香京julia种子在线播放

Daily atmospheric concentrations of allergenic pollen (pollen grains m⁻³) were available for six stations across Bavaria and a common time period of 1987 to 2017 (Figure 1, for further pollen site information see Supplementary Table 1). The station Oberjoch is alpine, while Munich (official pollen station name München) and Zusmarshausen are situated in the north of the alpine foothills, and Münnerstadt, Bamberg, Erlangen are in the warmer, north-western plateau region. Pollen monitoring was conducted with Hirst-type volumetric pollen traps, which started on average on Feb 8th (median Feb 5th). The data were collected, quality checked, and provided by the German Pollen Information Service (Stiftung Deutscher Polleninformationsdienst, PID). Details on the pollen monitoring can be found in Werchan et al. (34). We extracted seven pollen taxa (Alnus, Artemisia, Betula, Corylus, Fraxinus, Pinus, and Poaceae) for which corresponding flowering dates were available (see section Phenological Data).

For each station, start and end dates of the respective pollen seasons (SOS_P, EOS_P) were determined by the percentage method [see (35, 36)], by which the earliest and latest annual 2.5% of the pollen concentrations were eliminated. In addition, annual peak dates (PEAK_P) and annual pollen sums (SUM_P, or annual pollen integrals) were also derived. Supplementary Figure 1 allows a quick overview on basic descriptive statistics of the pollen seasons by the percentage method, which is recommended by Bastl et al. (37). For comparison, four other methods also implemented in the AeRobiology R package (38), namely the logistic method (39), the moving method (38), the clinical method (40) as well as the grains method (41), were equally applied for determining linear trends in start, peak, end of the pollen season as well as annual pollen sums. As there was a small residual risk that pollen monitoring was not started in time to always capture the start of the pollen season especially for Alnus and Corylus, we did a sensitivity analysis on SOS_P trends with two alternatively filtered data sets. We discarded SOS_P dates if the first day of operation already had pollen concentrations larger than four pollen grains m⁻³ and secondly, if the first day of operation was <10 days earlier than local flowering (see section Phenological Data and Supplementary Figure 5). For all five methods, we relied on the default setting of the AeRobiology R package of “lineal” imputation of missing values.

Daily pollen concentrations at SOS_P were classified according to Galán et al. (42), de Weger et al. (43), and Deutscher Wetterdienst (44) into intensity levels in order to assess the clinical relevance of long distance transport events.

Phenological data were retrieved from the German Meteorological Service (DWD) who is operating a volunteer network across Germany with about 1,200 sites in 2015 and only ~200 in Bavaria in 2017 (see Figure 1). We selected the phenological phases beginning of flowering of Artemisia vulgaris L. (common mugwort), Alnus glutinosa (L.) Gaertn. (European alder), Betula pendula Roth (European white birch), Corylus avellana L. (Common hazel), Fraxinus excelsior L. (European ash), Pinus sylvestris L. (Scots pine) as well as full flowering dates of two early grass species, Dactylis glomerata L. (Cock's foot or orchard grass) and Alopecurus pratensis L. (Meadow foxtail) for the period 1987–2017. No additional data filtering or quality assurance was performed since detailed observation instructions and automated data quality controls had been applied by DWD (45–47). Artemisia, Dactylis and Alopecurus are observed in plant communities of a site, for the latter two preferably on agricultural land. Flowering observations on the woody species are reported for a normal grown specimen at normal locations (microsites), preferably not in gardens and not on cultivated and garden varieties (45).

Annual phenological data were spatially interpolated following the proposed routine by DWD [(48, 49), Wolfgang Janssen, personal communication]. First Germany was divided into 30 overlapping circles of 1.95 degrees, then within each circle phenological onset dates (DOY) were regionally modeled by a multiple linear regression on elevation (h), longitude (lon) and latitude (lat) as independent variables (Equation 1).

The corresponding regional regression coefficients were assigned to the respective center point of the circle and then spatially interpolated using inverse distance weighting (IDW) from the nearest two to four circles. For all 1 km² pixels DOY were modeled based on geographical information and interpolated regression coefficients of the multiple linear regression model of equation 1. For all 70,609 1 km² pixels, trends by linear regressions for the period 1987–2017 were calculated as baseline flowering trends in Bavaria. For the six pixels of the pollen stations, the respective annual onset dates (SOS_F), as well as temporal trends, were extracted.

We assumed that during the main pollen season of a specific pollen type it is not possible to disentangle local and transported contributions, but that the start of emissions from local sources should be well-mirrored by the interpolated respective start of flowering dates of DWD. However, one had to take into account inherent uncertainties in the interpolated flowering dates [RMSE of up to 10 days according to (50)], phenological variations with micro- and meso-climate in the study area and pollen emission amounts depending on the actual land use.

Thus, we defined pollen (SOS_P) recorded <10 days from the interpolated flowering date at its 1 km² pixel (SOS_F) still as local and not necessarily as erroneous. Consequently, pollen season start dates equal to or more than 10 days earlier than the corresponding flowering dates at a station were considered as pre-season transport. Pollen season start dates equal to or more than 10 days later than local flowering hinted to either faulty pollen recording or lack of corresponding local sources (see classification rules in Table 1A). In addition we flagged trap operation starting dates after local flowering and discarded a few events which had more than ~20% of missing daily data in the pollen season, theoretically affecting the accuracy of the determined SOS_P dates. However, trends in pollen season start dates were largely unaffected by imputation of missing data as compared to raw data (see Supplementary Figure 5).

We applied this classification to the pollen data 1987–2017 of the stations Erlangen, Munich, and Oberjoch due to their superior data quality (see results Start of the Pollen and the Respective Flowering Season). For the high altitude station Oberjoch (see Supplementary Table 1), considerable local pollen sources should be present only for the taxa Pinus and Poaceae. Thus, for the other pollen taxa, generally certain sorts of transport phenomena were assumed and assigned to non-Bavarian sources if SOS_P was more than 10 days earlier than SOS_F at Munich and Erlangen. In general, similar SOS_P dates across stations (of not more than 2 days difference) also pointed to long-range transport events [(12); see Table 1A]. Post-season transport could not be identified in an equal manner since the end of flowering dates are not observed in the phenological program of DWD.

We compared the daily pollen concentrations at the start of pollen season dates (SOS_P) and compared them for two likely pollen sources (see Table 1A), i.e., pre-season transport and local sources per pollen taxon and site. We used two sided t-tests for testing the significance of differences.

The HYSPLIT (hybrid single-particle Lagrangian integrated trajectory) model was applied to study air transport and characterize potential pollen source regions (51, 52). The ERA5 dataset of the European Centre for Medium-Range Weather Forecasts (ECMWF) provided hourly forecast weather data at a 30-km resolution for the necessary underlying meteorological model. ERA5 data were downloaded from the Copernicus Climate Change Service (C3S)/Climate Data Store (CDS) and, for earlier tests, from ECWMF MARS. They were retrieved and converted to ARL, the format required by HYSPLIT, using a technically customized version of get_era5_cds.py and era52arl.f publicly provided by A. Crawford (NOAA/ARL) at https://github.com/amcz/hysplit_metdata. The 72 h HYSPLIT back trajectories were run at 3 h intervals from 00:00 to 21:00 h local time (LT) for the pollen sites for the period 2005 until 2015 due to computational restrictions. The backward trajectory calculations were started at an altitude of 500 m above ground level. We plotted across Europe the respective eight back-trajectories per day from Erlangen, Munich, and Oberjoch for all SOS_P dates as well as for the 2 days prior and after SOS_P.

The likelihood of pollen transport by HYSPLIT trajectories was assessed for all pollen taxa based on the earliest respective start of flowering dates (SOS_T) in the area covered by the 72 h back trajectories started at SOS_P as well as 1 and 2 days prior to SOS_P (see section HYSPLIT Transport Modeling). Taxa related species distributions following Caudullo et al. (53), Sofiev (32), and Holm et al. (54) (see Supplementary Figure 2) provided solid evidence for a complete pan-European distribution of each of the studied pollen taxa, except for parts of Spain and Portugal where Alnus, Betula, and Corylus pollen sources, of Italy where Betula and Pinus sources and of SE Europe where Pinus sources might be lacking. Since exact flowering dates for this large area were not available through the PEP725 database (55), we referred to the general movement of seasons throughout Europe (56): In early spring the green wave progresses from WSW to ENE, then from SW to NE and finally from SSW to NNE in late spring. In summer the movement is more south to north directed. Consequently, the annual start of flowering dates of Europe were extrapolated from the respective SOS_F dates of Erlangen and Munich (for Oberjoch we used the mean SOS_F of Erlangen and Munich). Extrapolation was made based on these season-specific speeds and directions of the green wave published (see Supplementary Table 2).

Then, the likelihood of pollen transport was assessed as medium if SOS_T was earlier or equal as SOS_F but later than SOS_P at the pollen receptor site, as high if SOS_T was even earlier or equal as SOS_P, and in all other cases as zero (trajectories only over areas with later flowering dates than at the receptor site). The overall transport was assumed to be confirmed (partly confirmed) if at least on 1 out of 3 days (SOS_P and 1 and 2 days earlier) it was classified as high (medium) (Table 1B).

Analyses were performed using R statistical software (57) together with AeRobiology package (38). The backward trajectories were processed using the openair package (58) within R.

Clear changes in key parameters of the pollen season were identified for the seven taxa across the six pollen stations (Figure 2), however only ~17% of the changes in the start, peak and end of the pollen seasons and ~27% of the changes in annual pollen sums, as assessed by the percentage method, were significant.

The early flowering taxa Corylus and Alnus mostly advanced their start and peak of season dates by up to 2 days year⁻¹. Taxa flowering in mid spring such as Betula, Fraxinus, and Pinus advanced their start and peak of season dates less, by ~ up to 0.5 days year⁻¹. Trends at the pollen station Bamberg were sometimes considerably stronger. Poaceae and Artemisia, flowering in late spring and full summer, exhibited uneven, i.e., advancing and delaying, trends with the exception of Poaceae peak dates strongly advancing in Munich, Erlangen and Bamberg. End of season dates were generally mostly unchanged. The choice of alternative approaches to the percentage method had no noticeable effect on the trends of key parameters of the pollen season (Figure 2) except SOS_P of Corylus and Alnus which were stronger at some stations and EOS_P which were weaker than by the alternative methods.

We therefore tested the robustness of SOS_P trends by the percentage method against possibly too late start of pollen monitoring after winter by two alternative data filtering methods. However, the results (see Supplementary Figure 5) indicated that only for the two early pollen taxa, namely Corylus and Alnus, trends slightly differed, but the main findings as described above were unaffected. Equally, we checked the effect of “lineal” interpolation of missing data during the pollen season on SOS_P trends; however, for most cases the differences between raw and imputed data were small or none except two cases (Alnus and Artemisia at Zusmarshausen).

Annual pollen sums (often significantly) increased across all stations for Corylus and decreased at all stations for Poaceae and Artemisia (the strong positive trend at Bamberg is due to a single outlier, see Supplementary Figure 3). We observed mixed and often insignificant changes in annual pollen sums for Alnus, Betula, and Fraxinus; however values for these three taxa at the southern stations (Zusmarshausen, Munich, and Oberjoch) were predominantly increasing, whereas for Pinus we derived more decreasing than increasing ones.

Trends largely deviating from the other stations mostly resulted from single year outliers at the start or end of the time series (see examples for Artemisia annual pollen at Bamberg and start of the pollen season at Zusmarshausen, Supplementary Figure 3).

The second method to derive trends in the start of the flowering season was via year-wise interpolated phenological observations of the DWD. We summarized the 70,609 results across all 1 km² pixels in Bavaria for each of the seven taxa, shown as violin density plots in Figure 2. Interestingly, the start of pollen season trends at the six stations were often outside the tails of the violin plots, thus largely deviating from the observed flowering trends, especially in the case of Corylus and Alnus, the early flowering species. Start of pollen season trends largely deviating from general flowering trends were found at Bamberg (more negative for Corylus, Alnus, Pinus, Fraxinus, more positive for Artemisia), equally at Münnerstadt (more negative for Corylus, more positive for Poaceae) and Zusmarshausen (more negative for Corylus, Alnus, Artemisia).

A closer look at the flowering and start of the pollen season time series revealed that pollen time series at these stations were considerably different from the corresponding flowering ones (Figure 3). Especially missing data for some years as well as failing procedures to determine the key parameters of the pollen season in the R package AeRobiology resulted in gaps in the time series and may have caused these pronounced differences, more prominent at the stations Münnerstadt, Bamberg, and Zusmarshausen. Time series and trends of flowering dates and start of pollen season dates at Oberjoch were comparable, however, the start of the pollen season was consistently earlier than the interpolated flowering dates due to the high elevation of this pollen station (see Supplementary Table 1). We, therefore, decided to study long-range transport, which mostly relies on the comparison of flowering and pollen dates, for the remaining three stations (Erlangen, Munich, Oberjoch). Supplementary Figure 4 shows in this respect two examples of strong outliers for the end of the Betula pollen season at Bamberg and Zusmarshausen.

Differences between pollen and flowering start dates at Munich and Oberjoch (2005–2015) and Erlangen (2005–2012) ranged between −88 and 53 days (Table 2). However, out of these 209 station ^* year ^* taxa combinations corresponding to the trajectory study (see section Transportation Modeled by HYSPLIT), 14 cases were not considered as reliable, eight of them related to Artemisia where the interpolated flowering dates seemed less robust (SOS_F at Oberjoch earlier than in the lowland) and six to Corylus, Alnus, and Poaceae at Erlangen where pollen recording obviously yielded wrong (i.e., too late) results. Overall, discrepancies between phenological flowering observations and airborne pollen measurements were common; their median across all stations and studied taxa was −15 days (even −17 days when excluding the non-reliable cases), indicating in general earlier airborne pollen than flowering onset dates.

At the high elevation station Oberjoch median differences per taxa were predominantly negative, decreasing during the year from −42 days (Corylus) or −43 days (Alnus) to −18 days (Poaceae), only Artemisia pollen was ~4 days later recorded than the interpolated flowering map suggested. At the two lowland stations of Erlangen and Munich, Corylus, Alnus, Betula, and Fraxinus pollen were measured earlier than the start of flowering was monitored (median differences −11, −16, −11, and −19 days, respectively). For Artemisia and Pinus, this median difference was −9.5 and −8 days, and for grass pollen even only +3 days. For Corylus, Alnus, Betula, Fraxinus, Pinus, and Poaceae at Erlangen and Munich (2005–2012), their median annual difference of SOS_P and SOS_F varied between −5.5 days (2006) and −15 days (2010).

The classification of likely pollen sources, as well as type of transport events based on the rules listed in Table 1, revealed the following findings for the period 2005 to 2015. For all pollen taxa (except Artemisia) and station ^* year combinations, the pollen sources were non-local pre-season in 68.7%, and local in 27.9% of the cases. For six cases at Erlangen (3.4%) it had to be assumed that the pollen recording of Corylus, Alnus, and Poaceae was obviously wrong. Naturally, all cases at the high elevation station Oberjoch were non-local pre-season transport.

Concerning all cases of non-local pollen sources, 57.7% of the pre-season transport could be regarded as long-range, for 42.3% a further differentiation was not possible. At Erlangen and Munich, local pollen sources of Pinus and Poaceae were more frequent than non-local ones whereas, for the other pollen taxa, non-local sources were predominant. For Artemisia, the interpolated flowering dates were less reliable and thus the respective numbers were 33.3% non-local, 40.0% local sources, and 26.7% likely wrong data (4 times at Oberjoch, and twice each at Erlangen and at Munich).

These above summarized results were largely confirmed by the full picture of all data available for 1987–2017, especially at the station Oberjoch. At Erlangen, pollen monitoring seemed to start too late since Corylus and sometimes Alnus pollen was recorded later than the local flowering estimates, similarly in fewer years for Munich. For Corylus (only Munich), Alnus, Betula, and Fraxinus at the lowland station Erlangen and Munich, pollen was recorded consistently earlier than local flowering (median differences between −11 and −19 days). Poaceae pollen was observed around the mean full flowering dates of the two grass species; however, the full data set revealed more years in which pollen was even recorded much later (≥10 days). Inherent variations in the start of the Artemisia pollen season, probably due to low pollen concentrations, led to high variability in their match to SOS_F dates.

For the selected years 2005 to 2015, we tried to confirm or reject the hypothesized transport events of Table 2 based on back trajectory modeling (see section HYSPLIT Transport Modeling) and the assumptions according to possible pollen sources (see section Likelihood of Transport Based on Back Trajectories and Figure 4).

The vast majority of the studied SOS_P dates (in day of the year, DOY) indicated that on the same day or the 2 days before pollen was likely to be transported from areas in Europe with earlier local flowering, often even equal or earlier than SOS_P [see example A in Figure 4 where Corylus pollen was likely to be transported from the NW of the Iberian Peninsula arriving 17 days before local flowering at Munich with medium pollen loads according to the classification of (44)]. Roughly 50% of these cases might have been due to long-range transport (as indicated by similar SOS_P dates across Bavaria, see Table 1).

For a quantitative analysis at the lowland stations Erlangen and Munich, we omitted Artemisia due to too high variability in pollen data and less reliable flowering interpolation, Poaceae since SOS_P almost perfectly matched SOS_F, and the four cases with too late Corylus and Alnus pollen monitoring at Erlangen (see Table 2, n = 90). In 54% of the 57 cases for non-local sources of Alnus, Betula, Corylus, Fraxinus, and Pinus, pre-season long-range transport could be confirmed by HYSPLIT back trajectories (Figure 4 example A). When SOS_P was much earlier than SOS_F, back trajectories could often only indicate likely transport since flowering in the source regions (SOS_T) was only earlier than SOS_F but not than SOS_P dates (35% of the cases, see example B where air masses originated from an area in the NW of France and SW of England with ~ 20 but not 43 days earlier flowering and transporting only small pollen loads). It was only for six out of 57 events (11%) with non-local pollen sources that a hypothesized transport could not be confirmed by back trajectory modeling [see example C where air masses came from NE directions with assumed later flowering dates and the daily pollen concentrations where according to Galán et al. (42) nil.]. In 23 of 33 cases with local pollen sources (70%), trajectories were again strongly pointing to additional pollen contributions by transport (see example D in which pollen was recorded almost the same day as local flowering, but additional long-range transport from SW France and NW Spain was very likely). In 30% (10 out of 33 cases) trajectories did not indicate any pollen transport in addition to the supposed local pollen sources [example E where air masses from the NE were unlikely to transport Betula pollen, daily pollen concentrations at SOS_P were classified as small according to Galán et al. (42), de Weger et al. (43), and Deutscher Wetterdienst (44)].

For a quantitative validation of pollen transport to Oberjoch, we discarded Artemisia as well as Poaceae yielding 55 (5 pollen species ^* 11 years) events (Figure 5). Since all pollen sources were defined as pre-season non-local (SOS_P at least 10 days earlier than interpolated SOS_F) (see section Pre-season Transport Events), we classified transport events into probably long-range from outside Bavaria (75%) and others. In only seven out of 55 cases (13%), it was not possible to confirm pollen transport by back trajectories (see example E in Figure 5 where air masses from the NW were unlikely to carry already Betula pollen). For 22 (12) of the 55 cases, pre-season long-range transport could be confirmed (partly confirmed) and for 10 (4) of the cases, pre-season transport could be confirmed (partly confirmed) by back trajectories from Oberjoch.

The daily pollen concentrations at SOS_P in Oberjoch associated with the examples provided in Figure 5 were either above the threshold of medium pollen loads (A Corylus, E Betula) or below (B Alnus, D, Fraxinus) according to respective thresholds given by Galán et al. (42), Deutscher Wetterdienst (44), and De Weger et al. (43). Daily pollen concentrations on SOS_P days with pre-season transport were not significantly different from respective amounts on SOS_P dates with local contributions (see Figure 6) although for early season pollen species (Corylus, Alnus, Betula, and Fraxinus) the mean concentrations seemed to be slightly higher. The clinical relevance of SOS_P pollen concentrations generally depended on taxa and site, less on likely pollen sources (pre-season transport or local contributions). For Corylus, Alnus, and Betula SOS_P pollen concentrations and Poaceae local contributions at Munich, the threshold of medium pollen loads was reached.