Laboratory evaluation of the scattering matrix of ragweed, ash, birch and pine pollens towards pollen classification

Pollens are nowadays recognized as one of the main atmospheric particles affecting public human health as well as the Earth’s climate. In this context, an important issue concerns our ability to detect and differentiate among the existing pollen taxa. In this paper, the potential differences that may exist in light scattering by four of the most common pollen taxa, namely 10 ragweed, birch, pine and ash, are analysed in the framework of the scattering matrix formalism at two wavelengths simultaneously (532 and 1064 nm). Interestingly, our laboratory experimental error bars are precise enough to show that these four pollens, when embedded in ambient air, exhibit different spectral and polarimetric light scattering characteristics, in the form of ten scattering matrix elements (five per wavelength), which allow identifying each separately. To end with, a simpler light scattering criterion is proposed for classifying among the four considered pollens by performing a principal component 15 (PC) analysis, that still accounts for more than 99 % of the observed variance. We thus believe this work may open new insights for future atmospheric pollen detection.

precipitating atmospheric pollen grains on an adhesive substrate prior to the microscopic observation of the deposited pollen grains. Though rather accurate, this historical methodology suffers from a too long time duration, with characteristic times ranging from a few hours to a week, much higher than the characteristic time for the impact of pollens on human health.
Another limitation of the historical methodology is that the pollens grains are counted one by one and identified by a scientist, which is time-consuming and may be operator dependent (Comtois et al., 1999). In this context, deep learning recently 35 improved the identification of pollen grains through microscopy (Daood et al., 2016) by automating the procedure to differentiate among existing pollens. Besides, new methodologies based on interferometry, light-scattering or laser-induced fluorescence have been developed. In the latter, the fluorescence spectrum and its life-time have been used from the UV to the NIR spectral range to identify pollens (Pan et al., 2011;Kiselev et al., 2013). Likewise, image recognition on pollen grains holographic images is also used (Giri et al., 2019;Sauvageat et al., 2020;Kemppinen et al., 2020) as an identification 40 methodology. Finally, light scattering by pollens has been studied in laboratory in aqueous solutions by (Bickel and Stafford, 1980) or when pollens are deposited on a substrate or an holder by (Surbek et al., 2011;Iwai, 2013;Raman et al., 2013;Nouri et al., 2018). Concerning airborne pollens, (Matsuda and Kawashima, 2018;Holler et al., 2016) studied forward and side scattering while polarization-dependent features of light scattering were first studied by our group by evaluating the scattering matrix of ragweed pollen in laboratory ambient air at near backscattering angles (Cholleton et al., 2020). More recently, 45 (Gómez Martín et al., 2021) evaluated the scattering matrix of cypress pollen far from the backscattering angle. Also, the depolarization ratio of several pollens has been studied in field by (Cao, 2010;Bohlmann et al., 2018;Sicard et al., 2021) using lidar measurements. Nevertheless, light scattering by pollens embedded in ambient air still remains a complex topic, mostly due to the complexity in size and shape of the pollen taxa. In particular, it is not a priori granted that pollen differing in size and shape exhibit different light scattering characteristics. Indeed, as underscored in (Cholleton et al., 2020), no analytical light 50 scattering numerical simulation exists for such complex objects so that a remaining issue is to predict how these grains scatter light. Therefore, a controlled-laboratory experiment is coveted to address light scattering by pollens with a precision allowing to identify each pollen.
In this paper, to complement the above airborne pollen studies, the scattering matrix of several airborne pollen taxa is for the 55 first time presented. The four considered pollen taxa are ragweed, ash, birch and pine, chosen for their important impact on human health. The laboratory experiment is carried out at two wavelengths simultaneously ( = 532 nm and = 1064 nm) and provides five scattering matrix elements per pollen taxa and per wavelength. Interestingly, our experimental error bars are precise enough to reveal the existing differences in light scattering by these four pollens, thus allowing to identify each separately. Ten scattering matrix elements are provided per pollen (five per wavelength). Hence, compared with our 60 previous contribution (Cholleton et al., 2020), three more pollens are here studied (ash, birch and pine) which allows discussing on airborne pollen classification using polarization-resolved light scattering. As an output, a principal component analysis is applied to provide a simpler light-scattering criterion for classifying among these four pollens. The classification can then be visualized on a simple 2D-plane representation. The novelty of the paper is hence two-fold. Firstly, precise light scattering https://doi.org/10.5194/amt-2021-272 Preprint. Discussion started: 21 September 2021 c Author(s) 2021. CC BY 4.0 License. characteristics of ragweed, ash, birch and pine are revealed in the form of ten scattering matrix elements (five per pollen per 65 wavelength). Secondly, an analysis is proposed towards a pollen classification. The paper is organised as follows. Section 2 presents the studied pollen samples, by characterizing their size and shape, based on scanning electron microscopic images taken in our laboratory. Section 3 details our laboratory methodology to precisely evaluate the scattering matrix of these pollens by detailing our error uncertainty. Section 4 presents the evaluation of the scattering matrix of ragweed, ash, birch and pine, and proposes the principal component analysis as an outlook as a visual 2D-representation to help classifying among the four 70 considered pollens. The paper ends with a conclusion and proposes outlooks. Figure 1 shows scanning electron microscopic images of the four studied pollen species, namely ragweed, ash, birch and pine.

Pollen samples
These microscopic images allow characterizing the size and the shape of these pollens: each pollen exhibits a very characteristic shape with comprising grains presenting a quasi-monodisperse size distribution. The four considered species 75 clearly differ in size and shape, from nearly spherical with a 20 µm volume equivalent diameter for ragweed to highly irregular shapes with a diameter larger than 60 µm for pine. Small-scale features exist at each pollen surface, such as granularity or apertures, which show the great complexity of these microscopic objects. These specific sizes and shapes are used for pollen recognition as extensively detailed in palynological databases (PalDat, 2021), as recalled in the introduction.

Ragweed 80
Ragweed or Ambrosia artemisiifolia pollen is amongst the most allergenic pollen in Europe and Northern America (Smith et al., 2013) with 15.8 million persons sensitized in Europe (Schaffner et al., 2020). Ragweed pollen induces particular allergenic reactions twice more often as other pollens (Thibaudon et al., 2014) with an annual economic cost of 7 billion euros in Europe (Schaffner et al., 2020). The ragweed season usually lasts from August to October and this duration is getting longer with expected increasing mean temperatures. Hence, ragweed pollen has been extensively studied in the literature (Smith et al., 85 2013;Hamaoui-Laguel et al., 2015;Zink et al., 2017) and extensive efforts are made for the observations and forecasts of its concentration (Prank et al., 2013). Ragweed pollen grains exhibit a characteristic shape, that bears some resemblance with coronavirus virions (Petrov, 2020), are nearly but non spherical, with a mean volume-equivalent diameter of 21 micrometres.
Its membrane exhibit three apertures (2 micrometres long), regular spikes (echinus) of about 1 micrometre length and a perforate structure, i.e. covered with holes of about 100 nanometres diameter. 90

Ash
Ash (fraxinus americana) pollen is a relevant source of allergenic reactions in Europe (Imhof et al., 2014) which has been underestimated for a long time as its bloom season overlap with that of birch. Moreover, it presents a high cross-reactivity from allergens from other plant species (Niederberger et al., 2002). Its blooming season lasts from March to May. Ash pollen https://doi.org/10.5194/amt-2021-272 Preprint. Discussion started: 21 September 2021 c Author(s) 2021. CC BY 4.0 License.
grains have an irregular shape, with a reticulate exine, i.e. ornamental elements arranged as a network of ridges with gaps of 95 the micron scale, and 3 colpi, i.e. thin elongated apertures regularly distributed along its equatorial region. Its volume equivalent diameter is about 31 micrometres.

Birch
Alongside with ragweed, birch (betula pendula) pollen is one of the most impacting in Europe (Biedermann et al., 2019). It also exhibits a cross-reactivity with other allergens. Depending on its location, the birch pollen season starts from March and 100 lasts until late June. Birch pollen grains are irregularly shaped, with sunken interpectoral areas when dried. Their mean volume equivalent diameter is about 28 to 29 micrometres. They present 3 protruding pores, i.e. three 2 µm large holes on its polar view. Birch pollen has been one off the first to be integrated in global pollen forecasts (Siljamo et al., 2013), using dispersion models.

Pine 105
In contrary to the previous taxa, pine (pinus strobus) pollen rarely cause allergies. It however remains an interesting study case, as it may impact the Earth's climate locally: (Spänkuch et al., 2000) reported that a high pine pollen concentration increased the down welling infrared flux up to 8 times the monthly means. Moreover, this pollen has been studied by optical diffraction tomography (Kim et al., 2018) to highlight its structure and morphological features, and evaluate its refractive index. At last, (Sicard et al., 2021) recently studied its vertical distribution in the atmosphere using lidar-derived profiles. As 110 shown Fig. 1(d), pine pollen is the largest of the studied samples with a mean diameter of 69 micrometers. Its shape largely differs from that of other pollens, as two sacci, i.e. two large air-filled bladders are attached to the center part (Schwendemann et al., 2007). The refractive index of pine at 532 nm has been estimated between 1.5 and 1.54 (Kim et al., 2018), while its imaginary part remains undocumented.

Pollen suspension 120
The studied pollen samples were supplied by Stallergenes Greer. Dry pollen powder has been suspended in the light scattering volume using a solid aerosol generator supplied with dried compresses air (RH < 10 %). The size distribution and concentration https://doi.org/10.5194/amt-2021-272 Preprint. Discussion started: 21 September 2021 c Author(s) 2021. CC BY 4.0 License. of the generated airborne ragweed pollen grains was measured using an Aerodynamic Particle Sizer (APS). As a consistency check, the retrieved ragweed pollen size distribution was found in agreement with the size specified by the supplier and also with above SEM observations issued from the state-of-the-art literature (PalDat, 2021). 125

Light scattering by pollens
We consider elastic light scattering by an ensemble of pollen grains suspended in ambient air with a polarized electromagnetic incident radiation of wavelength . For the pollen samples are nonspherical (see Fig. 1), the polarization state of the scattered radiation may differ from that of the incident radiation. To describe the polarization states of the incident and scattered waves, 130 the Stokes vector formalism (Mishchenko et al., 2002) is applied. The polarization state of the incident and scattered radiation is then described by a Stokes vector = [ , , , ] where , , , respectively relate to the total intensity, the linear degree of polarization in the scattering plane, and at 45° from it, and the degree of circular polarization. In the far-field single scattering approximation and assuming random orientation, we learn from light scattering textbooks (Mishchenko et al., 2002) that the incident and scattered Stokes vectors and relate by the so-called scattering matrix : 135 (1) where the scattering matrix elements ( , = 1 − 4) have been normalized with respect to the scattering phase function so that at most equals unity. 22 and 33 characterize how a linear polarization state is preserved during light scattering by the considered pollen while 44 relates if circular polarization is preserved during light scattering by the pollen grains. Off-diagonal 140 element 12 quantifies the amount of non-polarized light after scattering by pollens, while the other off-diagonal element 34 traduces the ability of pollens to convert a linearly polarized radiation to a circular polarization state throughout light scattering.
The scattering matrix elements describe how the polarization state of the incident radiation is modified during light scattering by the studied pollens grains. These matrix elements hence depend on the shape of the pollen grains which we here investigate.

Laboratory experimental set-up for pollen light scattering 145
The specific size and shape of each pollen taxon is addressed by considering the Fig. 2 laboratory experiment, which has proven efficiency for ragweed pollen detection (Cholleton et al., 2020). We here recall its main characteristics for the sake of clarity before discussing on its applicability to the detection of other pollens in the next paragraph. This laboratory polarimeter operates at the near backscattering angle = (177.5 ± 0.2)° and at two wavelengths = 532 nm and = 1064 nm simultaneously. A reflecting polarizing beam-splitter cube (PBC) is used so that it is the s-component of the scattered radiation 150 that is measured by the photodetector. This experimental set-up has been validated on spherical particles (Cholleton et al., 2020) for which the scattering matrix can be analytically computed by applying the Mie theory, by measuring the corresponding size distribution with particle optical sizers.  (Cholleton et al., 2020). The 177.5° scattering angle has been exaggerated to ease the reading. 0 ( ) and ( ) respectively stand for the incident and scattered wavevectors at wavelength . Pollens grains are embedded in laboratory ambient air as described Section 2.5. is the angle between the fast axis of the quarter-wave plate (QWP) and the scattering plane.
Following Fig. 2, the detected pollen scattered intensity can be obtained by considering the successive Mueller matrices 160 encountered by the incident radiation. After a few calculations (Cholleton et al., 2020), , we get at wavelength : where subscript stands for incident polarization state while 0 is a proportionality constant that accounts for the incident laser 165 power and the electro-optics efficiency. The pollen scattering matrix elements are retrieved from Eq. (2) by adjusting the detected scattered intensity as a function of the angle to retrieve the , , and coefficients, which only depend on the scattering matrix elements. To gain in accuracy, the polarization state of the scattered radiation is analysed for several incident polarization states, obtained by rotating an analyser (QWP), whose position is labelled by the angle . To retrieve all the pollen matrix elements, the polarization state of the scattered radiation is analysed for three successive incident polarization 170 states, namely ( ), (45 +) and ( ). To give an example, a ( )-polarized incident radiation allows retrieving 12 , 34 and 44 , as established in (Cholleton et al., 2020). Adding a (45 +)-polarized incident radiation allows evaluating 33 in addition.
Finally, 22 is retrieved from a ( )-polarized incident radiation. In summary, the following set of equations, established in (Cholleton et al., 2020), is applied to retrieve the pollen scattering matrix elements from adjusted coefficients , , and

Applicability of this set-up to the detection of other pollens
To study the sensitivity of our methodology to other pollens, three different Taxa are considered, characterized by the following set of scattering matrix elements: 180  a spherical pollen called 1, with normalized scattering matrix elements [ 22 = 1, 33 = 1, 44 = −1, 12 = 0, 34 = 0]. 1 is to our knowledge an hypothetical pollen since pollens are nonspherical (Hesse, 2009). It is here however considered for several pollens exhibits an overall spherical shape. Also, the spherical shape is the ideal shape to be detected and can hence be considered as a reference case for polarimetric studies. To fix ideas, we plotted in Fig. 3 the variation of as a function of the -angle of the analyser at a given wavelength , for 1, 2, 3 at the three successive incident polarization states ( ), (45 +), ( ). Interestingly, each pollen Taxon is associated with a specific light scattering curve, showing the sensitivity of our methodology. From 1 to 3, the ( )polarization curve ( ) exhibits varying minima to be related to variations in the 22 -scattering element. The same 195 conclusions can be drawn with the ( ) curve minima, which are determined by the 44 scattering matrix element. If the 45+ ( ) curve is -periodic whatever the considered taxon, its extrema, which depend on 12 , 34 and 33 , are also Taxondependent. As a result, the proposed methodology can potentially be used to differentiate pollens exhibiting different scattering matrix elements. For that however, the scattering matrix elements should be retrieved with precision, as discussed below.  = 1) so that any change in the scattered light intensity be due to polarimetric considerations. 205

Scattering matrix elements retrieval accuracy
Taxon 2 can then be distinguished from Taxon 3 using polarimetric light scattering if their corresponding Fig. 3 curves can be distinguished within our experimental error bars. To give an example of our ability to distinguish two pollen taxa, to a 1 %variation in the 44 scattering matrix element corresponds a 1 %-variation in the detected scattered light intensity. For Taxon 2 and 3 exhibit 44 -values differing by almost 50 %, the required precision can be reached. Moreover, in our experiment, the 210 evaluation of the , , and coefficients, and hence of the scattering matrix elements (see Eq. 2), relies on a large number of measurements data points (180 points per fitted curve). Special care has indeed been taken to specify our experimental error bars on the retrieved scattering matrix elements by considering both statistical and systematic errors. Polarization cross-talks in the analyser have been minimized by aligning a second polarizing beam-splitter cube in the detector. Wavelength crosstalks are also fully negligible, which is key for revealing the spectral dependence of the scattering matrix elements, for laser-215 line selective interference filters are used in the light detector. Also, the incident polarization state may slightly differ from the ( ), (45 +), ( ) polarization states. To quantify this uncertainty, we considered the following incident Stokes vector 0 = [1,1,2 , 2 ], where 2 and 2 represent deviations from the (p) incident polarization sate (a similar discussion is drawn for (45 +) and ( ) polarization states). At first order in 2 and 2 , the uncertainties on the scattering matrix elements [ 33 , 44 , 12 , 34 ] are [2 34 , 2 34 , 2 22 , 2 33 ] while the matrix element 22 remains unperturbed. As elements at most equal 220 unity, the uncertainties can be majored by their upper limit 2 or 2 , depending on the considered scattering matrix element. Besides, 2 is evaluated by recording the scattering curve corresponding to incident polarization states ( ) and ( ) since 2 = RC /( + RC )-LC /( LC + LC ). Finally, we used the outputs of a paper by Mishchenko (Mishchenko et al., 2007) 225 to check that the single scattering approximation was rather safe in our experiment where 1 〈 〉 ≫ 30 ( 1 is the wave vector in the surrounding medium and 〈 〉 is the average inter-pollen distance), while particle volume concentration remained lower that 1 %, hence considering a tenuous media. Statistical errors due to potential fluctuations in the grain number concentration in the scattering volume are also accounted for by normalizing the detected intensity by that of a photodetector placed at 170°scattering angle, which depends on the pollen grains number concentration. As a conclusion, from the measurements of the 230 scattered intensity at wavelength and , the above detailed method allows to precisely retrieve the scattering matrix with error bars that are sufficiently low to distinguish pollen Taxa through polarimetric light scattering, as detailed in the Section 4 below, dedicated to our laboratory results. an average of four measurements each composed of a sequence of 100 laser shots. In Fig. 4, the error bar affecting each data point is then the standard deviation of these four sequences. Moreover, the reproducibility of the observed minima during a full rotation of the analyser indicates that the shape of each considered pollen remained constant during the acquisition. Let us first discuss on the scattered light intensity by ragweed pollen. The ( )-polarization curve exhibits non-zero minima, showing that ragweed pollen grains are non-spherical, in agreement with the SEM observation. In the (45 +)-polarization curve, the 245 two successive local minima are not equal at wavelength , which proves that 34 is not null, as opposed to wavelength where, within our error bars, no significant differences occur in the successive local minima for ragweed for example.

Detected scattered light intensity by ragweed, ash, birch and pine 235
Moreover, the light scattering curves differ for the three other pollens (ash, birch, pine). For instance, pine pollen, whose successive local minima are identical in the (45 +)-polarization curve, exhibits a near zero 34 -value at both wavelengths as detailed below in Section 4.2. Likewise, the ( ) polarization curve of the ash pollen is nearly constant, which relates to a 44 -250 value of the matrix element close to zero.

Retrieved scattering matrix elements for ragweed, ash, birch and pine
The Fig. 4 experimental data points are adjusted with Eq. (2) to retrieve the coefficients to (see Fig. 4 fitted curves). The agreement between our measurement data points and the Eq.
(2) adjustment is notable. Therefore, we can apply the scattering matrix formalism to evaluate at each wavelength by applying Eqs.
(3), as presented in Fig. 5 with detailed numerical values 255 given in Table 1 (wavelength ) and in Table 2 (wavelength ). We first focus on the comparison of retrieved between the four pollens before discussing on the spectroscopic variations of these matrix elements. For our four pollens are nonspherical, all the retrieved 22 -values differ from unity. More interestingly, within our experimental error bars, each pollen exhibits a different 22 matrix element. Likewise, within our error bars, other diagonal matrix elements 33 and 44 , which are also not null and differ from one to another, specifically describe light scattering by each pollen. The 44 260 element is clearly higher for ragweed and birch, compared with ash and pine. Hence, among our set of four pollens, a set of elements can be unequivocally linked to a single pollen taxon. Off-diagonal elements 12 and 34 are nearly equal to zero within our error bars. Regarding now wavelength , the same conclusions can be drawn with 22 , 33 and 44 , i.e. this set of matrix elements unequivocally relates to a single pollen taxon. Interestingly, while 12 remains near-zero, 34 is not null for ragweed, ash and birch pollen. As underlined by (Bickel and Stafford, 1980), this matrix element may be relevant 265 for the study of pollen particles. The diagonal elements are generally larger at wavelength , which indirectly means easier to evaluate with a higher precision. The considered pollens exhibit a rather pronounced spectral variation except birch for which the matrix elements remain constant within error bars. The spectral dependence however remains complex to interpret, as the scattering matrix elements of pollens depend on the complex refractive index of pollen grains and the literature on pollen refractive indices is rather sparse, with, to our knowledge, no evaluation of their imaginary part and no spectral measurement 270 at wavelengths and .
https://doi.org/10.5194/amt-2021-272 Preprint. Discussion started: 21 September 2021 c Author(s) 2021. CC BY 4.0 License. Figure 4: Detected scattered light intensity by ragweed, ash, birch and pine pollens as a function of the of the -angle of the analyser for the three successive incident polarization states (p), (45+) and (RC) at wavelength (in green) and wavelength (in red). The detected light intensity is normalized so that × 0 = 1, as detailed Section 3.3. A break has been inserted to ease the reading. Each measurement data point is obtained by repeating 400 lasers shots and the uncertainty is calculated by following the methodology described in Section 3.4. https://doi.org/10.5194/amt-2021-272 Preprint. Discussion started: 21 September 2021 c Author(s) 2021. CC BY 4.0 License. Figure 5: Retrieved scattering matrix elements of ragweed, ash, birch and pine at 177.5° scattering angle at wavelengths and , with corresponding uncertainties evaluated by applying the methodology presented in Section 3. For each pollen taxon, the evaluation of the scattering matrix has been repeated with different samples: each evaluated scattering matrix lies within presented error bars.

Towards pollen identification using principal component analysis 290
As above explained, each of the four considered pollens exhibits its own set of ten scattering elements (five per wavelength), which allow identifying them among the considered set of four pollens. Pine can however be unequivocally identified using 34 only as being the only taxon for which 34 is null. Also, 22 may allow distinguishing pine from other taxa, however not ragweed from birch. Likewise, within our experimental uncertainties, 12 and 34 do not allow differentiating among the four considered pollens. Consequently, for pollen identification through polarized light scattering, the complete set of 295 scattering matrix elements must be generally taken into account, giving rise to the forty elements given in Tables 1 and 2. It however remains challenging and at least non intuitive to identify a pollen taxon based on these set of intricate forty scattering matrix elements. In the literature, decision trees have been applied for classification but overlaps exist due to experimental uncertainties, which limit the application of this classification methodology. Rather, we here propose to reduce the dimensionality of our system to a 2D-representation plane while taking into account experimental uncertainties. Therefore, the 300 goal of this paragraph is to provide a simpler light scattering criterion for classifying among the four considered pollens, based on a 2D-representation plane. To reduce the dimension of our four pollen dataset, a principal component analysis ( ) has been performed. Indeed, is a statistical method for dimension reduction and feature extraction (Jolliffe, 2005). There, each pollen is described by its two principal components, hereafter called 1 and 2 , which form an orthogonal basis of two uncorrelated components. Each measured scattering pattern can then be defined in terms of a combination of the s. 305 Interestingly, besides the advantage of a dimensionality reduction, the maximizes the variance in the ( 1 , 2 ) representation of the dataset. For that, we considered the explained variance, which measures the proportion to which a mathematical model accounts for the dispersion of a given data set. In our case, a threshold of 99 % explained variance is reached by taking into account only two principal components. As a result, Fig. 6 presents the projection of each pollen taxon in the new ( 1 , 2 )-orthogonal basis. The newly obtained coordinates 1 and 2 of each pollen account for our ten 310 retrieved scattering matrix elements (five per wavelength per pollen) and the corresponding error bars in Fig. 6 have been evaluated by propagating uncertainties on each . The explained variance of 1 is 78.8 % while that of 2 reaches 20.3 % so that the dimension reduction still accounts for 99.1 % of our experimental variance. Based on Fig. 6, a classification criterion among our four pollen dataset is to assign to each pollen taxon the area delimited by its error bar. To each measured scattering pattern (i.e. each set of ten evaluated matrix elements) then corresponds a single point in the ( 1 , 2 )-basis, which allows 315 identifying the pollen if the evaluated point lies in one of the Fig. 6 coloured areas, which provides a simple criterion for classifying among the four considered pollens, based on a 2D-representation plane. This methodology can interestingly be extended to other pollen taxa provided that precise laboratory measurements of its matrix elements are first accurately carried out, as performed in Section 3 for ragweed, ash, birch and pine. Still as is, our methodology allows identifying non equivocal regions in the basis, due to the sensitivity and accuracy of our laboratory work. Hence, our evaluation of pollens scattering matrix at two wavelengths, when associated with a principal component analysis, is a step towards identifying nonequivocal polarized light scattering characteristics of pollens, that work for the set of following taxa: ragweed, ash, birch and pine. Figure 6: Projection of the ten scattering matrix elements (five per wavelength) for each pollen on the new ( 1 , 2 )-orthogonal basis 325 using principal component analysis. Error bars have been evaluated by propagating the uncertainties on each scattering matrix element. Coloured regions correspond to the range of ( 1 , 2 ) where the pollen taxon can be unequivocally identified among our set of four pollens.

Conclusion and outlooks
In a context where pollens are highly impacting atmospheric bioaerosols with a high socio-economical cost, new 330 methodologies are required for monitoring and differentiating pollens. In this paper, the ability of polarized light scattering to differentiate a set of four pollens (ragweed, ash, birch and pine) is experimentally studied in laboratory at two wavelengths.
Each pollen taxon having a very specific size and shape, light scattering by these complex-shaped particles is challenging and cannot be easily numerically simulated. In this context, a controlled laboratory experiment, relying on the robust scattering matrix formalism, has been carried out, with special emphasis on the required precision to identify different spectral and 335 polarimetric characteristics of the four considered pollens. This laboratory experiment provides precise retrievals of the scattering matrix elements of each pollen taxon in the form of ten matrix elements (five per wavelength). Hence, a set of ten retrieved matrix elements has been drawn per pollen. Interestingly, within our experimental uncertainties, the four considered pollens, when embedded in ambient air, exhibit clearly different light scattering characteristics, which allowed to differentiate between each pollen taxon from the set of forty matrix elements (five per wavelength per pollen). Finally, to reduce the 340 dimensionality of our system to a 2D-representation plane while taking into account our experimental uncertainties, a principal https://doi.org/10.5194/amt-2021-272 Preprint. Discussion started: 21 September 2021 c Author(s) 2021. CC BY 4.0 License. component analysis is here proposed. Though the dimensionality is reduced, this projection of each pollen taxon in the new ( 1 , 2 )-orthogonal basis interestingly accounts for more than 99 % of our experimental variance, which allow to provide a simple criterion for classifying among the four considered pollens, based on a 2D-representation plane. There are multiple outlooks to this work. To remain focused on laboratory work, which is currently in short supply, the evaluation of the scattering 345 matrix for a larger set of pollens taxon should be considered. If the four chosen pollen taxa are among the most impacting pollens, this work could be extended to other pollens by following our methodology. We may expect our methodology to be applicable to other sets of pollen taxa, provided that the same accuracy is experimentally achieved. The laboratory experiment should then be carried out. As well, extension of this work to other wavelengths, to pollen mixtures, or to other scattering angles is also an interesting outlook of this work.