Next Article in Journal
Discovery of Large Methane Emissions Using a Complementary Method Based on Multispectral and Hyperspectral Data
Previous Article in Journal
Analytical Approach to Enhancing Efficiency of Silt Loading Collection in EPA Vacuum Sweep Method Using K-Means Clustering
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Atmosphere
Volume 16
Issue 5
10.3390/atmos16050531
atmosphere-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Validation of the Automatic Real-Time Monitoring of Airborne Pollens in China Against the Reference Hirst-Type Trap Method

by
Yiwei Liu
1,†,
Wen Shao
2,†,
Xiaolan Lei
1,
Wenpu Shao
1,
Zhongshan Gao
3,
Jin Sun
1,
Sixu Yang
4,
Yunfei Cai
5,
Zhen Ding
6,
Na Sun
7,
Songqiang Gu
8,
Li Peng
4,* and
Zhuohui Zhao
1,4,9,*
1
NHC Key Laboratory of Health Technology Assessment, Key Laboratory of Public Health Safety of the Ministry of Education, Department of Environmental Health, School of Public Health, Fudan University, Shanghai 200032, China
2
Shanghai Chenshan Botanical Garden, Shanghai 201602, China
3
Institute of Immunology and Allergy Research Center, School of Medicine, Zhejiang University, Hangzhou 310058, China
4
Shanghai Key Laboratory of Meteorology and Health, Shanghai Meteorological Service, Shanghai 200030, China
5
Department of General Management and Statistics, Shanghai Environment Monitoring Center, Shanghai 200235, China
6
Department of Environment and Health, Jiangsu Provincial Center for Disease Control and Prevention, Nanjing 210009, China
7
Guanghua Hospital Affiliated to Shanghai University of Traditional Chinese Medicine, Shanghai 200052, China
8
Shanghai Pudong New Area Meteorological Bureau, Shanghai 200135, China
9
Shanghai Key Laboratory of Meteorology and Health, Typhoon Institute/CMA, IRDR International Center of Excellence on Risk Interconnectivity and Governance on Weather/Climate Extremes Impact and Public Health WMO/IGAC MAP-AQ Asian Office Shanghai, Fudan University, Shanghai 200438, China
*
Authors to whom correspondence should be addressed.
†
These authors contributed equally to this work.
Atmosphere 2025, 16(5), 531; https://doi.org/10.3390/atmos16050531
Submission received: 16 March 2025 / Revised: 19 April 2025 / Accepted: 23 April 2025 / Published: 30 April 2025
(This article belongs to the Section Air Quality)
Browse Figures
Versions Notes

Abstract

Background: There is a lack of automatic real-time monitoring of airborne pollens in China and no validation study has been performed. Methods: Two-year continuous automatic real-time pollen monitoring (n = 437) was completed in 2023 (3 April–31 December) and 2024 (1 April–30 November) in Shanghai, China, in parallel with the standard daily pollen sampling(n = 437) using a volumetric Hirst sampler (Hirst-type trap, according to the European standard). Daily ambient particulate matter and meteorological factors were collected simultaneously. Results: Across 2023 and 2024, the daily mean pollen concentration was 7 ± 9 (mean ± standard deviation (SD)) grains/m3 by automatic monitoring and 8 ± 10 grains/m3 by the standard Hirst-type method, respectively. The spring season had higher daily pollen levels by both methods (11 ± 14 grains/m3 and 12 ± 15 grains/m3) and the daily maximum reached 106 grains/m3 and 100 grains/m3, respectively. A strong correlation was observed between the two methods by either Pearson (coefficient 0.87, p < 0.001) or Spearman’s rank correlation (coefficient 0.70, p < 0.001). Compared to the standard method, both simple (R2 = 0.76) and multiple linear regression models (R2 = 0.76) showed a relatively high goodness of fit, which remained robust using a 5-fold cross-validation approach. The multiple regression mode adjusted for five additional covariates: daily mean temperature, relative humidity, wind speed, precipitation, and PM10. In the subset of samples with daily pollen concentration ≥ 10 grains/m3 (n = 98) and in the spring season (n = 145), the simple linear models remained robust and performed even better (R2 = 0.71 and 0.83). Conclusions: This is the first validation study on automatic real-time pollen monitoring by volumetric concentrations in China against the international standard manual method. A reliable and feasible simple linear regression model was determined to be adequate, and days with higher pollen levels (≥10 grains/m3) and in the spring season showed better fitness. More validation studies are needed in places with different ecological and climate characteristics to promote the volumetric real-time monitoring of pollens in China.

1. Introduction

With global warming, there has been an intensified production of airborne pollens, extension of the pollen pollination period, and changes in airborne allergenic components [1]. Studies in North America showed a significant 20% increase in airborne pollen concentrations in 2018 compared to 1990, and a 20-day extension of pollen seasons [2]. In Switzerland, the rising temperature has led to an earlier start for spring pollen species, with some species experiencing up to a 29-day advancement [3]. In the United Kingdom, it was reported that Betula pollen seasons were becoming more severe due to warmer temperatures, while Quercus pollen seasons started earlier despite not becoming more severe [4]. The impact of climate change on pollens and pollen seasons varies, depending on regions and the specific pollen species.
Airborne pollen allergens are the common cause of respiratory allergic diseases, in particular for allergic rhinitis (AR) or hay fever [5]. The extension of the pollination period and increased allergenic pollens aggravate allergic respiratory diseases, and they further affect patients’ quality of life and respiratory health [6]. A retrospective analysis in Japan found an increase in pollen counts was significantly associated with a rise in seasonal AR outpatients [7]. The month of birth was correlated with sensitization to aeroallergens too. Children born in certain months showed a higher prevalence of specific IgE towards certain allergens and allergic diseases [8]. In Sweden, birch pollen was the dominant cause of pollen sensitization in 4-year-old children, and those who were sensitized to birch also had a higher proportion of sensitization to other inhalant allergens [9]. After thunderstorms, pollens emerged as the predominant cause of allergic rhinitis (AR) and asthma, resulting in a significant surge in demand for medication in both clinical emergency and outpatient settings [10].
To accurately reflect variations in pollen counts in a timely and convenient manner, it is essential to implement automatic real-time monitoring for pollen concentrations. Such prompt and effective monitoring of pollen concentrations is necessary to provide data for precise forecasting and alerts. Currently, the most widely accepted method of pollen monitoring internationally is the volumetric pollen counting method, primarily using Hirst-type traps [11]. It is recommended by the European committee as the standard pollen monitoring method. In the past few years, automated pollen monitoring devices have also been invented [12], making real-time monitoring of airborne pollens feasible and more convenient. A series of studies have been carried out in Switzerland comparing the automatic pollen counting method with the manual Hirst-type trap sampling method [13].
In Japan, real-time pollen counting methods using the Yamatronics KH3000 instrument (Yamatronics, Kanagawa, Japan) and the traditional Hirst-type trap sampling method showed a favorable correlation with each other [14]. To our knowledge, no verified real-time pollen monitoring approach has been reported in China so far. Gravity sedimentation remains the primary technique, including the painstaking enumeration of pollen grains under microscopes. In China, most pollen sampling employs a canopy-type gravity sedimentation sampler equipped with microscope slides covered in glue. These samplers are positioned on the rooftops of buildings between two and six floors high, where they gather pollen that descends to the sampler. The slides are exchanged every 24 h, and the accumulated pollen on the slides is dyed, preserved, and subsequently categorized and enumerated under an optical microscope. Daily observations are conducted to track the fluctuations of various kinds of pollen over distinct dates or seasons [15]. The pollen concentrations obtained from the gravity sedimentation method represent concentrations per area, rendering them incompatible with the internationally recognized volume concentration values.
In this study, we aimed to set up a validated, real-time, and convenient pollen monitoring method. By comparing the real-time monitoring of pollen concentrations and standard manual pollen counting using the volumetric Hirst samples [16], we aimed to validate a model for real-time daily airborne pollen volumetric concentration against the standard method. The influencing meteorological factors and airborne particulate matter were considered and adjusted appropriately. This trial is the first validation study in China for real-time pollen monitoring. Such monitoring will help spread information on pollen in a timely manner for clinical doctors and community populations and allow for alerts to be provided in advance so that susceptible patients or populations can take preventive measures. It will also help in optimizing medical resources in advance to deal with potentially urgent demands for medication in susceptible subjects.

2. Materials and Methods

2.1. Pollen Monitoring and Sampling Settings

Automatic real-time pollen monitoring and manual sampling by the standard method were completed at the Meteorological Bureau of Pudong New Area, Shanghai, China (31°13′33″ N, 121°33′32″ E). The place was adjacent to Shanghai’s Century Park which had a rich variety of flowers and trees representing local plants and flowers. Specifically, automatic real-time pollen monitoring was performed by the Yamatronics KH3000 instrument on the roof of the 3rd building’s floor, 12 m above the ground within the height range recommended by Rojo [17]. Pollen sampling by the standard Hirst-type trap method was performed on the same roof 3 m apart to reduce mutual turbulence interference.
The pollen monitoring and sampling were performed over a total of 534 days, from 4 April to 31 December 2023, and from 1 April to 30 November 2024. By removing the days with unavailable or ineffective pollen data, the effective days for pollen concentrations were 483 days (90.4%) and 482 days (90.3%). Out of these days, 437 days (81.8% and 90.6%) were matched between the two methods on the same days, consisting of 239 days in 2023 and 198 days in 2024, respectively.

2.2. Pollen Monitoring and Sampling Methods

2.2.1. Real-Time Monitoring by Yamatronics KH3000

The Yamatronics KH-3000 is a fully automated pollen monitoring device that counts pollen number per m3 of air (grains/m3) per minute. The daily concentration per volume of air (grains/m3) was calculated by summing up all pollen grains in a whole day divided by the total volume pumped in (m3). It utilized scattered laser beams from semiconductor materials to detect and calculate airborne pollens. Upon passing through the inlet, high-density particles were eliminated by the dust filtration apparatus. The air, comprising residual particles primarily consisting of pollen grains, was injected into the optical system and irradiated by a semiconductor laser beam. Two detecting elements (PIN-PD) captured the dispersed light, one for forward scattering and the other for side scattering. The disparity between the side and forward scattering intensity of the scattered light was mostly associated with the particle morphology and surface texture. The scatter plot of lateral scattering intensity versus forward scattering intensity defined a rectangular area called the “extract window”. Through this extraction window, particles could be distinguished, allowing for efficient and reliable identification of pollen and other particles (e.g., fiber dust). Details on the instrument’s system can be found in the previous article [18].
The standard inlet flow rate of the Yamatronics KH3000 was 4.1 L/min. During the sampling period, the flow rate was checked regularly at least once every two weeks. The KH3000 pollen monitoring instrument was equipped with an airflow meter. When the center of the small ball in the flow meter was aligned with the marked scale, it indicated that the instrument’s intake airflow was normal. The Yamatronics KH3000 has been widely applied in Japan for its high reliability and low cost of monitoring and maintenance. It is used in the Hanako-san system managed by the Japanese Ministry of the Environment.

2.2.2. Standard Pollen Sampling by Hirst-Type Trap

The volumetric Hirst sampler [16] was operated to collect pollen samples according to the European standard [19]. This sampler contained a rotating drum that pumped air at a flow rate of 10 L/min. The air inlet flow rate was regularly checked by a non-resistance flowmeter at least once every two weeks. The pollen sample traps were collected once a week and carefully cut into 7 splits corresponding to 7 continuous days. The whole instrument was cleaned and maintained at least once a month to ensure it was in normal working condition.

2.3. Manual Counting of Pollens Under Microscopes

According to the European standards [19], pollens in the Hirst-type trap samples were counted in four longitudinal lines with 400× magnification, accounting for about 15.8% of the total surface of the slide, meeting the requirement of at least 10% by the standard. Two lines were in the upper part and the other two lines were in the lower part, and they were symmetrical to the middle line. The daily pollen numbers were obtained by multiplying to the whole day’s sampling surface area.
All pollen samples collected by the Hirst-type trap method were stained with toluidine blue before counting under the fluorescence microscope.

2.4. Meteorological Data and Ambient Airborne Particles

Hourly averages of meteorological factors (air temperature (Temp, °C), relative humidity (RH, %), precipitation (Precip. mm/h), wind speed (WS, m/s), air pressure (AP, Pa), and ambient particles PM2.5 and PM10 (aerodynamic diameter equal to or less than 2.5 μm and 10 μm, μg/m3) were collected from the Shanghai Pudong Monitoring Station. Daily averages were calculated based on the hourly data across 2 years and in each year in 2023 and 2024, respectively. This station was 2.6 km away from the sampling site.

2.5. Data Quality and Statistical Analyses

Daily monitoring data with at least 16 h a day (the monitoring time length exceeding two-thirds of the entire day) were considered as effective and were then included in the analysis. The manual counting of pollen numbers under the microscope was carried out by professionally trained personnel independently. They were blind to the real-time data provided by the KH3000 instrument to ensure the reliability of validation.
Daily averages and the standard deviation (SD) of pollen concentrations (grains/m3) were calculated, as well as the median, minimum, and maximum values. Data on meteorological factors and PM were also described. Pearson correlation and Spearman’s rank correlation were applied to analyze the correlations between the two methods.
Both the simple and multiple linear regression models were constructed for the real-time pollution monitoring, in which the multiple linear regression model was optimized for the best set of variables with the highest goodness of fit. The coefficient of determination (R2) was calculated for each model. R2 represented the proportion of change in the dependent variable (the Hirst-type data) predicted from the independent variables (the real-time monitoring data) and/or other variables. The slope was the parameter that represented the rate of concentration change in the Hirst-type pollens relative to the rate of concentration change by the real-time monitoring instrument (KH3000). The p-value < 0.05 was considered statistically significant, and two-sided significance tests were applied. Meteorological factors (Temp, RH, and AP and so on), PM2.5, and PM10 were selected and optimized in the multiple linear regression model to find the best model with the highest goodness of fit.
To evaluate the performance of simple and multiple linear regression models, we employed K-fold cross-validation. The dataset was randomly split into K equal-sized subsets, known as folds. In each iteration, one of these subsets acted as the validation set, while the remaining K-1 subsets were applied to train the models. This cycle was repeated K times, guaranteeing that each subset was used for validation at least once. We selected K as 5, which aligned with standard practices in the literature and was a suitable size, given the size of the whole dataset. After completing the five iterations, we averaged the performance metrics, coefficient of determination (R2), root mean squared error (RMSE), and mean absolute error (MAE) to obtain a reliable assessment of the models and identify the best performance.

3. Results

3.1. Effective Data Collection

During the whole monitoring period, 90.4% of days (n = 483 days) had effective data on daily pollen numbers by Hirst trap sampling, and 90.3% of days (n = 482 days) had real-time monitoring data on daily pollen numbers by KH3000. Specifically, in 2023, the effective data accounted for 89.3% (259 days) and 94.1% (273 days) of the days, respectively, while in 2024, they were 91.8% (224 days) and 85.6% (209 days) (Table 1). The unavailable data were due to the inevitable time intervals during the regular replacement of sample tapes in the Hirst-trap instrument, unexpected electrical power failures in the sampling field, and the discontinuity of the wireless network for data transmission to the data server.
Daily meteorological factors and PM were calculated for each day in parallel with pollen monitoring and sampling (Table S1). With a subtropical maritime monsoon climate in Shanghai, the average daily temperature was 22.3 ± 6.8 °C with an average RH of 83.6 ± 13.8%. The average daily precipitation was 3.2 mm with a large SD of 25.4 mm. Compared to 2023, the climate in 2024 became warmer (average Temp 23.9 ± 6.0 °C in 2024 vs. 22.0 ± 7.4 °C in 2023) and more humid (86.3 ± 13.4% in 2024 vs. 81.3 ± 13.8% in 2023) but with much less precipitation (0.2 ± 0.7 mm in 2024 vs. 5.8 ± 34.2 mm in 2023).

3.2. Descriptions of Daily Pollen Concentrations by Two Methods

After excluding days with less than 16 h of measurements, we obtained a total of 437 daily averages able to be matched between two methods: 239 from 2023 and 198 from 2024. In this matched dataset across 2 years (n = 437), the daily average pollen concentration was 8 ± 10 grains/m3 (mean ± SD) by the manual Hirst-type trap sampling method, with a median of 5 grains/m3 and a range of 0 to 100 grains/m3 (Table 2). By the real-time pollen monitoring, it was 7 ± 9 grains/m3, with a median of 5 grains/m3 and a range of 0–106 grains/m3. By Pearson correlation and Spearman’s rank correlation analyses, significant correlations were observed between the two methods (p < 0.001), with the correlation coefficients (CC) reaching 0.87 and 0.70, respectively.
In each year (Table 2), a significant correlation between the two methods was observed (p < 0.001), both in 2023 (Pearson CC 0.91 and Spearman’s rank CC 0.68) and in 2024 (Pearson CC 0.76 and Spearman’s rank CC 0.62). The CC values were numerically higher in 2023 than in 2024. For the daily averages, a higher annual daily average pollen level was observed in 2023 than in 2024, by both the manual sampling method (10 ± 12 grains/m3 in 2023 vs. 6 ± 8 grains/m3 in 2024) (p < 0.01) and by the real-time automatic monitoring method (9 ± 11 grains/m3 in 2023 vs. 4 ± 5 grains/m3 in 2024) (p < 0.01), as determined using the Wilcoxon signed-rank test. By plotting the temporal variation in daily pollens measured by the two methods in 2023 and 2024, both curves exhibited a similar variation pattern, with the pollen peak appearing in the spring season in April and in May (Figure 1).
This process was repeated five times, with each fold serving as the test set exactly once.
Grouped by daily pollen levels <10 grains/m3 or ≥10 grains/m3, days with pollens higher than ≥10 grains/m3 accounted for 22.4% (n = 98), and the CC between the two methods was higher than that in the other group with daily pollen < 10 grains/m3. Further, days in the spring season (April–June) had higher average levels and higher CC levels than those in non-spring seasons (Table 2).

3.3. Validation Model for the Automatic Real-Time Pollen Monitoring

By simple linear regression, a strong linear relationship was obtained for the automatic real-time pollen monitoring, Y = 0.95 + 0.94 X, in which Y stood for the standard manual Hirst-trap type data and X for the real-time monitoring data. R2 was 0.76 (p < 0.01) (Figure 2).
By multiple linear regression, we tested and optimized for the best validation model by selecting covariates among meteorological factors including Temp (°C), RH (%), Precip. (mm/h), WS (m/s), and AP (Pa), as well as airborne particles PM10 (µg/m3) and PM2.5 (µg/m3). First, the Pearson correlation analysis was conducted for all potential variables (Figure S1). Due to the high mutual correlation between PM2.5 and PM10 (r = 0.73), PM10 remained since PM10 had a higher positive correlation with pollen concentrations in both methods (r = 0.33 and 0.36, respectively) than PM2.5 (r = 0.12 and 0.15). Second, we assessed multi-collinearity among the potential variables by calculating the variance inflation factor (VIF) values. Those with VIF ≥ 10 were excluded from the model. Third, we compared the Akaike information criterion (AIC) and Bayesian information criterion (BIC) values across different models with different combinations of variables (Table S2). The one with the lowest AIC and BIC was chosen as the best one with a high goodness of fit. Finally, the multiple linear regression model for automatic daily pollen concentration was determined, with five independent variables included: automatic real-time daily pollen concentrations, Temp (TEM), PM10, RH, and WS:
YDaily pollen (grains/m3) = 2.75 + 0.93 × KH3000Daily + 0.01 × PM10 Daily + 0.04 × TEMDaily − 0.02×RHDaily − 0.79 × WSDaily
The model demonstrated a substantial level of predictive accuracy, with an R2 value of 0.76, indicating that approximately 76% of the variance in the actual pollen concentration could be explained by the multiple regression model (p < 0.01). As evidenced by the alignment of the data points along the diagonal reference line, the predicted values were well matched to the actual values (Figure 3).
The above two regression models demonstrated comparable explanatory power, with R2 achieving 0.758 in the simple linear regression model and a marginally higher R2 of 0.761 in the multiple regression model. By five-fold cross-validation, almost identical predictive accuracy was obtained in both models with the mean R2 equal to 0.75, and the simple model displayed slightly fewer error metrics (MAE 3.15 vs. 3.19, and RMSE 4.80 vs. 4.82) (Table 3).

3.4. Model Robustness in Subgroups of Days

To test the robustness of the simple linear validation model (Figure 4), we reconstructed the model in subgroups of days with daily pollens <10 or ≥10 grains/m3 according to the Hirst-type trap method. The results showed that in days with ≥10 grains/m3 (n = 98), the regression model performed much better (R2 = 0.71) than in those with <10 grains/m3 (R2 = 0.24, n = 339).
The second subset was for days in the spring and non-spring seasons. In the spring season (April–June, n = 145), the simple linear model achieved an R2 of 0.83, while in non-spring periods (July–December, n = 292), the model had a reduced goodness of fit with an R2 of 0.40.

4. Discussion

In this study, the automatic real-time monitoring method for daily pollen concentration (KH3000 instrument) was validated in a typical Yangtze River Basin city in China. Through almost 2 years of continuous parallel sampling and comparisons against the international standard Hirst-type manual method, a simple linear regression model for real-time pollen monitoring was proven to be acceptable, reliable, and feasible, with the coefficient of determination R2 reaching 0.76 (p < 0.01). Specifically, the model performed even better in days with ≥10 grains/m3 and in the spring season. A multiple linear regression model was also constructed by optimizing the adjusted variables, despite no obvious improvement being observed compared with the simple linear regression model. Our study provided the first reference model for automatic real-time monitoring of daily pollens (grains/m3) in mainland China. We believed it would help promote the monitoring of daily pollens in a more convenient and efficient way.
Traditional gravity sampling combined with manual counting under a microscope is still the primary method in China for pollen collection and measurement [5]. While it is widely used at a lower cost, it has significant limitations. First, manual counting is time-consuming and labor-intensive, requiring a high level of skill from the operator. This makes it challenging to meet the demands for high-throughput and long-term continuous monitoring. Second, the traditional method of pollen monitoring can only be calculated as area concentration data (grains/mm2), which does not align with the volume concentration data (grains/m3), the latter of which is more accepted internationally [20]. In the context of a warming climate and increased prevalence of allergic respiratory diseases, including AR or hay fever [21], a convenient and reliable method for pollen monitoring is in great demand. Furthermore, the high temporal resolution in real-time pollen monitoring is a necessity for timely forecasting and for providing alerts related to pollen exposure, needed by both clinical doctors and susceptible community populations.
The Hirst-type trap sampling of pollens is now recognized internationally as a reference method for pollen monitoring, which provides volumetric concentration data. This active collection method can accurately reflect the airborne pollen concentrations [22].
However, the Hirst-type trap sampling still requires manual counting, making it difficult to realize the automation of data processing at a high temporal resolution.
To address the need for more convenient and higher temporal resolution pollen monitoring, several types of automated pollen monitoring instruments have been developed. These include optical counting methods [23], image recognition techniques [24], and laser particle size measurement techniques [24]. The significant advantages of these automated monitoring approaches include a reduction in manual operation, the improvement of monitoring efficiency, and real-time data collection. However, existing automatic pollen monitoring instruments also face challenges, such as the accuracy of detection and classification, being affected by algorithms and environmental factors [25].
The KH3000 automatic pollen monitor utilized in this study had high temporal resolution and detection accuracy [26], underscoring its superiority in data processing and real-time monitoring of pollen concentration changes. It provided a vital reference for pollen monitoring efforts. However, no systematic comparative study on automatic monitoring instruments has been conducted in China against traditional manual methods, since the local plantation and ecological systems can vary across regions and countries. As far as we know, this study represented the first effort to verify the effectiveness and applicability of automated pollen monitoring by the KH3000 automatic monitor in China.
In our analysis, the coefficient of determination was similar to the pollen concentration results obtained from two instruments previously conducted in Japan (R2 = 0.83) [14]. It was also noteworthy that automatic measurements, particularly those obtained using the KH3000 instrument, tended to underestimate total pollen concentrations in comparison to the standard Hirst method. This discrepancy might be attributed to the fixed “extract window” setting of the KH3000, which might result in the omission of pollen particles outside of the designated size range. In contrast, the Hirst method allowed for the identification of all pollen particles that adhered to the adhesive tape after undergoing color staining for microscopic observation. The rare periods with a higher pollen level observed by the KH3000 instrument occurred during the peak pollen days in 2023. We considered that this was likely due to the underestimation of pollen numbers by the Hirst-type method, since on these days with an extremely high number of pollens, pollens on the tape overlapped with each other. As a result, the number counted under the microscope might be lower than the actual levels. Despite these limitations, it is important to note that we found a significant correlation between the measurements from both methods, indicating that they were fairly aligned with each other.
With an acceptable goodness of fit in the simple linear model across 2 years of parallel sampling, we noticed that in the subset of days with higher daily pollen concentrations ≥10 grains/m3, the real-time monitoring was more in agreement with manual observations than in days with daily pollen concentrations <10 grains/m3. Similarly, better performance was also observed in the spring season (average daily pollen ≥ 10grains/m3) compared to non-spring seasons (average daily pollen < 10 grains/m3). This discrepancy might be due to several reasons. First, at higher concentrations, the signal-to-noise ratio improved, allowing the optical sensors to more effectively distinguish pollen from abiotic particles [25]. The second reason is related to pollen species and morphology. In spring, Shanghai has a dominant pollen species, primarily tree pollen, with a narrower range of particle sizes and similar morphological characteristics. This reduced the possibility of misidentification by the KH3000. In contrast, in other seasons, there is a mixture of pollen from trees, grasses, and weeds, which vary significantly in particle size and morphology. This variation led to more diverse optical scattering patterns, reducing the specificity of detection. A third reason is the potential for error in the Hirst-type method. Although the Hirst-type trap is widely used as a reference pollen monitoring instrument, it is not considered as the absolute “gold standard”. Additionally, manual counting involved counting the four long axes of the pollen sample bands, and this could lead to errors in the data obtained by the Hirst-type trap, particularly when pollen concentrations were low [27]. A fourth reason is related to possible frequent rainfall and relative humidity in non-spring seasons. In non-spring seasons, increased precipitation and higher relative humidity can cause pollen to fragment, making it difficult for the KH3000 to accurately identify the particles [25]. In several previous studies, the performance of KH3000 was only evaluated when the daily pollen was >10 grains/m3 or in pollen peak seasons [28].
Our study has several strengths. The validation was robust and supported by almost 2 years of parallel sampling and monitoring data, which covered both the seasonal and inter-annual variations. Second, it adopted a double-blind method to ensure the objectivity of comparison between manual counting and automatic real-time monitoring. Third, the manual counting under the microscope was completed by fixed professional personnel to reduce objective bias among different subjects. Fourth, both simple and multiple linear regression models were tested and compared to select the most effective and feasible model.
There were also some limitations. The monitoring site in this study was in Shanghai, a city at a closer latitude, and with a more similar climate, to Japanese cities than most other cities in China. Our results provided evidence of feasibility, which might be more suitable to apply in cities with similar latitudes or climate. However, it is uncertain whether results can be extrapolated to other cities with different vegetation and climate characteristics. Second, due to our inability to access the backend data of the KH3000, particularly the specific “pollen window” data, the further classification and counting of pollen types was not be possible for validating the pollen type-specific counting by the KH3000. Thus, there is still a need to verify the accuracy and reliability of specific pollen species concentrations between KH3000 and Hirst-type data. From the perspective of pollen prevention, there are significant differences in the allergenicity of different pollen species. So, future research should incorporate artificial intelligence technology to develop the automatic real-time identification of pollen species, not only the total pollen numbers, to enhance the precise monitoring and forecasting of pollen species.

5. Conclusions

This study verified for the first time the effectiveness and applicability of real-time automatic pollen monitoring in both simple and multiple linear regression models in China. Through an almost 2-year-long parallel sampling and monitoring period, the results were robust for the whole year and performed better on days with higher pollen levels (≥10 grains/m3) and in the spring pollen season. This study will promote pollen monitoring in a more efficient and convenient way, in particular in places with similar vegetation and climate as in Shanghai, China.

Supplementary Materials

The following supporting information can be downloaded at https://www.mdpi.com/article/10.3390/atmos16050531/s1: Figure S1: Pearson correlation among pollen concentrations, meteorological factors, and ambient air pollutants (n = 437); Table S1: Summary of daily averages of meteorological factors and ambient particulate matter (n = 437) in the same time periods as pollen sampling; Table S2: Comparing the regression model fitness for daily pollen concentrations using the Akaike information criterion (AIC), Bayesian information criterion (BIC), adjusted R2, and variance inflation factor (VIF).

Author Contributions

Conceptualization, Z.Z., L.P. and Y.L.; methodology, Y.L., W.S. (Wenpu Shao)., L.P., Z.G. and Z.Z.; software, Y.L., Y.C., L.P. and S.Y.; validation, Y.L., S.Y., S.G., J.S., Z.G. and Z.Z.; formal analysis, Y.L., X.L. and W.S. (Wen Shao).; investigation, Y.L., X.L. W.S. (Wen Shao)., S.G., Y.C., J.S., L.P., N.S., and Z.Z; resources, Z.G., S.G., L.P. and Z.Z.; data curation, Y.L., S.Y., X.L., J.S., Y.C. and Z.D. writing—original draft preparation, Y.L. and Z.Z; writing—review and editing, Y.L., W.S. (Wen Shao)., L.P. and Z.Z.; visualization, Z.D., S.G., L.P., N.S., and Z.Z.; supervision, L.P. and Z.Z.; project administration, L.P. and Z.Z.; funding acquisition, L.P. and Z.Z. All authors have read and agreed to the published version of the manuscript.

Funding

This study was funded by the National Natural Science Foundation of China (No. 82473591), the Shanghai 3-year Public Health Action Plan (grant numbers GWVI-11.2-XD11, GWV-11.1-39, GWVI-11.1-01), the Shanghai B&R Joint Laboratory Project (No. 22230750300), the Shanghai International Science and Technology Partnership Project (No. 21230780200).

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The original contributions presented in this study are included in the article/Supplementary Materials. Further inquiries can be directed to the corresponding author(s).

Conflicts of Interest

The authors declare no competing financial interest.

References

  1. Ravindra, K.; Goyal, A.; Mor, S. Pollen allergy: Developing multi-sectorial strategies for its prevention and control in lower and middle-income countries. Int. J. Hyg. Environ. Health 2022, 242, 113951. [Google Scholar] [CrossRef] [PubMed]
  2. Anderegg, W.R.L.; Abatzoglou, J.T.; Anderegg, L.D.L.; Bielory, L.; Kinney, P.L.; Ziska, L. Anthropogenic climate change is worsening North American pollen seasons. Proc. Natl. Acad. Sci. USA 2021, 118, e2013284118. [Google Scholar] [CrossRef]
  3. Gehrig, R.; Clot, B. 50 Years of Pollen Monitoring in Basel (Switzerland) Demonstrate the Influence of Climate Change on Airborne Pollen. Front. Allergy 2021, 2, 677159. [Google Scholar] [CrossRef]
  4. Adams-Groom, B.; Selby, K.; Derrett, S.; Frisk, C.A.; Pashley, C.H.; Satchwell, J.; King, D.; McKenzie, G.; Neilson, R. Pollen season trends as markers of climate change impact: Betula, Quercus and Poaceae. Sci. Total Environ. 2022, 831, 154882. [Google Scholar] [CrossRef] [PubMed]
  5. Buters, J.T.M.; Antunes, C.; Galveias, A.; Bergmann, K.C.; Thibaudon, M.; Galán, C.; Schmidt-Weber, C.; Oteros, J. Pollen and spore monitoring in the world. Clin. Transl. Allergy 2018, 8, 9. [Google Scholar] [CrossRef] [PubMed]
  6. Blando, J.; Bielory, L.; Nguyen-Feng, V.N.; Diaz, R.; Jeng, H.A. Anthropogenic Climate Change and Allergic Diseases. Atmosphere 2012, 3, 200–212. [Google Scholar] [CrossRef]
  7. Wakamiya, S.; Matsune, S.; Okubo, K.; Aramaki, E. Causal Relationships Among Pollen Counts, Tweet Numbers, and Patient Numbers for Seasonal Allergic Rhinitis Surveillance: Retrospective Analysis. J. Med. Internet Res. 2019, 21, e10450. [Google Scholar] [CrossRef]
  8. Saitoh, Y.; Dake, Y.; Shimazu, S.; Sakoda, T.; Sogo, H.; Fujiki, Y.; Shirakawa, T.; Enomoto, T. Month of birth, atopic disease, and atopic sensitization. J. Investig. Allergol. Clin. Immunol. 2001, 11, 183–187. [Google Scholar]
  9. Ghunaim, N.; Wickman, M.; Almqvist, C.; Soderstrom, L.; Ahlstedt, S.; van Hage, M. Sensitization to different pollens and allergic disease in 4-year-old Swedish children. Clin. Exp. Allergy 2006, 36, 722–727. [Google Scholar] [CrossRef]
  10. Davies, J.M.; Thien, F.; Hew, M. EPIDEMIC THUNDERSTORM ASTHMA Thunderstorm asthma: Controlling (deadly) grass pollen allergy. BMJ—Br. Med. J. 2018, 360, k432. [Google Scholar] [CrossRef]
  11. Lucas, R.W.; Bunderson, L. A Review of Pollen Counting Networks: From the Nineteenth Century into the Twenty-first Century. Curr. Allergy Asthma Rep. 2024, 24, 1–9. [Google Scholar] [CrossRef]
  12. Jalbert, I.; Golebiowski, B. Environmental aeroallergens and allergic rhino-conjunctivitis. Curr. Opin. Allergy Clin. Immunol. 2015, 15, 476–481. [Google Scholar] [CrossRef]
  13. Tummon, F.; Adamov, S.; Clot, B.; Crouzy, B.; Gysel-Beer, M.; Kawashima, S.; Lieberherr, G.; Manzano, J.; Markey, E.; Moallemi, A.; et al. A first evaluation of multiple automatic pollen monitors run in parallel. Aerobiologia 2024, 40, 93–108. [Google Scholar] [CrossRef]
  14. Kawashima, S.; Clot, B.; Fujita, T.; Takahashi, Y.; Nakamura, K. An algorithm and a device for counting airborne pollen automatically using laser optics. Atmos. Environ. 2007, 41, 7987–7993. [Google Scholar] [CrossRef]
  15. Wu, S.; Sun, A.; Shen, X.; Luo, X.; Li, X. The dispersal and deposition characteristics of airborne pollen and its response to meteorological factors in northern Beijing, China. Aerobiologia 2025. [Google Scholar] [CrossRef]
  16. Hirst, J.M. An automatic volumeteric sport trap. Ann. Appl. Biol. 1952, 39, 257–265. [Google Scholar] [CrossRef]
  17. Rojo, J.; Oteros, J.; Perez-Badia, R.; Cervigon, P.; Ferencova, Z.; Gutierrez-Bustillo, A.M.; Bergmann, K.C.; Oliver, G.; Thibaudon, M.; Albertini, R.; et al. Near-ground effect of height on pollen exposure. Environ. Res. 2019, 174, 160–169. [Google Scholar] [CrossRef]
  18. Kawashima, S.; Thibaudon, M.; Matsuda, S.; Fujita, T.; Lemonis, N.; Clot, B.; Oliver, G. Automated pollen monitoring system using laser optics for observing seasonal changes in the concentration of total airborne pollen. Aerobiologia 2017, 33, 351–362. [Google Scholar] [CrossRef]
  19. EN 16868:2019; Ambient Air-Sampling and Analysis of Airbone Pollen Grains and Fungal Spores for Networks Related to Allergy-Volumetric Hirst Method. CEN-CENELEC Management Centre: Brussels, Belgium,, 2019; p. 20.
  20. Maya-Manzano, J.M.; Smith, M.; Markey, E.; Clancy, J.H.; Sodeau, J.; O’Connor, D.J. Recent developments in monitoring and modelling airborne pollen, a review. Grana 2021, 60, 1–19. [Google Scholar] [CrossRef]
  21. Karatzas, K.; Katsifarakis, N.; Riga, M.; Werchan, B.; Werchan, M.; Berger, U.; Pfaar, O.; Bergmann, K.C. New European Academy of Allergy and Clinical Immunology definition on pollen season mirrors symptom load for grass and birch pollen-induced allergic rhinitis. Allergy 2018, 73, 1851–1859. [Google Scholar] [CrossRef]
  22. Levetin, E.; McLoud, J.D.; Pityn, P.; Rorie, A.C. Air Sampling and Analysis of Aeroallergens: Current and Future Approaches. Curr. Allergy Asthma Rep. 2023, 23, 223–236. [Google Scholar] [CrossRef]
  23. Landsmeer, S.H.; Hendriks, E.A.; de Weger, L.A.; Reiber, J.H.; Stoel, B.C. Detection of pollen grains in multifocal optical microscopy images of air samples. Microsc. Res. Tech. 2009, 72, 424–430. [Google Scholar] [CrossRef]
  24. Ning, H.; Wang, H.; Chen, Y.; Wang, X. Analysis of airborne pollen and the visiting rate of patients with allergic diseases in spring in urban area of Beijing. Chin. Arch. Otolaryngol.-Head Neck Surg. 2021, 28, 98. [Google Scholar]
  25. Tummon, F.; Bruffaerts, N.; Celenk, S.; Choël, M.; Clot, B.; Crouzy, B.; Galán, C.; Gilge, S.; Hajkova, L.; Mokin, V.; et al. Towards standardisation of automatic pollen and fungal spore monitoring: Best practises and guidelines. Aerobiologia 2024, 40, 39–55. [Google Scholar] [CrossRef]
  26. Muradil, M.; Okamoto, Y.; Yonekura, S.; Chazono, H.; Hisamitsu, M.; Horiguchi, S.; Hanazawa, T.; Takahashi, Y.; Yokota, K.; Okumura, S. Reevaluation of pollen quantitation by an automatic pollen counter. Allergy Asthma Proc. 2010, 31, 422–427. [Google Scholar] [CrossRef]
  27. Milic, A.; Addison-Smith, B.; Van Haeften, S.; Davies, J.M. Analysis of quality control outcomes of grass pollen identification and enumeration: Experience matters. Aerobiologia 2021, 37, 797–808. [Google Scholar] [CrossRef]
  28. Zhang, F.; Krafft, T.; Zhang, D.; Xu, J.; Wang, W. The association between daily outpatient visits for allergic rhinitis and pollen levels in Beijing. Sci. Total Environ. 2012, 417–418, 39–44. [Google Scholar] [CrossRef]
Atmosphere 16 00531 g001aAtmosphere 16 00531 g001b
Figure 1. Temporal variation in daily pollen concentration (grains/m3) in 2023 and 2024, respectively. The blue and red lines refer to the daily pollen concentrations by the Hirst-trap method and real-time instrument (KH3000) around 12 m above the ground, respectively.
Figure 1. Temporal variation in daily pollen concentration (grains/m3) in 2023 and 2024, respectively. The blue and red lines refer to the daily pollen concentrations by the Hirst-trap method and real-time instrument (KH3000) around 12 m above the ground, respectively.
Atmosphere 16 00531 g001aAtmosphere 16 00531 g001b
Atmosphere 16 00531 g002
Figure 2. Simple linear regression models for daily pollen concentration for automatic real-time pollen monitoring (n = 437). Note: The model (n = 437) was constructed against the pollen data using the reference Hirst-trap method in 2023 and 2024. The dashed line is the reference line y = x.
Figure 2. Simple linear regression models for daily pollen concentration for automatic real-time pollen monitoring (n = 437). Note: The model (n = 437) was constructed against the pollen data using the reference Hirst-trap method in 2023 and 2024. The dashed line is the reference line y = x.
Atmosphere 16 00531 g002
Atmosphere 16 00531 g003
Figure 3. Scatter plot of predicted vs. actual values in multiple regression analysis. Note: This scatter plot illustrates the relationship between the predicted values from the multiple regression model and the actual values. Each point on the plot represents a pair of predicted and observed values for a given observation of multiple factors. X-axis (predicted values): The values predicted by the regression model. Y-axis (actual values): The observed values from Hirst. The dashed line is the reference line y = x.
Figure 3. Scatter plot of predicted vs. actual values in multiple regression analysis. Note: This scatter plot illustrates the relationship between the predicted values from the multiple regression model and the actual values. Each point on the plot represents a pair of predicted and observed values for a given observation of multiple factors. X-axis (predicted values): The values predicted by the regression model. Y-axis (actual values): The observed values from Hirst. The dashed line is the reference line y = x.
Atmosphere 16 00531 g003
Atmosphere 16 00531 g004
Figure 4. Simple linear regression models for daily automatic real-time pollen concentrations in subsets of days with pollens < 10 or ≥ 10 grains/m3 and in spring or non-spring seasons, respectively. The dashed lines represent the reference line y = x.
Figure 4. Simple linear regression models for daily automatic real-time pollen concentrations in subsets of days with pollens < 10 or ≥ 10 grains/m3 and in spring or non-spring seasons, respectively. The dashed lines represent the reference line y = x.
Atmosphere 16 00531 g004
Table 1. Summary of data collection by standard pollen sampling (Hirst-type trap) and automatic real-time monitoring (KH3000 instrument) across 2 years and in 2023 and 2024, respectively.
Table 1. Summary of data collection by standard pollen sampling (Hirst-type trap) and automatic real-time monitoring (KH3000 instrument) across 2 years and in 2023 and 2024, respectively.
MethodsTime PeriodsInstrument TypeAvailable Data (%)Time ResolutionEffective Sampling Days (n)
TotalHirst-type trap2023.4.3–12.31
2024.4.1–11.30		Pollen sampling90.41 h483
KH3000Real-time monitor90.31 min482
2023Hirst-type trap2023.4.3~2023.12.31Pollen sampling89.31 h259
KH3000Real-time monitor94.11 min273
2024Hirst-type trap2024.4.1~2024.11.30Pollen sampling91.81 h224
KH3000Real-time monitor85.61 min209
Table 2. Average daily pollen concentrations (mean ± SD, grains/m3) by standard Hirst-trap and automatic real-time monitoring method across 2 years and stratified by year, pollen levels and pollen seasons, respectively.
Table 2. Average daily pollen concentrations (mean ± SD, grains/m3) by standard Hirst-trap and automatic real-time monitoring method across 2 years and stratified by year, pollen levels and pollen seasons, respectively.
Daily Pollen Concentrations (Grains/m3) Matched Between Two Methods
MethodsnMean ± SDMinMedianMaxCC
TotalHirst-type trap4378 ± 1005100
KH30004377 ± 9051060.87 ***/0.70 ***
2023
(April–December)		Hirst-type trap23910 ±1206100
KH30002399 ±11061060.91 ***/0.68 ***
2024
(April–November)		Hirst-type trap1986 ± 80464
KH30001984 ± 503430.76 ***/0.62 ***
Daily pollen
<10 grains/m3		Hirst-type trap3394 ± 30410
KH30003394 ± 604160.49 ***/0.54 ***
Daily pollen
≥10 grains/m3		Hirst-type trap9820 ± 151014100
KH30009817 ± 162111060.86 ***/0.59 ***
Spring
(April–June)		Hirst-type trap14512 ± 1507100
KH300014511 ± 14161060.91 ***/0.78 ***
Non-springHirst-type trap2925 ± 50534
(July–December)KH30002925 ± 505520.63 ***/0.64 ***
CC: correlation coefficient. Data available and matched on the same days for the two methods were summarized in this table, and their CCs were calculated by Pearson correlation/Spearman correlation analysis. *** p-value < 0.001 refers to the statistical significance of correlation between the standard method (Hirst-type trap) and automatic real-time monitoring method (KH3000).
Table 3. Five-fold cross-validation in simple and multiple linear regression models for daily pollen concentrations.
Table 3. Five-fold cross-validation in simple and multiple linear regression models for daily pollen concentrations.
Simple Linear Regression Multiple Linear Regression
FoldR2RMSEMAER2RMSEMAE
10.774.363.210.847.094.08
20.745.063.030.773.402.53
30.726.173.250.645.203.35
40.805.843.460.763.552.82
50.764.112.880.764.853.16
Mean0.754.803.150.754.823.19
R2 refers to R-squared value of the model; RMSE: root mean squared error; and MAE: mean absolute error. Note: This approach involved partitioning the dataset into five equally-sized folds. In each fold, one subset was used for testing, while the remaining k − 1 subsets were used for training the model.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Liu, Y.; Shao, W.; Lei, X.; Shao, W.; Gao, Z.; Sun, J.; Yang, S.; Cai, Y.; Ding, Z.; Sun, N.; et al. Validation of the Automatic Real-Time Monitoring of Airborne Pollens in China Against the Reference Hirst-Type Trap Method. Atmosphere 2025, 16, 531. https://doi.org/10.3390/atmos16050531

AMA Style

Liu Y, Shao W, Lei X, Shao W, Gao Z, Sun J, Yang S, Cai Y, Ding Z, Sun N, et al. Validation of the Automatic Real-Time Monitoring of Airborne Pollens in China Against the Reference Hirst-Type Trap Method. Atmosphere. 2025; 16(5):531. https://doi.org/10.3390/atmos16050531

Chicago/Turabian Style

Liu, Yiwei, Wen Shao, Xiaolan Lei, Wenpu Shao, Zhongshan Gao, Jin Sun, Sixu Yang, Yunfei Cai, Zhen Ding, Na Sun, and et al. 2025. "Validation of the Automatic Real-Time Monitoring of Airborne Pollens in China Against the Reference Hirst-Type Trap Method" Atmosphere 16, no. 5: 531. https://doi.org/10.3390/atmos16050531

APA Style

Liu, Y., Shao, W., Lei, X., Shao, W., Gao, Z., Sun, J., Yang, S., Cai, Y., Ding, Z., Sun, N., Gu, S., Peng, L., & Zhao, Z. (2025). Validation of the Automatic Real-Time Monitoring of Airborne Pollens in China Against the Reference Hirst-Type Trap Method. Atmosphere, 16(5), 531. https://doi.org/10.3390/atmos16050531

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop