Comparative modeling approaches for predicting Olea and Quercus pollen seasons in Thessaloniki, Greece

Abstract

In the Mediterranean region, Olive (Olea europaea L.) is a primary source of airborne allergenic pollen, while Quercus contribute substantial quantities of pollen grains to the atmosphere, posing significant challenges in predicting their Main Pollen Seasons (MPS). This study addresses these challenges through the application of various predictive methodologies, including Thermal Time (TT) models, which integrate chilling and heat requirements, along with Partial Least Squares Regression (PLS), and Temperature-Photoperiod (TP) models. Pollen data from 2016 to 2022 in Thessaloniki, Greece, and meteorological data from ERA5 ECMWF reanalysis were analyzed. The results indicate that the mean duration of the Olea MPS was 58 days, starting in early April and peaking in mid-May, while the Quercus MPS lasted 55 days, starting in late March and peaking on average on April 25th. The predictive models were generally effective for forecasting the MPS, with a combination of TT and PLS methods providing the most robust predictions. PLS regression analysis highlighted the significant impact of spring and preceding autumn temperatures on the MPS. The chilling period for both pollen taxa typically begins in late November to early December, ending around early January, with an average requirement of about 1187 chilling hours. The heating requirements of the different predicting approaches for the start of MPS varied from 435 °C to 1113 °C for Olea and 544 °C to 1081 °C for Quercus. The results emphasize the importance of localized studies in aerobiology for accurate MPS predictions, which are crucial for public health planning and allergen management.

Introduction

Airborne pollen is a significant biological agent that can greatly impact human life of sensitized individuals during the flowering season¹. In European urban centers, research has shown that children are the most vulnerable group², with allergenic pollen affecting up to 40% of the population³ and, in some cases, reaching as high as 50%⁴. Pollen triggers a variety of allergic reactions, such as allergic rhinitis and asthma, while atmospheric pollen concentrations are projected to double every decade⁵. In the Mediterranean basin, pollen taxa as Olea and Quercus have significant health^3,6,7, ecological^8–10, and economic impacts^11,12, with high rates of flowering intensity comparing to other pollen taxa^7,13–20 and sensitization observed in countries like Spain^21–26, Portugal^27–29, Greece^30,31, Israel³², Turkey³³, France³⁴, Italy^35–38 and Balkan region³⁹. Therefore, estimating their main pollen seasons is essential for improving quality of life and mitigating allergic reactions.

Olive trees are well adapted to the Mediterranean climate, capable of withstanding high summer temperatures, low winter temperatures, and summer droughts. While primarily entomophilous, they exhibit secondary anemophily, a trait that has been enhanced by agricultural practices⁴⁰. Similar to other tree species, olive flowering intensity follows an alternating pattern, with years of high pollen production interspersed with years of low or even negligible production. These characteristic quasi-biennial cycles are readily apparent in observations^41,42. Their phenology is highly sensitive to temperature^43–48, which plays a crucial role in various phases of their reproductive cycle^49–53, involving bud formation during summer^54,55, dormancy in autumn^43,56–59, budburst in late winter^60–63, and flowering in late spring^45,64,65. This cycle is governed by the tree’s thermal requirements, including specific chilling temperatures needed to break dormancy^{11,54,57–60,66,67} and subsequent heat accumulation for flowering^{44,45,50,61,68–71}. Different olive cultivars exhibit variations in these thermal requirements, influenced by geographical factors such as latitude, longitude, and altitude^72–76.

Quercus species are predominant in many European evergreen forests⁹, playing a critical role in maintaining biodiversity and contributing to the ecological conservation of these ecosystems⁷⁷. These species can exhibit cross-reactivity with other allergens, significantly impacting susceptible populations^12,17. They are monoecious, with unisexual wind-pollinated flowers, and, in some cases, they can flower at different times throughout the season⁷⁸. Oak species can be classified into subgenera, sections, and species based on their morphological characteristics, while species and subspecies are primarily distinguished by variations in fruit traits^79,80. Studies in Southern Spain have revealed that oak trees can also exhibit a secondary flowering in autumn, driven by specific meteorological conditions such as low mean temperatures and high rainfall in September⁸¹. The reproductive buds are highly sensitive to temperature^7,51, with an earlier start to the spring season resulting from increased temperatures^5,9,82. The buds require chilling, and in spring higher temperatures are needed to initiate flowering, after a definite amount of heat has accumulated, as observed in tree species^10,83,84.

With trees’ dependence on temperature being direct, it is one of the main factors - along with precipitation - influencing bud production and flowering timing, which in turn affects the start of flowering season^{7,10,15,19,37,85–89}. Thus, olive and oak trees may also be considered by several studies as bioindicators of the impact of climate change^{9,10,50,69,90–93}. Climate change has had substantial impacts on the phenology of temperate plants, leading to shifts in the timing, intensity, and duration of their pollen seasons^{5,15,45,83,94–98}, while over the last 50 years, the majority of the Mediterranean macrobioclimate plants have shown alterations in their phenological behavior^99,100. Studies have highlighted that increased temperatures and altered precipitation patterns influence the onset and length of the overall pollen seasons^74,101. For instance, in the context of Quercus, it has been demonstrated that under a doubled CO₂ scenario, the pollination season could start on average one month earlier and airborne pollen concentrations will increase by 50% with respect to current levels⁹. Similar shifts have been observed for Olea, where earlier leafing and flowering are attributed to rising winter and spring temperatures¹⁰².

In the context of climate change, a key and complex goal of aerobiological research is to update and apply predictive models that can determine the onset and end of pollen seasons^{10,45,48,83,88,103–107}. Identifying the beginning of the pollen season is especially crucial for individuals allergic to specific pollen taxa, as it allows them to commence anti-allergy treatments in advance, thereby maximizing the treatment’s effectiveness¹⁰⁸. Identifying, in addition, the duration and the intensity of pollen seasons on the basis of criteria that have been established by the European Academy of Allergy and Clinical Immunology (EAACI), is a crucial factor for applying Allergen Immunotherapy as the only therapeutical method for sufferers^109,110. Importantly, there can be two approaches in the definition of a pollen season: one focusing on the period(s) within a calendar within which pollen induced symptoms are manifested and relevant medical actions are required, and one focusing on the plant phenology and on the release of the first pollen grains in the atmosphere, which is a necessary but not always sufficient condition for symptom exhibition. We are here focusing on the second approach. In this regard, the majority of previous studies have focused in temperature accumulation methods – Thermal Time models (TT), including chilling and heating requirements, in regions such as the Mediterranean^{10,50,59,83,108}, Poland⁸⁴, Germany⁹⁶, across Europe¹¹¹, Africa¹¹², and the United States¹¹³. Some of these studies also utilize photoperiod in combination with temperature^47,101 and regression models to determine the Main Pollen Seasons (MPS)^114–116. However, in Greece, particularly in Thessaloniki, studies on pollen taxa have primarily focused on monitoring and understanding health impacts while highlighting the importance of local pollen dynamics^30,31. These efforts have been largely observational, drawing data from the country’s oldest aerobiological station. In addition, research in this area includes studies on the allergenic properties of pollen taxa and field measurements^5,16,117,118 as well as on the modelling and forecasting of pollen concentration levels in the atmosphere^{106,119–122}.

Despite extensive research, gaps remain in the comparative approaches of modeling Olea and Quercus pollen seasons in Thessaloniki. This study aims to address these gaps for the first time in Greece by applying TT models, including Heat Units (HU), Growing Degree Days (GDDs), and accumulation of maximum temperature (T_max), in combination with chilling methods. These methods, augmented by Partial Least Squares Regression (PLS) and Temperature – Photoperiod models (TP), offer a robust approach to predicting the MPS. This comprehensive approach will enhance the ability to anticipate and manage the health impacts of pollen, while also contributing to broader ecological and climatological studies in the Mediterranean region.

Materials and methods

Area of study

Thessaloniki, the second-largest city in Greece, is situated in the northern part of the country and serves as a significant cultural and economic hub in the Mediterranean region^123,124. The geographic coordinates of Thessaloniki are approximately 40.6401° N latitude and 22.9444° E longitude, with an elevation of around 20 m above the sea level. The city is surrounded by diverse landscapes, including coastal areas along the Thermaic Gulf, as well as mountainous regions, which create varied microclimates that influence local vegetation patterns¹²⁵. The city’s climate is characterized by hot, dry summers and mild, wet winters, typical of the Mediterranean climate^126,127. These climatic conditions are conductive to the growth of various plant taxa, including Olea and Quercus, which are prominent in the local flora and significantly contribute to airborne pollen^118,128,129. The olive trees are extensively cultivated in large areas in the region of Halkidiki, located southeast of Thessaloniki, as well as along the coastal zone to the west of the city (Fig. 1a). Therefore, studying olive trees is crucial, not only because of their allergenic properties but also due to the prospect of long-range transport of their pollen grains¹³⁰. In contrast, oak trees are more abundant in the broader Thessaloniki area compared to olive trees, with the highest number of trees found in the nearby mountainous regions (Fig. 1b). The most commonly found species are Quercus coccifera, Quercus pubescens, Quercus frainetto, Quercus conferta, Quercus Robur, and Quercus Petraea^131–134. The study area encompasses the urban and surrounding rural regions in Thessaloniki, providing a comprehensive overview of pollen dynamics.

Fig. 1.

Distribution maps of fractional coverage of (a) Olea (CORINE Land Cover 2018–100 m) and (b) Quercus (Tree species maps for European forests 1 km – EFI¹³¹, taxa in the study area.

Pollen collection and data

Airborne pollen collection and analysis are vital components of aerobiological research, essential for understanding the phenology. Pollen data are collected by the Department of Biology at Aristotle University of Thessaloniki (AUTH), using a standardized method with a 7-day recording volumetric spore trap based on the Hirst design¹³⁵. The pollen monitoring station has been in continuous operation since 1987. This trap is situated about 30 m above ground level on the rooftop of the Biology Department at AUTH, providing a representative sampling of airborne particles^5,30,117. This device, following the European Aerobiology Society’s (EAS) guidelines by the standard method EN16868:2019¹⁶⁰ and the minimum requirements of the European Aeroallergen Network (EAN) for pollen monitoring described by Galán (2014)¹³⁷, ensures consistent and reliable data collection, which is essential for phenological and public health studies. Concentrations are reported as daily pollen grains per cubic meter of air (pollen grains/m³). This technique provides a precise and detailed record of pollen presence and density, crucial for phenological studies and public health monitoring.

Daily pollen concentrations from 2016 to 2022, excluding 2018 due to data unavailability, were utilized from the Open Data provided by the Municipality of Thessaloniki (MoT). The calculation of MPS was performed using the methodology described by Wozniak & Steiner (2017)¹³⁶ to exclude smaller signals from the dataset that may be due to count errors, followed by the application of the 95% method¹⁶¹. This approach identifies the days that collectively represent 95% of the annual total pollen concentration. The MPS begins on the day when the accumulated pollen concentration reaches 2.5% of the annual total and concludes on the day it reaches 97.5%. This method offers a robust framework for understanding the phenological behavior of airborne pollen.

Meteorological data

Hourly and daily meteorological data were obtained from ERA5, the latest reanalysis product from the European Centre for Medium-Range Weather Forecasts (ECMWF). ERA5 provides a comprehensive record of the global atmosphere, land surface, and ocean waves from 1950 onwards^139,138. The variables utilized in this study include 2-meter temperature (both maximum and minimum) in °C and total precipitation in mm, spanning from 2015 to 2022 in a resolution of 0.25° x 0.25°. The monthly temperature data are presented in Fig. 2. The year 2019 identified as the hottest year, having a mean temperature of 16.72 °C, and 2015 as the coolest, with a mean temperature of 15.45 °C. The mean annual precipitation across all years is approximately 28.67 mm. Additionally, photoperiod, calculated using latitude, date, solar declination, and hour angle at sunrise¹⁴⁰.

Fig. 2.

Mean maximum and minimum monthly temperatures for the period 2015–2022.

MPS prediction methods

Thermal time models (TT)

To predict the onset and ending of MPS for Olea and Quercus taxa, TT models incorporating both chilling and heat accumulation phases were employed. The chilling models utilized were the Chilling Hours (CH)⁶⁶ and the Chilling Units (CU)¹⁴¹.

The CH model quantifies the cumulative hours of cold temperatures necessary to break dormancy in trees. According to Aron (1983)⁶⁶, chilling temperatures are defined as air temperatures below 7.2 °C. This threshold is crucial for determining the effective chilling period, which commences on the first day in autumn when the mean daily temperature drops below 7.2 °C and is followed by subsequent low temperatures. Temperatures ranging from 0 to 7.2 °C are deemed most effective, while temperatures at or below freezing are less effective in breaking dormancy. This process typically begins in late autumn, around November. For Greece, due to elevated temperatures, this process starts in early to mid-December. From this point onward, CH is calculated using Arons’s (1983)⁶⁶ empirical equation. The chilling period ends when the accumulated hours satisfy the taxa specific requirement for dormancy break, generally occurring when CH peaks, around mid-winter, often in January.

The model considers the effectiveness of chilling temperatures, which can be influenced by various environmental factors such as solar radiation, cloud cover, and wind. High solar radiation can diminish the effectiveness of chilling temperatures; for instance, clear skies during the day can elevate bud temperatures, reducing the chilling effect accumulated during the night. Conversely, overcast conditions can enhance chilling effectiveness. Weinberger (1967)¹⁴² reported that chilling under foggy conditions is nearly 50% more effective than under clear skies. Lastly, strong winds can mix the air, disrupting temperature inversions and potentially decreasing the effectiveness of chilling by warming the buds during the day and cooling them at night.

The CU model, also known as the Utah model, provides a more nuanced approach by assigning different weights to temperatures within the effective chilling range. Richardson (1974)¹⁴¹ proposed this model to quantify the chilling requirement by accumulating chilling units (CU) over the dormant period. This model divides temperatures into ranges, each contributing differently to the chilling accumulation.

• 0 to 7.2 °C: Fully effective, each hour within this range counts as one chilling unit.

• 7.2 to 12.5 °C: Partially effective, each hour within this range counts as 0.5 chilling units.

• 12.5 to 15.9 °C: Slightly effective, each hour within this range counts as 0.25 chilling units.

• 15.9 to 18.0 °C: Negligible effect, each hour within this range counts as 0.1 chilling units.

• Above 18.0 °C: Negates previously accumulated chilling units, reversing the chilling effect at a rate of − 0.5 chilling units per hour.

Positive CU begin to accumulate just after the day in the fall when the largest negative accumulation is experienced. The chilling period ends for Olea when 825 units are reached⁶⁷, while for Quercus, it ends at 790 units, assuming that value from the calculated unit range of Richardson (1974)¹⁴¹. CU provides a detailed understanding of how different temperature ranges contribute to breaking dormancy. This weighted approach allows for a more accurate estimation of the chilling requirement, especially in regions with variable winter temperatures.

Following the completion of the chilling phase, the heat accumulation phase begins. The transition from the chilling phase to the heat accumulation phase is critical for predicting the onset of flowering and subsequent pollen release. Five different methods were used to determine the heat requirements.

1. Heat Units (HU) involves summing the differences between the daily mean temperature and the selected base temperature from the end of the chilling period until the start of the pollen season¹⁴³.

2. The approach of Growing Degree Days (GDDs) uses the mean of daily maximum and minimum temperatures, subtracting a base temperature. The formula [Eq. (1)] used is:

   1

   where T_max is the maximum daily temperature, T_min is the minimum daily temperature, and T_b is the base temperature. This method simplifies the calculation by considering only temperatures above the base temperature, disregarding all others, and summing the degree days from the end of the chilling period until the onset of the pollen season^45,50,112.​

3. Ribeiro’s (2006)⁴⁸ method for GDDs involves three different conditions for the calculation. These conditions use the daily mean temperature, integrating the base temperature. This approach provides a comprehensive way to measure heat accumulation, considering different conditions to predict the phenological events more accurately.

4. The Snyder (1985)¹⁴⁴ approach, uses a trigonometric sine curve to approximate diurnal temperature variations for calculating GDDs. This method integrates the area under the sine curve above a base temperature, allowing for an accurate estimation of heat accumulation necessary for plant development. The process involves collecting daily maximum and minimum temperatures, calculating the average daily temperature and temperature range, and then integrating the area under the sine curve above the base temperature using Snyder’s (1985)¹⁴⁴ formula. This approach adjusts the GDDs calculation by considering only the relevant heat accumulation and offers advantages in accuracy and flexibility, handling single or multiple threshold temperatures and accounting for daily temperature fluctuations.

5. The accumulated maximum temperature method, ΣT_max, involves summing the daily maximum temperatures above the base temperature⁴⁵.

The base temperatures can vary depending on the specific approach, the study conducted, and the on bio-geographical characteristics of the studied area⁵⁰. Studies conducted using phenological observations in the Mediterranean region, particularly on the Iberian Peninsula, have shown that the base temperature values for Olea range between 0 °C and 12.5 °C^{45,48,52,64,69,103,105}, while for Quercus, they range between 6 °C and 11 °C^7,81,83,108. In Greece, no similar study has been conducted especially for calculating threshold temperatures. Consequently, a base temperature of 5 °C was selected for olive trees, which is close to the mean of the reported values and consistent with the studies of Frenguelli (1989)¹⁰⁴, Alba (1998)¹⁴⁵ and Sofiev (2017)¹⁰³. For oaks, considering that the aforementioned studies focus on cities in Spain such as Malaga, Ourense, and Cordoba, which exhibit higher mean temperatures at the beginning of spring when heat accumulation starts (Compare the Climate and Weather in Thessaloníki, Málaga, Córdoba, and Ourense - Weather Spark), a base temperature of 3.5 °C was chosen. This value has been previously selected for similar forest tree species¹⁴⁶.

The chillR package in R was used to facilitate the calculation of chilling and heat requirements¹⁴⁷. The chillR package provides functions for processing temperature records into chilling and heat units, while it includes functions for generating and interpolating weather scenarios, conducting climate change analyses, and performing regression methods to relate phenological dates to temperature metrics. It is employed for the calculation of CU and GDDs utilizing the functions ‘chilling’, which calculates chilling and heat units from hourly temperature records, and ‘GDD’, which computes growing degree days based on specified base temperatures.

Partial least squares regression (PLS)

To develop predictive models for the main phenological phases of the trees, PLS regression was employed¹⁰¹. This statistical technique is effective for managing multicollinearity among predictors and performs well with datasets containing numerous explanatory variables. PLS regression was selected to maximize the covariance between the predictors (meteorological parameters) and the dependent variables (start and end MPS dates), resulting in an optimal linear model for forecasting. The meteorological variables were categorized seasonally: summer (June, July, and August), autumn (September, October, and November), and winter (December, January, and February). The PLS regression models were constructed using these seasonal meteorological variables along with bioclimatic indices, such as seasonal ombrothermic indices¹⁰¹. Each latent variable in the PLS model represented a key factor influencing the start and end of MPS dates for Olea and Quercus taxa. This methodology is being applied to Quercus taxa for the first time. All variables were standardized to ensure equal statistical weight across all phenological sampling points. Consequently, these models facilitate the study of the influence of climate on the biological development of trees, particularly in the context of climate change.

Temperature – photoperiod models (TP)

To predict the phenological stages of pollen taxa, TP models were implemented. These models integrate temperature accumulation and photoperiod thresholds, as described by¹⁴⁸. The development of reproductive buds in trees is characterized by the state of forcing (S_f), which accumulates daily based on the rate of forcing (R_f). The rate of forcing is calculated using the daily mean temperature (x_t), with equations adapted to the start and end dates of season [Eq. (2)]:

2

where d_{s, e}, and e_{s, e} are fitted parameters. The starting of the MPS occurs when S_f reaches a critical value (F*). The starting day for the accumulation of the rate of forcing (t₀) is determined by a critical photoperiod (P₀), which varies with the latitude of the studied area. This ensures that the model adapts to different geographical locations. The state of forcing (S_f) is given in Eq. (3).

3

The TP models were implemented using meteorological data, including daily temperature records and photoperiod calculations. Model parameters (F*, P₀, d_{s, e}, e_{s, e}) were fitted using the least-square method. The inclusion of a critical photoperiod as a threshold for starting temperature accumulation significantly improves the accuracy of the predictions, providing a valuable tool for forecasting phenological events over wide geographical scales.

Results and discussion

Analysis and predictive modeling of MPS

In the region of Thessaloniki, a comprehensive analysis spanning a 15-year period³⁰ and another study covering the past two decades⁵ have demonstrated that the total pollen spectrum of the city is significantly influenced by various taxa¹⁴⁹. Cupressaceae accounts for 24.9% of the total pollen, Urticaceae 13.6%, while Quercus (20.8%) and Olea (9.1%) rank second and fourth, respectively. Notably, Olea pollen exhibits high allergenicity, affecting 31.8% of the sensitive population³⁰.

The analysis of the MPS from 2016 to 2022 in Thessaloniki (Table 1), Greece, provides detailed data on the start, peak, and end of the MPS (days counting from 1st January), measured in Day Of Year (DOY), along with the season duration and the annual pollen index (APIn) as the annual sum of daily pollen concentrations¹⁵¹. The MPS for Olea in Thessaloniki commences in mid- to late March and concludes in early June, with daily concentrations remaining below 60 pollen grains/m². Indeed, the daily mean concentration for Olea, averaged over a calendar year, in Thessaloniki, has been found to be 2 pollen grains/m³, with a standard deviation of 9.6 pollen grains/m³, for a thirty year period (1987–2016¹⁵⁶). For Quercus, the MPS extends from mid-April to late-May, with daily concentrations in 2016 and 2021 reaching up to 670 pollen grains/m²¹²⁰. For olive trees, the earliest season start was observed on April 2 (DOY 22), 2022, with the season extending until June 11 (DOY 161), making it the longest season recorded within the studied period, although it did not correspond with a high APIn value. The highest APIn value was recorded in 2019, which also experienced the highest peak concentration at 152 pollen grains/m², more than twice the average peak value of 60 pollen grains/m².

Table 1.

Start, peak, end (DOYs), duration (days) of the MPS, and API (pollen grains/m³) for Olea and Quercus in Thessaloniki during the studied years.

Taxon	Year	Start (DOY)	Peak (DOY)	End (DOY)	MPS Duration (days)	APIn (pollen grains m− 3)
Olea	2016	94	130	142	48	352
	2017	100	135	155	55	249
	2019	99	148	157	58	499
	2020	98	145	154	56	351
	2021	89	142	151	62	267
	2022	92	145	161	69	284
Quercus	2016	83	110	138	55	4675
	2017	95	107	133	38	4117
	2019	102	121	150	48	2683
	2020	105	117	149	44	2043
	2021	97	121	147	50	5956
	2022	111	117	158	47	1160

Start: Beginning day of MPS; Peak: Day of highest pollen concentration in the MPS; End: Final day of MPS; MPS Duration: Total length of MPS; APIn: Annual Pollen Index in concentration (pollen grains/m³)¹⁵⁰.

In Thessaloniki, the mean MPS duration is approximately 58 days within the studied period, increasing to 77 days for a longer period of 30 years on the basis of a previous study¹⁵². The peak concentration usually occurs around mid-May, aligning with Cordero (2021)¹⁵⁰, who noted that the peak in Madrid occurs in the second half of May. Galera (2018)¹⁵³ found the highest average values in weeks 20 and 21 (May) in Cartagena, Spain. In addition, Aguilera (2015)⁷³ reported a peak in Andalusia on May 19, and similar peaks were observed across the Mediterranean Basin^75,154. The highest APIn value recorded in Thessaloniki is significantly lower than the average APIn value observed in Florence¹⁵⁵. These findings show a notable variation in pollen dynamics when compared to other Mediterranean locations, highlighting regional differences in phenological responses influenced by local conditions.

Regarding the Quercus taxa, in contrast to Olea, the mean APIn is 3439 pollen grains/m², indicating a significant abundance of its pollen grains in the air of Thessaloniki. The year 2016 recorded the highest APIn and had the longest duration of the MPS. The earliest start of the pollen season was observed on March 24th, 2016, while the latest end of the season was in 2022. Despite the extended season in 2022, the pollen index was at its lowest value. The peak of the MPS typically occurs near the end of April, with the mean peak count exceeding 300 pollen grains/m².

Comparative studies from the Iberian Peninsula highlight higher pollen production levels than Thessaloniki. For instance, Recio (2018)⁷ reported APIn values ranging from 1,925 to 9,643 pollen grains/m² in Malaga, with peak daily concentrations between 125 and 1,578 pollen grains/m². The earliest peak was recorded on March 25, and the latest on May 20. Similarly, López-Orozco (2021)¹⁵ reported an average APIn of 13,535 pollen grains/m² in southern Spain, while Fernández-Rodriguez (2016)¹⁵⁶ documented an average APIn of 14,308 pollen grains/m². These substantially higher APIn values suggest more favorable climatic conditions for Quercus pollen production and dispersal in southern Spain compared to Thessaloniki.

The analysis of predictive models for Olea indicates that the models are generally effective in forecasting the MPS (Table 2). The PLS model excels in predicting the starting dates, demonstrating exceptional accuracy due to its optimization capabilities. For predicting the ending phases, both the TT (vi) and PLS models perform well. In the case of Quercus, the TT methodologies exhibit minor deviations and satisfactorily capture the MPS. The PLS method stands out, demonstrating exceptional accuracy in predicting both the onset and ending phases. Specifically, the TT (iv and v) models and TP perform very well in predicting the end of the MPS. However, certain years present challenges for all models. For Olea, no model accurately predicts the start and end of the MPS for the year 2022, which recorded the longest season observed, while the average duration of MPS is 58 days. Similarly, for Quercus, the early start in 2016 and the shortest season end in 2017 were not well predicted by any model. This indicates a limitation in the models’ ability to accurately forecast MPS during years with significant deviations from the average conditions. Overall, the PLS method performs consistently well across the two different taxa. Given the varying strengths of each model, a combination of methods—particularly TT and PLS—would likely provide the most robust predictions for future MPS analyses. Utilizing these methods together can ensure more reliable and accurate predictions for both the start and end of the pollen seasons, thereby improving the forecasting accuracy.

Table 2.

Observed DOYs (Obs.) and predicted deviations from observed values for the start and end dates of pollen seasons for Olea and Quercus taxa in Thessaloniki.

Taxon		Year	Obs. (DOY)	TT											PLS	TP
				i	ii	iii	iv	v	vi	vii	viii	ix	x	xi
Olea	Start	2016	94	− 8	− 9	− 8	− 10	− 1	0	− 1	− 1	− 1	+ 3	− 4	0	− 4
		2017	100	+ 2	+ 1	+ 2	0	+ 6	0	+ 1	+ 1	0	+ 2	+ 4	0	0
		2019	99	− 1	0	+ 1	− 1	+ 2	− 4	− 1	− 1	− 1	− 4	+ 3	0	− 6
		2020	98	0	+ 5	+ 5	+ 4	+ 2	+ 2	+ 5	+ 5	+ 5	+ 2	+ 9	0	− 7
		2021	89	+ 4	+ 6	+ 7	+ 5	+ 8	+ 3	+ 16	+ 15	+ 17	+ 4	+ 22	0	+ 2
		2022	92	+ 13	+ 15	+ 13	+ 14	+ 15	+ 9	+ 24	+ 23	+ 23	+ 20	+ 26	0	− 4
	End	2016	142	+ 1	+ 2	+ 3	+ 3	+ 3	+ 7	+ 7	+ 3	+ 8	+ 10	+ 5	+ 5	− 1
		2017	155	0	− 1	− 1	− 1	+ 1	0	− 1	− 4	0	+ 1	0	+ 1	− 2
		2019	157	− 4	− 2	− 2	− 1	− 3	− 4	− 2	− 5	0	− 3	0	0	− 3
		2020	154	0	+ 4	+ 4	+ 4	− 2	− 2	+ 5	+ 1	+ 6	+ 1	+ 6	+ 1	+ 4
		2021	151	− 1	+ 1	+ 1	+ 3	− 1	+ 6	+ 7	+ 4	+ 9	+ 6	+ 9	+ 2	+ 2
		2022	151	+ 11	+ 10	+ 10	+ 10	+ 11	+ 9	+ 10	+ 6	+ 10	+ 9	+ 11	+ 9	+ 11
Quercus	Start	2016	83	+ 4	+ 3	+ 1	0	+ 6	+ 11	+ 11	+ 11	+ 11	+ 14	+ 5	+ 4	+ 5
		2017	95	+ 5	+ 6	+ 5	+ 3	+ 5	+ 4	+ 6	+ 7	+ 6	+ 4	+ 6	+ 1	+ 6
		2019	102	− 4	− 4	− 4	− 6	− 6	− 5	+ 3	− 3	− 3	− 6	− 4	− 1	− 2
		2020	105	− 5	− 4	+ 5	− 6	− 8	− 3	− 2	− 1	− 2	− 4	− 2	− 2	− 6
		2021	97	− 3	− 3	− 4	− 5	− 4	+ 7	+ 10	− 10	+ 10	− 1	+ 3	0	+ 3
		2022	111	+ 3	+ 5	+ 4	+ 3	0	− 5	+ 3	− 3	+ 2	− 7	+ 3	− 4	− 9
	End	2016	138	− 1	− 2	− 25	− 3	+ 1	+ 5	+ 2	+ 1	+ 2	+ 7	+ 2	+ 3	− 5
		2017	133	+ 14	+ 14	+ 14	+ 13	+ 16	+ 14	+ 12	+ 15	+ 12	+ 15	+ 16	+ 4	+ 9
		2019	150	− 3	− 3	− 2	− 4	− 2	− 3	− 4	− 4	− 4	− 2	0	− 2	− 1
		2020	149	− 1	− 1	0	− 2	− 3	+ 1	− 1	0	0	0	+ 4	− 1	0
		2021	147	− 3	− 2	− 2	− 3	− 2	+ 4	+ 2	+ 2	+ 2	+ 4	+ 7	− 1	+ 3
		2022	158	− 3	− 3	− 3	− 4	− 3	− 6	− 6	− 6	− 6	− 7	+ 10	− 5	− 8

Start: Beginning day of MPS; End: Final day of MPS; TT: Thermal Time models. ⁱChilling Hours (CH), Heat Units (HU), ⁱⁱChilling Hours (CH), Growing Degree Days (GDDs), ⁱⁱⁱChilling Hours (CH), Growing Degree Days (GDDs)⁴⁸, ^ivChilling Hours (CH), Growing Degree Days (GDDs)¹⁵⁷, ^vChilling Hours (CH), Accumulated Maximum Temperature (Tmax), ^viChilling Units (CU), Heat Units (HU), ^viiChilling Units (CU), Growing Degree Days (GDDs), ^viiiChilling Units (CU), Growing Degree Days (GDDs)⁴⁸, ^ixChilling Units (CU), Growing Degree Days (GDDs)¹⁵⁷, ^xChilling Units (CU), Accumulated Maximum Temperature (Tmax), ^xiChillR package; PLS: Partial Least Squares Regression; TP: Temperature-Photoperiod model

The effectiveness of the TP model, particularly in predicting the start of the Olea pollen season and the MPS of Quercus, is attributed to the model’s ability to accurately determine the state of forcing (Eq. 3) using appropriate parameters (Table 3). The photoperiod (P₀) and forcing temperature (F*) values in the current study are higher than those reported in the models of Garcia-Mozo (2009)¹⁴⁸, which range from 9.49 to 13.63 for P₀ and from 34.9 to 126.0 for F* across different localities in Spain and Italy. This suggests that in Thessaloniki, a longer daylight period and different thermal thresholds are necessary to initiate the pollen season for Olea and Quercus taxa, indicating a unique regional adaptation. Additionally, the parameters d_s and e_s exhibit positive values, in contrast to the typically negative values found in Garcia-Mozo’s (2009)¹⁴⁸ models. The d_e parameter is closer to zero, suggesting that this factor might be less influential in Thessaloniki. The e_e values, however, fall within the observed range, indicating some consistency in the end-of-season dynamics across different regions.

Table 3.

Parameters of TP model for Thessaloniki.

Taxon	P o	ds	de	es	ee	F*
Olea	13.5	5.0	− 0.40	− 13.7	− 12.3	80.1
Quercus	13.8	3.0	− 1.0	− 15.1	− 14.4	70.5

P₀: Photoperiod; d_s: model parameter for the start; e_s: model parameter for the start; d_e: model parameter for the end; e_p: model parameter for the end; F*: Forcing temperature

Analysis of PLS regression

The PLS regression analysis reveals that both Olea and Quercus pollen seasons are significantly influenced by temperature variables across different seasons. Total precipitation also affects the timing of MPS, as seen in Fig. 3. For the start of the Olea, critical variables include the maximum and minimum temperatures during the spring season. These results are consistent with previous studies which found that temperatures significantly impact phenological events^52,97 and that start dates of the MPS strongly depend on spring temperatures^{46,91,104,145,150,158}. Concerning the end of, the PLS model identifies winter temperatures as significant predictors. This may suggest that the limited dataset size can amplify the influence of certain variables, rendering meteorological factors statistically significant that might not otherwise affect the season’s end. In the case of Quercus season, the analysis indicates that winter temperatures and April precipitation, the month preceding the onset of flowering, are crucial for predicting the start of the MPS. This aligns with findings from Recio (2018)⁷, who found significant associations between Quercus pollen seasons and spring meteorological conditions, including positive correlations with temperature. However, it differs from Kasprzyk (2009)⁸⁴, who reported that the start of oak pollen seasons is significantly correlated with March minimum temperatures. This discrepancy is likely due to climatic differences, as Kasprzyk’s⁸⁴ study was conducted in Poland and included Quercus species such as Q. rubra, Q. coccinea, and Q. palustris, which are not abundant in Greece. Instead, the Quercus species abundant in Greece, such as Q. robur and Q. petraea¹³¹, also contribute to the pollen season patterns observed in this study. Similar to the Olea season, the end is best predicted by considering the precipitation and temperatures of the previous winter. These insights improve the accuracy of pollen season predictions, particularly regarding the onset of MPS, and underscore the need for further research into the effects of seasonal temperature and precipitation patterns in ending of the seasons, as explored in the study by¹¹⁸.

Fig. 3.

Regression coefficients of important variables for Olea and Quercus taxa.

Peak prediction of MPS

Figure 4, presents a comparison of the observed peak DOYs for Olea and Quercus with PLS and TP models for the years 2016 to 2022. The PLS model closely matches the observed peak days for both Olea and Quercus, typically deviating by only a few days. For instance, in 2016, the PLS model predicts the peak for Olea at DOY 135, slightly later than the observed 131, and for Quercus, it predicts 110, almost identical to the observed 111. The TP model also demonstrates good predictive capability but with slightly larger deviations in some years for both taxa. For example, in 2022, the TP model predicts a peak of 148 for Olea, which is later than the observed 145, and a peak of 125 for Quercus, which is later than the observed 117. Overall, the PLS model consistently shows closer alignment with the observed values, indicating it may be more reliable for predicting peak pollen seasons In Thessaloniki.

Fig. 4.

Comparison of peak DOYs for Olea and Quercus taxa with predicted values from PLS and TP models.

Chilling and heating requirements

The analysis of chilling dates using the CH and CU approaches provides insights into the periods during which chilling starts and ends for Olea and Quercus in Thessaloniki (Table 4). The CH methodology, a common approach used for trees requiring dormancy, indicates the same onset and end of the chilling period for these trees. In contrast, the CU approach, refined using the methods of Orlandi (2002)⁶⁷ and Richardson (1974)¹⁴¹, offers a more tailored determination of the end of the chilling season for these specific taxa, as detailed in Section “Thermal time models (TT)”.

Table 4.

Chilling dates for Olea and Quercus using CH and CU approaches.

	CH		CU
Year	Start	End	Start	End
				Olea	Quercus
2016	06/12/15	02/01/16	24/10/15	17/01/16	15/01/16
2017	29/11/16	01/01/17	28/10/16	06/01/17	03/01/17
2019	29/11/18	02/01/19	25/10/18	14/01/19	11/01/19
2020	04/12/19	09/01/20	23/11/19	21/01/20	19/01/20
2021	01/12/20	02/01/21	07/11/20	20/01/21	15/01/21
2022	13/12/21	10/01/22	12/10/21	07/01/22	05/01/22

Chilling Hours (CH); Chilling Units (CU); Start: Beginning day of MPS; End: Final day of MPS.

Based on the CH approach, the chilling period for both Olea and Quercus typically begins in late November to early December each year, indicating a relatively stable pattern of chilling initiation during early winter. The end of the chilling period generally occurs around early January and the mean value required for the fulfilment of chilling is 1187 h. The CU approach, indicates a chilling start from mid-October to late November, with the end of the chilling period varying between early and late January.

In comparison, various studies have highlighted the differences in chilling requirements and periods across different regions. For Quercus, Garcia-Mozo (2000)⁸³ reported that in Cordoba, the chilling requirement is achieved from mid-December to the end of January, with a mean value of 348 h. Similarly, Jato (2002)¹⁰⁸ found that in Ourense, Spain, the chilling period usually starts in mid-November, with the mean start date being November 27th. The chill requirement in Ourense was typically reached during the first 15 days of January, with mean chilling hours amounting to 873 h. For olive trees, Orlandi (2002)⁶⁷ documented that the average chilling requirement for a Spanish cultivar was approximately 997 h, about half of the 1848 h recorded for an Italian one. This indicates significant regional variability. In addition, there is some alignment in the chilling periods of Thessaloniki with Rojo (2020)¹⁵⁹, who identified the critical period for chilling accumulation in Toledo from mid-November to late December (November 10 to December 20). Galán (2001)⁴⁵ also reported that in Cordoba, chilling for Olea occurs from mid-December to the end of January, with a mean chilling time of 348 h, further illustrating regional variations in chilling periods and requirements.

The heating requirements for the start and end of the MPS in Thessaloniki, as presented in Table 5, provide valuable results into the thermal conditions necessary for the phenological events in Olea and Quercus taxa. The table outlines the mean heating values calculated using TT models, presented in Table 2. This study introduces the use of heat accumulation for predicting the end of the season for the first time. For the start of the season, the heating requirements for Olea range from 435 °C to 1113 °C, and for Quercus, from 544 °C to 1081 °C. The v and x models show the highest values due to the Tmax accumulation method used. For the end of the season, the heating requirements for Olea range from 1060 °C to 2164 °C, and for Quercus, from 1163 °C to 2063 °C.

Table 5.

Heating requirements for the start and end of MPS.

Taxon		TT models
		i	ii	iii	iv	v	vi	vii	viii	ix	x	xi
Olea	Start	824	445	452	446	1113	782	435	438	440	1028	471
Quercus		849	561	550	544	1081	825	563	570	566	1067	574
Olea	End	1751	1106	1112	1123	2164	1735	1110	1060	1135	2145	1133
Quercus		1653	1186	1196	1177	2063	1646	1163	1169	1172	2035	1241

All values are expressed in °C.

Start: Beginning day of MPS; End: Final day of MPS; Thermal Time (TT) models; ⁱChilling Hours (CH), Heat Units (HU), ⁱⁱChilling Hours (CH), Growing Degree Days (GDDs), ⁱⁱⁱChilling Hours (CH), Growing Degree Days (GDDs)⁴⁸, ^ivChilling Hours (CH), Growing Degree Days (GDDs)¹⁵⁷,. ^vChilling Hours (CH), Accumulated Maximum Temperature (Tmax), ^viChilling Units (CU), Heat Units (HU), ^viiChilling Units (CU), Growing Degree Days (GDDs), ^viiiChilling Units (CU), Growing Degree Days (GDDs)⁴⁸, ^ixChilling Units (CU), Growing Degree Days (GDDs)¹⁵⁷, ^xChilling Units (CU), Accumulated Maximum Temperature (Tmax), ^xiChillR package.

Analysis with other Mediterranean studies reveals significant differences. For Olea, Sofiev (2017)¹⁰³ reported GDDs of 275 °C in the Mediterranean Basin. Galán (2001)⁴⁵ in Cordoba, found that the average accumulated HU needed to induce flowering was 119.5 °C days, and an average heat accumulation expressed as GDD was 209.9 °C. Alba (1998)¹⁴⁵ in Spain noted HU mean values of 656–881 °C and GDDs averaging 778 °C. Ribeiro (2006)⁴⁸ in Portugal identified a mean HU of 1597 °C. For Quercus, Garcia-Mozo (2000)⁸³ found HU of 76 °C and mean GDDs of 127.3 °C in Cordoba. Recio (2018)⁷ in Malaga, observed an accumulation of approximately 796 °C GDDs, while Jato (2002)¹⁰⁸ in Ourense, Spain, reported an accumulated Tmax mean value of 935 °C and GDDs mean value of 263 °C. These values are generally lower than those found in Thessaloniki, consistent with the lower base temperatures used in this study (Section “Thermal time models (TT)”).

Conclusions

This study provides a comparative analysis of the pollen seasons of Olea and Quercus taxa in Thessaloniki, Greece, employing a range of predictive models to enhance the accuracy of MPS forecasts. The findings reveal significant local climatic influences on the pollen seasons, with Thessaloniki exhibiting longer MPS compared to other Mediterranean regions. The application of multiple modeling approaches, including TT, PLS, and TP models, demonstrates robust predictive capabilities for both the start and end of the pollen seasons. The PLS model shows accuracy in predicting the starting dates for Olea and performs well for both the start and end phases of Quercus, while TT models, particularly models iv and v, along with the TP model, are effective in predicting the end of the MPS for Quercus. Combining TT and PLS methods is recommended for more robust and accurate future predictions of MPS, enhancing the overall forecasting reliability. Chilling and heating requirements were critical factors, with the chilling period typically starting in late autumn and ending in early winter, while the heating requirements varied significantly, influencing the onset of the pollen seasons. PLS regression analysis identified key meteorological variables, such as spring and preceding winter temperatures and precipitation, significantly affecting MPS timing. These insights are vital for developing effective public health strategies to mitigate the impacts of pollen-related allergies and for advancing our understanding of the ecological and climatic factors influencing pollen dynamics in the Mediterranean region. Future research should focus on refining these models and expanding their application to other taxa and geographic areas to further improve predictive accuracy and public health outcomes.

Author contributions

Conceptualization, S.P. and D.M.; methodology, S.P.; investigation, S.P.; formal analysis, S.P.; resources, S.P.; writing—original draft preparation, S.P., K.K. and D.M.; writing—review and editing, S.P., K.K., S.K., A.P. and D.M.; supervision, D.M.; project administration, D.M.

Funding

The research leading to these results received funding from “Support for enhancing the operation of the National Network for Climate Change” (CLIMPACT), National Development Program, General Secretariat of Research and Innovation (2023ΝΑ11900001 - Ν. 5201588).

Data availability

The pollen observations of the Department of Biology of the Aristotle University of Thessaloniki are provided from the MoT at https://envdimosthes.gr/fisika-aerioallergiogona/, accessed on 23 January 2025.

Declarations

Competing interests

The authors declare no competing interests.