PM samples used for the present study have been collected during three distinct projects, which are described in the following. The locations of the corresponding measurement sites are presented in Fig. 1a and b, while site classifications, sampling periods, and numbers of available samples are summarised in Tables 1 and 2.

The first measurement campaign (QAMECS, Air Quality in the Grenoble Area: Assessment of Environment, Behaviour and Health) focussed on the PM10 loading and composition at various sites within the Grenoble metropole (France), as part of the MobilAir air quality programme (Borlaza et al., 2021a, b). In these campaigns, three sites were monitored over two 1-year periods (2017–2018 and 2020–2021). As the largest metropolis in the Alps, Grenoble is home to around 450 000 inhabitants. The city itself is situated within an Alpine valley: the centre is at relatively low altitude (between 200 and 600 m above sea level) and is surrounded by multiple separate mountain ranges, namely Chartreuse (to the north), Belledonne (east), and Vercors (south and west). These ranges heavily inhibit horizontal air movement, leading to unique meteorological conditions and favouring the formation of temperature inversions, trapping pollutants within the valley, especially during winter. During this study, a PM10 sampling campaign was undertaken in the Grenoble metropole at three sites, each representing a different urban site typology: Les Frênes (LF, urban background), Vif (peri-urban), and Caserne de Bonne (CB, urban centre). All three sites are within 15 km of one another (Fig. 1b).

Secondly, PM10 and PM2.5 samples could be obtained from a monitoring campaign at the Observatoire Pérenne de l'Environnement (ANDRA-OPE), in northern France (http://ope.andra.fr/index.php?, last access: 15 March 2021). Samples have been collected continuously for about a decade at this site (Golly et al., 2019; Borlaza et al., 2021c), but cellulose measurements were conducted and presented in this work for the years 2016, 2017, and 2020 only. PM10 and PM2.5 samples were taken on alternate days. The ANDRA-OPE site is situated 230 km east of Paris, on a rise in between lows of the Paris basin and the mountains in the department of Vosges (OPE-ANDRA Atmospheric Station, 2021). It is subject to persistent westerly prevailing winds and is surrounded by significant agricultural activities but is notably distant from towns (> 25 km) and small villages (> 4 km).

Last but not least, simultaneous PM10 and PM2.5 filter samples were taken across five sites in Switzerland, as part of an Empa (Swiss Federal Laboratories for Materials Science and Technology) monitoring campaign (Grange et al., 2021). These sites varied in characteristics and were sampled for 1 year (from June 2018 to May 2019). Two rural sites, Magadino and Payerne, are included within the study. The former is located south of the Alps, whilst the latter is found on the northern plateau roughly 50 km from the nearest city of Bern. Filter samples were also taken from urban sites within three of the most populous cities in Switzerland: Basel, Bern, and Zurich. Zurich has a similar topography to the Grenoble metropole, whilst the traffic-impacted site in Bern resides within a “street canyon”; thus both sites may also experience inhibited air movement. The monitoring site in Basel is within a suburban area, located in an open and park-like environment. It is not expected to be impacted by accumulation effects.

At each of the nine sites used for the present study, daily (24 h) PM sample collection periods were conducted according to Tables 1 and 2 (starting at 00:00 or 09:00 local time) with an average 3 d sampling interval within the Grenoble metropole, 4 d interval for the Swiss sites, and 6 d interval for the ANDRA-OPE monitoring site. Additional samples for PM10 were collected daily during 9 weeks in summer 2017 in OPE and Grenoble and measured for cellulose but are not considered in this study (Samaké et al., 2020). The PM collection was performed using high-volume samplers (Digitel DA80, 30 m3 h-1) onto 150 mm diameter pure quartz fibre filters (Pall Tissuquartz 2500 QAT-UP, diameter 150 mm). Excluding the Swiss sites, filters were pre-fired at 500 ∘C for 12 h before use to avoid organic contamination, and all were handled under strict quality control procedures. After collection, samples were wrapped in aluminium foil or sterile parchment, sealed in Ziploc plastic bags, and stored at < 4 ∘C until use for chemical analyses. Blank filters were collected to determine the detection limit (DL) and to check for the absence of contamination during sample transport, setup, and recovery.

All PM10 filters from the nine monitoring locations were analysed for cellulose, while PM2.5 filter samples have been analysed at three of the monitoring locations available. The PM10 and PM2.5 filter samples were subjected to several other chemical analyses in order to quantify their major chemical components and tracers used in this study.

This cellulose quantification method was subjected to a repeatability test, in order to quantify the uncertainties with respect to glucose content within the filter punches. Briefly, a high-volume sampler (Digitel DA80, 30 m3 h-1) was used to collect PM10 onto a pre-fired quartz fibre filter (Pall Tissuquartz 2500 QAT-UP, diameter 150 mm) on the roof of the laboratory and sampled a total of 615.1 m3 of air on 15 March 2021. Ten filter punches of 21 mm were then taken and subjected to the same cellulose-to-glucose enzymatic procedure as for normal samples. It is important to state that we assume constant concentrations of both native glucose and cellulose within the filter, as well as the same enzymatic cellulose-to-glucose conversion efficiency for all 10 filter punches. Each filter punch was then analysed three times using the same HPLC-PAD method to monitor repeatability in terms of both cellulose hydrolysis and PAD glucose concentration measurements. After hydrolysis, the total glucose content of the 10 filters was found. The variability (relative standard deviation, RSD) was small, ranging from 0.7 %–5.7 % for the three repeats of the same filter sample. The RSD of the glucose content within the 10 filter punches was calculated to be 9.9 %. For a 95 % confidence in the uncertainty estimate, the uncertainty in the measurement was therefore found to be 20 % at a maximum.

In order to check for potential contamination of filters during transport, sampling, and storage, blank filters were taken across the nine sites. Within the Grenoble metropole, blank filters were taken at Les Frênes and then applied to Caserne de Bonne and Vif (labelled QAMECS in Table 3). Further, blanks filters were taken at ANDRA-OPE on both PM10 and PM2.5 sampling days. With regards to the Swiss sites (Empa), blanks were taken from each sampling site, and an average glucose concentration was taken from across the five locations.

Glucose concentrations calculated in the blanks were then subtracted from measured glucose concentrations within each sample. After, any sample that then yielded a negative concentration of glucose was deemed to be lower than the quantification limit (< QL), representing 5.2 % of all samples. Table 3 summarises the concentrations of cellulose on the blank filters, which has been converted from the blank glucose concentration and the average sampling volume taken across the series. QL varied according to the site, from 0.53 to 13.4 ng m-3. In subsequent analyses of monthly, seasonal or annual concentrations (Sect. 3.1–3.3 and 3.6), any sample that was deemed < QL was assigned a cellulose concentration of [Blank] / 2. This prevents an artificial increase in average cellulose concentrations.

Figure 2 illustrates the annual averages of cellulose concentrations across our nine sites (in orange), as well as previous data from the literature (in blue), sorted by site typology and sampled particle size. The bars represent either annual cellulose averages (if sampling lasted greater than 1 year) or cellulose averages for the designated sampling period. We believe that the roughly 4440 samples (excluding the ones produced within our study) considered in this literature survey represent a nearly complete database of cellulose concentrations in PM available in the literature. A tabulated version of the results from within the study can be found in Table 4. An expanded version of Table 4, also including previous literature results, can be found in Table S1 in the Supplement. The evolution of cellulose concentrations across the respective sampling periods for our study has further been included in the Supplement (Fig. S1).

The concentrations measured in this study are in the same order of magnitude as those reported in the literature for previous measurement campaigns. This is generally the case for both seasonal averages and overall maximum concentrations, in both coarse- and fine-mode aerosol (Sánchez-Ochoa et al., 2007; Caseiro, 2008; Yttri et al., 2011a, b). As shown in Fig. 4, annual cellulose concentrations in PM10 in our study ranged from 29.3 ± 38.4 ng m-3 (Bern) to 284.3 ± 224.8 ng m-3 (Payerne), and in PM2.5 it ranged from 15.9 ± 15.0 ng m-3 (ANDRA-OPE) to 118.1 ± 76.5 ng m-3 (Payerne). This annual average PM10 cellulose concentration taken at Payerne is higher than any previously recorded in the literature by roughly 50 ng m-3.

Moreover, results obtained at Payerne evidenced three episodic (high cellulose concentration) spikes (3 June, 13 July, and 29 July – highlighted in red in Fig. S1) which exceeded any maximum episode found in the literature by at least 160 ng m-3 (Sánchez-Ochoa et al., 2007; Caseiro, 2008; Winiwarter et al., 2009). One striking feature of the overall concentration evolution at Payerne is the high cellulose concentrations at the beginning in June 2018 and the surprisingly low concentrations in April and May 2019 (Fig. S2). Another high-concentration episode exceeding those found in the literature was documented at the rural site of ANDRA-OPE. The episodic concentration of 2027 ng m-3 (7 July 2018 – highlighted in red in Fig. S1) is almost double that of any other measurement, including those generally obtained in the present study. Samaké et al. (2020) recently reported at the same site a noticeable increase in concentrations of PBAP tracers, cellulose included, during harvest in late summer 2017. However, given that the concentration spike in 2018 originated during early July, the middle of the European summer, it is not sure that this new episode can be correlated with agricultural activity.

Figure 3 presents the comparative monthly average concentrations of cellulose in PM10 and PM2.5 taken at the three sites of Payerne, Zurich, and ANDRA-OPE, respectively (overall concentration evolutions presented in Fig. S2 in the Supplement). Cellulose concentrations in PM10 are consistently much higher than those in PM2.5, with an annual average of PM2.5 representing between 18 % and 42 % of that in PM10 for the three sites. However, very large fluctuations in this monthly ratio can be observed, particularly for the two rural sites (Payerne and ANDRA-OPE). This is primarily due to changes in PM10 cellulose concentrations, as those within PM2.5 remained largely consistent. Further, considering the overall evolution in Fig. S2, episodic PM2.5 concentrations still generally remain well below the PM10 cellulose concentrations around the same period. It seems that some process is largely impacting the source strength of atmospheric plant debris within PM10, particularly in the rural sites. In the city of Zurich, the cellulose PM2.5 / PM10 ratio remained relatively constant, just like the concentrations themselves. The comparatively low cellulose concentrations at ANDRA-OPE for 2020 (both PM10 and PM2.5) are discussed, as part of Sect. 3.7, in the interannual comparison. No ratio is provided at ANDRA-OPE, as PM2.5 and PM10 measurements were completed on different days, as opposed to simultaneous PM10 and PM2.5 sampling at Payerne and Zurich.

Importantly, across the three sites, less than 30 % of atmospheric cellulose was found within PM2.5, on average. This large dataset of size-resolved cellulose concentrations confirms that plant debris predominantly resides within the coarse aerosol mode (Sánchez-Ochoa et al., 2007; Yttri et al., 2011a). Thus, the remainder of this work will solely discuss PM10 data to understand atmospheric cellulose and its behaviour.

Previous studies either indicate a temporal variation with cellulose concentration maxima during the spring and summer seasons (Sánchez-Ochoa et al., 2007) or show very minimal seasonality (Caseiro, 2008). The following discussion will take these observations into account by presenting the results in terms of seasonal averages. Seasons were defined in 3-month periods: December–February (winter), March–May (spring), June–August (summer), and September–November (autumn). At the nine sites investigated, our PM10 cellulose measurements were above the limit of detection across all seasons. Figure 4 illustrates these seasonal cellulose concentrations (ng m-3) for the nine locations. Numerical values of seasonal means and ranges are tabulated as part of Table S1 (Supplement).

In general, the seasonal pattern exhibited here shows higher cellulose concentrations during summer and autumn, likely due to increased temperature and humidity increasing the activity of soil and litter decomposers as well as improving the quality of the litter composition. For example, the nitrogen content of leaves is shown to be greater in warmer temperatures, which leads to better conditions for leaf degradation by microbial action (Liu et al., 2006; Verma et al., 2018). It should be stated that this hypothesis would require further experiments, including specific field measurements linking soil and litter state and plant debris emission. The general trend above is exhibited at all rural sites and some urban locations (Bern, LF, and Vif). However, the extent to which these concentrations exceed the other seasons varied greatly. Normalised seasonal concentrations for each site can be found in Fig. S3 to show this variability. Considering this general seasonality, a summer–autumn maximum in cellulose concentrations deviates from the spring–summer maximum suggested by Sánchez-Ochoa et al. (2007). This may be a result of the different particle size fractions measured as part of their sampling campaign (i.e. PM2, PM2.5, or PM10), compared to the consistent PM10 measurements used in this study. This might also be due to the presence of three high-altitude, mountainous sites comprised within the six sites investigated by Sánchez-Ochoa et al. (2007). Large standard deviations are also noticed at the two rural sites of ANDRA-OPE and Payerne, especially during the summer months. This implies a significant variability in the source of atmospheric cellulose at these sites, especially when compared to the more urban locations showing smaller standard deviations and therefore a smaller flux from the cellulose source.

Whilst this is the general case, there are notable exceptions. Both the urban centres of Zurich and CB show very little seasonal variability compared to their more rural counterparts. Cellulose concentrations in Basel (suburban) also show minimal seasonality, but this may be due to concentrations being too small to exhibit a full seasonal pattern. This is surprising, given the close proximity of the site to a park-like area with trees and gardens. The lack of seasonality in urban settings, however, is consistent with the findings of Caseiro (2008). Additionally, Caseiro (2008) provided some evidence of cellulose concentrations at urban sites being greater than for nearby rural or background sites, with residential areas being an intermediate case. Within our Grenoble-based dataset as a comparison, CB (urban) does indeed exhibit cellulose concentrations marginally higher than the urban-background site of LF and significantly higher than Vif (peri-urban).

Alongside Basel, Caserne de Bonne also deviates from the general trend of summer–autumn maxima in cellulose concentrations observed across the other seven locations investigated here. Reasons for this are unclear, but this is suggestive of a source change in atmospheric plant debris or an additional source being present at some urban locations that may mask the typical seasonality. Given that these locations are urban in character, the weak seasonal variations may owe to anthropogenic activity. This will be investigated in Sect. 3.5.

To determine the overall importance of cellulose contribution to PM, the percentage contribution of cellulose carbon to total organic carbon (cellulose C to OC) was determined. Figure 5 illustrates this seasonal average percentage contribution. Table S3 summarises numerically the overall average and seasonal percentage contributions and the ratio of cellulose C contribution during winter and summer. Also highlighted is the maximum contribution of cellulose C to OC experienced over the respective sampling periods at each site.

The highest contributions to OC were typically found at rural sites, potentially due to fewer local sources of OC in rural sites compared to more urban locations. In fact, the annual contribution to OC found at Payerne (5.9 ± 4.4 %) is the highest found in the literature. However, the annual average for the urban site of Zurich is also in a high range, at 3.8 ± 2.9 %. Regarding seasonal contributions, the rural sites in this study show a significantly different seasonal pattern compared those found in the study by Sánchez-Ochoa et al. (2007). Here, we see a noticeably smaller contribution of cellulose C to OC during winter compared to summer. This is reflected in the respective winter / summer ratios of cellulose C contribution: the values in this study range between 0.36–0.45, in comparison to 4.2 and 0.3 at the PM10 rural and high-altitude sites used in their study (Sánchez-Ochoa et al., 2007).

While seasonal contributions appear to be moderate in most cases, the contribution of cellulose C within episodes can be much more significant. It is also worth noting that these contributions to OC are derived from free-cellulose concentrations. Thus, the contribution to overall OC will be higher when considering total cellulose. At sites with typically lower seasonal contributions (Basel, Bern, and LF), the episodic contributions reached between roughly 4.1 % and 6.3 %. However, at the sites that illustrated a much higher seasonal average contribution to OC, the maximum contributions during episodes were found to be between 16.1 % at Zurich and 19.7 % at Payerne. These maximum contributions (detailed in Table S3) are similar to those found at the background sites by Sánchez-Ochoa et al. (2007). These values stand to highlight the substantial contribution that atmospheric plant debris can have on atmospheric composition. In other words, PBAPs, as well as plant debris in particular, can contribute greatly to OM and must be considered within all future characterisation and source apportionment studies.

Lastly, the contribution of coarse-mode (PM with diameter less than 10 µm and greater than 2.5 µm) cellulose C to coarse-mode OC was evaluated at the three sites that completed both PM10 and PM2.5 analysis (ANDRA-OPE, Payerne, and Zurich). This can be seen in Table S4 in the Supplement. As PM2.5 data for ANDRA-OPE were only available for the 2020 sampling campaign, PM10 data from 2016 and 2017 were excluded. Table S4 shows a contribution of coarse cellulose C to be 3.16 % at ANDRA-OPE, which is of a very similar magnitude to that of the overall cellulose C contribution to OC. This is potentially due to the significant reduction in cellulose source strength at the ANDRA-OPE site during the year of 2020, compared to the years prior. This will be discussed in Sect. 3.7. However, at both Payerne and Zurich, the annual contributions to coarse OC are notably higher (11.02 % and 13.04 %, respectively) than that of overall cellulose C to OC (5.88 % and 3.76 %, respectively). From these data, we can see that plant debris makes up a significant component of the coarse fraction of OM within these two datasets.

To further evaluate the potential sources of plant debris into the atmosphere, correlations between cellulose and other source-specific tracers were investigated. This is the first cellulose field study to investigate these correlations with other tracers. Briefly, three specific sources have been hypothesised in the literature: direct biogenic emissions, unpyrolysed cellulose during domestic biomass burning, and anthropogenic resuspension and milling of plant debris (Sánchez-Ochoa et al., 2007; Caseiro, 2008; Yttri et al., 2011a, b). The chemical tracers used as proxies for these sources in this study are (i) glucose and polyols; (ii) levoglucosan; and (iii) EC, Ca2+, and Ti, respectively. A suite of correlation coefficients (Spearman's rank correlation, Rs) was created for each site to monitor variations in correlations between site types using daily samples. Spearman's rank correlation was used in this section to better account for anomalous results between different datasets (e.g. cellulose vs. polyols). A value of 1 indicates a perfect positive correlation, and a value of -1 indicates a perfect negative correlation. Table 5 shows the strength of the cellulose–tracer correlation at individual sites across the entire sampling period. A full table, inclusive with the number of data points (n) and p values for each correlation, plus Rs values within each season, can be found in the Supplement (Table S5).

Seasonal cellulose variations show neither a similar pattern across all sites nor one that is consistent across different regions and scales. This trend, or lack thereof, was expressed numerically using correlation coefficients (R2) of monthly concentration averages for the groups of sites that were sampled at the same time. As shown in Table 6, the correlations between sites within the Grenoble metropole (CB, LF, and Vif) are low to moderate. This is also the case for the Swiss sites, which span a much larger spatial range compared to the Grenoble-based sites. The lack of a shared temporal variability seems to indicate that the major sources of plant debris are most likely to be local to each site. It may also suggest that several mechanisms impacting ambient cellulose concentrations contribute to different degrees according to the investigated site (Caseiro, 2008; Winiwarter et al., 2009; Borlaza et al., 2021a). Moderate correlations between the traffic-impacted location in Bern with the two rural sites of Magadino and Payerne were the highest among the Swiss sites. Regardless, these values are not indicative of a common source. The Grenoble-based sites of LF and Vif do seem to show a slight exception, producing an R2 value close to 0.7 (p < 0.0001). The three monitoring locations within the Grenoble metropole are within 15 km of one another, so a common source of atmospheric plant debris on local scales of this magnitude remains possible.

The R2 values in Table 6 were compared to correlations between monthly mean concentrations of the so-called polyol fraction (i.e. sum of arabitol, mannitol, and sorbitol) for the same set of locations (Samaké et al., 2019a; Borlaza et al., 2021a; Grange et al., 2021). In contrast to cellulose, polyols show common temporal variations, with R2 correlations ranging from 0.4–0.91 and 0.95–0.98 (p < 0.0001) within the groups of Swiss and Grenoble-based sites, respectively (Tables S6 and S7). Polyols are used as chemical tracers for fungal spores, a very common class of PBAPs, and here provide a nearly perfect example of a PBAP class displaying homogenised concentration variations over time at a regional scale. This suggests a single common source of polyols that is impacted similarly by external factors across all locations, especially at short range e.g. within the Grenoble area. This was also suggested by Borlaza et al. (2021b) during their PMF study and by Samaké et al. (2019a) as part of their study across all of France. Moreover, Samaké et al. (2020, 2021) evidenced that the presence of fungi and bacteria in ambient air is mostly related to a limited number of microorganism species only, which vary from one climatic region to the next.

The stark contrast between the two sets of chemical tracers (cellulose vs. polyols) highlights the rather local nature of atmospheric plant debris and its sources. Given that meteorology is relatively consistent on a short to medium scale (< 200 km), it would be expected that plant debris emissions would impact all sites of a given area similarly. However, heterogeneous distribution of the diverse plant species at the city (or regional) scale might induce specific temporal variations in the emissions of plant debris at the local scale. Therefore, the lack of correlation in cellulose datasets may result from site-to-site differences in the dominant sources (flora) or emission processes of ambient plant debris (Caseiro, 2008).

Cellulose concentrations were measured over two separate time periods, 2017–2018 and 2020–2021, in Grenoble and over three separate time periods, 2016, 2017, and 2020, at ANDRA-OPE. These multiple datasets (with a similar number of data points) gave us the opportunity to assess the interannual variations in atmospheric plant debris, in the same regions. This provided the possibility of combining the various analyses used in the above sections as part of a more small-scale, holistic investigation.