As a result of a project coordinated by the research group of the University of Szeged, an article dealing with the prediction of the total daily pollen concentration on a global scale has been accepted for publication in the prestigious journal "Allergy".

From our faculty, PhD Dean Edit Mikó, Lilit Czibolya and Prof. Dr. László Makra emeritus professor are among the authors. Lilit Czibolya is a graduate student of our faculty.

Pollen has been associated with allergic symptoms in sensitized individuals, having a direct negative impact on the overall quality of life of a substantial part of the population. Globally, the prevalence of allergenic pollen-induced asthma in different countries has been considered to range from 1% to 18% of the population. In 2016, it was estimated that more than 340 million people had asthma globally with 420,000 deaths associated with asthma at the global level. By 2025, the number of people with asthma may increase by an additional 100 million, and asthma occurrence in urban populations is projected to increase from 45% to 59%.

Given the extent of current and future climate uncertainty, there is a need to strengthen predictive pollen forecasts.

The aim of the study was (1) to develop and apply accurate and effective forecasting techniques to reliably predict daily total pollen concentrations on an expanded geographic scale, (2) to evaluate the overall order of importance of influencing variables in determining future pollen concentrations, and (3) to explore the physical-environmental associations for pollen responses as a function of the geographical location at a global level.

The pollen database of only a total of 23 cities on five continents met the strict data selection requirements, and these data formed the basis of our investigations.

Geographical distribution/location of the aerobiological stations that provided data for the analysis. The colours refer to the number of years considered with the available daily pollen concentrations. [median: 22 years; min: 5 years, Santiago (Chile); max: 38 years, Brussels (Belgium)].

The study aimed to use CatBoost (CB) and Deep Learning (DL) models for predicting the daily total pollen concentration up to 14 days in advance for the investigated 23 cities. The model includes the projected environmental parameters, recent concentrations (1, 2 and 4 weeks), and the past environmental explanatory variables, and their future values.

Modelled R² for each day, averaged over the 23 cities. (The R² metrics are computed using only the daily total pollen concentrations greater than 0.) (*Based on the environmental variables, including the daily total pollen concentrations of the past 7, 14, and 28 days.)

The best pollen forecasts include Mexico City (R²(DL_7)≈0.7, and Santiago (R²(DL_7≈0.8) for the 7th forecast day, respectively; while the weakest pollen forecasts are made for Brisbane (R²(DL_7)≈0.4) and Seoul (R²(DL_7)≈0.1 for the 7th forecast day.

Differences in the model performances (R² for each city averaged over days) for the best CB and DL models. DL_7 (orange) outperforms CB_7 (blue) for 14 cities. Brisbane (Australia) and Bursa (Turkey) present the largest differences in the performances. (The R² metrics differences are computed using only daily total pollen concentrations greater than 0.)

The global order of the five most important environmental variables in determining the daily total pollen concentrations is, in decreasing order: the past daily total pollen concentration, future 2 m temperature, past 2 m temperature, past soil temperature in 28-100 cm depth, and past soil temperature in 0-7 cm depth.

Global order of feature importance of the environmental variables resulting from the CB_7 model (orange: values of the past variables importance; blue: values of the future variables importance). The order of feature importance is averaged over cities and forecast days to describe variables effect on the daily total pollen concentration forecasts. Top: only the five most important (or top-ranked) variables are shown. Bottom: all variables are reported excluding the daily total pollen concentration variable, for visualization purposes. (Note that pollen importance is set to 1, and all other variables importance are given relative to the pollen importance (e.g., 0.2 means as important as one-fifth.)

City-related clusters of the most similar distribution of feature importance values of the environmental variables only slightly change on consecutive forecast days for Caxias do Sul, Cape Town, Brisbane, and Mexico City, while they often change for Sydney, Santiago, and Busan.

Clustering results for all combinations of city-day. The numbers in the squares are the serial numbers of the 19 clusters from 0 to 18, comprising groups of various environmental variables. A value of -1 indicates points classified as noise from the DBSCAN algorithm (points that are not part of a group of at least 4 points). For example, cluster 0 indicates that the same features (i.e., the same environmental variables) are important for 17 cities for predicting the daily total pollen concentration for day 1, i.e. for the next day. At the same time, for day 2 the same environmental variables in cluster 0 are important only for 6 cities. Larger squares indicate a higher predictive performance for the city-day combination (smallest square: Seoul (South Korea), day 12, R²=0; biggest square: Mexico City (Mexico), day 1, R²=1; median size square: Busan (South Korea), day 7, R²=0.506). The dendrogram at the top indicates cities that are similar in terms of the clusters they are assigned to.

This new knowledge of the ecological relationships of the most remarkable variables importance for pollen forecast models according to clusters, cities and forecast days is important for developing and improving the accuracy of airborne pollen forecasts.

The parameters of the article are as follows:

László Makra, Luca Coviello, Andrea Gobbi, Giuseppe Jurman, Cesare Furlanello, Mauro Brunato, Lewis H. Ziska, Jeremy J. Hess, Athanasios Damialis, Maria Pilar Plaza Garcia, Gábor Tusnády, Lilit Czibolya, István Ihász, Áron József Deák, Edit Mikó, Zita Dorner, Susan K. Harry, Nicolas Bruffaerts, Ann Packeu, Annika Saarto, Linnea Toiviainen, Maria Louna-Korteniemi, Sanna Pätsi, Michel Thibaudon, Gilles Oliver, Athanasios Charalampopoulos, Despoina Vokou, Ewa Maria Przedpelska-Wasowicz, Ellý Renée Guðjohnsen, Maira Bonini, Sevcan Celenk, Cumali Ozaslan, Jae-Won Oh, Krista Sullivan, Linda Ford, Michelle Kelly, Estelle Levetin, Dorota Myszkowska, Elena Severova, Regula Gehrig, María Del Carmen Calderón-Ezquerro, César Guerrero Guerra, Manuel Andres Leiva-Guzmán, Germán Darío Ramón, Laura Beatriz Barrionuevo, Jonny Peter, Dilys Berman, Connie H. Katelaris, Janet M. Davies, Pamela Burton, Paul J. Beggs, Sandra María Vergamini, Rosa María Valencia-Barrera, Claudia Traidl-Hoffmann, 2024: Forecasting daily total pollen concentrations on a global scale. Allergy, doi: 10.1111/all.16227