for nowcasting. However, the site-specific approach led to an uneven distribution of surface observations. Hence, the network was supplemented by four research-grade AWSs during the FDP, to fill data gaps and provide a more even coverage to the north and south of the venues. This was important both for the tracking of wind shifts, and for high-resolution data assimilation performed within the ANC. There was a network of 33 AWS stations within the Sydney area, including one offshore. An extensive network of 121 rain gauges was used both in real time for hydrological applications and for validation of radar products.

Wind and temperature data from commercial aircraft (Aircraft Meteorological Data Relay, AMDAR) were available. A wind profiler capable of measuring winds down to about 450 m AGL every 15 min was deployed at Sydney Airport (Vincent et al. 1998). One of the primary purposes of this was to provide wind measurements overnight when the airport is closed (and AMDAR unavailable) and before the morning BoM radiosonde flight. These data were also assimilated into the ANC. The low-level capability of this instrument has been improved since the FDP so that it now provides reliable wind information down to 300 m AGL.

The data from all the radars were fed into the bureau's 3D-Rapic radar communications and display system. A modified version of this 3D-Rapic software then supplied the data to the FDP systems in their native formats. The 3D-Rapic system includes a 3D visualization system, psuedo-range–height indicator scans (RHIs, essentially spatial cross sections along the radar beam) and displays of both reflectivity and Doppler velocity. The TITAN (Dixon and Wiener 1993) cell identification and tracking is integrated within the system to provide objective guidance. An example of the operational display is shown in Fig. 2. This total system is the primary display used by the bureau's severe weather forecasters for the detection of severe weather. The forecasters can set the threshold reflectivity that defines a cell and usually have several instances running. A 45-dBZ threshold is often used for tracking more severe convection. Other data are displayed in the bureau's Australian Integrated Forecast System (AIFS): a system similar to the Advanced Weather Interactive Processing System (AWIPS; e.g., Seguin 2002) used in the United States.

Fig. 2.

Example of the BoM operational 3D-Rapic radar data display showing storms and forecast cell positions generated by the TITAN algorithms. The purple polygons are the past positions and the red polygons the forecast position at 10-min intervals. The forecasters can display tracks for all or selected cells and display past history and/or forecast positions. A cell is defined by the 45-dBZ boundary. This example shows just one selected track. The color palette has been enhanced to show only reflectivities greater than 20 dBZ. The red arrow numbers are storm speed in km h−1

Fig. 2.

Example of the BoM operational 3D-Rapic radar data display showing storms and forecast cell positions generated by the TITAN algorithms. The purple polygons are the past positions and the red polygons the forecast position at 10-min intervals. The forecasters can display tracks for all or selected cells and display past history and/or forecast positions. A cell is defined by the 45-dBZ boundary. This example shows just one selected track. The color palette has been enhanced to show only reflectivities greater than 20 dBZ. The red arrow numbers are storm speed in km h−1

The forecasters also had access to the full numerical weather prediction (NWP) system fields for a variety of BoM models including limited area models at 5-, 12.5-, and 37.5-km resolution (Puri et al. 1998), a global system at 75-km resolution, and limited output from a variety of international systems such as the European Centre for Medium-Range Weather Forecasts (ECMWF), the Met Office, and U.S. models.

6. Radar system issues

Three radars were used during the FDP. The operating parameters of the three weather radars are given in Table 3. The original BoM radar network included the Kurnell Doppler and Wollongong conventional radars. All radar sites represent compromises and the bureau systems are no exception. The Wollongong radar has a relatively wide beamwidth and is located at a relatively high elevation and suffers from significant ground clutter. The Kurnell Doppler radar was deployed in part to support aviation applications. The only site that was available in a suitable location for these applications suffers from its location being on the coast and subject to sea clutter and extensive ground clutter over suburban Sydney. The Kurnell radar also has limited clear-air coverage over western Sydney where storms are often interacting with sea breeze and other local circulations because of its distance from this area. All the radars suffer significant blockage over the mountains around Sydney. The C-band polarimetric research radar (C-Pol; Keenan et al. 1998) was deployed in western Sydney at Badgeries Creek (hereafter referred to as the Badgeries Creek radar) to provide improved coverage over western Sydney, and a different view over the mountains, as well as dual-Doppler coverage over much of the Sydney urban area.

Table 3.

Operating parameters and characteristics of the three radars used in the FDP. Values in brackets were the pre-FDP settings

Operating parameters and characteristics of the three radars used in the FDP. Values in brackets were the pre-FDP settings
Operating parameters and characteristics of the three radars used in the FDP. Values in brackets were the pre-FDP settings

a. Calibration

With the presence of three radar systems with significant overlapping of their coverage, calibration is a major issue. This is an issue with all weather radars used for quantitative rainfall measurement and was the subject of a workshop in 2001 (Joe and Smith 2001). The Badgeries Creek radar (Keenan et al. 1998) was calibrated using a solid target (sphere) supplemented by solar calibration and absolute measurements within the system. The two operational radars used solar calibration and internal measurements.

Early in the project it became evident there was some difference in the observed radar reflectivity between the radars. The magnitude of this difference was difficult to quantify. Differences would be expected in individual cases because the radar reflectivity can vary significantly over small ranges and the three radars had different operating parameters. In particular, the operating wavelength, beamwidth, and range resolution for the radars are different and for a given point in space there will be differences in the sampling volume and beam elevation for each of the radars. Attenuation effects may also vary for the different radars with the 5-cm wavelengths used by the radars at Kurnell and Badgeries Creek having more severe attenuation than the Wollongong 10-cm radar. Thus, differences are expected depending on the location of the echoes relative to the radars, the intensity of the echoes, and whether they resulted from convective cells or stratiform rain.

In an effort to check the calibration and quantify the differences, a constant altitude plan position indicator (CAPPI) scan from each of the radars was interpolated to a Cartesian grid with a spatial resolution of 500 m and then remapped to a common origin centered between the three radars. Only data within 40 km of the common origin were considered in order to reduce beamwidth effects. The difference in the observed radar reflectivity at each of the grid points was then calculated and the distribution was examined to identify any bias. If no data were available for either radar at a grid point, the difference was set to a null value. Small errors in the remapping to a common origin will result in some misalignment of the radar echoes but sensitivity tests showed this does not significantly bias the distribution of the reflectivity difference. Based on the biases that were identified, the calibration of the radars was reexamined and corrections made. In this way the calibration was improved such that the mean difference in the radar reflectivity was consistently within 2 dBZ.

The procedure followed is illustrated in Figs. 3a–c, which shows the remapped CAPPI scans for an elevation of 2133 m for 1030 UTC 26 September 2000 for the Kurnell, Badgeries Creek, and Wollongong radars. In general there is good agreement between the radars with the main difference evident in the Wollongong data. This shows no data in the area at the south end of the main line of echoes because this area is too close to the radar (for the CAPPI at this altitude). Quality control algorithms on this radar have also resulted in some pixels within the main line of echoes being set to null values. The differences in reflectivity between the radars were calculated and are shown in Fig. 4. They show distributions that are approximately normal and mean differences that are less than 2 dBZ. Some asymmetries are visible in these plots and they are associated with some attenuation effects. This level of agreement requires careful calibration measurements.

Fig. 3.

CAPPI scans at a height of 2133 m for the three Sydney radars for 26 Sep 2000. The radars are (a) Badgeries Creek (C-Pol), (b) Kurnell (middle), and (c) Wollongong

Fig. 3.

CAPPI scans at a height of 2133 m for the three Sydney radars for 26 Sep 2000. The radars are (a) Badgeries Creek (C-Pol), (b) Kurnell (middle), and (c) Wollongong

Fig. 4.

Histograms of the reflectivity differences from the three radars, with (a) Badgeries–Kurnell, (b) Wollongong–Badgeries, and (c) Wollongong–Kurnell for the CAPPI scans in Fig. 3 

Fig. 4.

Histograms of the reflectivity differences from the three radars, with (a) Badgeries–Kurnell, (b) Wollongong–Badgeries, and (c) Wollongong–Kurnell for the CAPPI scans in Fig. 3 

The Wollongong radar has a greater beamwidth than the Kurnell and Badgeries Creek radars that will result in a greater azimuthal extent of radar echoes and a reduction in the maximum reflectivities. For the purposes of these comparisons, the impact of these factors was evaluated by applying a filter to the Kurnell and Badgeries Creek data that averaged the reflectivity in azimuth to simulate a beamwidth comparable to the Wollongong radar. The difference was recalculated and there was a small change in the distribution. The impact on the mean difference was less than 1 dBZ.

b. Location and boundary layer coverage

With a strong focus on detecting wind shift boundaries, clear-air coverage was of paramount importance. The Wollongong 10-cm radar is generally unable to see significant clear-air echoes, at least partly because of its high location, broad beamwidth, and low-level clutter at the lowest elevation. The Kurnell 5-cm Doppler radar was located specifically for observing boundary layer convergence layers. Both it and the Badgeries Creek radar routinely monitored the sea-breeze movement across the Sydney basin. It was also possible to see other convergence lines in western Sydney from Badgeries Creek on a regular basis. It should be noted that the clear-air coverage here, as in other locations, is highly seasonal and has a strong diurnal modulation. Clear-air echoes are primarily associated with scatter from insects and are generally seen only when the temperature is greater than about 10°C (Wilson et al. 1994). Thus, in the early morning clear-air echoes are quite weak and limited in spatial extent. The area of echoes builds up through the day, but is often very patchy and weak behind the sea breeze. Near dusk, there were echo “blooms” of very intense clear-air echoes as nocturnal insects emerged. The temperature dependence means that the radar is least useful for clear-air coverage in the winter. The clear-air capability enabled the display of fine lines of reflectivity and provided short-term nowcasts of wind shifts that were superior to those using the mesonet alone. This enabled the forecasters to better estimate the movement of the sea-breeze front, an example of which is shown in Fig. 5. This clearly shows the distortion of the front with the frontal propagation being retarded over the higher terrain to the north and south of the Sydney basin. These clear-air data were used to track the sea-breeze convergence lines as they moved across Sydney as well as the detection of other topographically forced convergence lines.

Fig. 5.

Example of the sea breeze at 0630 UTC 3 Oct 2000 shown in the operational display with (left) Kurnell and (right) C-Pol data (reflectivity on the bottom, radial velocity on the top). No scale is given for the reflectivity as it was purely clear-air signals and this display highlights the presence of clear-air signals. Note the bowing of the sea breeze as its propagation is retarded over the high terrain. Quantitative estimates of convergence are available comparing the velocities at the gust front with those behind. The beginning of a land breeze is also visible on the coast in the C-Pol Doppler data. The area of uniform reflectivity offshore of Kurnell is sea clutter

Fig. 5.

Example of the sea breeze at 0630 UTC 3 Oct 2000 shown in the operational display with (left) Kurnell and (right) C-Pol data (reflectivity on the bottom, radial velocity on the top). No scale is given for the reflectivity as it was purely clear-air signals and this display highlights the presence of clear-air signals. Note the bowing of the sea breeze as its propagation is retarded over the high terrain. Quantitative estimates of convergence are available comparing the velocities at the gust front with those behind. The beginning of a land breeze is also visible on the coast in the C-Pol Doppler data. The area of uniform reflectivity offshore of Kurnell is sea clutter

c. Clutter

These radars are located near mountains and near a large city ensuring that ground clutter is a significant issue. For clear-air measurements the second tilt (elevation angle of 1.5°) often showed the cleanest convergence lines, simply because of the reduced clutter. Digital clutter filters are used on the radars, with corrections to the reflectivity for errors induced by the filters estimated within the system. Some analysis systems used for the FDP [Spectral Prognosis (S-PROG; Seed 2003) and Nowcasting and Initialisation for Modelling Using Regional Observation Data (Nimrod; Golding 1998)] made use of CAPPIs calculated at a height of 1.5 km to reduce the effect of clutter on the precipitation estimates (using data from higher radar elevations). The presence of the Blue Mountains (see Fig. 1) also produces considerable blockage as well as clutter contamination.

The additional complication for the operational radars was sea clutter. The automatic systems again masked this out. The sea clutter was distinctive in its uniform intensity structure with an apparent intensity of ∼10 dBZ and the Doppler velocities were close to the clear air in magnitude. Occasionally, anomalous propagation would produce spectacular increases in the magnitude of the sea clutter that could reach 40 dBZ. When this occurred, a distinctive wave structures appeared in the clutter (see Fig. 6 for an example). This usually persisted for several hours.

Fig. 6.

Kurnell radar image showing a sea-breeze signature almost on the coast and spectacular sea clutter that was probably enhanced by anomalous propagation

Fig. 6.

Kurnell radar image showing a sea-breeze signature almost on the coast and spectacular sea clutter that was probably enhanced by anomalous propagation

There were some preliminary quality control algorithms employed at the Kurnell radar site. Data that had signal levels less than 1 dB above the receiver noise level and had an autocorrelation of the echo signal intensity from pulse to pulse that was less than a threshold value (0.22) were rejected. These parameters were set subjectively by examining the clear-air signals so that the clear-air velocities looked reasonably smooth, but not such that too much data were removed. More rigorous settings produced data with less noise, but at the cost of the clear-air coverage that is vital for detecting wind shift boundaries. The resulting data mask was then applied to both the reflectivity and velocity data during the FDP to ensure both fields were consistent in their data coverage. However, since the end of the FDP, the autocorrelation threshold was applied only to the velocity data and not the reflectivity. This was because application of the autocorrelation threshold resulted in missing data during intense storms when the spectral width of the echoes was very wide. Spectral processing and allowing for wide intense spectra would improve this situation.

d. Velocity aliasing issues

The prime source of velocity data for the systems was from the Kurnell dual pulse repetition time (PRT) radar. The radar was operated with a 1000/750-Hz pulse repetition frequency (PRF) producing an effective Nyquist velocity of 40 m s−1. The raw velocity data collected with the radar had significant speckle because of the effect of random measurement errors and real spatial gradients (May and Joe 2001; Joe and May 2003). This speckle caused great problems for the automated systems with unacceptable rates of false detections of mesocyclones, tornadoes, and microbursts.

Velocity fields were largely corrected by applying a median filter to the velocity data. The simple algorithm consisted of replacing the raw target pixel by the median of a 3 × 3 array of velocity points centered on the target. This was only applied if there was a valid velocity estimate for the target pixel. An example from a tornado observation is shown in Fig. 7. Clearly the field is smoothed slightly, but the important meteorological features are retained. Without this error correction the random speckle caused havoc with the automated feature detection algorithms.

Fig. 7.

Radar data from an elevation of 5.5° observed at 0505 UTC 3 Nov 2000 with the Kurnell 5-cm radar of a mesocyclone case. The figure shows (top) the reflectivity with a distinct hook echo, (middle) raw velocity data, and (bottom) median filtered velocity data. The median filtered data were supplied to the automatic algorithms and all the mesocyclone detectors were triggered by this event and an F1 tornado was observed. Positive values denote velocity away from the radar

Fig. 7.

Radar data from an elevation of 5.5° observed at 0505 UTC 3 Nov 2000 with the Kurnell 5-cm radar of a mesocyclone case. The figure shows (top) the reflectivity with a distinct hook echo, (middle) raw velocity data, and (bottom) median filtered velocity data. The median filtered data were supplied to the automatic algorithms and all the mesocyclone detectors were triggered by this event and an F1 tornado was observed. Positive values denote velocity away from the radar

The only cases seen where the filtered dual-PRT radar data were noisy occurred with high velocities at low elevations over the Blue Mountains. This is a result of the peak in the signal spectrum of the velocity data in the individual beams being aliased and merging with the ground clutter. The combination of the clutter filters “notch” frequency and the severe clutter produced unusable data in these situations. This is discussed in more detail by Joe and May (2003).

The Badgeries Creek polarimetric radar was operated at a fixed PRT, so Doppler velocity aliasing became a significant issue. The main system that was using these data was the ANC with its model adjoint assimilated winds. The ANC employed an automatic dealiasing scheme, but this was not always successful. However, postanalysis with improved algorithms (e.g., James and Houze 2001) shows considerable promise. In general it is highly recommended that dual-PRT operation with error correction schemes be adopted for operational 5-cm wavelength radar operation (Joe and May 2003).

e. Scanning strategy

Timeliness of the data is essential for any nowcast system. With the focus on severe weather the Kurnell radar was collecting volume scans every 5 min. The other two radars were on 10-min scans. This allowed the collection of long-range data with the Wollongong 10-cm system and high-quality polarimetric data with Badgeries Creek. Badgeries Creek (C-Pol) transmits alternate horizontally (H) and vertically (V) polarized pulses. The need for a reasonable number of pulses at each polarization limits the allowable antenna rotation rate compared with a conventional Doppler radar. This is one reason why alternate schemes such as transmitting the H and V polarizations simultaneously are being explored for operational applications (Doviak et al. 2000).

The faster update cycle for the Kurnell radar was a compromise. The 5-min update cycle limited the number of elevations in the volume to 11 compared with 15 for the 10-min cycle and a long-range surveillance scan was only performed using the wider-beamwidth Wollongong radar. However, there were significant gains with respect to storm forecasting. Figure 8 shows a comparison of the accuracy of the forecast tracks using the TITAN software. For this analysis, a storm was defined using a reflectivity threshold of 45 dBZ and forecasts were verified against subsequent TITAN analyses, ensuring a consistent approach for detection and verification was used. This clearly shows the positive impact of the faster update cycle for storm tracking, with an effective increase of about 5–10 min in lead time for a given track error rate. This increase is twofold. The more rapid updates allow more accurate forecasts, and successive updates are also given more frequently increasing the effective lead time by another 5 min. Additional modifications to the cell tracking routines are being performed to update the forecasts as data from individual radar tilts arrive, potentially adding a further few minutes of lead time.

Fig. 8.

TITAN forecast track errors. Forecasts are verified against subsequent TITAN analyses. Solid lines are for 5-min volume scans and dashed for 10-min volumes. Values for storms with a maximum reflectivity in two ranges are given

Fig. 8.

TITAN forecast track errors. Forecasts are verified against subsequent TITAN analyses. Solid lines are for 5-min volume scans and dashed for 10-min volumes. Values for storms with a maximum reflectivity in two ranges are given

7. Discussion

The previous sections have described the observing systems in place to support the operational forecasters and the FDP analysis systems. In this section, the question to be addressed is how well the combination of systems allowed the forecasters to address the forecast issues described in section 2.

Some specific forecast issues that were faced during the FDP included the following system detection and tracking issues:

  1. knowing if existing storms would move off the mountains into the Sydney basin;

  2. knowing if existing storms along the coast would move inland;

  3. forecasting convective cell movement where significant variations in storm motion were being observed, affecting both quantitative precipitation estimation and cell tracking; and

  4. detecting and forecasting local circulations such as the motion of the sea-breeze front.

And more fundamental dynamical issues, which are more difficult to address, included the following:

  • 5. forecasting storm initiation, growth, or dissipation within the domain as they moved toward Sydney, with emphasis on anticipating if storms would form on the sea-breeze front or develop along the east slope of the Blue Mountains with moist easterly flow (this was particularly significant after the tornado on 3 November);

  • 6. the transition from ordinary to severe thunderstorms; and

  • 7. diagnosing the severity of the weather at specific sites (a detection issue).

Issues 1–3 are effectively storm tracking problems and in principle are tractable with one well-placed radar, although the storm evolution certainly requires additional data and analysis including systems such as the ANC (Crook and Sun 2002). The utility of the FDP systems for automatically identifying, tracking, and diagnosing weather events is discussed in detail in companion papers (e.g., Wilson et al. 2004, in this issue), but a couple of general comments can be made. The tracking of thunderstorms and the diagnosing of storm severity certainly requires quality-controlled radar data for the automated FDP algorithms. The 5-min radar update cycle showed significant improvement in tracking storms and this is even more apparent with fast-moving storms.

The outstanding issues related to radar data quality control and data analysis for the automated algorithms, affect both the detection rate and the false alarm rate of severe weather phenomena. High spatial resolution is needed to resolve specific severe weather features such as bounded weak echo regions and hook echoes. This indicates that range resolution of about 250 m is required. This resolution is somewhat arbitrary, but based on our experience, it is sufficient to resolve cloud features out to ranges where the physical width of the radar beam becomes the limiting factor. Coarser pulse widths blur many of the important features for detecting severe weather signatures.

Beyond this, there are limitations for forecast skill that lie in scientific issues such as the fundamental storm predictability. Similar comments apply to issue 4 (detecting and forecasting local circulations), although the data from the surface network are even more important than for issues 1–3 (relating to cell movement) and probably need to be extended beyond what was available in Sydney for optimal analysis. This is particularly true at ranges near the limit of clear-air detection. We note that the Badgeries Creek radar was particularly useful in this regard during the FDP (but unfortunately this radar is no longer in Sydney). The clear-air coverage is seasonal, and the clear-air signals during winter are weak and the coverage is very limited.

Issues 5 and 6 are really related to questions about convective initiation and to evolution issues. As is discussed in detail by Wilson et al. (2004), these critical forecast problems remain some of the most challenging in the field. While there was considerable success with the FDP systems during the 3 November tornado event, future progress will be intimately linked to true mesoscale analysis methods, such as were being tested with the ANC. However, serious scientific issues remain on several levels. These include basic scientific understanding of convective processes, the ability to observe the required variables at the required resolution, and how to assimilate the available data at the highest resolution. For example, the humidity field is clearly a key parameter, but how much detail is needed with respect to its mesoscale structure? Providing some answers to these questions are among the aims of the International H2O project (Parsons et al. 2000).

The problems are even more serious with the main initiation area is over the mountains. Having said this, one of the prime difficult issues facing forecasters is the development or decay of the storms as they moved over the Sydney basin and interacted with other convergence lines. A case study of this is in the paper discussing the 3 November tornado in this issue (Wilson et al. 2004), but we consider that clear-air coverage over the whole domain of interest is required—in this instance requiring at least two Doppler radars. A probable major limitation is the relatively coarse observations of water vapor and temperature.

Several of the FDP systems focused on issue 7, the diagnosis of severe weather. In many ways the radar network was designed to address this particular issue for storm warning and aviation applications. The radar systems in place generally support this activity well given the inherently small-scale nature of many severe weather elements (Ebert et al. 2004).

Most of the demonstration systems tested during the FDP were developed for use in the data-rich North American environment including frequently updated high-resolution satellite data that were simply unavailable in Sydney. Their translation to Sydney exposed system design features that were “hard coded” to their “native” environments. Some were designed to incorporate high temporal resolution satellite data and dense surface networks. With considerable effort from all participants, the systems worked satisfactorily in Sydney, producing some notably successful forecasts. But, the Sydney experience suggests that many of the FDP systems would suffer degraded performance in environments that offer less data-gathering infrastructure than was available during the FDP.

8. Conclusions

The FDP had a profound impact on the routine operations and development of the BoM radar systems. The transition from manual qualitative interpretation of radar imagery to automated forecast and analysis aids here required a far greater effort on quality control. These more stringent requirements relate to both the support of the automated algorithms and the move to quantitative precipitation estimation. As can be seen in Table 3, the Doppler radar is also being operated with a larger Nyquist interval and at much higher spatial resolution. Both these characteristics are needed for the automated algorithms and to support manual forecaster analysis and insight. Accurate calibration and integration of the radar data with rain gauge measurements are crucial (e.g., section 6a).

The project is continuing to have significant implications for the BoM systems. Not just the adaptation and adoption of components of the individual FDP systems, although this is happening, but the need for high quality data in these automated systems is also leading to further developments. For example, a comprehensive radar reflectivity quality control package has been developed (Seed and Pegram 2001) and is being integrated into the radar display and analysis package. This is an essential step, not just for the methodologies discussed in this special issue, but for such activities as the direct application of radar data to drive hydrological models for flood forecasting and for assimilation into NWP systems.

One of the important lessons from the FDP was the need to consider the end-to-end requirements for the whole forecasting system. This includes not just optimizing the observing network for the forecast problems and displays to assist the forecaster's analysis, but considering the whole process from raw data quality control, to system design and providing the means to utilize and edit the automated products directly into forecasts (e.g., Bally 2004, in this issue).

The basic aims of a system need to be an integral part of network design. A mixture of systems is always going to be necessary. The limitations of each system must always be kept in mind. For example, there is limited clear-air radar coverage in Sydney over a significant part of the year. Furthermore, if convergence lines in western Sydney are deemed to have important implications for severe weather detection, a permanent radar installation in the area is warranted. The data from Badgeries Creek were important in the detection of boundaries in western Sydney, which was an important part of forecasts for wind shifts during bushfires and for monitoring the interaction between the sea breeze and a supercell thunderstorm that spawned a tornado in suburban Sydney on 3 November. Data assimilation systems are becoming a key component for weather analysis on the mesoscale, an example of which is the adjoint model within the ANC system. With more detailed and complex data, the requirements of such systems must be considered as they will play an increasing role in the synthesis of weather information by the forecaster and understanding the state of the atmosphere.

Acknowledgments

The project was conducted with the direct and indirect support of many agencies and individuals. Many people contributed to the success of the project. The participation of the Auto-nowcaster and NCAR personnel was made possible by the U.S. Weather Research Program and NSF base funds at NCAR. As a host, the BoM provided considerable ongoing and infrastructure support for the 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 Programme Sydney 2000 Forecast Demonstration Project (WWRP S2000 FDP).
Wea. Forecasting
,
19
,
168
178
.
Bally
,
J.
,
2004
:
The Thunderstorm Interactive Forecast System: Turning automated thunderstorm tracks into severe weather warnings.
Wea. Forecasting
,
19
,
64
72
.
BoM
,
1995
:
Report on the 21 January 1991 Sydney severe thunderstorm.
Bureau of Meteorology, Melbourne, Victoria, Australia, 26 pp
.
Colquhoun
,
J. R.
,
D. J.
Shepherd
,
C. E.
Coulman
,
R. K.
Smith
, and
K. L.
McInnes
,
1985
:
The southerly buster of southeastern Australia: An orographically forced cold front.
Mon. Wea. Rev.
,
113
,
2090
2107
.
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
.
Doviak
,
R. J.
,
V.
Bringi
,
A.
Ryzhkov
,
A.
Zahrai
, and
D.
Zrnić
,
2000
:
Considerations for polarimetric upgrades to operational WSR-88D radars.
J. Atmos. Oceanic Technol.
,
17
,
257
278
.
Ebert
,
E. E.
,
L. J.
Wilson
,
B. G.
Brown
,
P.
Nurmi
,
H. E.
Brooks
,
J.
Bally
, and
M.
Jaeneke
,
2004
:
Verification of nowcasts from the WWRP Sydney 2000 Forecast Demonstration Project.
Wea. Forecasting
,
19
,
73
96
.
Eilts
,
M.
, and
Coauthors
,
1996
:
Severe weather warning decision support system.
Preprints, 18th Conf. on Local Severe Storms, San Francisco, CA, Amer. Meteor. Soc., 536–540
.
Golding
,
B. W.
,
1998
:
Nimrod: A system for generating automated very short range forecasts.
Meteor. Appl.
,
5
,
1
16
.
Holland
,
G. J.
,
A. H.
Lynch
, and
L. M.
Leslie
,
1987
:
Australian east-coast cyclones. Part I: Synoptic overview and case study.
Mon. Wea. Rev.
,
115
,
3024
3036
.
James
,
C. N.
, and
R. A.
Houze
Jr.
,
2001
:
A real-time four-dimensional Doppler dealiasing scheme.
J. Atmos. Oceanic Technol.
,
18
,
1674
1683
.
Joe
,
P. I.
, and
P. L.
Smith
Jr.
,
2001
:
Summary of the radar calibration workshop.
Preprints, 30th Int. Conf. on Radar Meteorology, Munich, Germany, Amer. Meteor. Soc., 174–176
.
Joe
,
P. I.
, and
P. T.
May
,
2003
:
Practical operations and data characteristics of dual-PRT weather radars.
J. Atmos. Oceanic Technol.
,
20
,
429
442
.
Keenan
,
T. D.
,
1999
:
Hydrometeor classification with a C-band polarimetric radar.
Preprints, 29th Int. Conf. on Radar Meteorology, Montreal, QC, Canada, Amer. Meteor. Soc., 184–187
.
Keenan
,
T. D.
,
K.
Glasson
,
F.
Cummings
,
T. S.
Bird
,
J.
Keeler
, and
J.
Lutz
,
1998
:
The BMRC/NCAR C-band polarimetric (C-POL) radar system.
J. Atmos. Oceanic Technol.
,
15
,
871
886
.
Keenan
,
T. D.
, and
Coauthors
,
2003
:
The Sydney 2000 World Weather Research Programme Forecast Demonstration Project: Overview and current status.
Bull. Amer. Meteor. Soc.
,
84
,
1041
1054
.
Lapczak
,
S.
, and
Coauthors
,
1999
:
The Canadian National Radar Project.
Preprints, 29th Int. Conf. on Radar Meteorology, Montreal, QC, Canada, Amer. Meteor. Soc., 327–330
.
May
,
P. T.
, and
P.
Joe
,
2001
:
The production of high quality Doppler velocity fields for dual PRT weather radar.
Preprints, 30th Int. Conf. on Radar Meteorology, Munich, Germany, Amer. Meteor. Soc., 286–288
.
McInnes
,
K.
,
1993
:
Australian southerly busters. Part III: The physical mechanism and synoptic conditions contributing to the development.
Mon. Wea. Rev.
,
121
,
3261
3281
.
Mueller
,
C.
,
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
.
NSW Rural Fire Service
,
1998
:
A state ablaze—The January 1994 fires. 31 pp.
[Available from NSW Rural Fire Service Community Education Section, Unit 3/175 James Ruse Dr., Rosehill, NSW 2142, Australia.]
.
Parsons
,
D.
,
T.
Weckwerth
, and
M.
Hardesty
,
cited 2000
:
Scientific overview document for the International H2O Project (IHOP_2002).
.
Pierce
,
C. E.
, and
P. J.
Hardaker
,
2000
:
GANDOLF: A system for generating automated nowcasts of convective precipitation.
Meteor. Appl.
,
8
,
341
360
.
Potts
,
R. J.
,
T. D.
Keenan
, and
P. T.
May
,
2000
:
Radar characteristics of storms in the Sydney area.
Mon. Wea. Rev.
,
128
,
3308
3319
.
Puri
,
K.
,
G. S.
Dietachmayer
,
G. A.
Mills
,
N. E.
Davidson
,
R.
Bowen
, and
L. W.
Logan
,
1998
:
The new BMRC Limited Area Prediction System, LAPS.
Aust. Meteor. Mag.
,
47
,
203
223
.
Seed
,
A. W.
,
2003
:
A dynamic and spatial scaling approach to advection forecasting.
J. Appl. Meteor.
,
42
,
381
388
.
Seed
,
A. W.
, and
G. G. S.
Pegram
,
2001
:
Using Kriging to infill gaps in radar data due to ground clutter in real-time.
Proc. Fifth Int. Symp. on Hydrological Applications of Weather Radar, Kyoto, Japan, Kyoto University, A-12
.
Seguin
,
W. R.
,
2002
:
AWIPS—An end-to-end look.
Preprints, Interactive Symp. on the Advanced Weather Interactive Processing System (AWIPS), Orlando, FL, Amer. Meteor. Soc., J47–J50
.
Vincent
,
R. A.
,
S.
Dullaway
,
A.
MacKinnon
,
I. M.
Reid
,
F.
Zinc
,
P. T.
May
, and
B.
Johnson
,
1998
:
A VHF boundary layer radar: First results.
Radio Sci.
,
33
,
845
860
.
Vivekanandan
,
J.
,
S. M.
Ellis
,
R.
Oye
,
D. S.
Zrnic
,
A. V.
Ryzhkov
, and
J.
Straka
,
1999
:
Cloud microphysics retrieval using S-band dual-polarization radar measurements.
Bull. Amer. Meteor. Soc.
,
80
,
381
388
.
Wilson
,
J. W.
,
T. M.
Weckwerth
,
J.
Vivekanandan
,
R. M.
Wakimoto
, and
R. W.
Russell
,
1994
:
Boundary layer clear air radar echoes: Origin of echoes and accuracy of derived winds.
J. Atmos. Oceanic Technol.
,
11
,
1184
1206
.
Wilson
,
J. W.
,
E. E.
Ebert
,
T. R.
Saxen
,
R. D.
Roberts
,
C. K.
Mueller
,
M.
Sleigh
,
C. E.
Pierce
, and
A.
Seed
,
2004
:
Sydney 2000 Forecast Demonstration Project: Convective storm nowcasting.
Wea. Forecasting
,
19
,
131
150
.

Footnotes

Corresponding author address: Dr. Peter May, BMRC, GPO Box 1289K, Melbourne VIC 3001, Australia. Email: p.may@bom.gov.au

1

“Severe thunderstorm advices” are issued as an alert to the public, emergency services, and other organizations that severe thunderstorms are likely to develop or extend into a (possibly broad) specific area over the next few hours. These may extend out to 5 h.