1. Introduction
The purpose of this paper is threefold. The first intent is to provide insights into the convective storm nowcasting challenges specific to Beijing, China. The second intent is to present the results of nowcasts made using a diverse variety of state-of-the-art nowcasting systems during the 2008 Beijing Summer Olympic Games. The third is to provide comments on the present status of nowcasting convective storms, and possible future directions based on these experiences.
A nowcasting demonstration, sanctioned by the World Meteorological Organization’s (WMO) World Weather research Program (WWRP), was conducted in Beijing, China, during the 2008 Summer Olympic Games. This demonstration was called the Beijing 2008 Forecast Demonstration Project (B08FDP). A similar FDP was conducted in Sydney, Australia, during the Sydney Summer Olympics Games (Keenan et al. 2003). The purposes of the FDPs are to accelerate the worldwide development of nowcasting systems and to demonstrate the benefits of nowcasting products to weather-sensitive activities like outdoor sporting events. The stated overall objective for the B08FDP was to demonstrate the benefits of state-of-the-art nowcasting systems for mitigating high-impact weather. The focus was to use the latest science and technology to predict convective weather in the next 6 h, with particular emphasis on the 0–3-h period. The B08FDP brought together nowcasting systems from Australia, Canada, Hong Kong, China, and the United States. The systems were deployed at the Beijing Meteorological Bureau (BMB) of the China Meteorological Administration (CMA). The authors believe the participating systems are generally representative of present operational capabilities of nowcasting systems worldwide.
CMA was tasked with providing nowcasting services to the Olympic Games and invited the WWRP to conduct an FDP. It was agreed that the WWRP would sponsor the installation of several nowcast systems to provide short-term guidance products. Meteorologists associated with the various FDP nowcasting systems would also help train select Chinese forecasters in nowcasting methods. Then, during the Olympic Games, FDP forecasters would provide real-time nowcasting guidance to BMB forecasters. For expedience, the FDP convective nowcasts were made available on the Web for optional use within BMB forecast operations. The BMB forecasters were responsible for all official weather forecasts for the Olympics.
The term nowcasting is used to emphasize that the forecasts are for the 0–6-h time period on a spatial scale of no more than a few kilometers, with an update frequency of ≤1 h. Nowcasting plays an effective role in many tactical weather support operations including aviation weather, marine weather, wildfire weather, severe local storms, and flooding. In recent years there has been a growing desire by both the public and private sectors to receive nowcasts to support almost any outdoor activity, particularly sporting events, to improve safety and facilitate cost effectiveness. For example, forecasters have been actively involved for many years in providing severe storm warnings in effort to protect life, property, and economic interests. The issuance of such warnings is a form of nowcasting. To a large extent, severe storm warnings are based on forecaster interpretation of radar displays, with the help of automated algorithms that identify storm circulation patterns, hail probabilities, and storm tracks, as well as other indicators of storm character and trend (Polger et al. 1994).
Because of the growing volume of increasing higher-resolution information (e.g. satellites, Doppler radars, and mesonet stations) available to operational forecasters, there has been pressure on the research community to develop automated nowcasting systems to assist in the rapid and continuous processing of data. The first automated nowcasting system was implemented at McGill University in Montreal, Quebec, Canada, in 1976 (Bellon and Austin 1978). The history of nowcasting system development is described in Browning (1982) and Wilson et al. (1998). The nowcasting systems demonstrated in Sydney 2000 (Pierce et al. 2004) were a mixture of radar echo extrapolation systems, an expert system, and systems that blended extrapolation and numerical models. The primary finding reported by Wilson et al. (2004) was that improved skill over extrapolation occurred when boundary layer convergence lines, as identified by a forecaster, were ingested into an expert system known as the National Center for Atmospheric Research (NCAR) Auto-Nowcaster (Mueller et al. 2003). This system utilized conceptual models to forecast storm evolution. Through the use of fuzzy logic (McNeill and Freiberger 1993), forecast parameters derived from diverse datasets are combined to yield short-term convective forecasts.
During the 8 yr between the Sydney and Beijing Olympics, advances in computing technology have allowed numerical weather prediction models to use increased spatial resolutions (3–5 km), obviating the need for cumulus parameterizations, and the assimilation of a broader suite of higher-resolution nonconventional data. Three of the B08FDP nowcasting systems utilized such numerical model forecasts and “blended” them with radar echo extrapolation forecasts.
In preparation for the B08FDP, an examination of radar, satellite, sounding, and mesonet station data from 2006–07 was conducted. Based on these studies, it was apparent that the local terrain around Beijing would cause significant nowcasting challenges for both extrapolation techniques and numerical models. To provide quality nowcasts, it was evident that conceptual models of storm evolution needed to be developed for Beijing (and vicinity) along with situational forecast rules. Once refined, these rules were utilized by the FDP forecasters and are presented in section 2.
Section 3 briefly describes the weather events during the intensive 1–24 August period of the FDP. The six weather events that were most challenging to nowcast and also impacted or had the potential to impact the games are the focus of this paper. Section 4 offers a very short description of each of the nowcasting systems and the nowcasting process. Section 5 describes each of the six challenge events and presents the nowcasting skill of the FDP systems. Section 5 also describes the subjective evaluation criteria that were used for these six cases to determine the success–failure of the nowcasting systems. Section 6 is a summary of the present status of nowcasting convective storms, and section 7 outlines future directions.
2. Background studies
In this section we first present some general statements about convective storm behavior in the vicinity of Beijing. Second, we describe some studies from 2006 and 2007 that were used to help establish heuristic nowcasting rules, which are presented at the end of the section.
a. General statements
The metropolitan area of Beijing is on a flat plain located at the foot of the Yan Shan Mountains. The mountains have an elevation of 1.0–2.5 km and they extend roughly in a northeast-to-southwest line to the west and north of the city (see Fig. 1). Beijing is at an altitude of only 30 m and is open to the south and east to the influx of very warm moist air from the Bohai Sea. Significant forecast challenges present themselves in the vicinity of Beijing in response to this very humid air impinging on the nearby mountains. Thunderstorms frequently initiate over the mountains and move to the southeast. Sometimes these storms dissipate upon reaching the plains, but other times they grow and organize into major squall lines. A variety of boundary layer convergence lines frequent the plains and often play a significant role in storm initiation and evolution. Convective storms often form along the sea-breeze front that is generated by the Bohai Sea (see Fig. 1), which is 140 km southeast of Beijing; however, they seldom reach Beijing. The urban heat island effect (Dixon and Mote 2003) of Beijing also appears to be occasionally a contributing cause for triggering thunderstorms.
b. 2006–07 studies
In preparation for the formal FDP, studies were conducted with data collected from 2006 and 2007. Data were available from the Beijing (BJRS), Tianjin (TJRS), and Shijiazhuang (SJZRS) S-band radars; radar locations are shown in Fig. 1. These three radars have operating capabilities very similar to the Weather Surveillance Radar-1988 Dopplers (WSR-88Ds) in the United States. Clear-air return from insects is very extensive, often beyond 100 km in range, thus making it possible to observe boundary layer winds and detect and monitor the movement of boundary layer convergence lines (Wilson et al. 1994). In addition, satellite imagery, 1-hourly weather station reports (wind, temperature, and dewpoint), and data from twice-daily1 radiosonde launches from Beijing were available.
Table 1 lists 15 identified forecast challenges derived from eight convective precipitation episodes during 2006. An analysis of factors affecting storm evolution for these cases has been presented in Wilson et al. (2007). The study focused on the general synoptic situation, an estimate of the atmospheric stability conditions during the time of the convective events, the low-level wind relative to the terrain, and an assessment of whether the transition between the plains and the mountains affected the evolution of storms. It is surmised that factors affecting storm evolution could be changes in atmospheric stability from the mountains to the plains, and blocking or lifting of southerly flow by the mountains. In 8 of the 15 forecast challenges, it was hypothesized that influences from the impacts of mountains directly contributed to the challenge. As indicated in Table 1, thunderstorms often occur in the Beijing area when there is a synoptic trough approaching from the west or a Mongolian vortex at upper levels and the boundary layer winds are southerly. Monitoring the stability of the air mass over the plains is important for anticipating whether storm initiation or dissipation will be likely near the foothills. Small-scale topographical and meteorological features play a major role in the evolution of convective storms in the Beijing area.
Weather events from 2007 were also analyzed in a similar fashion and used to train forecasters. Prior to 2007, few China forecasters had experience in operational nowcasting; thus, there was a need to develop the necessary conceptual models and forecast rules particular to the Beijing area.
c. Heuristic nowcast rules
Based on examination of the 2006–07 data and nowcasting experience from other locations (Wilson and Mueller 1993; Wilson and Megenhardt 1997; Mueller et al. 2003; Lima and Wilson 2008), the following nowcasting rules were established and utilized by the FDP nowcasters.
1) Situation A: Storms moving from the mountains to the plains
A frequent dilemma for forecasters involves deciding whether convective storms that are extrapolated to move from the mountains to the plains will dissipate, stay the same, or intensify. The scenario depicted in Fig. 2 shows a particularly challenging event that occurred on 12 June 2006. A line of stratiform precipitation with some embedded convection (Fig. 2a) suddenly intensified (Figs. 2b and 2c) into a convective line as it approached the foothills, but then dissipated as it moved onto the plains (Fig. 2d). The intensification was likely the result of the influx of moist air teamed with convergent flow along the foothills of southwesterly winds above 700 m. Conversely, the subsequent rapid dissipation, which occurred when the convective line reached the plains, was likely the result of storm-relative updrafts ingesting the very stable air present over the plains below 700 m. Below is the rule set for situation A.
Nowcast dissipation if
storms are isolated or they are part of a poorly organized cluster, or
there is no obvious gust front with a line of storms, particularly if a recent sounding over the plains shows stable conditions.
Nowcast no change or intensification if
a recent sounding over the plains in advance of line of storms is conditionally unstable and growing cumulus clouds are prevalent over the plains or
a strong gust front is observed to move off the mountains or is observed with the storms over the mountains. An exception would be if a recent sounding indicates stable conditions on the plains.
Nowcast intensification if
a gust front is moving onto the plains and there is evidence of instability over the plains such as existing storms or growing cumulus.
2) Situation B: Storm initiation over the foothills of Beijing associated with warm, humid low-level SSE winds
Storm initiation along the foothills near Beijing, or over Beijing, is very challenging in that there are almost no early indicators before rain begins in the vicinity. These cases occur primarily with warm, humid, low-level southerly winds. Frequently, a weak short-wave trough is approaching from the west. Figure 3 is an example of storm initiation along the foothills that occurred on 9 July. While the surface observing stations showed very light northerly flow, the radar indicated southeasterly winds immediately above the surface and flowing up the slope of the foothills. The Beijing sounding showed only a small amount of convective inhibition (CIN), which should have easily been overcome by the forced lifting of the air along the foothills. The forecast challenge concerned why initiation occurred when it did; there was no obvious precursor in the wind information. For this reason the maximum lead time in nowcasting the convective rainfall in Beijing would be limited to <1 h. Nowcasting the formation of this line would be limited to <30 min by monitoring radar and satellite infrared cloud-top temperatures (Roberts and Rutledge 2003). After formation of the line, radar extrapolation would provide no more than 30-min lead time to reach Beijing.
On two occasions during 2006 (27 June and 5 July), storm initiation occurred over Beijing. Figure 4 shows the initiation of nocturnal storms on 27 June between 2200 and 2400 LT. The radar Doppler velocities indicated there was a 15 m s−1 southeast flow just above the surface and the sounding at 2000 LT showed considerable CAPE with little CIN. Radar showed the first convective clouds and precipitation echoes formed on the northwest side of Beijing (Figs. 4a and 4b) close to the foothills. Other small convective cells also started to form away from the city (Fig. 4c). So, there were likely other contributing factors that may have included a possible urban heat island effect. However, the largest and most intense storm occurred over the city (Fig. 4d). Similar to the above 9 July case, successful nowcasting of storms over Beijing would have been challenging and based solely on observing rapid radar growth and decrease in satellite infrared cloud-top temperatures in the presence of favorable flow and atmospheric vertical instability. The rule set for situation B is presented next.
Nowcast initiation if
the mountains are not blocking the low-level SSE flow (Froude number <1) and rapidly growing cumulus are observed. By monitoring the Doppler velocity and the surface station winds, evidence of mountain blocking of the flow can be inferred. The blocking is evidence of stability.
3) Situation C: Dissipation of small echo clusters
There is a vast volume of literature that examines the lifetime of radar convective precipitation echoes dating back to Ligda (1953). In general, lifetime is related to cell size, organization, and the updraft forcing mechanism. Most individual cells live less than 30 min, whereas organized systems like supercells (which have rotating updrafts) and squall lines may last for many hours. The rule set for situation C follows next.
Nowcast dissipation within 60 min if
the cell diameter is less than about 10 km and the cells are not associated with a convergence line or
the convergence line and storm are moving apart.
4) Situation D: Storm initiation and growth associated with convergence lines
Boundary layer convergence lines (boundaries) were frequently observed in the vicinity of Beijing. These boundaries included synoptic fronts, sea breezes, gust fronts, and mesoscale boundaries of unknown origin. Particularly on the plains, these boundaries had a major influence on the initiation and evolution of storms. Figure 5 is an example of colliding boundaries on 5 July. In Fig. 5a, there are four gust fronts that collided and produced the large storm complex labeled D and E in Fig. 5b. Gust fronts GF3 and GF4 intersect and form storm E. All four gust fronts collide to form storm D. In addition to demonstrating the importance of colliding boundaries, this figure also demonstrates that storm longevity is increased when the boundary and storm are moving together in the same direction, which is the case for storm B and gust front G2. In contrast, gust front GF1 is moving away from storm A. Storm A has begun dissipation, and 35 min after Fig. 5b its reflectivity is <40 dBZ and is about one-half its full size. Here is the situation D rule set.
Nowcast initiation if
satellite and/or radar indicates rapidly growing cumulus along a boundary or
a convergence line is going to move through a field of cumulus clouds in an environment that does not have a strong capping temperature inversion.
Forecast intensification if
a convergence line is going to intersect an existing cell or
two boundaries are forecast to collide and a strong temperature inversion is absent.
Nowcast intensification or no change if
a storm is moving with about the same speed and direction as the convergence line, or is moving along the convergence line.
3. B08FDP precipitation events
As stated previously, the primary focus of this paper is the evaluation of the skill in predicting convective precipitation events during the B08FDP period (1–24 August) that had an impact on the Olympic Games. Impacts included postponing and rescheduling of weather sensitive outdoor events for the safety of athletes and spectators. Six cases were chosen as nowcasting challenges for study here. These six cases either impacted the Olympic events or were a significant potential threat. There were eight other days with precipitation on the plains between the hours of 1000 and 2200 LT, which approximates the time period Olympic events were held. Three of these eight days had showers that were associated with the sea-breeze front well south of Beijing. Three other days had large-scale dissipating rain events that did not reach Beijing. Another case featured a brief shower near Beijing and in the eighth case there was precipitation associated with storms on a convergence line east and south of Beijing.
Conservatively, we have chosen 35 dBZ to define a thunderstorm event using radar. The significance of each case is as follows:
2 August, thunderstorms threatened a rehearsal for the opening ceremony that was scheduled to start near the time the storms were projected to reach the Olympic Stadium;
8 August, thunderstorms threatened the opening ceremonies;
10 August, late afternoon and early evening heavy rain and thunderstorms disrupted venue events.
14 August, intense thunderstorms and heavy rain initiated in the Beijing area and postponed or delayed many events;
21 August, a line of thunderstorms threatened the nighttime competition; and
24 August, thunderstorms threatened the closing ceremonies.
4. B08FDP nowcasting systems and process
a. Nowcasting systems
The 11 different nowcasting systems that participated in the B08FDP are listed in Table 2. A very brief description of each individual system is provided; the reader is encouraged to see the references listed in Table 2 for more in-depth discussions. Five systems were primarily radar echo extrapolation systems (see Table 2). A sixth system [the Beijing auto-nowcaster (BJANC), which was based on the NCAR Auto-Nowcaster (Mueller et al. 2003)] was able to grow and dissipate extrapolated storms as well as initiate new storms. Three of the systems blended extrapolation with numerical model forecasts (labeled blending in Table 2). The finest spatial scale of the numerical models was between 3 and 5 km. The models cycled every 1–3 h. Data assimilation included conventional data and mesonet data with at least one of the systems assimilating satellite winds, profiler winds, aircraft temperature and winds, radar velocity azimuth display (VAD) winds, and GPS water vapor. All three models used a three-dimensional variational data assimilation (3DVAR) scheme.
Niwot is described here since it is a new system with no reference. Niwot is a nowcasting system developed by NCAR to produce 1–6-h nowcasts of radar reflectivity every hour. The forecasts are based on the merging and blending of radar reflectivity forecasts from the extrapolation of radar echoes with forecasts from the BMB Rapid Update Cycle Weather Research and Forecasting model (WRF/RUC). The model is based on a 3DVAR version of a 3-km WRF system (Chen et al. 2009). Radar echo extrapolation was a modified version of the Thunderstorm Identification, Tracking, Analysis and Nowcasting (TITAN; Dixon and Wiener 1993) and Tracking Radar Echoes by Correlation (TREC; Rinehart and Garvey 1978; Tuttle and Foote 1990) methodologies. The primary assumption used by Niwot is that the location of precipitation is best forecast by radar echo extrapolation and the numerical model provides skill in forecasting changes in the areal extent of the echo. Importantly, Niwot was designed to allow the operator complete flexibility to ignore, accept, or modify any combination of the forecasts from the numerical model, extrapolation, or blending of both.
Thunderstorm Interactive Forecast System (TIFS) had the ability to automatically produce probabilistic precipitation products based on an ensemble of equally weighted forecast products from BJANC, the Canadian Radar Decision System (CARDS), the Global–Regional Assimilation Prediction System Severe Weather Integrated Forecasting Tool–extrapolation (GRAPES–SWIFT), the McGill Algorithm for Precipitation Nowcasting by Lagrangian Extrapolation (MAPLE), the Short-Term Ensemble Prediction System (STEPS), and Short-Range Warning of Intense Rainstorms in Localized Systems (SWIRLS). The FDP TIFS operator could make a final decision to modify the FDP combined forecast products before posting. TIFS automatically generated text and graphical thunderstorm warnings on a display that BMB forecasters could view.
b. B08FDP nowcasting process
There was a FDP forecaster representing each system. These representatives would arrive at a consensus nowcast and it would be passed on to Chinese meteorologists who had received training in the various nowcasting systems and who could translate the information into the native language. They would then pass the translated consensus nowcast on to the BMB forecasters who were responsible for providing the official forecasts for the Olympic venues. This procedure tended to reduce misunderstandings due to language differences.
The FDP human forecast process contained three phases. The first phase was prior to storm initiation. During this phase the FDP forecasters used the numerical models to provide an outlook of when and where initiation might take place. Phase two occurred once storm initiation began. At this point, the FDP forecasters used the extrapolation systems to get a quick first guess of forecast storm positions. The third phase occurred almost simultaneous with the second and continued throughout the remainder of the event. In this phase, the forecaster applied the heuristic rules in section 2 to the forecast storm evolution.
5. B08FDP nowcast challenge cases and evaluation procedure
Discussed below are the meteorological situations and forecast products for the six nowcasting challenges selected in section 3. Tables 3 –8 provide a summary of the nowcasts by available systems for each of the six cases. Because of space limitations, figures of satellite imagery and FDP system nowcast products will only be presented for the 2 August case. The other cases will be described more briefly using a limited set of figures. In each case the wind flow, stability, and precipitation distribution will be conveyed at critical forecast points. The nowcasting systems are divided into two groups, those that provide forecasts for ≥1 h (GRAPES–SWIFT–blend, GRAPES–SWIFT–extrapolation, SWIRLS, WRF/RUC, Niwot–extrapolation, Niwot–human, TIFS) and those that provided forecasts for ≤1 h (MAPLE, STEPS, BJANC, CARDS).
Evaluation of the forecast accuracy for these six events is subjective. Objective techniques are not well suited for evaluating rare events and assessing the usefulness of nowcasting guidance information to forecasters. For example, an objective technique would score a forecast as correct if no rain were forecast at the Olympic stadium and none occurred. However, if a squall line were approaching the stadium and dissipated just before reaching the stadium and the nowcasting system did not even nowcast that a squall line would occur, this would be considered a failure of the system. If the forecasting system does not capture the general character and distribution of the present precipitation pattern, the forecaster will not have confidence in the forecast.
The primary emphasis here is on the 0–3-h forecast period; obviously, for those systems that only forecast for the 0–1-h period the emphasis is on the first hour. Typically, the focus was on forecasting the start, duration, and intensity of precipitation at the Olympic stadium. It is these three parameters that are considered in defining success. A successful >1-h nowcast for the start or end time of precipitation at the Olympic stadium would have an error of less than 1 h. If moderate to heavy rain were forecast at the stadium and only a brief light shower occurred, an objective verification technique might consider that to be a correct forecast but here it would be recorded as a significant error in intensity. Table 9, at the end of the section, summarizes the subjective evaluations that were made for each technique for each case.
a. 2 August: Rehearsal for opening ceremonies
A rehearsal for the opening ceremonies for the Olympics was planned for the evening of 2 August starting at about 1700 LT. Throughout the early and midafternoon, thunderstorms were forming over the mountains and moving toward Beijing. The forecast challenge at 1500 LT was to provide the organizers responsible for the rehearsal notification if there would be thunderstorms–rain between about 1700 and 1900 LT over the Olympic stadium.
Figure 6 shows the evolution of the radar echoes from 1500 to 1800 LT. The storms, with one exception, dissipated as they moved from the mountains to the plains. The exception was one storm 30 km northeast of Beijing along the base of the foothills that lived for a short time on the plains along a gust front. Between 1500 and 1700 LT the FDP forecasters were consistently advising the BMB forecasters that the thunderstorms would likely dissipate as they moved off the mountains and would not interfere with the rehearsal.
The rationale supporting forecast dissipation was predicated on the nowcast rules for situation A (section 2b). The storms were only moderately organized (see Fig. 6), which often results in dissipation when moving from the mountains to the plains. More important, the sounding released from Beijing at 1300 LT indicated stable conditions over the plains with a CIN value of −87 J kg−1 and CAPE of 97 J kg−1. This was even after the sounding had been modified to account for the dry moisture bias2 that has been documented by Wang and Zhang (2008). In this marginal instability situation for thunderstorms, the near-surface dewpoint was closely monitored since it could lead to enhancement in CAPE. The surface mesonet stations, around Beijing, during midafternoon had an average dewpoint of 17°C, which is only about ½° higher than the dewpoint of the near-surface lifted parcel from the 1300 LT sounding. Additional evidence of continued stability over the plains was provided by the lack of convective activity on satellites and the Beijing and Tianjin radars. The Chinese Fēngyún-2C and -2D (FY2C) and (FY2D) visible satellite images indicate very few cumulus clouds over the plains in advance of the mountain storms (see Fig. 7). It could be argued that small cumulus would be very difficult to observe in the satellite data because of the 5-km resolution. Yet radar, having the ability to observe the presence of small nonprecipitating cumulus within about 50 km (Knight and Miller 1993; Mueller and Wilson 1994; Wilson et al. 1994), also showed no evidence of cumulus formation over the plains. In addition the TJRS radar and satellite imagery did not indicate any cumuli along a sea-breeze front southeast of Beijing, which was propagating inland from the Bohai Sea (Fig. 7). The weather stations in the vicinity of the sea breeze showed surface conditions very similar to those around Beijing. All these factors gave added confidence to the FDP forecaster that stable conditions existed over the plains and that the mountain storms would dissipate as they moved over the plains.
Two of the nowcasting systems (GRAPES–SWIFT–blend, Fig. 8a; Niwot–extrapolation, Fig. 8c) available to the forecaster at 1500 LT suggested there would be rain or even thunderstorms between 1700 and 1800 LT (see Tables 3 and Fig. 8). The nowcast from the GRAPES–SWIFT–blend system available at 1530 LT was nowcasting rain at the stadium between 1730 and 1800 LT (Fig. 8a). The WRF/RUC used in Niwot–human did not forecast storms over the plains throughout the entire day. However, it did forecast a few storms over the mountains but they did not resemble the observed and unfolding conditions so this gave the forecaster little confidence in its skill for the day. As can be seen in Fig. 8d, the human forecaster (Niwot–human) dissipated all rain over the plains for the reasons discussed above.
By 1600 LT, only the GRAPES–SWIFT–blend and GRAPES–SWIFT–extrapolation were forecasting rain between 1700 and 1900 LT. Niwot–extrapolation and SWIRLS were now forecasting the storms to pass east of the stadium. By 1700 and 1800 LT all systems, including the ≤1-h systems, depicted no rain over the stadium between 1700 and 1900 LT.
Figure 9 shows the tracks of the storm centroids identified by TITAN. The short-lived nature of most storms and the absence of storms over the plains are clear, with the one exception noted above.
In summary, during the period from 1500 to 1530 LT, the consensus from the automated ≥1-h nowcast systems was that precipitation would likely occur over the stadium during the rehearsal. However, the FDP forecasters using the nowcasting rules consistently and correctly forecast the mountain thunderstorms to dissipate before reaching the stadium. All the ≤1-h nowcasting systems were consistently forecasting that the storms would miss the stadium. Since all the ≤1-h nowcasts were the same, they are not listed in Table 3.
b. 8 August: Storms threaten opening ceremonies
August 8 was the night of the spectacular Olympic Games opening ceremonies. Storms began forming in the mountains west and southwest of the stadium by early afternoon, moving in an east-northeast direction. By the time the ceremonies began (2000 LST), the leading edge of one storm cluster was 60 km to the southwest of the stadium, on the edge of the foothills, and poised to move toward the stadium. The FDP forecasters were predicting that this storm cluster would dissipate when it moved over the plains and would not affect the opening ceremonies. Figure 10 shows the evolution of the storms. As can be seen in Fig. 10, the storms did begin to dissipate when they reached the plains and completely dissipated before reaching Beijing. After 2400 LT storms reformed east of the city where radar analysis indicates outflow boundaries from the storm of concern, and from storms in the mountains northeast of the city, collided.
At 1900 LT (1 h before the opening) the only ≥1-h nowcasting systems available were the Niwot systems, which were nowcasting no rain at the stadium during the ceremonies. The GRAPES–SWIFT system later became available at 2030 LT and it also nowcast no rain. The extrapolation techniques and model techniques were either extrapolating the storms slowly so that rain did not reach the stadium until after 2300 LT or they were extrapolating them to move south of Beijing. The FDP forecasters were less certain than on 2 August that the storms would dissipate when reaching the plains. The main reason for forecasting dissipation was that the cells were not well organized and a gust front had not yet been observed (situation A, in section 2b), although a weak gust front did form later. Uncertainty for dissipation existed because both the 0700 and 1900 LT soundings indicated appreciable instability over the plains with virtually no CIN to overcome; the CAPE values were between 1700 and 2400 J kg−1. The forecasters felt that even if the storms did not dissipate, they would not reach the stadium until after the ceremonies concluded.
In summary, although the air over the plains was unstable, all available nowcasting systems were correct in not forecasting precipitation over the stadium during the opening ceremonies.
c. 10 August: Large-scale rain band disrupts afternoon events
There were two forecast challenges on this day: first, forecasting the cessation of early morning light rainfall so that the tennis matches could begin and, second, forecasting if heavier rainfall would impact the Beijing venues later in the day. Separate from the light morning rain event over Beijing, a large north–south rainband associated with a synoptic-scale system developed and intensified west of Beijing. The tennis venue was located just north of the Olympic stadium. FDP nowcasts for the morning light rain event were based on radar echo extrapolation, which forecast precipitation ending at least an hour too soon since the movement of the rain stalled just a few kilometers south of the tennis stadium.
The threat for additional heavier rain was present only 30 km to the west (Fig. 11a). This rainband was moving very slowly eastward. A small intense convective rainband developed between 1400 and 1500 LT about 40 km southwest of the stadium (Fig. 11b). This band then moved over the stadium between 1630 and 1730 LT (Fig. 11c). The main rainband moved in around 1840 LT (Fig. 11d), but was preceded by a period of a little more than an hour of no precipitation. The challenge was to forecast when heavy rain with possible lightning would move over the stadium and postpone evening track and field events.
With the exception of WRF/RUC and GRAPES–SWIFT/blend, the FDP ≥ 1-h nowcasting systems generally provided good guidance (Table 5a). SWIRLS provided a 3-h lead time on the start of the heavy rain at the stadium. GRAPES–SWIFT–extrapolation and the Niwot systems provided 1.5–2-h lead times. The forecasts by WRF/RUC and GRAPES–SWIFT–blend were disregarded since in one case the rainband was forecast to move in the wrong direction, and in the other case it indicated rain was already occurring at forecast time. Table 5b shows that the ≤1-h extrapolation systems were generally accurate. Once the heavy rainband to the southwest formed, it was correctly projected to move with a constant speed.
In summary, the nowcasts for the start time of the heavy afternoon rain, which were based on model input, ranged from very good for the SWIRLS 3-h forecast to poor for the WRF/RUC and GRAPES–SWIFT–blend. The 1- and 2-h extrapolation forecasts were accurate. The Niwot–human nowcasts simply accepted the extrapolation forecasts.
d. 14 August: Heavy afternoon thunderstorms postpone outdoor events
This convective case had the most impact during the Olympic period and developed in close proximity to Beijing. At about 1115 LT thunderstorms initiated roughly 20 km south and east of the Olympic stadium (Fig. 12a). Figure 13, which shows the same time as Fig. 12a, indicates there is general convergence of airflow into the Beijing area. The Doppler velocities from the Beijing radar, which is located in the southeast corner of Beijing, showed the unusual signature of radial velocities approaching the radar from all directions (Fig. 13a). This convergence signature was observed on several other days, suggesting an urban heat island affect that may be responsible for initiating convective storms. Figure 13b is a graphic illustration of this converging wind field that was obtained from the single Doppler radar retrieval system called the Variational Doppler Radar Analysis System3 (VDRAS). This system was running in real time during the FDP and is described by Sun and Crook (1998, 2001) and Sun et al. (2010).
The small storm just south of Beijing in Fig. 12a developed where the maximum convergence is observed in Fig. 13b. This storm continued to grow and generated a gust front that initiated other storms. Strom evolution continued over the next hour, resulting in a rather large storm in the vicinity of the radar (Fig. 12b). The gust front from this storm then collided with a gust front from a cluster of storms over the foothills northeast of Beijing and produced a significant area of storms northeast of Beijing (Fig. 12c). This large area of thunderstorms propagated west and southwest producing heavy rain over the stadium starting at about 1330 LT (Fig. 12d). Rain of varying intensity persisted over the stadium for the next 4–5 h, postponing all scheduled outdoor sporting events.
The guidance from the ≥1 h nowcasting systems on the timing of storm initiation in the vicinity of Beijing was 1–3 h slow (see Table 6a), but still useful in increasing the FDP forecasters’ confidence that storms would likely initiate. Some of the venues south of Beijing were impacted by storms as early as 1230 LT. After the storms initiated it was a challenge to forecast when the Olympic stadium would be affected. At the stadium from 1300 to 1400 LT there were some very light intermittent showers; the heavier and continuous rain began at 1400 LT and lasted for about 4 h. The ≤1-h nowcast systems were roughly 1 h too slow (see Table 6b) in forecasting the start time of the more continuous heavier rain. This was primarily because there was initiation taking place on the southern and western edges of the westward-moving storm system. The BJANC system has the capability to forecast the initiation and growth. However, while doing an excellent job of forecasting the major initiation northeast of Beijing, it did not do as well at capturing the more modest initiation on the westward-propagating system. This is because the gust front on the southwestern side of the storm was not entered by the BJANC operator, whereas the two colliding boundaries had been entered.
In summary, the ≥1 nowcasting systems were 1–3 h too slow in forecasting thunderstorm initiation in the vicinity of Beijing during the early afternoon. Once the storms began to form, the ≤1-h nowcasting systems were too slow in predicting the start time at the stadium because of initiation on the leading edge of the storm complex. Unfortunately, Niwot–human nowcasts were not provided because of technical problems with the ingestion of the WRF/RUC data. The rules for storm initiation and intensification in situation C (section 2b) were very applicable for this event.
e. 21 August: Nighttime events threatened by line of mountain thunderstorms
By midafternoon, a squall line developed to the northwest in the mountains (Figs. 14a and 14b). The nowcasting challenge was whether this squall line would reach the Olympic stadium to interrupt the nighttime track and field events. Table 7a shows that at 1730 LT GRAPES–SWIFT–blend was nowcasting that the squall line would reach the stadium by 2030 LT. This forecast was in rough agreement with the Niwot–extrapolation. At that time the question was whether the squall line would survive the transition from mountains to plains. The radar indicated outflow from the storms but was only moderately well organized. The sounding released at 1300 LT showed that the lower atmosphere was saturated from the early morning stratiform rain and cloud. The lapse rate was slightly more stable than moist adiabatic. The satellite showed that clouds were thinning over the plains but it was late afternoon and little heating could be expected. Thus, it was summarized, by the FDP forecasters, that it would remain stable over the plains. The Niwot–human nowcast (see Table 7a) accepted the extrapolation but dissipated the intensity as the squall line moved off the mountains (rule 2.b.1). By 1900 LT the squall line was on the edge of the foothills and showing signs of dissipation (Fig. 14c). At this time, most of the extrapolation systems were correctly forecasting a brief period of light to moderate rain starting about 2000 LT (see Table 7b) with the bulk of the heavier precipitation north of Beijing. The Niwot–human forecast again accepted the extrapolation, but decreased the intensity (Table 7b). As Fig. 14d shows, the squall line rapidly dissipated and light rain briefly fell at the stadium between 2000 and 2015 LT. Assessing the stability over the plains was again critical in nowcasting the dissipation of the squall line as it moved to the plains.
In summary, of the two models available one was accurate in timing but too high in intensity and the other was not accurate. The timing estimates of the extrapolation forecasts were accurate but the intensities were too high because they did not account for dissipation as the storms moved from mountains to the plains. The Niwot–human nowcast correctly decreased the intensity of the extrapolated echoes, but was still too high in magnitude. Again, the forecast rules for storms moving off the mountains were very useful.
f. 24 August: Rain threat for closing ceremony
The closing ceremony was to be conducted between 2000 and 2300 LT. There was considerable concern during the afternoon that thunderstorms might occur during that period. Table 8 shows the WRF/RUC nowcast available at 1500 LT was predicting thunderstorms over Beijing between 1800 and 2000 LT and the GRAPES–SWIFT–blend 3-h forecast available at 1500 LT was also forecasting storms over Beijing at 1800 LT. In addition to the model runs there were other factors that suggested thunderstorms could affect the closing ceremonies. The Beijing sounding released at 1300 LT had a CAPE of 3459 J kg−1 and CIN of −39 J kg−1 (after correction for dry bias). A boundary layer convergence line was positioned east of Beijing (Fig. 15a) and satellite and radar indicated cumulus along portions of this boundary. Also, thunderstorms were already occurring in the mountains 90 km west of Beijing and moving toward the east.
By 1600 LT two thunderstorms had developed on the convergence line 80 km east of Beijing and a squall line was developing 200 km to the northwest moving toward the southeast. All of these factors, along with the potentially very unstable sounding, prompted concern (rule 2.b.4). However, for Beijing the concern was not immediate for several reasons. The storms in the mountains west of Beijing had slowed their eastward progression, and the storms to the east on the convergence line (and any new ones that might develop) were expected to move away from Beijing. Additionally, the convergent line was slowly moving east and north away from Beijing, and the WRF/RUC was not predicting any rain over Beijing after 2000 LT (Table 8).
After 1600 LT the squall line to the northwest continued to build and intensify with its extrapolated motion taking it well east of Beijing. The mountain storms west and southwest of Beijing continued to develop but were no closer to Beijing and were gradually moving southward. None of the ≤1-h nowcasting systems was anticipating storms anywhere near Beijing. So by 1800 LT, there was little concern that rain would affect the closing ceremonies. Figure 15b shows the radar reflectivity field at 2000 LT, the start of the closing ceremonies. While there is considerable storm activity, it is not near Beijing. The line of storms to the southwest moved onto the plains as expected given the unstable conditions there (rule 2.b.1). The storms northwest of Beijing were the remnants of the earlier squall line and their motion carried them well east of Beijing.
This case shows that while in general the conditions were ripe for thunderstorms, with two of the models indicating storms during the ceremonies, there was never an immediate concern that storms would affect the closing. Had the convergence line been stationary over or west of Beijing or the mountain thunderstorms been closer, there would have been more reason for concern.
g. Case summary
Table 9 provides a measure to the overall accuracy of the various nowcasting systems and is divided into the ≥1-h and ≤1-h nowcasting systems. As described in section 5, the nowcasts are subjectively evaluated based on how well they (a) depicted the general pattern and distribution of the precipitation, (b) determined the start and end of precipitation, and (c) represented the intensity of the precipitation. A letter, defined in Table 9, is used to represent the skill of each system. As expected, the overall accuracy results of the ≤1-h nowcasting systems are higher than those of the ≥1-h nowcasting systems. The case-to-case inconsistency in the accuracy of the ≥1-h nowcasting systems is apparent, with the exception being the consistently good accuracy of the Niwot–human system, which depended heavily on the heuristic nowcasting rules in section 2c.
6. Summary
The summary comments presented below are primarily based on analyses of the six challenge events presented in this paper rather than a comprehensive, comparative, statistical analysis for the total FDP period. Such comprehensive analyses often obscure or dilute the actual accuracy for the high-impact rare events (Stephenson et al. 2008; Roberts et al. 2007). While daily observations of the various forecast systems and interactions between B08FDP personnel and the Olympic forecasters are not discussed here, they do influence the following comments.
None of the automated nowcasting systems provided a sufficiently consistent level of accuracy to allow forecasters to comfortably disseminate the forecast to users without human oversight. However, the nowcasts provided by the FDP forecasters to the BMB Olympic forecasters were generally useful and accurate. This was because forecasters examining high-resolution observations and automated forecast products such as those available during the B08FDP can use their physical reasoning and pattern recognition capabilities to assess data quality, evaluate automated forecast guidance, and apply broad meteorological reasoning to formulate forecasts superior to those generated automatically. This is not to say that the automated nowcasts were not useful; as described in section 4b, the FDP forecasters made use of them during the nowcasting process.
The following were problems with the automated nowcasting systems. Similar to what was concluded for the Sydney 2000 FDP (Wilson et al. 2004), the skill of the extrapolation forecasts is often low even for nowcasts of 1 h or less due to the considerable initiation, growth, and decay that can take place over a short period of time; this was particularly the case in the Beijing area because of the influence of local terrain on storm evolution. Nevertheless, for nowcast periods ≤3 h the extrapolation nowcasts were generally more accurate than the numerical model or blending nowcasts. None of the models assimilated high-resolution radar reflectivity or Doppler velocities. As reported previously (Wilson and Roberts 2006), it appears that without the knowledge of the actual precipitation field at model initialization time, particularly for convective events with weak synoptic forcing, it will not be possible to provide accurate nowcasts on the very short-term time and space scales. While progress is being made in assimilating high-resolution radar data (Xue et al. 2007; Sun and Zhang 2008; Smith et al. 2008), it was not yet possible during the B08FDP. The blending techniques also produced inconsistent nowcasts because they were heavily dependent on the model to forecast initiation growth and decay.
Improved nowcasts will depend on advances in at least three areas: first and foremost is improved understanding of physical processes on the micro- and mesoscales, second is high-resolution observation of boundary layer winds and three-dimensional temperature and moisture fields, and third is major improvements in numerical model forecasts on the nowcast time and space scale. Each of these is addressed below.
Our understanding of convective storm initiation and evolution has improved greatly as the result of a number of research field experiments [e.g., the Convection Initiation and Downburst Experiment, CINDE (Wilson et al. 1988); the Convection and Precipitation/Electrification experiment, CaPE (Wakimoto and Lew 1993); the International H2O Project, IHOP (Weckwerth et al. 2004); the Convective Storm Initiation Project, CSIP (Browning et al. 2007); and the Convective and Orographically Induced Precipitation Study, COPS (Wulfmeyer et al. 2008)]. However, substantial improvements will depend on understanding the roles of inadequately observed phenomena such as gravity waves, local variations in kinematic and thermodynamic variables, and precipitation microphysics.
The importance of high-resolution observations for nowcasting was particularly evident for storms moving from the mountains to the plains (situation A, section 2b) and for storm initiation along the foothills and over Beijing (situation B). Nowcasting these situations would have benefited from higher-resolution observations of stability and three-dimensional boundary layer winds. The VDRAS analysis system, which was implemented for the FDP (shown in Fig. 13), showed considerable potential for providing the desired high-resolution monitoring of wind and thermal boundaries. Future heuristic nowcasting rules could utilize these VDRAS analysis fields as well as numerical models. A long unfulfilled wish is a method for monitoring the strength of the temperature inversions that cap convection. Inclusion of frequent vertical profiles of temperature and moisture information obtained from commercial aircraft takeoffs and landings (Moninger et al. 2003) would help fill this need near airports.
The numerical model nowcasts often suffered from inaccuracies in the initial location of the precipitation and errors in its evolution. These problems arise from numerous sources; we speculate they include (a) a lack of high-resolution assimilation of the precipitation, wind, and stability fields; (b) difficulties in handling scale imbalances between relatively sparse sounding data and dense surface mesonet stations and eventually dense radar data when they become operationally available; (c) inaccurate prediction of secondary convection resulting from storm outflows; (d) parameterization of convective and microphysical processes, (e) difficulties in handling planetary boundary layer processes particularly in complex terrain; and (f) inherent fast–nonlinear error growth at convective scales.
7. Future
For the foreseeable future numerical models will not likely provide 1- and 2-h nowcasts better than those from extrapolation or expert systems. An exception might be strongly forced large-scale convection where local influences are small (Wilson and Roberts 2006). Stensrud et al. (2009) has proposed for 2020 that it may be possible to provide up to 90-min probabilistic severe storm warnings using numerical weather prediction. This possibility would most likely be applicable to strongly forced synoptic situations and not for the weakly forced synoptic situations, like those discussed in this paper, where local forcing dominates. With the possible exception of these strongly forced situations, the best way to improve the 0–2-h forecasts is likely to be via an expert forecaster–computer system; the BJANC and NCAR auto-nowcaster are such systems. However, it requires a development period to modify localized forecast rules and fuzzy logic to the available observations and local climatology. Outside of severe storm warnings, which are very important, nowcasting of convective storms by weather services worldwide is limited. The experiences from the FDPs indicate that the capability exists to provide users with nowcasts of convective storms that could greatly improve operational efficiency and public safety. Since numerical models are least effective during the nowcast period, it is only natural for the forecasters to further fill this void. As more emphasis is placed on nowcasting, localized heuristic nowcasting rules will develop.
The 3–6-h nowcast period will depend on methodologies that blend extrapolation and numerical model techniques. However, as discussed above, progress here will depend on improvements in numerical modeling on the nowcasting scale.
Acknowledgments
The B08FDP was a great success, which was the result of the tireless work of many individuals associated with the Beijing Meteorological Bureau and the China Meteorological Administration. Not only were the facilities and data delivery exceptionally good, the Chinese were perfect hosts to the visiting scientists associated with each of the nowcasting systems. Specific recognition is made to Xie Pu, Jianjie Wang, Yingchun Wang, Feng Liang, Yubin Wang, Debin Su, Ron Kong, and Feng Gao. The first author would particularly like to thank Mingxuan Chen of IUM/BMB for his personal support and advice throughout the entire B08FDP process.
Beth Ebert of the Centre for Australian Weather and Climate Research kindly reformatted and stored the nowcasts on the B08FDP server. NCAR software engineers Sue Dettling, Dan Megenhardt, and Dave Albo were instrumental in installing Niwot and supporting the first author with his software needs from the beginning of the B08FDP project through the analysis period. We thank Tammy Weckwerth and Stan Trier of NCAR for providing thoughtful reviews of the original version of this paper. We thank Tracy Emerson for the final preparation of the figures.
We thank the three anonymous formal reviewers whose comments were particularly helpful for improving the paper. The NCAR effort was funded from the NCAR Short Term Explicit Prediction Program, which is supported by NSF funds.
REFERENCES
Bally, J., 2004: The Thunderstorm Interactive Forecast System: Turning automated thunderstorm tracks into severe weather warnings. Wea. Forecasting, 19 , 64–72.
Bellon, A., and Austin G. L. , 1978: The evaluation of two years of real time operation of a short-term precipitation forecasting procedure (SHARP). J. Appl. Meteor., 17 , 1778–1787.
Browning, K. A., 1982: Nowcasting. Academic Press, 256 pp.
Browning, K. A., and Coauthors, 2007: The Convective Storm Initiation Project. Bull. Amer. Meteor. Soc., 88 , 1939–1955.
Chen, M. X., Fan S.-Y. , Zhong J. , Huang X.-Y. , Guo Y.-R. , Wang W. , Wang Y. , and Kuo B. , 2009: A WRF-based rapid updating cycling forecast system of BMB and its performance during the summer and Olympic Games 2008. Proc.Symp. on Nowcasting and Very Short Range Forecasting, Whistler, BC, Canada, WMO, 3.14. [Available online at http://www.nowcasting2009.ca/images/stories/Presentations/WSN09_Presentations.htm].
Chen, M. X., Gao F. , and Kong R. , 2010: Introduction of system for auto-nowcasting convective storm and its performance in support of the Beijing Olympics (in Chinese). J. Appl. Meteor. Sci., 21 , 395–404.
Dixon, M., and Wiener G. , 1993: TITAN: Thunderstorm Identification, Tracking, Analysis and Nowcasting—A radar-based methodology. J. Atmos. Oceanic Technol., 10 , 785–797.
Dixon, P. G., and Mote T. L. , 2003: Patterns and causes of Atlanta’s urban island-initiated precipitation. J. Appl. Meteor., 42 , 1273–1284.
Joe, P., Falla M. , Van Rijn P. , Stamadianos L. , Falla T. , Magosse D. , Ing L. , and Dobson J. , 2002: Radar data processing for severe weather in the National Radar Project of Canada. Preprints, 21st Conf. on Severe Local Storms, San Antonio, TX, Amer. Meteor. Soc., 221–224.
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.
Knight, C. A., and Miller L. J. , 1993: First radar echoes from cumulus clouds. Bull. Amer. Meteor. Soc., 74 , 179–188.
Laroche, S., and Zawadzki I. , 1995: Retrievals of horizontal winds from single-Doppler clear-air data by methods of cross correlation and variational analysis. J. Atmos. Oceanic Technol., 12 , 721–738.
Leduc, M., Joe P. , Falla M. , Van Rijn P. , Lapczak S. , Ruddick I. , Ashton A. , and Alsen R. , 2002: The July 4 2001 severe weather outbreak in southern Ontario as diagnosed by the new radar data processing system of the National Radar Project of Canada. Preprints, 21st Conf. on Severe Local Storms, San Antonio, TX, Amer. Meteor. Soc., 170–173.
Li, L., Schmid W. , and Joss J. , 1995: Nowcasting of motion and growth of precipitation with radar over a complex orography. J. Appl. Meteor., 34 , 1286–1300.
Li, P. W., and Lai E. S. T. , 2004: Applications of radar-based nowcasting techniques for mesoscale weather forecasting in Hong Kong. Meteor. Appl., 11 , 253–264.
Liang, Q., Feng Y. , Deng W. , HU S. , Huang Y. , Zeng Q. , and Chen Z. , 2010: A composite approach of radar echo extrapolation based on TREC vectors in combination with model-predicted winds. Adv. Atmos. Sci., 27 , 1119–1130.
Ligda, M. G., 1953: The horizontal motion of small precipitation areas as observed by radar. Dept. of Meteorology Tech. Rep. 21, Massachusetts Institute of Technology, Cambridge MA, 60 pp.
Lima, M. A., and Wilson J. W. , 2008: Convective storm initiation in a moist tropical environment. Mon. Wea. Rev., 136 , 1847–1864.
McNeill, D., and Freiberger P. , 1993: Fuzzy Logic: The Revolutionary Computer Technology that is Changing Our World. Simon and Schuster, 319 pp.
Moninger, W. R., Mamrosh R. D. , and Pauley P. M. , 2003: Automated meteorological reports from commercial aircraft. Bull. Amer. Meteor. Soc., 84 , 203–216.
Mueller, C. K., and Wilson J. W. , 1994: Reply. Wea. Forecasting, 9 , 663–664.
Mueller, C. K., Saxen T. , Roberts R. , Wilson J. , Betancourt T. , Dettling S. , Oien N. , and Lee J. , 2003: NCAR Auto-Nowcast system. Wea. Forecasting, 18 , 545–561.
Pierce, C. E., and Coauthors, 2004: The nowcasting of precipitation during Sydney 2000: An appraisal of the QPF algorithms. Wea. Forecasting, 19 , 7–21.
Polger, P. D., Goldsmith B. S. , Przywarty R. C. , and Bocchieri J. R. , 1994: National Weather Service warning performance based on the WSR-88D. Bull. Amer. Meteor. Soc., 75 , 203–214.
Rinehart, R. E., and Garvey E. T. , 1978: Three-dimensional storm motion detection by convective weather radar. Nature, 273 , 287–289.
Roberts, R. D., and Rutledge S. , 2003: Nowcasting storm initiation and growth using GOES-8 and WSR-88D data. Wea. Forecasting, 18 , 562–584.
Roberts, R. D., Nelson E. , Pinto J. , Saxen T. , and Phillips C. , 2007: Status of nowcasting thunderstorm initiation: Where do we go from here? Preprints, 33rd Conf. on Radar Meteorology, Cairns, QLD, Australia, Amer. Meteor. Soc., 5.3. [Available online at http://ams.confex.com/ams/pdfpapers/123659.pdf].
Seed, A. W., 2003: A dynamic and spatial scaling approach to advection forecasting. J. Appl. Meteor., 42 , 381–388.
Smith, T. L., Benjamin S. G. , Brown J. M. , Weygandt S. S. , Smirnova T. , and Schwartz B. E. , 2008: Convection forecasts from the hourly updated, 3-km High Resolution Rapid Refresh Model. Preprints, 24th Conf. on Severe Local Storms, Savannah, GA, Amer. Meteor. Soc., 11.1. [Available online at http://ams.confex.com/ams/pdfpapers/142055.pdf].
Stensrud, D. J., and Coauthors, 2009: Convective-scale warn-on-forecast system: A vision for 2020. Bull. Amer. Meteor. Soc., 90 , 1487–1499.
Stephenson, D. B., Casati B. , Ferro C. A. T. , and Wilson C. A. , 2008: The extreme dependency score: A non-vanishing measure for forecasts of rare events. Meteor. Appl., 15 , 41–50.
Sun, J., and Crook N. A. , 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 Crook N. A. , 2001: Real-time low-level wind and temperature analysis using single WSR-88D data. Wea. Forecasting, 16 , 117–132.
Sun, J., and Zhang Y. , 2008: Analysis and prediction of a squall line observed during IHOP using multiple WSR-88D observations. Mon. Wea. Rev., 136 , 2364–2388.
Sun, J., Chen M. X. , and Wang Y. C. , 2010: A frequent-updating analysis system based on radar, surface, and mesoscale model data for the Beijing 2008 Forecast Demonstration Project. Wea. Forecasting, 25 , 1715–1735.
Turner, B. J., Zawadzki I. , and German U. , 2004: Predictability of precipitation from continental radar images. Part III: Operational nowcasting implementation (MAPLE). J. Appl. Meteor., 43 , 231–248.
Tuttle, J. D., and Foote G. B. , 1990: Determination of the boundary layer airflow from a single Doppler radar. J. Atmos. Oceanic Technol., 7 , 218–232.
Wakimoto, R. M., and Lew J. K. , 1993: Observations of a Florida waterspout during CaPE. Wea. Forecasting, 8 , 412–423.
Wang, J., and Zhang L. , 2008: Systematic errors in global radiosonde precipitable water data from comparisons with ground-based GPS measurements. J. Climate, 21 , 2218–2238.
Weckwerth, T. M., and Coauthors, 2004: An overview of the International H2O Project (IHOP_2002) and some preliminary highlights. Bull. Amer. Meteor. Soc., 85 , 253–277.
Wilson, J. W., and Mueller C. K. , 1993: Nowcasts of thunderstorm initiation and evolution. Wea. Forecasting, 8 , 113–131.
Wilson, J. W., and Megenhardt D. L. , 1997: Thunderstorm initiation, organization, and lifetime associated with Florida boundary layer convergence lines. Mon. Wea. Rev., 125 , 1507–1525.
Wilson, J. W., and Roberts R. D. , 2006: Summary of convective storm initiation and evolution during IHOP: Observational and modeling perspective. Mon. Wea. Rev., 134 , 23–47.
Wilson, J. W., Moore J. , Foote B. , Martner B. , Rodi A. , Uttal T. , and Wilczak J. , 1988: Convection Initiation and Downburst Experiment (CINDE). Bull. Amer. Meteor. Soc., 69 , 1328–1348.
Wilson, J. W., Weckwerth T. M. , Vivekanandan J. , Wakimoto R. M. , and Russell R. W. , 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., Crook N. A. , Mueller C. K. , Sun J. , and Dixon M. , 1998: Nowcasting thunderstorms: A status report. Bull. Amer. Meteor. Soc., 79 , 2079–2099.
Wilson, J. W., Ebert E. , Saxen T. , Roberts R. , Mueller C. , Sleigh M. , Pierce C. , and Seed A. , 2004: Sydney 2000 Forecast Demonstration Project: Convective storm nowcasting. Wea. Forecasting, 19 , 131–150.
Wilson, J. W., Chen M. , Wang Y. , and Wang L. , 2007: Nowcasting thunderstorms for the 2008 Summer Olympics. Preprints, 33rd Int. Conf. on Radar Meteorology, Cairns, QLD, Australia, Amer. Meteor. Soc., 12B.3. [Available online at http://ams.confex.com/ams/pdfpapers/123018.pdf].
Wong, W. K., and Lai E. S. T. , 2006: RAPIDS–Operational blending of nowcast and NWP QPF. Proc. Second Int. Symp. on Quantitative Precipitation Forecasting and Hydrology, Boulder, CO, NCAR–WMO. [Available online at http://www.mmm.ucar.edu/events/qpf06/QPF/Session3/RAPIDS_wong.pdf].
Wulfmeyer, V., and Coauthors, 2008: The Convective and Orographically Induced Precipitation Study. Bull. Amer. Meteor. Soc., 89 , 1477–1486.
Xue, J., Feng Y. , Zeng Q. , Hu S. , Liang Q. , Wang Y. , and Huang Y. , 2009: An overview of the GRAPES-SWIFT nowcasting system and its performance in the B08FDP. Preprints, Fourth Workshop of WWRP B08FDP/RDP, Guangzhou, China, WMO, 57–61.
Xue, M., Droegemeier K. K. , and Weber D. , 2007: Numerical prediction of high-impact local weather: A driver for petascale computing. Petascale Computing Algorithms and Applications, D. A. Bader, Ed., Chapman Hall, 103–124.
Yeung, L. H. Y., Wong W. K. , Chan P. K. Y. , and Lai E. S. T. , 2009: Applications of the Hong Kong Observatory Nowcasting System SWIRLS-2 in support of the 2008 Beijing Olympic Games. Proc. Symp. on Nowcasting and Very Short Range Forecasting, Whistler, BC, Canada, WMO, 1.5. [Available online at http://www.nowcasting2009.ca/images/stories/Presentations/WSN09_Presentations.htm].
Topography in the vicinity of Beijing. The terrain elevation is color coded (m) on the scale to the right. The 200-m contour line is in black and is used in many of the following figures to differentiate between the plains to the south and mountains to the north. The faint blue rectangle outlines the forecast verification area. The location of Beijing is represented by the yellow rings, which are the ring roads within Beijing. The locations of the radars are given by their letter codes.
Citation: Weather and Forecasting 25, 6; 10.1175/2010WAF2222417.1
Evolution of a convective line on 12 Jun 2006 as it moves from the mountains to the plains. The colors are radar reflectivities in dBZ at a radar elevation angle of 1.5° given by the scale on the right. The green and blue colors (<15 dBZ) near the radar are primarily returns from insects. The colors >15 dBZ mostly represent rain intensity. The black line marks the boundary between the plains and foothills (200-m-elevation contour). The faint yellow circles represent the ring roads of Beijing.
Citation: Weather and Forecasting 25, 6; 10.1175/2010WAF2222417.1
Radar reflectivity images at 1.5° elevation angle showing the late evening initiation of a line of thunderstorms along the foothills on 9 Jul. The scales and markings are as in Fig. 2.
Citation: Weather and Forecasting 25, 6; 10.1175/2010WAF2222417.1
Nocturnal initiation of storms over Beijing on 27 Jun 2006: (a) 2149 LT, radar reflectivity at 6.1°; (b) 2207 LT, reflectivity at 6.1°; (c) 2237 LT, reflectivity at 1.5°; and (e) 2331 LT, reflectivity at 1.5°.
Citation: Weather and Forecasting 25, 6; 10.1175/2010WAF2222417.1
An example of storm initiation and enhancement with colliding gust fronts on 5 Jul during a 36-min time period. The letters refer to individual storms and GF stands for gust fronts. The radar elevation angle is 1.5°.
Citation: Weather and Forecasting 25, 6; 10.1175/2010WAF2222417.1
Time evolution of radar reflectivity (scale on the right), for an elevation angle of 1.5° on 2 Aug 2008, as thunderstorms move from the mountains toward the Olympic stadium where the rehearsal for the opening ceremonies was to start at about 1700 LT. The yellow circle is the location of the stadium. The black line is the 200-m-elevation contour. Shown are (a) 1500, (b) 1600, (c) 1700, and (d) 1800 LT. The thin enhanced reflectivity line in (c) extending over the stadium is the gust front produced by the thunderstorm.
Citation: Weather and Forecasting 25, 6; 10.1175/2010WAF2222417.1
Satellite visible cloud image at 1451 LT 2 Aug 2006. As in Fig. 5, the black line is the 200-m-elevation contour. The location of the Olympic stadium is denoted by the yellow circle. The blue line is the location of the sea-breeze front discussed in the above text. Note that there are no obvious convective clouds over the plains.
Citation: Weather and Forecasting 25, 6; 10.1175/2010WAF2222417.1
Forecast guidance for 1800 LT 2 Aug 2008 available to the forecaster at 1500 LT. (a) GRAPES–SWIFT 3-h extrapolation, (b) GRAPES–SWIFT 3-h blend, (c) Niwot 3-h extrapolation, and (d) human modification of (c). All of the images are scaled to be spatially the same. Note that the reflectivity color scales for GRAPES–SWIFT and Niwot are different (see scales on the right). The Niwot scale starts at 35 dBZ compared to 5 dBZ for GRAPES–SWIFT.
Citation: Weather and Forecasting 25, 6; 10.1175/2010WAF2222417.1
Centroid tracks of convective cells >35 dBZ on 2 Aug 2008. The light tan line running SW to NE is the 200-m-elevation contour. Beijing is located within the yellow circle.
Citation: Weather and Forecasting 25, 6; 10.1175/2010WAF2222417.1
Time evolution of radar reflectivity, at an elevation angle of 1.5°, covering the period of the opening ceremonies, which began at 2000 LT 8 Aug 2008. The yellow circle marks the location of the stadium. Shown are (a) 1800 (2 h prior to start of the opening ceremonies), (b) 2130, and (c) 2200 LT 8 Aug 2008, as well as (d) 0100 LT 9 Aug 2008 (after the ceremonies).
Citation: Weather and Forecasting 25, 6; 10.1175/2010WAF2222417.1
Time evolution of radar reflectivity at an elevation angle of 1.5° that impacted the Olympic Games on the afternoon of 10 Aug 2008. The yellow circle covers the location of Olympic venues in the Beijing area. Shown are (a) 1300, (b) 1517 (note new short line forming SW of Beijing), (c) 1705 (new line has moved over Beijing), and (d) 1841 LT (primary band has reached Beijing).
Citation: Weather and Forecasting 25, 6; 10.1175/2010WAF2222417.1
Time evolution of radar reflectivity, at an elevation angle of 1.5°, for 14 Aug 2008 as storms initiated in the Beijing area and disrupted outdoor Olympic events. The yellow circle covers Olympic venues in the Beijing area. Shown are (a) 1117, (b) 1217 (the thin white lines indicate the locations of gust fronts), (c) 1317, and (d) 1435 LT.
Citation: Weather and Forecasting 25, 6; 10.1175/2010WAF2222417.1
(a) Doppler velocities as at an elevation angle of 0.5° observed by the Beijing radar at 1117 LT 14 Aug 2008 (same time as in Fig. 12a). The radar is located on the southeast side of Beijing. Note that while the data are noisy, essentially all of the velocities, from all directions, are toward the radar at 3–9 m s−1 (green–blue colors). This indicates converging airflow at the radar. The white areas are locations where ground clutter has been removed. (b) VDRAS wind analysis at a height of 187 m for the same time as in (a). The brown colors represent convergence and the green colors divergence. The convergence values are in steps of 0.1 m s−1 km−1. The maximum mustard color at the center of the convergence is 0.3 m s−1 km−1; thus, the initial brown color is 0.1 m s−1 km−1.
Citation: Weather and Forecasting 25, 6; 10.1175/2010WAF2222417.1
Time evolution of radar reflectivity on 21 Aug 2008 as an advancing line of thunderstorms threatens to disrupt outdoor Olympic events. The yellow circle represents the location of Olympic venues in the Beijing area. Shown are (a) 1700, (b) 1800, (c) 1900, and (d) 2000 LT.
Citation: Weather and Forecasting 25, 6; 10.1175/2010WAF2222417.1
Radar reflectivities prior to and at the start of the closing ceremonies on 24 Aug 2008. The Olympic stadium is within the yellow circle. (a) Antenna elevation is 0.5° at 1505 LT; a convergence line (enhanced reflectivity line) extends through the domain with SE winds on the east side and SSW winds on the south side. (b) At 2000 LT showing the presence of thunderstorms but all at least 100 km from Beijing. Note for clarity purposes that in (b) only reflectivities >15 dBZ are shown.
Citation: Weather and Forecasting 25, 6; 10.1175/2010WAF2222417.1
Nowcasting challenges from the summer of 2006 in the vicinity of Beijing. Yes in column four means the terrain was suspected to have a significant effect on how the convection evolved. Barrier flow means the mountains were suspected to be blocking or diverting the flow.
FDP nowcasting systems available for analysis in this paper. The forecast parameters in column 3 are 1, reflectivity, precipitation rate, precipitation accumulation; 2, quantitative precipitation estimation; 3, convergence lines; 4, wind; 5, initiation, growth decay; and 6, hail, strong winds.
Nowcast systems ≥1 h on 2 Aug 2008.
Nowcast systems on 8 Aug 2008 for (a) ≥1 h and (b) ≤1 h.
Nowcast systems for 10 Aug 2008 at (a) ≥1 h and (b) ≤1 h.
Nowcast systems for 14 Aug 2008 at (a) ≥1 h and (b) ≤1 h.
Nowcast systems for 21 Aug 2008 at (a) ≥1 h and (b) ≤1 h.
Nowcast systems for 24 Aug 2008 at ≥1 h.
Summary of nowcasting systems’ level of accuracy for the six challenge cases: (a) for ≥1-h and (b) ≤1-h nowcasting systems.
This was increased to 4 times a day during the summers of 2007 and 2008.
Beijing used the Shang-E radiosonde, which has been shown (Wang and Zhang 2008) to have a significant dry bias. To account for this bias the dewpoints below heights of 1.5 km have been increased by 3°C for all the Beijing soundings used in this study. Since the dry bias at higher heights does not affect the computation of CIN and CAPE dewpoint adjustments were not made at higher levels.
VDRAS produces three-dimensional fields of wind velocity and temperature perturbations based on dynamically balancing, within a numerical model, available observations from radiosondes, profilers, mesonets, and Doppler radar reflectivities and radial velocities.