Intercomparison studies and potential of pre-load algorithms for training and validating automated sampling systems in aerobiology

Background: Monitoring airborne pollen measurements depends on accurate and reproducible pollen recognition and analysis. The traditional volumetric Hirst method for pollen monitoring is based on European standards. This method demands skilled technicians, and it is a time-consuming process. Furthermore, this method has a 5–10 days delay in the data reported. To overcome these problems, a transition from manual to modern automatic methodologies is needed.  A Hirst-type trap has been used as a baseline and maintains historical time sequences of measured data to validate automatic samplers.

Methods: We compared both three-hourly and daily data of selected pollen types: Cupressus, Fraxinus, Pinaceae, Platanus, Poaceae, and Quercus, detected with a Hirst-type trap with a parallel retrieved concentration from three different automatic samplers: Pollen Sense APS-330, Hund BAA-500, and Swisens Poleno Jupiter. The pollen measurement campaign was conducted from 01 January 2024 to 03 June 2024 in Córdoba under the Mediterranean climate. The automatic samplers are based on different sampling and analysis methods, such as imaging-based identification, fluorescence spectroscopy and/or holographic imaging. We calculated Pearson's linear correlation coefficients (r) and daily ratio among Hirst and automatic measurements. 

Results: These results indicates the statistical significant differences between Hirst and automatic samplers (p ≤ 0.001) with strong correlation coefficient (r) for APS-330 data with r > 0.74 (3h) and r > 0.89 (daily) for Quercus, r > 0.77 (3h) and r >0.83 (daily) for Cupressus, and r > 0.65 (3h) and r > 0.70 (daily) for Poaceae ; the Hund BAA-500 data shows r > 0.63 (3h) and r > 0.84 (daily) for Cupressus, r > 0.55 (3h) and r > 0.56 (daily) for Poaceae at statistically highly significant (p ≤ 0.001), and for Platanus at moderately significant (p ≤ 0.01) with r > 0.50 (daily) and non-significant (p > 0.05) with r > 0.13 (3h). As with Swisens Poleno Jupiter, the weak correlation results with  r > 0.61 (3h) and r > 0.85 (daily) (p ≤  0.001) for Platanus, r > 0.38 (3h) and r > 0.67 (daily)  (p ≤ 0.001) for Poaceae, and r > 0.17 (3h) (p ≤ 0.001) and r > 0.36 (daily) (p ≤ 0.01) for Quercus. The average daily ratio for APS-330 was 2.31, for Hund BAA-500 was 2.77, and for Swisens Poleno Jupiter was 3.18.

Conclusion: This research study gives first insights into the Mediterranean environment. The results are from the pre-loaded algorithms and the companies did not include southern categories (e.g. Olea, Morus), which could lead to false positivity for already trained categories. We observed comparable concentrations provided by Hirst and both APS-330 and Hund BAA-500. The concentrations are not similar due to different measuring techniques of automatic samplers, we always noted a similar distribution curve, taking into account the use of scaling factors for the application of a homogenization index. However, further intercomparison studies, in particular, after the training of local categories and refinement of the algorithms based on digital reference datasets (DRD), the automatic samplers would show potential.

Keywords: Automated biomonitoring; Validation; Intercomparison studies; Airborne pollen