This paper investigates the mechanisms involved in the dispersion, structure, and mixing in the vertical column of atmospheric pollen. The methodology used employs observations of pollen concentration obtained from Hirst samplers (we will refer to this as surface pollen) and vertical distribution (polarization-sensitive lidar), as well as nested numerical simulations with an atmospheric transport model and a simplified pollen module developed especially for this study. The study focuses on the predominant pollen type,
Pollen is a very important biological structure present all over the world. It functions as a container in which the male gametophyte generation of the angiosperms and gymnosperms is housed and is responsible of the gene flow. To be functional, mature pollen must be transported from the place where it is generated to the female structures of a flower of the same species, through a process named pollination. Several pollination types exist, with one of them being anemophily. Anemophily occurs when pollen grains are passively transported by the air. In this case, pollen behaves as biogenic aerosol and constitutes a substantial fraction of the mass of particulate matter in the air during the flowering season. Consequently, pollen can have strong health effects, causing allergenic rhinitis and asthma. The study of the pollen transport in the atmosphere is a relevant topic, not only because it allows the evaluation of the potential risks for human health and the prevention of its effects but also because it will possibly provide a better understanding of the spatial distribution of the species (Belmonte et al., 2008; Schmidt-Lebuhn et al., 2007; Sharma and Kanduri, 2007; Smouse et al., 2001).
In the last few decades, several works have studied the transport of pollen
species in the atmosphere, developed different numerical models, and
compared the results with in situ surface observations. Such models include
a source term and a dispersion module. The source term characterizes the
pollen emission considering the start, end, and duration of the pollen season
(e.g., Sofiev et al., 2013) and the diurnal profiles of the emission fluxes
(e.g., Helbig et al., 2004; Sofiev et al., 2013) with parameterizations
derived from statistical analysis of available observations (surface pollen
counts and meteorological variables). The seminal works of Helbig et al. (2004), Schueler et al. (2005), Schueler and Schlünzen (2006), and Sofiev
et al. (2006) showed the value of such models for studying the transport of
pollen and their application as forecasting tools. Different models have
been developed since then to model the transport of birch (Sofiev et al.,
2006; Vogel et al., 2008; Efstathiou et al., 2011; Zink et al., 2013; Zhang
et al., 2014), ragweed (Efstathiou et al., 2011; Zink et al., 2013; Prank et
al., 2013; Wozniak and Steiner, 2017), grass (Zhang et al., 2014; Wozniak
and Steiner, 2017), olive (Zhang et al., 2014; Sofiev et al., 2017),
broadleaf tree pollen (Helbig et al., 2004; Wozniak and Steiner, 2017), and
evergreen needleleaf tree pollen (Wozniak and Steiner, 2017) over regional
domains, with horizontal resolutions ranging from 50 to 10 km. Although
advancements in the field have been achieved, current models still present
significant limitations to reproduce the life cycle of pollen, as highlighted
in the model intercomparison works of Sofiev et al. (2015, 2017), where
different transport models were used to study the pollination season of
birch and olive with large variability among them. Nowadays, the Copernicus
Atmosphere Monitoring Service (CAMS) regional production provides forecasts
at the European continental scale of pollen concentration for birch, olive,
grasses, ragweed, and alder using a multi-model ensemble approach
(
Scattering coupled to depolarizing properties, on the one hand, and emission
of fluorescence spectra when excited with UV radiation of some chemical
substances contained in pollen and bioparticles in general, on the other
hand, are the two main properties of pollen grains that allow their remote
detection in the atmospheric column with lidar techniques. The first
property makes elastic, polarization-sensitive lidar systems powerful
instruments for the detection of atmospheric pollen. The number of articles
from the lidar community dealing with this topic has increased in recent
years (Sassen, 2008; Noh et al., 2013; Sicard et al., 2016a; Bohlmann et
al., 2019; Shang et al., 2020; Bohlmann et al., 2021). In particular, Shang
et al. (2020) developed a method to retrieve the linear depolarization ratio
of the pollen (or the mixture of pollen) present from measurements of the
particle backscatter coefficient and depolarization ratio and
Ångström exponent. The second property of pollen to emit
characteristic fluorescence spectra has been known for a few years. Such
spectra have been detected by the technique of the so-called laser-induced
fluorescence lidars (Sugimoto et al., 2012; Sharma et al., 2015; Wojtanowski
et al., 2015; Rao et al., 2017; Saito et al., 2018; Richardson et al.,
2019). Very recently, fluorescence returns produced by atmospheric pollen
excited at 355 nm were also measured with broadband filters and combined
with multi-wavelength, multi-depolarization Raman retrievals (Veselovskii
et al., 2020, 2021). Some authors even detected fluorescence effects of the
aerosol of biogenic origin with the water vapor channel of an elastic/Raman
lidar system (Immler et al., 2005). Although the use of lidar-derived,
range-resolved information on the optical properties of pollen is available,
it has never been used in a generalized manner for validating modeling
experiments trying to reproduce the pollen release and transport in the
atmosphere, nor has it been used as a complementary tool to understand
the pollen vertical dispersion, distribution, and mixing. Some tentative
exercises, published in reviewed proceedings of international conferences,
using modeling to help understand the observed vertical distribution of
The objective of the paper is to improve our understanding of pollen
vertical distribution in the atmosphere by combining in situ concentration
(Hirst), columnar optical property (lidar) measurements, and dispersion
modeling. The paper is focused on the mechanisms responsible of the pollen
vertical dispersion and mixing and how the pollen vertical structure impacts
its horizontal transport. For that reason, the pollen type selected is
Pollen grain daily concentration was measured by the Aerobiological Network
of Catalonia (Xarxa Aerobiològica de Catalunya – XAC) at seven sites around Catalonia (NE Spain). At the Barcelona site (2.165
Pollen samples are obtained using volumetric suction pollen trap based on
the impact principle (Hirst, 1952), the standardized method in European
aerobiological networks (Galán et al., 2014). The Hirst sampler (Hirst,
1952) is calibrated to handle a flow of 10 L of air per minute, thus
matching the human breathing rate. Pollen grains are impacted on a
cylindrical drum covered by a Melinex film coated with silicon fluid
(LANZONI srl®) as trapping surface. The drum rotates at 2 mm per hour; so, each 48 mm represents 24 h of continuous sampling. The drum is changed weekly, and the exposed tape is cut into pieces, with each one corresponding to 1 d. Pollen grains are counted under a light microscope at
Pollen (XAC) and meteorological (XEMA) surface stations used in
this study.
The profiles of the particle backscatter coefficient and the volume and
particle depolarization ratios were measured with the Barcelona micro-pulse
lidar (MPL) system (Sigma Space Corporation; model MPL-4B). The system is part of the MPLNET (Micro-Pulse Lidar Network;
The total lidar signal,
By adapting the notations of Flynn et al. (2007) to ours, one can formulate
the linear volume depolarization ratio,
We also calculated the vertical height,
All the MPL retrievals presented in this work were performed with in-house algorithms and not with the MPLNET processing, since the system only entered the network in 2016.
The dispersion of the airborne pollen in the atmosphere was modeled with the Multiscale Online Nonhydrostatic AtmospheRe CHemistry model (MONARCH; Pérez et al., 2011; Jorba et al., 2012; Badia and Jorba, 2015; Badia et al., 2017). The MONARCH model is a fully online multiscale chemical weather prediction system for regional and global-scale applications, with telescoping nest capabilities, developed at the Barcelona Supercomputing Center (BSC). The system is based on the meteorological Nonhydrostatic Multiscale Model on the B grid (NMMB; Janjic and Gall, 2012), widely verified at the National Centers for Environmental Prediction (NCEP). The MONARCH model couples online the NMMB with the gas-phase and aerosol continuity equations to solve the atmospheric chemistry processes in detail. The model is designed to account for the feedbacks among gases, aerosol particles, and meteorology. Currently, it can consider the direct radiative effect of aerosols while neglecting dynamic cloud–aerosol interactions. Different chemical processes were implemented following a modular operator splitting approach to solve the advection, diffusion, chemistry, dry and wet deposition, and emission of atmospheric constituents. Meteorological information is available at each time step to solve the chemistry. In order to maintain consistency with the meteorological solver, the chemical species are advected and mixed at the corresponding time step of the meteorological tracers, using the same numerical schemes implemented in the NMMB. The advection scheme is Eulerian, positive definite, and monotone, maintaining a consistent mass conservation of the chemical species within the domain of study (Janjic and Gall, 2012).
In this work, the model has been enhanced with a new pollen module that
allows the study of the life cycle of different pollen types. The numerical
schemes used for the aerosols have been extended for pollen. The pollen type
largely predominant during the pollination event analyzed in this study was
The MONARCH model implements a mass-based aerosol scheme that has been
extended to pollen bioaerosols. Table 2 summarizes
the main characteristics of the implementation.
Databases and
Mass-based growth factor (
The emission scheme implemented in MONARCH is based on the concepts of the
parameterization of Helbig et al. (2004), with some modifications. It
computes the vertical emission flux of
The available number of pollen grains per tree during a season is a highly
uncertain parameter. Several works provide estimates of this parameter for
pines. Tormo Molina et al. (1996) calculated the annual production of pollen grains for three individual pine trees (
Once
The weather-dependent function
To study the dispersion of
The simulations cover the period 20 March to 2 April 2015, which comprises
the pollination event under study in this work. The meteorological initial
and boundary conditions are obtained from the ERA-5 reanalysis at 30 km
horizontal resolution and 6 h frequency. No boundary conditions for the
Note that the
We propose performing a closure study between in situ and column measurements by constraining the lidar retrieval in the first part of the vertical profile to the concentration measurement of the Hirst collector. An important added value is that the closure study also fixes parameters needed for the conversion of the model output. In order to compare the lidar retrieval to measurements made near the ground level, we consider the first lidar measurement (225 m) to be a proxy of what it would be near the ground level. This hypothesis is somehow validated by the fact that the lidar vertical distribution, as it will be shown later, is rather flat (concentrations barely vary) as one comes closer to the ground. Let us note that the Hirst collector is situated on the roof of a building at 23 m above ground level. According to Rojo et al. (2019), the effect of height on pollen concentrations is mainly determined by differences within the first 10 m above ground. In the case of the model, we consider the first model layer, the center of which varies between 24 and 24.4 m over all the simulations considered here.
From the analysis of the samples of the Hirst collectors, the concentration
of the pollen type
The lidar instrument measures the particle backscatter coefficient from
which the total pollen contribution, i.e., the total pollen backscatter
coefficient,
The mass concentration of Eq. (10) can be converted into number
concentration of the pollen grain
Finally, the model provides mass concentration for the pollen grain
The lidar number concentration,
Each day, for the 24 h (
Summary of the daily values of wf and
As an illustration of how the lidar and Hirst number concentration agree with
the values of
In order to quantify for the pollen type of interest in this study,
The event of interest took place between 27 and 31 March 2015, the most
intense period of the
Figure 5c also shows the quicklook plot, also called time–height plot, of the volume depolarization ratio of the lidar. It is not a quantitative plot, since both the molecules and the particles contribute to the volume depolarization, but it is an indicator of the presence (or not) of depolarizing particles (here, pollen grains). In Fig. 5c, the dark green areas represent the molecular level (only detectable at the high temporal resolution of the quicklook during nighttime because of the reduced background signal – as opposed to daytime), light green areas represent low-depolarizing particles (urban background), and yellow/orange areas represent high-depolarizing particles, i.e., pollen. The vertical distribution of the airborne pollen also shows a clear diurnal cycle, with usually no or weak nighttime activity in the upper
layers, where the pollen grains are probably staying very close to the ground. The diurnal cycle is marked by an increase in the amplitude and height of the volume depolarization ratio, starting around 10:00 UTC, and a decrease, starting before 16:00 UTC. This diurnal pattern is observed on each single day of the pollination event. On the first 4 d, the volume depolarization ratio has come back to its background value (
In order to understand the dynamics of
Maps of the surface
To assess the representativeness of the model results, we have compared them
with the hourly meteorological and daily aerobiological observations
described in Sect. 2. Here, we focus on surface observations. The quality of the results is quantified by means of classical statistics (Pearson correlation coefficient,
Regarding the meteorological results, Table 5 presents the statistics for the 9 km base case run and all seven stations (results for 3 and 1 km in the Tables S1 and S2 in the Supplement), and Fig. 7 shows the temporal evolution of the same variables (except concentration) at the three model resolutions for Barcelona and Bellaterra/Sant Cugat (results at the other five stations in Fig. S1 in the Supplement). The statistics indicate a good agreement
for the temperature, irradiance, and relative humidity in most of the stations (correlations above 0.9 and low bias), while higher errors are observed in the wind. Overall, the results are within the typical performance range of mesoscale models in the area of study (i.e., Jiménez-Guerrero et al., 2008), with surface winds being one of the most difficult variables to reproduce in coastal regions with a complex terrain like the one under study. On the one hand, the model results in coastal sites may show higher errors in
temperature, winds, and relative humidity. These sites are close to the sea–land interface, and small inaccuracies in the representation of the
coastline may have a strong impact in the results. The Tarragona site is a clear example where the temperature results are degraded with the 1 km domain compared with the upper nests (see the Supplement) as a result of the land and sea grid cells represented in each model resolution surrounding the site. An excessive influence of marine air masses will result in a lower thermal amplitude and overestimated relative humidity. On the other hand, better statistics are obtained at inland sites where the topography is properly captured by the model resolution. Most sites show RMSE and bias below 2.6 and
The Barcelona site is representative of a coastal station, located few kilometers off the shore, while Bellaterra/Sant Cugat is an example of an inland site not affected by major mountain ranges (i.e., the Pyrenees). At both stations an improvement is detected in most meteorological variables with the increase in resolution (see Fig. 7). The model reproduces the daily cycle of the temperature and relative humidity and captures the cloudiness observed during 29M and 30M. As noted in the statistics, more disagreement is seen in the wind speed, and the different resolutions show larger variability compared with the other variables. The wind results in the Barcelona site are consistent among the model resolutions. Some systematic bias is observed in the wind speed at the end of the day and during the first hours of the following day. The model overestimates the calm winds observed during nighttime, and it tends to underestimate the morning peak (i.e., transitions from 27M to 28M, 29M to 30M, and 30M to 31M). This results in lower relative humidity and higher temperatures during the calm periods compared with observations, although this does not impact the wind direction that is well captured most of the time. The results at the Bellaterra/Sant Cugat site show a clear improvement with the resolution. The site is located in a long valley surrounded by two mountain ranges that are better represented in the model by increasing the resolution. The overestimated wind during nighttime in Bellaterra/Sant Cugat results in the model wind veering from southwest to northwest (i.e., 27M and 29M), while observations show a calm, stagnated situation. Such an error leads to a model underestimation of the relative humidity and an overestimation of the temperature there. On the night of 31M, the observations show weak easterly winds, while the model develops northwesterly winds. Albeit that the limitations identified, the model captures the meteorology of the event under study reasonably well.
The main driver of the pollen release in the atmosphere is the wind. The
period of study is not characterized by intense surface winds. This means that wind speed peaks are below 10 m s
The daily average of the
Statistics (Pearson correlation coefficient
Comparison of the model and observations at the
Daily average
In this section, we discuss the hourly resolution results of the surface
concentration at the Barcelona XAC site where specific high-temporal
resolution measurements were available for the period 27M to 31M.
Figure 9 shows the time series of the hourly
To compare the performances of the model at different horizontal resolutions,
Table 6 presents the day-by-day statistics for the hourly
Comparison of the model forecast and the observations of
Day-by-day statistics (root mean square error (RMSE), Pearson
correlation coefficient,
The study of the mechanisms responsible for the pollen transport and the
analysis of the model performances for predicting the vertical distribution of
Before analyzing the results, we present the latitudinal
and longitudinal cross sections of the model
Vertical cross section of the
In order to avoid misleading results caused by the averaging of low
(nighttime) and high (daytime) concentration profiles, a first analysis is
made with the diurnal (defined as the average of the hourly values between
09:00 and 17:00 UTC) statistical indicators. In Table 7, we report the diurnal average of FB and
The diurnal correlation coefficient is higher than 0.08 in all cases
(Table 7) and varies between 0.08 and 0.92. The lowest values of
Diurnal (09:00–17:00 UTC) statistics (fractional bias,
A lot of variability in FB is also observed as a function of the domain
resolution considered for the simulation (Table 7).
In all cases, the simulations at 1 km resolution give lower diurnal FB
than at 9 and 3 km resolution. Except on 29M, the ay for which the 9 and 3 km resolution yields similar FB, the 3 km resolution gives a slightly lower FB than the 9 km one. The correlation coefficient can also vary
significantly from one resolution to another; a difference of up to 0.80
(enhanced simulation on 29M) is observed between the 9 (
The diurnal standard deviation ratio (Table 7) also
illustrates the underestimation of the model during the diurnal hours on 27M
and 31M (SDR
The effect of the deposition and sedimentation on the quantity of pollen
grains transported vertically is studied by means of the two simulations
defined in Sect. 3.3 as the base case (100 % deposition; 100 % sedimentation) and enhanced (50 % deposition; 50 % sedimentation) simulations. The results are reported diurnally in Table 7 and daily in the top plots of Fig. 11. The study on the sedimentation is motivated by the high sedimentation velocity of large
To have a closer look to the performance of the model in the column with
respect to time, we plot, in Figs. 12, 13, and 14, the hourly evolution of the profiles of
For each of the 5 d of the event (columns), the daily mean
modeled vs. observed vertical profile (top row) of the
Hourly evolution of the vertical profile of
Same as Fig. 12 for 29M.
Same as Fig. 12 for 30M.
As seen in Sect. 5.1 and in Fig. 7 in particular, the model has some systematic error that induces an overestimation of nocturnal wind and underestimation of the increase in wind during the morning. The afternoon peak is likely due to the fact that pollen stays aloft longer than it does in reality, or that there is an excessive emission flux upwind during the previous hours. This behavior of the model is presented in Figs. 12, 13, and 14. According to the lidar, the pollen layer grows between 09:00–14:00 UTC, following a typical development of the convective boundary layer. During that period, the top of the layer is well reproduced by the model with excessive mobilization of pollen grains, depending on the horizontal resolution and hour. Note the good agreement in some specific profiles, see, e.g., on 30M at 12:00–13:00 UTC for 3 km (Fig. 14). The sink of the layer is observed between 14:00 and 15:00 UTC on 28M and 29M (the model predicts it 1 h later) and between 13:00 and 14:00 UTC on 30M (the model predicts it 2 h later). The delay of the model in predicting the pollen layer drop in the afternoon might be linked to the fact that the model delays the decay of the convective boundary layer. In the afternoon, and especially for the 1 km resolution, the model predicts a strong and steep decrease in concentration towards the surface. The same result is observed with the surface concentration in Sect. 5.2, where the simulation at 1 km resolution shows a decrease in concentration (which leads to an underestimation of the model; see Fig. 9), followed by sudden peaks that correlate well with the wind speed. Again, it shows the higher sensitivity to wind speed and direction of the 1 km simulation compared to the 3 and 9 km simulations. Also, more structures are visible at 1 km resolution than at the other resolutions. The finer the resolution, the more vertical structures can be seen. However, the structures visible at the 1 km resolution are not always reproducing reality (see, e.g., on 29M at 14:00 UTC; Fig. 13).
The Hirst observations are much more variable than the model concentration
and the meteorology. Although following a general trend, the Hirst
concentrations oscillate up and down most of the time and throughout the whole day (see the black circles in the hourly plots of Figs. 12, 13, and
14, and the black line in Fig. 9), while the model concentration usually
steadily increases in the morning and decreases in the afternoon after
15:00–17:00 UTC. Such a difference between model and observations is not visible in the meteorological variables (Fig. 7). In fact, for some variables, the opposite happens, e.g., the model wind speed is more
variable than the observation. The high variability in the observed
This paper combines pollen concentration surface (Hirst) and columnar
(lidar) measurements, as well as an atmospheric transport model, with a
simplified pollen module especially implemented for this study to improve
our understanding of pollen vertical dispersion, distribution, and mixing in
the atmospheric column. The pollen type under study is
Nested numerical simulations at 9, 3, and 1 km horizontal resolution are
deeply analyzed, in addition to the effects of sedimentation and dry deposition. The model used is the Multiscale Online Nonhydrostatic AtmospheRe CHemistry model (MONARCH) developed at BSC. The three model resolutions follow the evolution of the measurements in general. The 1 km run tends to simulate sudden rises and falls in the concentrations compared with the 9 and 3 km, which are able to maintain some background concentrations in the air. Although the simulation at 1 km may improve the model score in places with complex topography, the meteorological errors in the wind have larger impacts in such a high-resolution configuration. In this sense, the combination of the three resolutions provides complementary information to advance our understanding in key driving processes, but 9 or 3 km simulations might be preferred for specific forecasting applications. The largest discrepancies between measured surface (Hirst) and column(lidar) concentrations occur during the nighttime as no pollen is detected in the column, whereas it is present at the surface. This is likely due to the limitation of the lidar for measurements below 225 m where most of the stable boundary layer resides. Simulated profiles at different resolutions show large variability throughout the event of study. During nighttime, the model tends to overestimate the amount of pollen available in the stable boundary layer compared with the Hirst concentration, but matches its top when it is detected by the lidar. Such an overestimation may be attributed to systematic errors in the wind speed during midnight. This points out that the wind is the main driver of the nighttime/early morning pollen activity. A 50 % decrease of the sedimentation/deposition parameter increases the daily column concentration 10 %–17 %. Decreasing the deposition/sedimentation in the model is not enough to significantly change the results on a daily basis. Both parameters only have a limited impact on the vertical concentration, suggesting that other processes are more relevant to reproduce the measurements. These mechanisms might have a larger impact during nighttime and would deserve further investigation. Finally, from our model results, the vertical structure of the pollen is mainly controlled by the vertical mixing within the boundary layer and the sensitivity of the emission scheme to winds. In general terms, the model matches the depth of the pollen layer during stable conditions and the growth of the convective boundary layer, but systematic biases are detected in the second half of the day, with persistent overestimation within the convective boundary layer. During the latter period, lidar profiles do not extend up to the top of the boundary layer, probably because of gravity effects that are not well represented by the model.
The results of the study emphasize the tremendous importance of the
completeness of the tree spatial distribution, density, and type. Even in a
relatively small geographical region, the behavior of the pine trees is
significantly heterogeneous due to the presence of different pine species,
different micro-climates, and meteorological conditions. The assumption made
in the model that the availability of pollen grains in the pine trees (
The proposed methodology requires a Hirst sampler nearby a polarization-sensitive lidar, first, for the confirmation of the presence and the type of pollen observed and, second, for the conversion of the lidar-derived backscatter coefficient into number concentration (through the retrieval of the specific extinction cross section). To apply the methodology at sites with no Hirst sampler, but at least knowing the most probable predominant pollen type present, a look-up table would be needed. Such a table can be obtained by applying the methodology at a site with both a polarization-sensitive lidar and a Hirst sampler on a large number of pollen loads with different predominant pollen types and over a relatively long period of time. This is a guideline for future work.
The MONARCH source code is available at
The meteorological data from XEMA are available at
The supplement related to this article is available online at:
MS and OJ conceived the study. OJ, MA, and MS designed the pollen emission scheme. OJ developed the pollen module in NMMB-MONARCH and conducted the model experiments. RI and MA provided the tree density maps. MS and AC provided the lidar measurements, and MS conducted the comparison with the model. JB and CdL provided the pollen measurements and the information related to pollen, phenology, and aerobiology. JJ computed the model statistics at the surface, and OJ conducted the comparison with the model. MS and OJ prepared the paper, with contributions from all co-authors.
The contact author has declared that neither they nor their co-authors have any competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The authors thankfully acknowledge the computer resources at MareNostrum 4 and the technical support provided by BSC (grant nos. RES-AECT-2019-3-0001 and RES-AECT-2020-1-0007). The authors also thank the Meteorological Service of Catalonia for providing the meteorological measurements. The MPLNET staff at NASA GSFC are warmly acknowledged, for the continuous help in keeping Barcelona MPL system and the data analysis up to date. Jose Maria Baldasano is acknowledged as the principal investigator (PI) of the Barcelona MPL.
The lidar data analysis has been supported by funding from the H2020 program from the European Union (grant nos. 654109, 778349, and 871115), the Spanish Ministry of Economy, Industry and Competitiveness (grant no. CGL2017-90884-REDT), the Spanish Ministry of Science and Innovation
(grant no. PID2019-103886RB-I00), and the Unity of Excellence “María de
Maeztu” financed by the Spanish Agencia Estatal de Investigación (grant no. MDM-2016-0600). Modeling activities have been supported by funding from the Ministerio de Ciencia, Innovación y Universidades, as part of the
BROWNING project (grant no. RTI2018-099894-BI00) and ACTRIS-España
(grant no. CGL2017-90884-REDT). Airborne pollen data sampling and analyzing have been supported by funding from sponsors of the Catalan Aerobiological Network (LETI Pharma, Diputació de Tarragona, Servei Meteorològic de
Catalunya, Diputació de Lleida, Sociedad Española de
Alergología e Inmunología Clínica (SEAIC), Societat Catalana
d'Al
This paper was edited by Susannah Burrows and reviewed by two anonymous referees.