2.1 Observations of pollen

For the purposes of this study, we obtained hourly average grass pollen observations using standard methods for a 34-hour period beginning on 20 November at 04:30 UTC and ending on 21 November at 13:30 UTC. The grass pollen measurements derive from a Hirst-type pollen trap (Burkard Manufacturing) situated on a roof-top at Melbourne University (144.965°E -37.797°N, 43 m elevation). Pollen is collected on slides coated with Sylgard adhesive, then prepared with Calberla’s stain to make observation easier and counted using light microscopy. The number of whole grass pollen and broken or empty pollen grains (called pollen ‘shells’ hereafter) was assessed across each hourly transect. We could not distinguish with confidence the taxa of these pollen shells; thus a total count is presented. Usual practice is to count pollen across 24-hour transects, as the process is manually demanding; for this study we counted vertical transects to obtain hourly measurements around the 21 November event, with pollen counts converted into grains m^-3 (S1 Table in S1 File). 24-hour pollen counts on either side of 21 November were too low to consider hourly subdivision (S2 Table in S1 File). The respirable SPPs are too small to impact on the volumetric spore sampler and are not detectable with standard light microscopy procedures.

2.2 The Victorian Grass Pollen Emissions Model

The Victorian Grass Pollen Emissions Module (VGPEM1.0) treats whole pollen as 35 μm diameter particles with a density of 1000 kg m^-3 [23]. The model uses the satellite derived Enhanced Vegetation Index product to estimate the seasonal variation in grass pollen emissions, together with hourly meteorological inputs for the shorter-term variations. High temperatures and low relative humidity (RH) promote pollen emission. The horizontal resolution of VGPEM is 3 km across Victoria, using 306 x 306 grid cells. The temporal resolution is hourly, time-stamped at the beginning of the hour.

VGPEM is an emissions module inside the CSIRO Chemical Transport Model (C-CTM) [29], which provides the atmospheric dispersion and deposition capabilities. In this work we drive the C-CTM with meteorological forecasts from three different weather models to gain an ensemble view representing uncertainty in the meteorological conditions. The first model is the numerical weather prediction version of the Australian Community Climate and Earth System Simulator (ACCESS) [30]. The ACCESS domain used here is nested within a global ACCESS run and does not require separate boundary conditions. The second model is the Conformal Cubic Atmospheric Model (CCAM) [31] using boundary conditions from ERA-Interim [32]. The third model version 4.1.1 of the Weather Research and Forecasting model (WRF) [33] using boundary conditions from National Centers for Environmental Prediction Final reanalyses (NCEP FNL) [34]. Details of the meteorological models used are given in S3 Table in S1 File. CCAM and WRF meteorological capabilities were evaluated across south east Australia and found to reproduce temperatures and local scale meteorological features well, with wind speeds and water vapour mixing ratios within benchmark ranges [35].

The C-CTM has previously been used with both ACCESS and CCAM and has an interface designed for both inputs. The vertical dimensions of ACCESS and CCAM are structured differently; ACCESS uses hybrid model levels (to 29 km on 67 levels), whilst CCAM is output on pressure levels (to 2.2 hPa on 35 levels). WRF is not interfaced with the C-CTM but is similarly structured on pressure levels like CCAM and was therefore reformatted to be used with the CCAM input interface. The C-CTM resolves to 11 km for ACCESS and 444 hPa in CCAM/WRF (~6 km). For consistency, data are stored for all models to a height of ~5 km.

In this paper we refer to the meteorological results from each of the driving models by their names, e.g. ACCESS, CCAM and WRF. The different results of whole pollen, pollen shells and SPPs predicted by the C-CTM and driven by each of these meteorological models will be referred to as ACCESS^C-CTM, CCAM^C-CTM and WRF^C-CTM.

2.3 In-atmosphere humidity induced grass pollen rupturing process

We represent grass pollen rupturing following the theoretical approach of Wozniak et al. [36]. They describe in-atmosphere rupturing, which takes place using an 80% RH threshold. Wozniak et al. [36] treat this rupturing mechanism as an emission, whereas in this work we treat the in-atmosphere rupturing as an instantaneous process depending on the concentration of whole grass pollen grains present in any grid cell (g m^-3). Note here that the C-CTM carries pollen in the model as a mass concentration of aerosol, which is converted to grains m^-3 using the mass of one ryegrass pollen grain, m_pol = 22.4 × 10⁻⁹ g derived in Emmerson et al. [23]. Once the pollen concentration is converted to grains m^-3, χ, we calculate the number of whole grass pollen grains ruptured, N_rupt (grains m^-3) if at least one pollen grain per m³ is present (1 grain m^-3 = 0.0224 μg m^-3) and when the RH reaches 80% at any model grid cell in space and time: (1) where F_rupt is the fraction of whole grass pollen grains that rupture (= 0.7), based on the fraction of ryegrass pollen grains that rupture in water after 5 minutes [15]. F_rupt is the same for all simulations in the absence of literature on other rates of ryegrass pollen rupturing. We assume that 1 empty shell is produced per whole pollen grain. Using N_rupt we calculate the number of SPPs generated by multiplying by the number of SPPs produced per grain of pollen, n_spg. The measurements of Suphioglu et al. [16] on ryegrass showed around 700 SPPs per whole pollen grain of a size range 600 nm– 2.5 μm. We will use 700 SPPs per whole pollen grain with an SPP diameter of 600 nm which is the lower end of Suphioglu et al’s [16] measurements. The SPPs have the same density as whole grass pollen. This yields a mass of one SPP, m_SPP = 1.13 × 10⁻¹³ g and a fall speed of 0.01 cm s^-1. Upon rupturing, we convert the number of SPPs generated per m³ back into a mass concentration (using m_SPP) to be carried by the C-CTM. This mass concentration is used twice; to be lost from the tracer containing the mass of whole pollen, and to be added to a tracer containing the mass of SPPs. We also track the number of whole pollen grains that rupture (N_rupt), as there are no observations of SPPs, only the pollen shells.

2.4 Humidity induced on-plant mechanical rupturing

We also include Wozniak et al’s [36] on-plant mechanical rupturing, M_rupt of pollen on the plant, which occurs when exposed to moisture from humidity and precipitation [11,15]. Similar to Wozniak et al. [36], we treat mechanical rupturing as an emission occurring at the surface only (g m^-2 s^-1): (2) where P_fx is the emission rate of grass pollen from the plant (g m^-2 s^-1) predicted by VGPEM, and m_SPP/m_pol is the mass ratio of SPPs to whole grass pollen grains. The terms f_PR and f_RH are the emission activity factors for precipitation and RH, respectively. The idea is that mechanical rupturing will occur on the fraction of grass pollen left on the plant when humidity conditions are too high to warrant direct pollen emission to the atmosphere. The emission activity factors are calculated in the model using a logistic function, f_l of the form: (3) Where y is the variable of interest, e.g. RH or precipitation, for rate parameter α and location parameter c. For RH, the function follows Sofiev et al. [37]: (4)

The rate and location parameters will give f_l = 0.95 at 50% RH and 0.05 at 80%. f_baseline represents the fraction of pollen emitted at very low RH (= 0.33) and was used in Emmerson et al. [23].

The precipitation function f_PR: (5) where the logistical function parameters result in values of 0.95 for no precipitation falling and 0.05 for precipitation at 0.5 mm h^-1. The terms α_RH and α_PR are both negative, meaning the pollen emission decreases as either the humidity or precipitation rates increase.

2.5 Using strong wind gusts as a mechanism for in-atmosphere rupturing

Given the strength of winds in the storm gust front it seems likely that a wind driven rupturing process could represent the pollen shell observations. Visez et al. [38] simulated the action of wind blowing birch pollen against solid surfaces and found higher concentrations of SPPs associated with stronger wind speeds. Here we test two variations; one using a function of wind speed, f_WS described in Sofiev et al. [37], the other using wind speed (WS) directly to rupture the pollen. The function of wind speed takes the form: (6) where U_sat is a saturation wind speed of 5 m s^-1 which constrains the pollen emission, and f_baseline is 0.33 as before. The U_sat constraint suggests that high winds can only promote pollen release if pollen is available on the plant. The function takes the form of a curve, beginning at 0.33 for zero wind speed and asymptotes to 1 at wind speeds >15 m s^-1. The pollen rupturing occurs using a variation to Eq 1 with no threshold value: (7)

In the second case of rupturing directly with wind speed, we ensure windspeeds are above a threshold of U_sat. This ensures that rupturing only occurs at higher wind speeds, as f_WS will allow rupturing even at negligible wind speeds. The threshold U_sat value of 5 m s^-1 is equivalent to f_WS = 0.75. Grass pollen rupturing then occurs similar to Eq 1, but with the condition of RH replaced with the U_sat threshold.

2.6 On-plant mechanical rupturing using wind gusts

A prerequisite of in-atmosphere rupture is the presence of airborne pollen that is already disperse. We adjust Wozniak et al’s description of mechanical rupturing from Eq 2 to represent a wind speed induced emission of 600 nm particles from the ground, replacing (1-f_PRf_RH) with either f_WS or wind speed as the part of the definition, e.g.: (8)

This process considers the resuspension of previously deposited SPPs.

2.7 Electrical charge associated with low relative humidity

When air is humid, the moisture aids the transfer of electrical charges between atoms. At lower RH the electrical transfer is less effective and higher charges can build up. Vaidyanathan et al. [17] showed that electrical activity could trigger release of allergenic material, and observed Bermuda grass pollen rupture occurring when subjected to voltages of 10 kV m^-1. Taylor et al. [39] thought the same was likely of ryegrass pollen, given its high allergenic properties and association with thunderstorm asthma, even in air with an RH of 40%. We explore the hypothesis that pollen particles could become electrically charged when the RH is low. In this simulation, we use Eq 1 again to represent this process but use a threshold RH value of ≤ 30% to start pollen rupturing.

2.8 Lightning

The presence of lightning strikes has been associated with increased asthma presentations generally [40], but not necessarily in conjunction with an epidemic event.

Lightning is not simulated by any of our meteorological models and are incorporated into the pollen simulations from lightning observations. The number of lightning occurrences per hour, n_lightning, was re-gridded to our 3km resolution based on data from the World Wide Lightning Location Network (WWLLN) comprising of observations from a ground-based network of sensors [41]. The WWLLN includes both cloud-to-ground and cloud-to-cloud lightning, but preferentially detect the stronger cloud-to-ground occurrences as noted by Virts et al. [42]. If lightning occurred in any grid cell where whole pollen grains were also present, then this was used as the rupturing trigger. Eq 1 was again modified to include the lightning count, meaning that rupturing was stronger if there were more lightning occurrences: (9)

2.9 Dust

We examined existing C-CTM tracers representing mineral dust to replicate the action of particles being swept into the atmosphere by the strong winds in the storm inflow and outflow. The dust parameterisation relies on individual particles being saltated from land designated as bare ground and having a low soil moisture [43]. Three dust tracers are summed to represent PM₁₀ (<2.5, 2.5–5 and 5–10 μm diameter).

In total, VGPEM has been set up to run with eight pollen rupturing mechanisms, and the C-CTM dust tracer described above. The different rupturing thresholds and rupturing equations are compared below in Table 1. In each run, the model will carry whole pollen, pollen shells and SPPs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Thresholds and variables used in each of the model experiments.

https://doi.org/10.1371/journal.pone.0249488.t001

3.1 Humidity induced grass pollen rupturing

Using the hypothesised rupturing process with an 80% RH threshold does not yield a sharp rise in the modelled pollen shells (or the SPPs) at the time of the storm front, as we would expect to observe given the huge increase in respiratory problems (Fig 2b). Key to the Hughes et al. [18] finding was that SPP increases were associated with high rates of rainfall coincident with thunderstorm activity. The difference in the Melbourne storm was the absence of rain. Lack of rain is the major factor in why so many people were outside at the time and were exposed. The models show that humidity induced rupturing was strongest at other times during the week e.g. on 18 and 19 November (Fig 2b), and also tends to occur each night rather than during the day (due to lower temperatures). Inclusion of the on-plant mechanical rupturing (Fig 2c) adds thousands of pollen shells into the atmosphere, but the process reinforces the wrong timing of the rupturing (at night) rather than contributing to the TA event. Using the RH ≥80% rupturing mechanism in the models under-predicts the pollen shell concentrations by a factor of ~30 for the in-atmosphere only case, whilst inclusion of the mechanical rupturing overpredicts by a factor of ~80.

Prior to the arrival of the gust front, observations from the closest surface weather station to Melbourne University (Olympic Park) showed very low RH of 18% (Fig 3a). Three other weather stations upwind of the airflow to the north (Essendon and Melbourne Airports) and west (Laverton) also observed RH between 18–19%. Humidity induced rupturing at the surface on the day of the storm could only occur when RH reached 80% some five hours after the storm had passed.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Time series in (a) 2m relative humidity (b) 2 m temperature and (c) 10 m wind speed at Melbourne Olympic Park, the closest automatic weather station to the pollen observation site.

The vertical dashed line indicates the time of the storm, the horizontal dashed line indicates the 80% RH threshold.

https://doi.org/10.1371/journal.pone.0249488.g003

Differences in the timing of observed and predicted grass pollen concentrations could be due to errors in the modelled meteorological forcing. Of the meteorological models driving VGPEM, ACCESS and WRF capture the timing of the humidity increase and associated sudden temperature decrease (Fig 3b) after the storm passes better than CCAM, though wind speeds predicted by all models are lower than was observed prior to the storm (Fig 3c). Both CCAM and WRF tend to predict stronger wind speeds than ACCESS, and only CCAM predicts the correct wind speed at the time of the storm. All models predict RH above 80% at night on 19 and 21 November, whilst ACCESS also has a high RH at night on 20 November, coinciding with the timing of rupturing in Fig 2b and 2c.

Radiosonde soundings from Melbourne Airport (20 km north of central Melbourne) show the vertical profiles in RH at 00:00 UTC and 12:00 UTC; six hours before and after the storm (Fig 4a and 4c, respectively). None of the models track the observations precisely, and the shape of the modelled RH profiles tend to be better after the storm at 12:00 UTC. Vertical profiles of RH in the models are shown at 06:00 UTC (Fig 4b) and all show an increase in RH around 600 hPa which is where we expect the storm clouds to be. The high observed humidity at 600 hPa ±6 hours either side of the storm could reasonably be expected to persist throughout the day, so we infer the predictions of the meteorology models are reasonable at 06:00 UTC.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4.

(a-c) radiosonde observations at Melbourne Airport compared to model predictions for the 21 November 2016 of relative humidity from the surface to 0 hPa in each meteorological model. (d) pollen concentrations in VGPEM1.0 driven by each meteorological model. Dashed lines in (b) and (d) represent the thresholds for rupturing at 80% RH and 1 grain m^-3, respectively. The maximum height of modelled pollen differs because CCAM and WRF are output on pressure levels, whilst the ACCESS height is converted to hPa from the 5 km model level.

https://doi.org/10.1371/journal.pone.0249488.g004

We plot the 80% RH threshold in Fig 4b at the time of the storm, showing only ACCESS and WRF reach the required 80% RH at around 600 hPa (Fig 4b). In Fig 4d the corresponding vertical profile in whole grass pollen at Melbourne Airport shows whole grass pollen grains present throughout the lowest 5 km of the modelled atmosphere. The 1 grain m^-3 threshold is lost at ~800 hPa in ACCESS^C-CTM and CCAM^C-CTM or ~2 km, and 630 hPa in WRF^C-CTM. Vertical concentrations of whole grass pollen decrease exponentially from the surface, with most (62% average across models) situated in the first 40 m. It is only in the WRF^C-CTM profile that the whole pollen may intersect with RH conditions being high enough for rupture, noting that observations plotted in Fig 4 were above Melbourne Airport whereas rupturing may have occurred at other locations in the model domain.

Large differences exist between the models in the whole pollen concentrations produced with altitude and highlights how running an ensemble is beneficial. Whilst our results, and those from Wozniak et al. [36] show an exponential decline in whole pollen concentrations with altitude, measurements at 2000m altitude in the mountainous Thessaloniki region of Greece show grass pollen 2.6 times less than the surface concentrations [47]. Our models produce whole pollen at 2000 m in the range 185–1770 times less than the surface, whereas the Wozniak et al [36] model produces >500 times less in the north east region of the USA in spring. We tested whether artificially increasing the model vertical whole pollen concentration in the RH ≥ 80% experiment would impact the rate of rupturing at the time of the storm. We applied a linear relationship such that the ratio of surface to 2000 m whole pollen concentrations is 2.6:1. The results are largely similar, with the RH ≥80% threshold producing more pollen shells and SPPs in the models outside of the storm period (and mainly at night) using this mechanism (S1 Fig in S1 File).

The vertical time series to 5 km from each model at Melbourne University shows whole grass pollen concentrations are maintained for some time before and after the storm (Fig 5a–5c), indicating availability of pollen to be ruptured. We still see most of the whole pollen concentrated close to the surface. There are very low concentrations of SPPs throughout the ACCESS^C-CTM and CCAM^C-CTM 5 km model atmospheres before and at the time of the storm (Fig 5d and 5e). The pre-gust front air was well mixed and turbulent, sweeping SPPs out of the Melbourne air, with no opportunity for resupply as the RH levels were between 10–30% to a height of 2 km in each of the models. The WRF^C-CTM model shows SPPs present at the time of the storm, though this only translates into a pollen shell concentration of 2.2 grains m^-3 at the surface (Fig 2b). Modelled pollen shell concentrations were higher on 19 and 20 November, but these dates did not coincide with health impacts.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Time series with altitude plots of whole grass pollen (a–c), SPPs to log base 10 (d–f), and relative humidity (g–i) in the atmosphere above the pollen counting site in Melbourne.

ACCESS^C-CTM is shown in the left hand panels (a,d,g), CCAM^C-CTM in the middle panels (b,e,h) and WRF^C-CTM in the right hand panels (c,f,i). Vertical dashed line indicates the time of the storm.

https://doi.org/10.1371/journal.pone.0249488.g005

To summarise our initial findings, we find that the RH throughout the atmosphere at central Melbourne to be well below the 80% required for pollen rupturing at the time of the storm on 21 November. The key finding is that more SPPs are produced regularly at other times when RH tends to be higher, such as at night rather than during a hot late spring day. The high risk of false positives does not make the RH ≥80% mechanism a good one for the prediction of thunderstorm asthma. On the day of the storm, SPPs were only produced in the models some hours after the storm had passed, as the humidity increased. This is too late to have caused the spike in emergency calls at 07:00 UTC in Melbourne. The condition of the 80% RH threshold required for pollen rupturing is also a condition that suppresses whole pollen emissions from the plant in VGPEM via the closing of anthers at high humidity. In the laboratory rupturing experiments cited earlier, pollen was observed to rupture after full submersion in water, or humidified for hours, and it is not clear whether a brief intersection of pollen and high humidity is enough to cause rupturing in the atmosphere.

3.2 Alternative mechanisms for the 2016 Melbourne event

We know that a process occurred in the atmosphere at 06:30 UTC on 21 November 2016, causing a high concentration of respirable particles to be inhaled. In this section we focus on the results of the other model experiments that could possibly explain how a sudden source of SPPs might have occurred. Each of the panels (b-g) in Fig 6 shows each experiment producing a high concentration of pollen shells, some of which produce peaks in line with the pollen shell observations (highest on 20 November, gives higher correlations e.g. (r = 0.86–0.91 in Fig 6d), and some which produce higher peaks at the time of the storm (with corresponding poor correlations, e.g. (r = 0.48–0.54 in Fig 6g).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Time series at Melbourne in (a) dust as PM₁₀, and (b-g) pollen shells for the seven labelled rupturing experiments.

Note the modelled pollen shell concentrations are on the right-hand side y-axes in panels (b-g) and are not similar. s = slope of the linear regression and r = r correlations between modelled and observed pollen shells (or dust), given in same colours as the legend. Vertical dashed line indicates the time of the storm.

https://doi.org/10.1371/journal.pone.0249488.g006

Fig 6a is different and shows the time series in the model dust tracers compared to hourly PM₁₀ measurements from Footscray, the closest air quality monitoring site to the pollen observations. The dust tracers are normalised to the observed peak value for visual comparison, as dust only represents one fraction of total PM₁₀. This experiment shows the three meteorological models simulate the gusty winds well, and that wind-driven processes would cause particles to spike in the Melbourne atmosphere during the storm. The timing of the modelled CCAM^C-CTM and WRF ^C-CTM dust peaks coincides with the peak in hourly PM₁₀ measurements (Fig 6a), with ACCESS^C-CTM peaking slightly later. The peak of 699 μg m^-3 in the 5-minute observations (not shown) occurred at 06:55 UTC as the storm front passed and PM₁₀ remained above 100 μg m^-3 for 20 minutes. There was no increase in observed PM_2.5 mass concentrations across Melbourne air quality stations (at Footscray, Alphington and Geelong South), although whole pollen and dust are generally found in more coarse size fractions. Assuming SPPs were responsible for the 2016 Melbourne event, the PM_2.5 observations at the three Melbourne stations suggest SPPs were too small to contribute much mass to PM_2.5.

We now investigate wind driven rupturing processes. The time series in modelled and observed pollen shells fit quite well using both f_WS and WS parameterisations (Fig 6b and 6c), suggesting more whole pollen ruptured on the evening of 20 November. However, ACCESS^C-CTM predicts a slightly higher peak in pollen shells at the time of the storm, whilst CCAM^C-CTM and WRF^C-CTM predict more pollen shells the day before. More pollen shells result from f_WS than the WS experiment because rupturing occurs without a threshold.

On-plant mechanical rupturing using f_WS produces the best correlations between the model and observed pollen shells of all our experiments (r = 0.86–0.91 in Fig 6d). Whilst on-plant mechanical rupturing using f_WS might therefore provide the best parameterisation of pollen shell concentrations in the atmosphere, it does not explain the severe respiratory impacts of the storm. On-plant mechanical rupturing using WS directly with a threshold of 5 m s^-1 (Fig 6e) produces a peak in pollen shells at the time of the storm in all three models (CCAM^C-CTM also produces a larger peak a few hours prior). The vertical time series in wind speed at Melbourne (not shown) shows wind speeds of up to 18 m s^-1 throughout the lowest 5 km of the atmosphere, starting at 12:00 UTC on 20 November and remaining high above 2 km throughout 21 November. The downward transport of horizontal momentum in the thunderstorm downdraft likely contributed to the gusty winds that occurred on 21 November.

As f_WS yields a value between 0 and 1 and the wind speed can be any value, the pollen shell concentrations resulting from the WS on-plant mechanical rupturing experiment are much higher (Fig 6d and 6e). In terms of the ratio of in-atmosphere to on-plant mechanical rupturing, mechanical rupturing provides 95–96% of the total concentration of pollen shells comparing Fig 6 panels (b) and (d) for f_WS across all models and 80–94% comparing Fig 6 panels (c) and (e) for WS. Note here that as mechanical rupturing takes place before the pollen leaves the plant, the threshold of whole pollen grains in the air does not apply. Therefore, there are many thousands of pollen shells resulting from the mechanical rupturing experiments, as there is currently no model cap to the amount of pollen on the plants. The purpose of the experiment was to investigate whether a rapid increase in pollen shells could be achieved at the correct time. A parameter representing the SPP reservoir can be added when measurements of SPP concentrations are available.

We now review the possibility of pollen rupturing due to electrical activity when the RH is low. In the hours prior to the storm, RH was low in the observations and in all the models, particularly at the surface, and was < 20% in CCAM and WRF up to a height of ~2 km. The timings of the peaks produced by the RH ≤ 30% experiment (Fig 6f) are very similar to the in-atmosphere f_WS experiment (Fig 6b), but with lower correlations (r = 0.56–0.76 Fig 6f). A peak in pollen shells is produced at the time of the storm, but it is not as large as the peak produced the day before.

Electrical activity due to lightning is now explored as a trigger for rupturing. Fig 7 shows that the pattern of lightning strikes traversed from north west Victoria to south of the city of Melbourne during the day of the storm, and that there were no counts of lightning in the city itself. In this experiment, whole pollen was ruptured to the west of Melbourne, and the pollen shells were transported to the monitoring site (Fig 6g). The models continue to predict two peaks in the time series of pollen shells. CCAM^C-CTM and WRF^C-CTM produce larger peaks in pollen shells at the time of the storm. However, the predicted concentration of pollen shells in Melbourne is low, partly because the location of the lightning was west of Melbourne, and partly because the parameterisation is dependent on the number of lightning occurrences per grid cell per hour. In this domain the maximum in any hour is only 6.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Location of lightning occurrences in 6-hourly blocks starting late 20 November into 21 November 2016.

The pink pixels coincide with timing of the storm passing Melbourne. Basemap from Igismap.com 2020.

https://doi.org/10.1371/journal.pone.0249488.g007

This paper has focussed on the timeseries of pollen shells measured at Melbourne University. Modelling has shown that lightning or a mechanism based on wind speed provides the best explanation of these measurements. We use Fig 8 to determine the spatial variation in surface SPPs produced at the time of the storm for four of the in-atmosphere rupturing experiments, f_WS, WS ≥ 5 m s^-1, RH ≤ 30% and lightning. Resulting SPP concentrations are in the range 1.4×10⁴–1.4×10⁵ m^-3. The aim is to determine if these mechanisms produce a field in SPPs that resembles the position of the storm front, allowing for possible small timing differences related to the meteorology (c.f. Fig 3). There are only SPPs present in the WRF^C-CTM atmosphere at this time using the RH ≥ 80% threshold option, with the highest concentrations ~5×10³ m^-3 situated on the coast (S2 Fig in S1 File). The first three experiments (f_WS, WS ≥ 5 m s^-1 and RH ≤ 30%) cause similar SPP fields than the lightning experiment. The first three experiments show SPPs both behind and ahead of the storm front in all three models. The north-south line of the storm is better predicted in ACCESS^C-CTM, than CCAM^C-CTM or WRF^C-CTM which show alignment from the north-west to south-east direction. The only experiment to restrict the SPP field behind the storm front occurs using lightning in all three models. At 06:00 UTC the lightning occurred to the west of Melbourne, and the SPPs produced reached Melbourne in the following model timesteps (S3 Fig in S1 File). This is an artefact of the modelling process; in-atmosphere rupturing occurs at the end of each model time step after the airborne concentration of whole pollen is calculated. Transport then occurs at the beginning of the next time step. Following the pattern of lightning occurrences, the position of the peak SPPs incorrectly suggests that towns to the west of Melbourne (e.g. Creswick and Geelong) would receive the majority of the exposure. Although lightning indicates the presence of a thunderstorm, perhaps another meteorological variable of the storm such as diagnosis of strong atmospheric convergence lines might provide a better description of in-atmosphere rupturing [48].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 8. Sub-pollen particles (m^-3) in the surface atmosphere at 06:00 UTC.

Does not include mechanical rupturing. The heavy black line shows the approximate position of the storm front diagnosed from radar [19]. Panels (a-c) f_WS, (d-f) WS > 5 m s^-1 (g-i) RH < 30% and (j-l) lightning. Note the different scale for lightning colourbar. ACCESS^C-CTM is shown in the left hand panels (a,d,g,j), CCAM^C-CTM in the middle panels (b,e,h,k) and WRF^C-CTM in the right hand panels (c,f,I,l). Basemap from Igismap.com 2020.

https://doi.org/10.1371/journal.pone.0249488.g008