Temporal Variability and Geographical Origins of Airborne Pollen Grains Concentrations from 2015 to 2018 at Saclay, France

The study of the origin and dispersion processes associated with airborne pollen grains are important to understand due to their impacts on health. In this context, a Hirst-type spore trap was utilized over the period 2015–2018 to monitor ambient pollen grains at Saclay, France, a receptor site influenced by both clean air masses originating from the Atlantic Ocean and polluted air masses under anticyclonic conditions. The objective of this work was to use ZeFir (a user-friendly, software tool recently-developed to investigate the geographical origin and point sources of atmospheric pollution) as a method to analyse total and allergenic airborne pollen grain concentrations. Strong interannual variability was exhibited for the total pollen grains concentrations and it was determined that this was mainly driven by Betulaceae pollen, with a general increasing trend displayed. The start of the pollen season was seen to be triggered by particular synoptic conditions after a period of dormancy and two maximums were displayed, one in April and a second in June. Results from the ZeFir tool, fed with on-site hourly meteorological and pollen measurements, demonstrate that the dominant pollen grains inputs to Saclay are favoured by non-prevailing winds originating from East and North in association with dry air, moderate winds, mild temperature and enhanced insolation.


Introduction
The concentration of Bioaerosols in the atmosphere, including airborne pollen grains, has been monitored for several decades all over the world and especially in Europe. This is due to their impact on climate and health. [1,2]. In populated areas, pollen concentrations are monitored mainly for allergy prevention purposes [3][4][5]. It is acknowledged that changes in land uses, as well as growing urbanization and global warming is affecting the biosphere and modifying biodiversity [6,7]. Atmospheric concentrations of pollen grains are often associated with reproduction mechanisms, where two major processes are involved. The predominant one is a reproduction mechanism using  [42,43]. The Counting and identification of the pollen grains were performed by the French Monitoring Network of Aerobiology (Réseau National de Surveillance Aérobiologique, RNSA, Brussieu, France). The RNSA is a member of the European Aeroallergen Network (EAN) and follows the recommendations of [44] concerning the minimum requirements for the counting procedure. The quality assurance (QA) and the quality control (QC) of such measurements were reported in Reference [45] during an International Intercomparison on 15 sites. The standard analytical method used accounts for 10% of the surface area of each daily sample.

Pollen Grains Identification and Counting
The VPPS 2000 spore trap was located on the roof of the observatory at 15 m above the ground without any neighbouring vegetation. The air was continuously pumped through a 14 × 2 mm orifice and particulate matter was impacted on a rotating drum containing a 19-mm wide cellophane tape coated with silicone. The drum containing the impacted pollen grains was changed every week at the same time. Sampling was always facing the prevailing winds and the flow rate was 10 L per minute (LPM). A clockwork mechanism rotated the drum continuously at a speed of 2 mm per hour. After 7 days of sampling, the tape, which contained the impacted particles, was removed and cut into seven equal parts. Each part corresponded to one day of sampling and was coloured with a colour-fixing reagent based on Fuchsin (Prolab Diagnostics, Switzerland), facilitating the differentiation of pollen grains from fungal spores via optical microscopy. This optical reading and counting was done using an optical microscope, (Realux, France) at ×400 of magnification ( Figure 2). The standardized technique used to count the pollen grains was longitudinal. The results were available in intervals of two-hour averages and were representative of regional concentrations [46,47].  [42,43]. The Counting and identification of the pollen grains were performed by the French Monitoring Network of Aerobiology (Réseau National de Surveillance Aérobiologique, RNSA, Brussieu, France). The RNSA is a member of the European Aeroallergen Network (EAN) and follows the recommendations of [44] concerning the minimum requirements for the counting procedure. The quality assurance (QA) and the quality control (QC) of such measurements were reported in Reference [45] during an International Intercomparison on 15 sites. The standard analytical method used accounts for 10% of the surface area of each daily sample.

Pollen Grains Identification and Counting
The VPPS 2000 spore trap was located on the roof of the observatory at 15 m above the ground without any neighbouring vegetation. The air was continuously pumped through a 14 × 2 mm orifice and particulate matter was impacted on a rotating drum containing a 19-mm wide cellophane tape coated with silicone. The drum containing the impacted pollen grains was changed every week at the same time. Sampling was always facing the prevailing winds and the flow rate was 10 L per minute (LPM). A clockwork mechanism rotated the drum continuously at a speed of 2 mm per hour. After 7 days of sampling, the tape, which contained the impacted particles, was removed and cut into seven equal parts. Each part corresponded to one day of sampling and was coloured with a colour-fixing reagent based on Fuchsin (Prolab Diagnostics, Switzerland), facilitating the differentiation of pollen grains from fungal spores via optical microscopy. This optical reading and counting was done using an optical microscope, (Realux, France) at ×400 of magnification ( Figure 2). The standardized technique used to count the pollen grains was longitudinal. The results were available in intervals of two-hour averages and were representative of regional concentrations [46,47].
Equation (1) gives the estimated concentration (Q) of the pollen grains sampled.
where: SA is the Analysed Surface of one segment (mm 2 ), V is the sampling volume (m 3 ), ST is the total surface of one segment (mm 2 ) and n the number of counted grains. Equation (1) gives the estimated concentration (Q) of the pollen grains sampled.
where: SA is the Analysed Surface of one segment (mm 2 ), V is the sampling volume (m 3 ), ST is the total surface of one segment (mm 2 ) and n the number of counted grains.

Geographical Origins
The investigation of the geographical origins of pollen has been performed by coupling ambient concentrations with onsite measured wind data. A Two-dimension Non-parametric Wind Regression (NWR), originally developed by [48], has successfully been applied to various online air quality datasets in the literature. However, to our knowledge, it is the first time this tool designed for air quality applications is applied for the investigation of the geographical origins of atmospheric pollen. Equation (2) below describes the calculation of NWR.
where E is an estimated concentration at a wind direction θ and speed υ; Wi, Yi and Ci are respectively input wind direction, wind speed and concentration data measured at ti; K1 and K2 are two Kernel functions; and σ and h the smoothing factors. The calculation basically consists in a weighing average of concentration, where the weighing coefficients are determined through Kernel functions, whose widths are controlled by σ and h. In classical NWR as described in Equation (2), fixed σ and h values are applied. However, the (daily) temporal resolution of our dataset (daily averages) makes NWR unsuitable for this study due to atmospheric variability. Indeed, wind data that are associated with concentrations are assumed to be statistically representative but daily mean values of wind speed and direction may not be representative of the variety of wind conditions which could occur during a particular day. Therefore, if wind conditions are not stable at ti, the corresponding concentration should be downweighted to better account for atmospheric stability. For this reason, we chose a variant of NWR, called the Sustained Wind Incidence Method (SWIM) developed by [49]. Although the general principle remains the same, a scalar weight is applied to concentrations, depending notably on wind direction standard deviation.
Equation (3) below describes the calculation of SWIM.
where  represents wind direction standard deviation. This actually allows to efficiently downweight daily concentration values associated with high atmospheric variability during that day. Wind direction standard deviation was estimated by the 1-pass Yamartino equations [50]. This entire study

Geographical Origins
The investigation of the geographical origins of pollen has been performed by coupling ambient concentrations with onsite measured wind data. A Two-dimension Non-parametric Wind Regression (NWR), originally developed by [48], has successfully been applied to various online air quality datasets in the literature. However, to our knowledge, it is the first time this tool designed for air quality applications is applied for the investigation of the geographical origins of atmospheric pollen.
Equation (2) below describes the calculation of NWR.
where E is an estimated concentration at a wind direction θ and speed υ; Wi, Yi and Ci are respectively input wind direction, wind speed and concentration data measured at ti; K1 and K2 are two Kernel functions; and σ and h the smoothing factors. The calculation basically consists in a weighing average of concentration, where the weighing coefficients are determined through Kernel functions, whose widths are controlled by σ and h. In classical NWR as described in Equation (2), fixed σ and h values are applied. However, the (daily) temporal resolution of our dataset (daily averages) makes NWR unsuitable for this study due to atmospheric variability. Indeed, wind data that are associated with concentrations are assumed to be statistically representative but daily mean values of wind speed and direction may not be representative of the variety of wind conditions which could occur during a particular day. Therefore, if wind conditions are not stable at ti, the corresponding concentration should be downweighted to better account for atmospheric stability. For this reason, we chose a variant of NWR, called the Sustained Wind Incidence Method (SWIM) developed by [49]. Although the general principle remains the same, a scalar weight is applied to concentrations, depending notably on wind direction standard deviation.
Equation (3) below describes the calculation of SWIM.
where δ represents wind direction standard deviation. This actually allows to efficiently downweight daily concentration values associated with high atmospheric variability during that day. Wind direction standard deviation was estimated by the 1-pass Yamartino equations [50]. This entire study was performed with ZeFir, a user-friendly tool for wind analysis [39]. More information can be found here: https://sites.google.com/site/ZeFirproject.

Results
In this section we present the results on the annual variability, seasonal and daily variability of total pollen grains concentration measured at Saclay. These results are presented as to compare this site to similar sites in the literature and to give context to the work before using Zefir.

Interannuality of the Total Pollen Grains
During the four-year observation period presented in this study (2015-2018), 72 taxa were identified in the air of Saclay. The taxa correspond to the lowest taxonomic level, which could be identified in optical microscopy. The results presented in Figure 3 for this period show a clear year-to-year variability, as the total concentration of grains increased from +15% between 2015 and 2016 to +107% between 2017 and 2018. The trend, estimated by regular linear regression using the Annual Pollen Sum (Y) (now called Annual Pollen Integral, APIn) and the year time data (X) showed a global increase in the concentrations (Y = 6.807.7 X + 45,915.5; R 2 = 0.20). This result should be refined with at least two more years of observations and compare to similar increasing trends reported in the literature [2].
Remote Sens. 2018, 10, x FOR PEER REVIEW 5 of 25 was performed with ZeFir, a user-friendly tool for wind analysis [39]. More information can be found here: https://sites.google.com/site/ZeFirproject.

Results
In this section we present the results on the annual variability, seasonal and daily variability of total pollen grains concentration measured at Saclay. These results are presented as to compare this site to similar sites in the literature and to give context to the work before using Zefir.

Interannuality of the Total Pollen Grains
During the four-year observation period presented in this study (2015-2018), 72 taxa were identified in the air of Saclay. The taxa correspond to the lowest taxonomic level, which could be identified in optical microscopy. The results presented in Figure 3 for this period show a clear yearto-year variability, as the total concentration of grains increased from +15% between 2015 and 2016 to +107% between 2017 and 2018. The trend, estimated by regular linear regression using the Annual Pollen Sum (Y) (now called Annual Pollen Integral, APIn) and the year time data (X) showed a global increase in the concentrations (Y = 6.807.7 X + 45,915.5; R 2 = 0.20). This result should be refined with at least two more years of observations and compare to similar increasing trends reported in the literature [2]. Our results are in accordance with those observed for total pollen grains records in Europe. For example in Lublin (Poland), the total sum for 2001 was 62163 Nb#/m 3 and for 2002 50238 Nb#/m 3 [51]. The global increase of pollen grains in the air [52] previously showed an augmentation of 100 grains per year (for Quercus spp). Most of the literature published showed similar results but with a focus on specific species to calculate the trends and biannual variations [11,[53][54][55][56].

Seasonality of Total Pollen Grains Concentration at Saclay
The monthly distribution of pollen concentrations averaged over the four years of measurements displays a clear seasonal cycle as illustrated in Figure 4. P90, P75, P25, P10 represent the 90th, 75th, 25th and 10th percentiles respectively. The Main Pollen Season (MPS) is defined as the duration time when pollen is present in the atmosphere in significant concentrations at a location. It defines the main season starts and ends. The MPS used in this work is the period, when the time of the sum of Our results are in accordance with those observed for total pollen grains records in Europe. For example in Lublin (Poland), the total sum for 2001 was 62163 Nb#/m 3 and for 2002 50238 Nb#/m 3 [51]. The global increase of pollen grains in the air [52] previously showed an augmentation of 100 grains per year (for Quercus spp). Most of the literature published showed similar results but with a focus on specific species to calculate the trends and biannual variations [11,[53][54][55][56].

Seasonality of Total Pollen Grains Concentration at Saclay
The monthly distribution of pollen concentrations averaged over the four years of measurements displays a clear seasonal cycle as illustrated in Figure 4. P90, P75, P25, P10 represent the 90th, 75th, 25th and 10th percentiles respectively. The Main Pollen Season (MPS) is defined as the duration time when pollen is present in the atmosphere in significant concentrations at a location. It defines the main season starts and ends. The MPS used in this work is the period, when the time of the sum of daily mean pollen concentrations reaches 5% of the total sum until the time when the sum reaches 95%. Thus representing 90% of the pollen season. The monthly mean found for the MPS in our study starts in February (81 Nb#/m 3 ) and ends in September (29 Nb#/m 3 ).
The first maximum was observed in April (703 Nb#/m 3 ) followed by a rapid decrease in May (236 Nb#/m 3 ) and a secondary maximum in June (335 Nb#/m 3 ), followed by a slow decrease during July and August (data available in Table A1). In total, the pollen season is significant over 9 months of the year (from February to September), representing 75% of the year and 98% of the total pollen grains concentrations. It is noteworthy that 2018 exhibits very high concentrations and drives up the statistical values but similar seasonal starts/ends point and variability patterns were also observed for the 2015-2017 dataset.
Remote Sens. 2018, 10, x FOR PEER REVIEW 6 of 25 daily mean pollen concentrations reaches 5% of the total sum until the time when the sum reaches 95%. Thus representing 90% of the pollen season. The monthly mean found for the MPS in our study starts in February (81 Nb#/m 3 ) and ends in September (29 Nb#/m 3 ).
The first maximum was observed in April (703 Nb#/m 3 ) followed by a rapid decrease in May (236 Nb#/m 3 ) and a secondary maximum in June (335 Nb#/m 3 ), followed by a slow decrease during July and August (data available in Table A1). In total, the pollen season is significant over 9 months of the year (from February to September), representing 75% of the year and 98% of the total pollen grains concentrations. It is noteworthy that 2018 exhibits very high concentrations and drives up the statistical values but similar seasonal starts/ends point and variability patterns were also observed for the 2015-2017 dataset.

Daily Variability of Total Pollen Grains Concentration at Saclay
The Start of Pollen Season (SPS) used in this work follows the recommendation of [57] and has been applied to our data set, although this calculation is usually applied to species and not on the total pollen grains concentrations. During the period of observation, concentration exhibits a clear seasonality but also a strong daily variability as is illustrated in Figure 5. The identified pollen are characteristic of a degraded oceanic and semi-continental climate and were representative of this region including the Paris Area [58]. The pollen pattern at Saclay is different from one year to the next and this shall be due to variability of meteorological factors (temperatures, insolation, precipitation), which has been reviewed in Reference [59]. Regarding the MPS, the dates of the start season calculated for each of the four years were the 5 February 2015 (8 Nb#/m 3 ), 6 January 2016 (10 Nb#/m 3 ), 31 January 2017 (6 Nb#/m 3 ) and 9 January 2018 (16 Nb#/m 3 ). The calculation applied here was when the daily moving average on five consecutive days reached 5% of the annual mean [55]. This is in accordance of what is observed from the phenology observations for the early pollen season (Nadine Dupuy, personal communication).

Daily Variability of Total Pollen Grains Concentration at Saclay
The Start of Pollen Season (SPS) used in this work follows the recommendation of [57] and has been applied to our data set, although this calculation is usually applied to species and not on the total pollen grains concentrations. During the period of observation, concentration exhibits a clear seasonality but also a strong daily variability as is illustrated in Figure 5. The identified pollen are characteristic of a degraded oceanic and semi-continental climate and were representative of this region including the Paris Area [58]. The pollen pattern at Saclay is different from one year to the next and this shall be due to variability of meteorological factors (temperatures, insolation, precipitation), which has been reviewed in Reference [59]. Regarding the MPS, the dates of the start season calculated for each of the four years were the 5 February 2015 (8 Nb#/m 3 ), 6 January 2016 (10 Nb#/m 3 ), 31 January 2017 (6 Nb#/m 3 ) and 9 January 2018 (16 Nb#/m 3 ). The calculation applied here was when the daily moving average on five consecutive days reached 5% of the annual mean [55]. This is in accordance of what is observed from the phenology observations for the early pollen season (Nadine Dupuy, personal communication).
As shown in Figure 5, pollen concentrations can be extremely variable in time and intensity with regard to the annual maxima. As an example, in 2018, from one day to the next the concentrations can be seen to increase by 10 fold or more, an example of which can be seen between April 5th and 6th, 2018, as the daily concentrations spiked from 306 to 3932 Nb#/m 3 . As shown in Figure 5, pollen concentrations can be extremely variable in time and intensity with regard to the annual maxima. As an example, in 2018, from one day to the next the concentrations can be seen to increase by 10 fold or more, an example of which can be seen between April 5th and 6th, 2018, as the daily concentrations spiked from 306 to 3932 Nb#/m 3 .

Allergenic Pollen Grains Abundance and Variability at Saclay
While discussions of total pollen concentrations are of interest, certain pollen species and families can be considered more allergenic than others. In this section, we will focus on five tree species belonging to the Betulaceae and Oleaceae families, two shrub families (Cupressaceae-Taxaceae) and two herbaceous families (Poaceae and Urticaceae) because of their important allergenic properties (Tables 1 and 2). These pollen species are of particular relevance due to their health impacts [60,61] by the French Ministry of Sanitary Survey in the Paris region. For Poaceae all species are considered allergenic and cannot be accurately distinguished between each other via light microscope, hence all species have been condensed into one grouping.  Table 2. Table of the two families of spontaneous herbs with the strongest allergenic potential.

Species Families All
Poaceae Parietaria Urticaceae Figure 6 shows the relative abundance of those 9 allergenic species/families with respect to total pollen grains. Two thirds of the pollen grains can be considered allergenic (highlighted in red in Figure 6), while one third have no known impact on human health. This result indicates that allergenic species strongly drive the pollen pattern in the air at Saclay as that dominate overall concentrations. The tree and shrub pollen species, Alnus, Betula, Carpinus, Corylus, Fraxinus and

Allergenic Pollen Grains Abundance and Variability at Saclay
While discussions of total pollen concentrations are of interest, certain pollen species and families can be considered more allergenic than others. In this section, we will focus on five tree species belonging to the Betulaceae and Oleaceae families, two shrub families (Cupressaceae-Taxaceae) and two herbaceous families (Poaceae and Urticaceae) because of their important allergenic properties (Tables 1 and 2). These pollen species are of particular relevance due to their health impacts [60,61] by the French Ministry of Sanitary Survey in the Paris region. For Poaceae all species are considered allergenic and cannot be accurately distinguished between each other via light microscope, hence all species have been condensed into one grouping. Table 1. Tree and shrub species with the strongest allergenic potential.

Species Families
Alnus Oleaceae Table 2. Table of the two families of spontaneous herbs with the strongest allergenic potential.

All
Poaceae Parietaria Urticaceae Figure 6 shows the relative abundance of those 9 allergenic species/families with respect to total pollen grains. Two thirds of the pollen grains can be considered allergenic (highlighted in red in Figure 6), while one third have no known impact on human health. This result indicates that allergenic species strongly drive the pollen pattern in the air at Saclay as that dominate overall concentrations. The tree and shrub pollen species, Alnus, Betula, Carpinus, Corylus, Fraxinus and Juniperus, Cupressus combined together represent 38% of the allergenic pollen, while the grass pollen Poaceae and Urticaceae together account for 27%. Figure 7 displays the daily concentrations of the 9 allergenic pollen species/types over the 4 seasonal cycles between 2015 and 2018. The activity appears well featured in time exhibiting contrasting patterns: tree pollen first occur from February to May, followed by herbaceous pollen Remote Sens. 2018, 10, 1932 8 of 23 between June and September. It is noticeable that Poaceae disappears at the end of July while Urticaceae values remains high until the end of August, before decreasing in September. Our observations are in accordance with previous reports [62]. Tree and herbaceous pollen are the drivers for the 2 modes we previously reported, in April and June, respectively ( Figure 5).
Remote Sens. 2018, 10, x FOR PEER REVIEW 8 of 25 Juniperus, Cupressus combined together represent 38% of the allergenic pollen, while the grass pollen Poaceae and Urticaceae together account for 27%.  . The activity appears well featured in time exhibiting contrasting patterns: tree pollen first occur from February to May, followed by herbaceous pollen between June and September. It is noticeable that Poaceae disappears at the end of July while Urticaceae values remains high until the end of August, before decreasing in September. Our observations are in accordance with previous reports [62]. Tree and herbaceous pollen are the drivers for the 2 modes we previously reported, in April and June, respectively ( Figure 5). As illustrated by Figure 7, the allergenic species appear and disappear over the years with a repeatable/seasonal pattern. Thus, the allergenic pollen season followed this order: Corylus, Alnus, Cupressaceae-Taxaceae, Fraxinus, Betula, Poaceae and Urticaceae.   . The activity appears well featured in time exhibiting contrasting patterns: tree pollen first occur from February to May, followed by herbaceous pollen between June and September. It is noticeable that Poaceae disappears at the end of July while Urticaceae values remains high until the end of August, before decreasing in September. Our observations are in accordance with previous reports [62]. Tree and herbaceous pollen are the drivers for the 2 modes we previously reported, in April and June, respectively ( Figure 5). As illustrated by Figure 7, the allergenic species appear and disappear over the years with a repeatable/seasonal pattern. Thus, the allergenic pollen season followed this order: Corylus, Alnus, Cupressaceae-Taxaceae, Fraxinus, Betula, Poaceae and Urticaceae. As illustrated by Figure 7, the allergenic species appear and disappear over the years with a repeatable/seasonal pattern. Thus, the allergenic pollen season followed this order: Corylus, Alnus, Cupressaceae-Taxaceae, Fraxinus, Betula, Poaceae and Urticaceae.

Discussion
In this section the interannual, seasonal and daily variability of total pollen grains concentrations at Saclay are discussed and compared to the literature. For the purpose of investigating the origins and point sources of pollen grains impacting Saclay, the source-receptor integrated tool ZeFir was run.

Interannuality and General Increase of Pollen Concentrations
Although limited to 4 years, the strong interannuality of daily total pollen levels at Saclay seems to present a bi-annual cycle mainly influenced by Betulaceae species (Appendix A, Figure A1), in accordance with the results reported by [11,53,63] for Betula. Regarding the literature, the North Atlantic Oscillation (NAO), do not appear to influence the interannuality of Betula [64] but seems to be linked to synoptic weather patterns [19]. The general increase observed in the total pollen grains has also reported by [18], who investigated the effect of eutrophication in urban areas. However, to our knowledge, no clear link has yet been established in the literature with nitrogen originating from crop activities, or with atmospheric nitrogen from polluted environments. The combination of deposited ammonium nitrate attached to soil dust with species sensitive to nitrogen like Urticaceae remains to be investigated. Meanwhile, global warming as a whole has been pointed out by the research community as the key driver for the interannual pattern observed for given species, as reported by [65,66]. To understand the high concentration observed over the APS in 2018 we investigate different meteorological variables and the effect of rain in particular. The main reason why the 2018 APS exhibits very high concentration is the presence of high concentration of Fraxinus (major genus of Oleaceae family) and Betula as it is shown in Figure 7. For 2015, 2016, 2017, 2018, the abundance of Oleaceae was 8%, 2%, 3%, 11%, while the abundance of Betula was 13%, 27%, 13%, 34%, respectively, as illustrated by Figure A8 in Appendix A. The interannuality is therefore primarily driven by Betulaceae. As a winter drought could have had a significant effect on the tree pollen season, the effect of rain on the APS was investigated. It was noted that no strong variation in the total amount of rain was observed for our period of study (Table A3). The annual sum of rain cannot explain the maximum observed in 2018: 2015 (531.5 mm); 2016 (647.5 mm), 2017 (648.7 mm), 2018 (507.8 mm, excluding November and December). Further analysis was undertaken to examine the relationship between rainfall amounts in the 6-month period preceding the start of the flowering period of Betulaceae, that is, from August to January and defined as the Integrated Period (IP) as seen below (Table 3). The results plotted in Figure 8a show a comparable interannual variability in the precipitation pattern of the IP and in the APS of the following year. A correlation coefficient (R 2 = 0.97) was obtained when considering the IP precipitation illustrated in the Figure 8b. This result is consistent with the recent work of [66] and could be useful to estimate the severity of the pollen trees emissions and particularly Betulaceae.
The limited length of our dataset is an obstacle to statistical representativeness and will therefore need to be supported with further similar measurements for the next few years or compared with other observations, like those from the Paris station operated by RNSA. The limited length of our dataset is an obstacle to statistical representativeness and will therefore need to be supported with further similar measurements for the next few years or compared with other observations, like those from the Paris station operated by RNSA.

Pollen Seasonality Related to Air Temperature, Relative Humidity and Rain
Air temperature has been found to be a key factor for determining the start date of the pollen season, while humidity is known to explain variations in the atmospheric loadings [59]. The relationship between air temperature and relative humidity has been well described in the literature. Most of the available results however are related to studies performed on specific species for the purpose of evaluating the start and the dynamics of their seasonal cycle. Figure 9 associates the seasonality of T, RH and pollen found in Saclay over the period of observation. In accordance with previous works [67], warmer winters are followed by early onsets of the growing season. The starts of tree flowering mainly results from a balance between cold winter and warmer temperatures occurring in early spring which would participate to interrupt the vegetative state [63,68].

Pollen Seasonality Related to Air Temperature, Relative Humidity and Rain
Air temperature has been found to be a key factor for determining the start date of the pollen season, while humidity is known to explain variations in the atmospheric loadings [59]. The relationship between air temperature and relative humidity has been well described in the literature. Most of the available results however are related to studies performed on specific species for the purpose of evaluating the start and the dynamics of their seasonal cycle. Figure 9 associates the seasonality of T, RH and pollen found in Saclay over the period of observation. In accordance with previous works [67], warmer winters are followed by early onsets of the growing season. The starts of tree flowering mainly results from a balance between cold winter and warmer temperatures occurring in early spring which would participate to interrupt the vegetative state [63,68]. The limited length of our dataset is an obstacle to statistical representativeness and will therefore need to be supported with further similar measurements for the next few years or compared with other observations, like those from the Paris station operated by RNSA.

Pollen Seasonality Related to Air Temperature, Relative Humidity and Rain
Air temperature has been found to be a key factor for determining the start date of the pollen season, while humidity is known to explain variations in the atmospheric loadings [59]. The relationship between air temperature and relative humidity has been well described in the literature. Most of the available results however are related to studies performed on specific species for the purpose of evaluating the start and the dynamics of their seasonal cycle. Figure 9 associates the seasonality of T, RH and pollen found in Saclay over the period of observation. In accordance with previous works [67], warmer winters are followed by early onsets of the growing season. The starts of tree flowering mainly results from a balance between cold winter and warmer temperatures occurring in early spring which would participate to interrupt the vegetative state [63,68].  The main variable associated with the beginning of the pollen season at Saclay is air temperature (more than 6 • C on average) and low relative humidity (below 75% on average). In August, while air temperature is high (more than 21 • C) and relative humidity increases (more than 75%) appear to cause the rapid end of the pollen season. In this study, the start of the pollen season calculated for the total pollen grains concentrations fluctuates on average for 12 days, spanning a period of 3 days to 30 days (e.g., Section 3.2.). However, temperature and relative humidity fail to explain the absolute concentrations of pollen grain. Regarding the decrease of pollen concentrations in May, one explanation can be found in the seasonality of the rainfall (Figure 10). This decrease in concentration could be due to the wet removal of tree pollens which are predominant in the air in this period of the year. Intense rainfalls occurs when the temperature was seen to increase. This double effect (water and temperature) is probably responsible of the enhanced grass growth in early June. The release of grass pollen occurred when the precipitation was generally lower and the temperature higher. Moreover, during this period the weather was unstable and the rain showers were short and intense on a daily basis as illustrated by Figure A7 in Appendix A. This last point could explain the increase of intense peaks due to plant stress during this period as illustrated in the Figure A8 in Appendix A. Interestingly, it does not affect the total concentration of grass pollen which remains relatively stable (Appendix A, Table A2).
Remote Sens. 2018, 10, x FOR PEER REVIEW 11 of 25 The main variable associated with the beginning of the pollen season at Saclay is air temperature (more than 6 °C on average) and low relative humidity (below 75% on average). In August, while air temperature is high (more than 21 °C) and relative humidity increases (more than 75%) appear to cause the rapid end of the pollen season. In this study, the start of the pollen season calculated for the total pollen grains concentrations fluctuates on average for 12 days, spanning a period of 3 days to 30 days (e.g., Section 3.1.3.). However, temperature and relative humidity fail to explain the absolute concentrations of pollen grain. Regarding the decrease of pollen concentrations in May, one explanation can be found in the seasonality of the rainfall (Figure 10). This decrease in concentration could be due to the wet removal of tree pollens which are predominant in the air in this period of the year. Intense rainfalls occurs when the temperature was seen to increase. This double effect (water and temperature) is probably responsible of the enhanced grass growth in early June. The release of grass pollen occurred when the precipitation was generally lower and the temperature higher. Moreover, during this period the weather was unstable and the rain showers were short and intense on a daily basis as illustrated by Figure A7 in Appendix A. This last point could explain the increase of intense peaks due to plant stress during this period as illustrated in the Figure A8 in Appendix A. Interestingly, it does not affect the total concentration of grass pollen which remains relatively stable (Appendix A, Table A2).

Wind Prevalence at Saclay
Wind pollination is an important feature regarding the seasonality of pollen [69]. In order to better understand the mechanisms of transport of pollen by the wind, we used the ZeFir sourcereceptor tool, which allows visualization of the possible spatial origins of pollen. At Saclay, prevailing winds come from West to South-West, with speeds generally ranging from 5 to 12 km/h ( Figure 11) and are associated with oceanic air masses. A second wind regime is characterized by North (5°) to South East (125°) winds, at speeds ranging from 2 to 7 km/h, bringing rather sunny skies, dry air and higher temperatures.

Wind Prevalence at Saclay
Wind pollination is an important feature regarding the seasonality of pollen [69]. In order to better understand the mechanisms of transport of pollen by the wind, we used the ZeFir source-receptor tool, which allows visualization of the possible spatial origins of pollen. At Saclay, prevailing winds come from West to South-West, with speeds generally ranging from 5 to 12 km/h ( Figure 11) and are associated with oceanic air masses. A second wind regime is characterized by North (5 • ) to South East (125 • ) winds, at speeds ranging from 2 to 7 km/h, bringing rather sunny skies, dry air and higher temperatures.

Origins and Point Sources of Pollen Grains
The calculation using ZeFir from this 4-year dataset led to an interesting discovery regarding the origin of total pollen grains concentrations. Indeed, the model designates a main origin from the northeast to East sector, which is not from that of the prevailing winds that are in the southwestern

Origins and Point Sources of Pollen Grains
The calculation using ZeFir from this 4-year dataset led to an interesting discovery regarding the origin of total pollen grains concentrations. Indeed, the model designates a main origin from the northeast to East sector, which is not from that of the prevailing winds that are in the southwestern sector, as illustrated by Figure 12. Thus, contrasting patterns can be clearly seen when comparing Figures 11 and 12. Highest concentrations at wind speeds higher than 15 km/h only suggest that the main pollen events may be advected over Saclay and the Ile-de-France region. Figure 11. Joint probability polar plot, equivalent to a wind rose from January 2015 to September 2018. The white circles represent a wind speed scale in kilometre per hour (km/h). The colour grid represents the estimated concentration (Nb#/m 3 ) for any wind speed and wind direction.

Origins and Point Sources of Pollen Grains
The calculation using ZeFir from this 4-year dataset led to an interesting discovery regarding the origin of total pollen grains concentrations. Indeed, the model designates a main origin from the northeast to East sector, which is not from that of the prevailing winds that are in the southwestern sector, as illustrated by Figure 12. Thus, contrasting patterns can be clearly seen when comparing Figures 11 and 12. Highest concentrations at wind speeds higher than 15 km/h only suggest that the main pollen events may be advected over Saclay and the Ile-de-France region. To better understand the origin of allergenic pollens, we tested the model for the different families to which the allergenic species belong. The results of the model indicated that for the Betulaceae family, Figure 13a, the main origin found is from an Easterly direction (60° to 120°) with a second origin from the North (20°) and were associated with strong winds between 15-20 km/h. Equally, for grass, Figure 13b, the main origin comes from the northeast sector (10° to 45°) but is associated with more moderate winds (10 to 15 km/h). For Cupressaceae and Taxaceae since these two families are not distinguishable in optical microscopy the model shows three origins (Figure 13c). The main one linked to Northerly directions (20°) and associated with very strong winds (more than 20 km/h), a second source from the South-East (90° to 110°), connected with moderate winds (less than 10 km/h) and a third in the southwestern sector, accompanied by moderate to strong winds (16 To better understand the origin of allergenic pollens, we tested the model for the different families to which the allergenic species belong. The results of the model indicated that for the Betulaceae family, Figure 13a, the main origin found is from an Easterly direction (60 • to 120 • ) with a second origin from the North (20 • ) and were associated with strong winds between 15-20 km/h. Equally, for grass, Figure 13b, the main origin comes from the northeast sector (10 • to 45 • ) but is associated with more moderate winds (10 to 15 km/h). For Cupressaceae and Taxaceae since these two families are not distinguishable in optical microscopy the model shows three origins (Figure 13c). The main one linked to Northerly directions (20 • ) and associated with very strong winds (more than 20 km/h), a second source from the South-East (90 • to 110 • ), connected with moderate winds (less than 10 km/h) and a third in the southwestern sector, accompanied by moderate to strong winds (16 km/h). With regard to the Oleaceae family, the origin is clearly from the northwest to the southwest (Figure 13d)and associated with strong winds exceeding 20 km/h. Figure 13 indicates that the main sources of highly allergenic pollen grains (Betulaceae, Poaceae, Urticaceae), which account for 51% of total pollen concentrations, come from the North and East sectors, while Cupressaceae, Taxaceae and Oleaceae, which represent 14% of this total have several origins from southeast to southwest and northwest. For non-allergenic families (trees, grasses and flowers, which account for 34% of total pollen concentrations), Figure A2 in Appendix A, shows the predominant local/regional origin for such pollen. Knowledge of the geographical origins and variability of these different species is crucial information for a better understanding of potential pollen health impacts in densely populated urban areas like Paris. It also clearly supports the need for long-term speciation within the total pollen burden. As different species/families had significantly different wind directions and speeds associated with their sampling. pollen concentrations), Figure A2 in Appendix A, shows the predominant local/regional origin for such pollen. Knowledge of the geographical origins and variability of these different species is crucial information for a better understanding of potential pollen health impacts in densely populated urban areas like Paris. It also clearly supports the need for long-term speciation within the total pollen burden. As different species/families had significantly different wind directions and speeds associated with their sampling. As a first conclusion, regarding the pollen origins and point sources, Figure 13 clearly shows that the origins and point sources for both the Betulaceae (Betula pollen is the main genus) and for Oleaceae (where Fraxinus is the main genus) are dictated by the wind speed and direction and fast changes in the wind sectors. Given that, both occur at the same time of the year this may not be surprising however, his results does show the importance of measuring the wind direction and the wind speed at the same site which pollen monitoring is performed. This is especially significant for forecasting applications as the aforementioned meteorological parameters cannot be achieved by back trajectories.

Wind Occurrence and Pollen Patterns
To further analyse, the relationship between pollen seasonality and the controlling meteorological factors influencing total pollen grains concentrations, the respective occurrences of winds directions at Saclay have been studied, based on hourly wind data from 2004 to 2018. Westerly As a first conclusion, regarding the pollen origins and point sources, Figure 13 clearly shows that the origins and point sources for both the Betulaceae (Betula pollen is the main genus) and for Oleaceae (where Fraxinus is the main genus) are dictated by the wind speed and direction and fast changes in the wind sectors. Given that, both occur at the same time of the year this may not be surprising however, his results does show the importance of measuring the wind direction and the wind speed at the same site which pollen monitoring is performed. This is especially significant for forecasting applications as the aforementioned meteorological parameters cannot be achieved by back trajectories.

Wind Occurrence and Pollen Patterns
To further analyse, the relationship between pollen seasonality and the controlling meteorological factors influencing total pollen grains concentrations, the respective occurrences of winds directions at Saclay have been studied, based on hourly wind data from 2004 to 2018. Westerly winds had a larger number of occurrences than Easterly winds from 0 • to 180 • as illustrated by Figure. For the West sector (180 • to 359 • ), results are presented in Appendix A, Figure A4, a maximum during summer (July and August) was noted with values ranging from 5149 to 7144 counts. Results for the East sector, more interestingly, highlight a far different trend, with a clear seasonality noted and exhibited by bi modal pattern visible (Figure 14), with maxima reported in March (4054 counts) and in October (3671 counts). We also observed that there is a decrease in May (3292 counts) followed by a small increase in the frequency (3389 counts) in June, comparable to levels in levels to February (3443 counts). Easterly winds show a decrease starting in July, reaching a minimum in August (2300 counts). The monthly mean wind direction shows similar variations, although the smoothing effect on the monthly means (in Appendix A, Table A2). No standard deviation can be applied on occurrence calculations.
Interestingly the Easterly wind occurrence pattern matches the seasonal pattern of total pollen grains concentrations from early spring to the end of the summer-during winter, since no significant source of pollen is known in the local atmosphere of Saclay, no significant relationship can be observed. This covariation between typical synoptic conditions and the pollen levels has been recently documented for Swedish cities by [19] where exceptionally high concentrations of Birch pollen were found to coincide with East North East winds and significant concentrations of pollutants. Also, [22] published work on the identification of the meteorological factors that influence the occurrence of airborne pollen concentrations.
Remote Sens. 2018, 10, x FOR PEER REVIEW 14 of 25 winds had a larger number of occurrences than Easterly winds from 0° to 180° as illustrated by Figure. For the West sector (180° to 359°), results are presented in Appendix A, Figure A4, a maximum during summer (July and August) was noted with values ranging from 5149 to 7144 counts. Results for the East sector, more interestingly, highlight a far different trend, with a clear seasonality noted and exhibited by bi modal pattern visible ( Figure 14), with maxima reported in March (4054 counts) and in October (3671 counts). We also observed that there is a decrease in May (3292 counts) followed by a small increase in the frequency (3389 counts) in June, comparable to levels in levels to February (3443 counts). Easterly winds show a decrease starting in July, reaching a minimum in August (2300 counts). The monthly mean wind direction shows similar variations, although the smoothing effect on the monthly means (in Appendix A, Table A2). No standard deviation can be applied on occurrence calculations. Interestingly the Easterly wind occurrence pattern matches the seasonal pattern of total pollen grains concentrations from early spring to the end of the summer-during winter, since no significant source of pollen is known in the local atmosphere of Saclay, no significant relationship can be observed. This covariation between typical synoptic conditions and the pollen levels has been recently documented for Swedish cities by [19] where exceptionally high concentrations of Birch pollen were found to coincide with East North East winds and significant concentrations of pollutants. Also, [22] published work on the identification of the meteorological factors that influence the occurrence of airborne pollen concentrations.

Pollen Point Sources, Long Range Transport and Allergen Transfers during Pollution Events
All runs confirmed that the various pollen seasons are associated with specific anticyclonic weather conditions (East and North East). Our results show that the main allergenic pollen grains reaching Saclay: (i) are transported by winds of 12 to over 20 km/h; and (ii) primarily originate from the East and North East sectors. These factors imply the following: (i) Relatively high wind speeds associated with Betulaceae, Urticaceae and Poaceae lead to two hypotheses that should be addressed. First, some taxa in Saclay may have been transported over long distances, as has been shown in other regions for specific pollen with respect to their aerodynamic properties (e.g., density, mass, granulometry). This factor remains to be characterized for our site. In particular, significant inputs from a remote area may occur before the local or regional pollen cycles

Pollen Point Sources, Long Range Transport and Allergen Transfers during Pollution Events
All runs confirmed that the various pollen seasons are associated with specific anticyclonic weather conditions (East and North East). Our results show that the main allergenic pollen grains reaching Saclay: (i) are transported by winds of 12 to over 20 km/h; and (ii) primarily originate from the East and North East sectors. These factors imply the following: (i) Relatively high wind speeds associated with Betulaceae, Urticaceae and Poaceae lead to two hypotheses that should be addressed. First, some taxa in Saclay may have been transported over long distances, as has been shown in other regions for specific pollen with respect to their aerodynamic properties (e.g., density, mass, granulometry). This factor remains to be characterized for our site. In particular, significant inputs from a remote area may occur before the local or regional pollen cycles have started. The different results obtained by season illustrated in Appendix A, Figure A3 show that the pollen can be transported over long distances, particularly during warmer temperatures and low relative humidity periods. This main result regarding long-range transport pollen is consistent with the work of [70][71][72]. Our study suggests that long-range transported pollen can affect pollen levels monitored at a receptor site under favourable synoptic conditions. Secondly, as pollen release from trees and plants is known to be both wind speed and relative humidity dependant [59], it requires consideration not only of potential remote areas of emission but also regional and local ones that complement the information provided by wind analysis.
(ii) In such atmospheric conditions, particulate matter pollution events regularly occur between December and April within the Paris region by [73,74] and reference therein. Inputs of pollen from the North East and East sectors might have to be considered together with inputs of long range transport of reactive gases and aerosols, which follow the same path in the same annual period, as previously reported, for Saclay [40]. The combination of atmospheric pollutants and allergenic pollen content is of great interest in understanding allergy in general and even asthmatic events coincident with thunderstorms during the pollen season [19,60,61]. Over their common atmospheric transport modes, pollen, gas and aerosols can possibly interact. In 2006, [17] published a study on the release mechanisms when grass pollen are submitted to high concentrations of gaseous pollutants and the presence of water. Both, [75,76] have showed that allergenic activity of pollen grains could be found in PM10 and below. Moreover, [77] showed that when birch trees are flowering and exposed to moisture followed by drying winds they can produce particulate aerosols from the nanometre to micrometre range, which contain the allergens. This last point is of importance as negligible amounts of particles above 10 micrometre (aerodynamic size) can penetrate deep into the lung all the way to the bronchi. In future studies it will be necessary to evaluate the transfer of allergenic cytoplasm components in small particles during pollution events and their transport by highly hydrated aerosols. This mechanism could possibly explain thunderstorm induced asthma episodes or the asthmatic bronchitis encountered in megacities or highly populated areas downwind atmospheric mixtures of both biological and anthropogenic pollution.

Conclusions
The daily observations made at Saclay over four years demonstrate that airborne pollen grains concentration exhibits a biannual cycle of total airborne pollen grains mainly driven by some species of the Betulaceae family. The general increasing trend in the annual sum of concentrations needs to be further examined given the limited observations expressed here and consequently compared with more established pollen monitoring sites such as those that have been located in Paris and operational for over 25 years (M. Thibaudon personal communication). During our observation period, strong variations in the interannual concentrations were observed. This interannuality seems to be closely linked to the total amount of rain water during the 6 months preceding the start of the pollen season. We noticed that the pollen season starts in early February under anticyclonic conditions (North and East wind regime) and ends in early September when the wind sectors were mostly coming from the Ocean (West regime). The first phase of the pollen season is due to tree pollen dispersions and the second phase to herbaceous species. The mean concentration during the nine month of pollen season was seen to be on average 200 Pollen/m 3 of air and mainly dominated by allergenic species which represents 65% of the total pollen grains. The daily concentrations measured fluctuated in intensity from day-to-day and are probably in relation to weather instability. The cycles of hot days followed by rainfall possibly produced, plant stresses, fast growing which can produce huge amount of submicronic biological allergens. This effect is more visible for regional species and particularly grass species. The burst of pollen concentrations can increase by up to 300% compared with the mean concentration of the whole pollen season. To validate our observations and in comparison with the studies available in the literature, the Sustainable Wind Incidence Method was used to couple atmospheric concentrations and wind data. The model shows that the wind prevalence was from the South-West whereas the pollen was associated with the North to East sector with strong winds. The use of ZeFir tool validates our observation regarding the type and the nature of the air masses implicated in the atmospheric pollen loading. We have determined that the origin of pollen grains impacting the region of Saclay were wind direction dependant: higher levels of pollen are observed when the winds are coming from the East and North East sectors, suggesting long range transport for Betulaceae and regional sources for Poaceae. As the local wind is very much affected by mesoscale weather phenomena, inverse dispersion modelling supported by a mesoscale Numerical Weather Prediction (NWP), and/or a footprint calculation for the observation site looks would be interesting for comparison studies. We also identified potential point sources that need to be confirmed by on-site inventories. Additionally, these pollen episodes are often associated with pollution events, which could increase the allergenic character of smaller particles. This last aspect must be further investigated by intensive campaigns to identify in which class of particle size the allergenic material is transferred. In summary, this study has shown that by combining our statistical analysis method with meteorological analysis as well as aerobiological data and pollution measurements we should be able to predict a realistic risk index on a daily basis to reduce the exposure of susceptible individuals during the pollen season for the region of Paris. Funding: This research was funded by CEA-NRBC-E grant number D34 and the APC was funded by PREVIPOL ANR project.

Acknowledgments:
The authors want to thank all the team from the Réseau National de Surveillance Aérobiologique (RNSA) and particularly Nadine Dupuy, Gilles Oliver, Charlotte Sindt and Jennifer Charbonnier. We are grateful to Francois Truong from LSCE for his help during the experiments.

Appendix A
for comparison studies. We also identified potential point sources that need to be confirmed by onsite inventories. Additionally, these pollen episodes are often associated with pollution events, which could increase the allergenic character of smaller particles. This last aspect must be further investigated by intensive campaigns to identify in which class of particle size the allergenic material is transferred. In summary, this study has shown that by combining our statistical analysis method with meteorological analysis as well as aerobiological data and pollution measurements we should be able to predict a realistic risk index on a daily basis to reduce the exposure of susceptible individuals during the pollen season for the region of Paris.
Author Contributions: R.S.E. has pioneered bioaerosol research at CEA/LSCE, he has conceived and designed the experiments and has lead the writing of this paper. D.B. has participated to the conception and design of the experiments, she has analysed the data and contributed to the writing of this article. B.G. contributed to the interpretation of the dataset in correlation with pollution and health impact and writing of the paper. J.-E.P. has developed the ZeFir tool and contributed to the interpretation of the outputs of wind analysis. J.P.B., M.T., V.G. have contributed to the interpretation of the data and writing of the paper. J.S. and D.O. contribute to supervise the science on pollen research and writing of the paper.
Funding: This research was funded by CEA-NRBC-E grant number D34 and x] and the APC was funded by PREVIPOL ANR project.
Acknowledgments: The authors want to thank all the team from the Réseau National de Surveillance Aérobiologique (RNSA) and particularly Nadine Dupuy, Gilles Oliver, Charlotte Sindt and Jennifer Charbonnier. We are grateful to Francois Truong from LSCE for his help during the experiments.