Forecast models and trends for the main characteristics of the Olea pollen season in Nice (south-eastern France) over the 1990–2009 period

Abstract

The olive tree, Olea europaea, is very common in the southeast of France and its pollen is recognised as one of the most important allergenic in the Mediterranean. This study allowed for the development of predictive models to calculate the main characteristics of the Olea pollen season over the last 20 years from a wide range of meteorological variables. Clear evidence of the relationship between the main features of the Olea pollen season and the temperature recorded during the months before the flowering period could be demonstrated. The mean temperature in February plays an important role in determining reproductive growth and anthesis. It seems that the mean temperature during autumn influences the pollen index of the next year's pollen season. Other environmental factors, such as global radiation and rainfall, may be of great influence in determining the onset and final date of pollination. The accumulated rainfall amount during the pollination period has a negative effect on the pollen index. This may be interpreted as the wash out of airborne pollen by raindrops. However, rainfall during the vegetative period has a positive effect on pollen production. The pollen quantities depend not only on meteorological conditions before pollen release, but also maybe on those prevailing during pollen release. Finally, we could demonstrate an upward trend in annual pollen production and a stability of the mean duration of the Olea pollen season. The increase in Olea pollen abundance coincides with a rise in air temperature over the last 20 years.

Keywords:

• aerobiology
• climate change
• forecast
• Mann–Kendall test
• Olea pollen

The olive tree, Olea europaea L. is one of the most characteristic features of Mediterranean flora and very common in south-eastern France. Its pollen has been recognised as one of the most important allergenic in the Mediterranean, where olive pollinosis is a widespread form of respiratory allergic disease (Hoxha, 2007). Advanced knowledge about annual crop production is highly desirable from both biological and economic points of view (Galán et al., 2004). Olive flower phenology is characterised by bud formation during summer, dormancy during autumn and early winter, budburst in late winter (February) and flowering in late spring (May–June). Weather conditions directly influence pollen counts by affecting pollen production, the beginning of flowering, the abundance of vegetation and by controlling the amount of pollen released daily into the air. One of the most important aspects of aerobiological studies is to obtain prediction models helping to establish the onset and characteristics of a pollen season. The forecast of the start of the pollen season has a particular importance because this information is very useful for accurate use of medicine for allergies and for planning of the patient's activities. Several studies have been published based on different methods for forecasting the start of the pollen seasons (Frenguelli et al., 1989; Emberlin et al., 1997, 2002; Alba & Díaz de la Guardia, 1998; Chuine, 2000; Fornaciari et al., 2000, 2005; Laaidi, 2001; Galán et al., 2001, 2005; Rodríguez-Rajo et al., 2003; Orlandi et al., 2004, 2005; Hoxha, 2007; Ribeiro et al., 2007; García-Mozo et al., 2008).

The main goal of this study was to investigate the influence of weather conditions on pollen counts by influencing pollen production, the beginning of anthesis, the abundance of vegetation and the amount of pollen released. Therefore, the development of predictive models of the Olea pollen season on the basis of pollen and weather data compiled over the last 20 years (1990–2009) for Nice in south-eastern France was analysed. Long-term trends in pollen atmospheric levels in Nice were also examined in order to study the relationship between climatic trends, climatic changes and the biological behaviour of the olive tree pollen as a possible bio-indicator species in order to investigate its potential to adapt to future climatic scenarios.

Materials and methods

The daily pollen count is provided by the French aerobiology network (RNSA) that is in charge of the analysis of biological particles in the air. In Nice, pollen data were recorded in the years 1990–2009 using a Hirst-type volumetric spore pollen trap, which was located on the roof of the faculty of medicine (7° 15′ E; 43° 42′ N), around 63 m above sea level (Figure 1). Nice has a Mediterranean climate with annual precipitation of 800 mm, comparatively high temperatures in winter (9.1 °C) and fresh ones in summer (maximum of 27.3 °C in July). Winter is characterised by mild days (11–17 °C), cool nights (4–9 °C) and variable weather. Days can either be sunny and dry or damp and rainy. Spring starts mild and rainy in late March, and is increasingly warm and sunny towards June.

Figure 1. Map of France with Nice in the southeast.

The region is exposed to large air pollution, ranging among the most influenced areas by the atmospheric pollution in Europe, and is characterised by an increased allergic risk for pollen due to the presence of a large number of cypress, plane and olive trees (Sicard et al., 2011). The pollution is partially caused by road traffic increase combined with the strong hot season of the Mediterranean climate. These different factors produce serious human health problems.

Pollen grains have been expressed as an average of pollen grains per m³ of air. In the case of meteorological data, they are given as sum and average for periods of ten, 20 and 30 days. Meteorological data were supplied by the French national meteorological service (Météo-France) from the pilot site situated near to the pollen trap [daily maximum (T _x) and minimum (T _n) temperatures in °C), relative humidity (RH in percentage), rainfall (R in millimetres), global radiation (GR in J/cm²) and mean temperature T _mean based on 24 hourly observation of T _x and T _n values]. In accordance with Galán et al. (2001) and García-Mozo et al. (2008), we defined the start of Olea pollen season as the date on which one pollen grain per m³ was recorded on five consecutive days (‘Method 1’). The pollen index is defined as the sum of the daily average pollen count per m³ of air during the pollen season. The annual pre-peak pollen index (PPI) is the sum of the daily pollen count in the air from the start of pollination to the peak date. The different dates are represented in term of number of days from 1 January.

Statistical methods used

In order to forecast the main characteristics of the Olea pollen season in Nice, we used various methods to analyse meteorological and pollen data for a period of 20 years (1990–2009).

Given the number of observations, a principal components analysis (PCA) was performed to detect significant relationships between aerobiological and meteorological variables. The results were used as a classification method of additional influencing variables on main characteristics of the Olea pollen season. The PCA allows for the detection of causal factors. Pearson correlation coefficients were calculated to test the relationship between the main characteristics of pollen season and meteorological variables. The stepwise regression analysis is conceived to build a statistical model describing the impact of meteorological factors on the main characteristics of the pollen season. We used an upward selection (or downward selection) consisting in starting with a model integrating only a constant and adding variable, one at a time, if they significantly improve the curve fitting. The root mean square error (RMSE) was used to measure the differences between values predicted by a model or an estimator and the actual values from the thing being modelled or estimated. The mean absolute error (MAE) is a quantity used to measure how close forecasts or predictions are to the eventual outcomes. To rank the goodness-of-fit of the models, Akaike's information criterion (AIC) has been computed. The AIC is based on the mean squared error in attempting to forecast one period beyond the end of the data, penalised by the number of parameters that need to be estimated. The Durbin–Watson statistic (DW) tests residuals in order to determine if there is a significant correlation based on the order in which they appear in the data file.

Forcing units

Heat requirement can be calculated and expressed in different ways; one of them are the growth degree-days (GDD in °C). The accumulation of heat is called ‘physiological time’ and degree-days are a measure of physiological time. One degree-day is equal to one degree above the threshold temperature (T _L) during 24 hours. Therefore, there are no degree-days when the threshold temperature is higher than the daily maximum temperature. When the threshold is lower than the daily minimum temperature, the number of degree-days can be estimated as the mean of the maximum and minimum temperatures minus the threshold, calculated according to the equation proposed by Rickman et al. (1983). The heat requirement (HR) was estimated as a function of the sum of the daily maximum temperatures from the end of the chilling period. The end date of the chilling period is fixed, thus we calculated the forcing units from 1 January and 1 March.

Detection and estimation of annual and seasonal trends

The Mann–Kendall test is a non-parametric statistical test to detect the presence of a monotonic increasing or decreasing trend within a time series in absence of any seasonal variation or other cycles. Many data in air quality have a seasonal variation. The Mann–Kendall test is, accordingly, limited to annual data to be free from the problem of seasonal variation and autocorrelation. The test is used for four different significance levels α (0.1, 0.05, 0.01 and 0.001). The literature using the Mann–Kendall test, recommended by the World Meteorological Organisation (WMO), is abundant and universal (Lehmann et al., 2005; Sicard et al., 2007, 2009, 2010). To estimate the trend, a consistent non-parametric estimator for the coefficients of a linear regression was suggested and modified by Sen (1968) to include the possibility of ties in the time series. In general, we calculate the confidence interval with two levels ϵ = 0.01 and 0.05 (99 and 95%) resulting in two different confidence intervals. The Seasonal Kendall test is an extension of the Mann–Kendall test for the detection of trends (Hirsch et al., 1991). The Seasonal Kendall test may be applied to data presenting some seasonality. This test also performs reasonably well when both the mean and the variance are seasonal. For estimating the magnitude of trends, the Seasonal Kendall slope estimator may be used (Gilbert, 1987). The Seasonal Kendall slope estimator is a generalisation of Sen's (1968) estimator of slope discussed in the previous paragraph.

Results

From PCA results, probability, RMSE, MAE and AIC values, we tried to predict main characteristics of Olea pollen season by means of linear regression studies, using the meteorological variables as pollen concentration estimators. The variables that gave the highest correlation were included in the model. In order to calculate the accuracy of the selected models, we selected two random years (1995 and 2009) to compare the predicted value, or date, by the selected model with the observed value.

The sum of Olea daily pollen count, as an average over the 1990–2009 period in Nice, was 2502±1248 pollen grains/m³. The year 1994 has a higher total sum with 5822 pollen grains/m³ and the year 1993 has the lowest pollen concentrations with only 978 pollen grains/m³. The Olea pollen is present in the air from May to July. The start of Olea pollen season varied considerably from one year to the next; it ranged from 10 April to 10 May. On average, the beginning of the season appears to occur on 25 April (n = 115 days) and the end on 16 July (n = 197 days). The mean length of Olea pollen season was 82 ± 22 days. The mean peak value was 320 ± 193 pollen grains/m³ and the mean peak date appears to occur on 31 May (n = 151 ± 8 days). The year 1994 has the higher peak value with 831 grains/m³ while the year 1993 has the lowest peak value with only 86 pollen grains/m³. The mean duration between onset and peak date was 36 days. The PPI varied between 215 pollen grains/m³ (1993) and 3578 pollen grains/m³ (1994) with an average of 1504 ± 876 pollen grains/m³ over the given period. Finally, the PPI represented about 57.6 ± 15% of the annual pollen count. Olea pollen constituted the fourth important number of pollen over the study years with a percentage that varied from 3.9% (2006) to 8.2% (2009).

Different threshold temperatures (4, 4.5, 5, 5.5, 6, 6.5, 7, 7.5, 8, 9 and 10 °C) were used to estimate the most accurate to estimate the GDD and the HR. We calculated the forcing units from 1 January and 1 March. The lowest standard deviation and variation coefficient were used to identify the most adequate threshold temperature. In this study, the threshold temperature is fixed to 5 °C and the 1 March as the beginning of the heat accumulation.

Characteristics of the Olea pollen season

Accumulative methods

Several cumulative methods have been used by different authors to predict the start of the Olea pollen season (e.g. Frenguelli et al., 1989; Galán et al., 2001). The methods having the highest correlation are presented in this study. First, we used the GDD and HR calculated from 1 March. It attempts to predict the start of the Olea pollen season using a mean of accumulated growing-degree days during the period of our study. Finally, the variable of accumulated maximum and mean daily temperature, based on T _n and T _x values, from 1 March have been studied.

The accumulated temperature needed to induce the onset of flowering ranged between 425 and 908 °C for Olea according to the used method (Table I). The variable accumulated maximum temperature from 1 March until the start date of Olea pollen season (r = 0.97) successfully predicted the start of the Olea pollen season with a delay of one day in 1995 (accuracy 99.1%). The accuracy achieved for 2009 was 89.1%. The maximum cumulative temperatures in the months preceding the Olea pollination could therefore play an important role. The relationship between the start of the Olea pollen season and the maximum temperature recorded during the months before the flowering period was clearly evidenced. The accumulated maximum temperature needed to induce the onset of flowering was 908 °C for Olea (Table I).

Table I. Accuracy of the models, based on the cumulative method, predicting the start date of the Olea pollen season

Start date	Accumulated HR with 5 °C threshold	Accumulated GDD with 5 °C threshold	Accumulated maximum temperature	Accumulated temperature (T n + T x)/2
Mean value	627 °C	425 °C	908 °C	707 °C
Correlation coefficient	0.95	0.91	0.97	0.96
Start date in 1995	119	119	119	119
Predicted start date 1995	119	nd	118	119
Accuracy	100%	nd	99.1%	100%
Start date in 2009	129	129	129	129
Predicted start date 2009	113	nd	115	113
Accuracy	87.6%	nd	89.1%	87.6%

The cumulated maximum temperature variable from 1 March, was able to predict the final date of the Olea pollen season in 1995 in a very satisfactory way with 96.5% accuracy while in 2009 the accuracy was even higher (98.0%). Also here a clear relationship between the end of the Olea pollen season and the maximum temperature recorded during the months before and during the flowering period is visible in the data. The accumulated maximum temperature needed to induce the final of the Olea pollen season was 2806 °C (Table II).

Table II. Accuracy of the models, based on the cumulative method, predicting the end and peak date of the Olea pollen season

End date	Accumulated maximum temperature	Peak date	Accumulated maximum temperature
Mean value	2806 °C	Mean value	2450 °C
Correlation coefficient	0.98	Correlation coefficient	0.84
End date in 1995	195	Peak date in 1995	158
Predicted end date 1995	202	Predicted peak date 1995	155
Accuracy	96.5%	Accuracy	98.1%
End date in 2009	199	Peak date in 2009	146
Predicted end date 2009	195	Predicted peak date 2009	151
Accuracy	98.0%	Accuracy	96.6%

The variable of accumulated maximum temperature from 1 January until the peak date of the Olea pollen season predicted the peak date in 1995 in a very satisfactory way with 98.1% accuracy while in 2009 slightly lower (96.6%; Table II), and a clear relationship between the peak date and the maximum temperature recorded during the months before and during the flowering period could also here been observed. The accumulated maximum temperature needed to induce the peak of the Olea pollination was 2450 °C (Table II).

Start date

The average of the daily minimum and mean temperature on days 341–360 (December) has the highest coefficient correlation (r = 0.61 and r = 0.62, respectively) with the start of the Olea pollen season. The accumulated minimum temperatures on days 31–40 counted from 1 January have a positive significant correlation with the start date (r = 0.60). This could be interpreted as a pattern whereby the higher the minimum temperatures from days 31–40, the more the start of the Olea pollen season will be delayed. The decisive role of the temperature for the start of Olea pollen season has been described by many authors (e.g. Alba & Díaz de la Guardia, 1998; Fornaciari et al., 1998; Galán et al., 2001). Temperatures in the periods preceding pollination play an important role in determining reproductive growth and anthesis. In all cases, one of the variables exerting the greatest effect on the onset of the pollen season was the minimum and mean temperature in February. The start of the Olea pollen season has a negative significant correlation with the global radiation in November (r = −0.63). In addition, the start of Olea pollen season has a positive significant correlation with the rainfall in October–December (r = 0.50–0.58). When the rainfall is abundant in the months prior to the pollination period, the Olea pollen season will start later. It is well documented for plants that they respond to low levels of rainfall with anticipating anthesis.

Most trees in temperate zones go after the active phase of growth in the spring/summer period through a phase of dormancy during the autumn. Usually, the beginning of dormancy is determined both by shortening of the photoperiod and lowering of temperature. These conditions generally take place during November, and the period of chilling accumulation continues until the second fortnight in December. Growth-arresting conditions are eliminated when buds are exposed to chilling temperatures for a certain period. The subsequent stage of dormancy is called ‘quiescence’, which is defined as the period in which the buds remain dormant due to unfavourable environmental conditions. This preliminary study suggests that there is an effect of air temperature on Olea pollination, but other environmental factors, such as global radiation, rainfall, may be of great influence in determining the onset of pollination in plants with a spring flowering (Rodríguez-Rajo et al., 2003). From the results obtained from the PCA, the correlation study and the multiple regressions with an upward selection, we were able to identify the most appropriate model with associated statistics (Tables III, IV). The selected predicting model enabled the forecast of the beginning of the Olea pollen season in 1995 with 98.3% accuracy and in 2009 with 93.8% (Table V).

Table III. Statistic parameters of the models predicting the main features of the Olea pollen season

	Parameters	Estimation	Error	Probability
Start date^a	Constant	101.660	18.912	0.0001
	Mean(301–330)GR	−0.058	0.018	0.0056
	Sum(31–40) T n	0.303	0.098	0.0078
	Sum(41–50)GR	0.004	0.001	0.0055
End date^b	Constant	264.518	20.403	0.0000
	Sum(141–160) R	−0.612	0.131	0.0004
	Sum(11–20) (T n+T x)/2	−0.624	0.199	0.0075
	Sum(31–60) R	0.153	0.060	0.0237
Duration^c	Constant	61.228	25.451	0.0295
	Sum(81–90) T n	0.780	0.295	0.0183
	Sum(341–360) T n	−0.393	0.119	0.0048
	Sum(31–60) R	0.095	0.099	0.0352
Pollen Index^d	Constant	13565.100	3528.810	0.0023
	Sum(61–80) T mean	−14.089	6.427	0.0388
	Sum(41–50) T x	131.860	37.694	0.0044
	Sum(41–50) T mean	−155.240	39.962	0.0022
	Sum(21–40) T x	−54.943	13.969	0.0020
	Sum(21–40) T n	23.054	7.932	0.0132
	Sum(1–10) R	9.743	3.626	0.0198
Peak date^e	Constant	248.344	16.060	0.0000
	Sum(361–365)GR	−0.004	0.001	0.0124
	Sum(91–120) T x	−0.130	0.033	0.0014
	Sum(31–40) T x	−0.150	0.069	0.0462
Peak value^f	Constant	2401.130	591.752	0.0014
	Sum(101–110)GR	−0.024	0.009	0.0175
	Sum(101–120) T x	−3.110	0.847	0.0028
	Sum(21–40) T x	−6.785	1.304	0.0002
	Sum(341–360)GR	0.059	0.013	0.0005
	Mean(61—80)GR	0.477	0.126	0.0023
Pre-peak pollen index^g	Constant	7296.670	2413.680	0.0091
	Sum(1–10) R	3.983	2.298	0.0310
	Sum(41–50) T x	82.274	33.983	0.0297
	Sum(41–50) T mean	−109.409	35.371	0.0079
	Sum(21–40) T x	−27.308	10.626	0.0222
^aStart date = 101.66 –  0.058 × Mean(301–330)GR + 0.303 × Sum(31–40) T n + 0.004 × Sum(41–50)GR;
^bEnd date = 264.52 –  0.624 × Sum(11–20) T mean –  0.612 × Sum(141–160) R + 0.153 × Sum(31–60) R;
^cDuration = 61.23 –  0.39 × Sum(341–360) T n + 0.78 × Sum(81–90) T n + 0.09 × Sum(31–60) R;
^dPollen Index = 13565 –  14.09 × Sum(61–80) T mean + 131.86 × Sum(41–50) T x –  155.24 × Sum(41–50) T mean –  54.94 × Sum(21–40) T x + 23.05 × Sum(21–40) T n + 9.74 × Sum(1–10) R;
^ePeak date = 248.34 –  0.13 × Sum(91–120) T x –  0.15 × Sum(31–40) T x –  0.004 × Sum(361–365)GR;
^fPeak value = 2401.13 –  0.024 × Sum(101–110)GR –  3.11 × Sum(101–120) T x –  6.78 × Sum(21–40) T x + 0.06 × Sum(341–360)GR + 0.48 × Mean(61–80)GR;
^gPre-peak pollen index = 7297 + 3.98 × Sum(1–10) R + 82.27 × Sum(41–50) T x –  109.41 × Sum(41–50) T mean–27.31 × Sum(21–40) T x.
Note: Sum(21–40) T x represents the sum of maximal temperature on days 21–40 from 1 January.

Table IV. Statistic parameters of the models predicting the main features of the Olea pollen season

Model	Sum of squares error	Mean square error	r ^2 (%)	SD of residuals	MAE	DW test	Residual autocorrelation
Start date	1480.54	493.51	78.46	5.39	3.87	1.79	0.059
End date	3929.67	1309.89	72.98	10.19	6.78	2.60	−0.321
Duration	5883.61	1961.20	64.10	14.83	11.55	1.81	0.029
Pollen index	2.33E+7	3.88E+6	83.04	629.72	374.50	1.77	0.019
Peak date	845.65	281.88	78.23	3.96	2.76	2.49	−0.260
Peak value	5879.40	1175.88	87.53	80.28	53.14	2.25	−0.132
Pre-peak pollen index	8.89E+6	2.22E+6	64.31	593.66	397.34	2.35	−0.189
Note: SD, standard deviation; MAE, mean absolute error; DW, Durbin–Watson statistic.

Table V. Accuracy of the models, based on the statistic method, predicting the main features of the Olea pollen season

	Start date	End date	Pollen index	Duration	Peak date	Peak value	Pre-peak pollen index Method 1	Pre-peak pollen index Method 2
Value in 1995	119	195	2773	76	158	258	1535	1535
Predicted value 1995	117	205	2307	66	155	273	635	1329
Accuracy	98.3%	94.9%	83.2%	86.8%	98.1%	94.2%	41.4%	86.6%
Value in 2009	129	199	2968	70	146	450	2349	2349
Predicted value 2009	121	205	3931	91	152	413	2217	2264
Accuracy	93.8%	97.0%	67.6%	70.0%	95.9%	91.8%	94.4%	96.4%

End date

The variables with the highest coefficient correlation are the accumulated daily maximum and minimum temperatures on days 181–200 during and at the end of the pollination period (r = 0.65 and r = 0.60, respectively). According to years, the end date ranged between day 173 and day 230. In order to predict the end date of the Olea pollen season, these variables have not been integrated in the models. The accumulated daily maximum temperature on days 101–120, the last 20 days of April, has a coefficient correlation of 0.55 with the end of the pollen season. The accumulated rainfall from days 31–60 (February) has a positive significant correlation with the end date (r = 0.48). When the rainfall is abundant in the months prior the Olea pollination period, the end date will be delayed. The end of the Olea pollen season has a negative correlation (r = −0.49) with the accumulated rainfall amount on days 141–160 (the last ten days of May and the first ten days of June). Then, the higher the rainfall amount from days 141–160 is during the pollination period, the more the end date of the Olea pollen season will be advanced. This preliminary study suggests that there is a significant effect of daily maximum and minimum temperatures and rainfall during and at the end of the pollination on the end date of the Olea pollen season. The retained model (Tables III, IV) enabled the forecast of the end date of the Olea pollen season in 1995 with 94.9% accuracy and in 2009 with 97.0% accuracy (Table V).

Pollen index

In order to forecast the pollen index of the Olea pollen season, the same combination of the meteorological parameters was used as for the start and the end date of the season. The variables with the highest coefficient correlation are the accumulated daily maximum and mean temperature on days 41–50, in February, (r = −0.59 and r = −0.69, respectively). The accumulated daily maximum and minimum temperature on days 21–40 (vegetative period; the last ten days of January and the first ten days of February) have a positive significant correlation with the pollen index of the Olea pollen season (r = 0.59). This could be interpreted as a pattern where the higher the minimum temperatures from days 21–40 are, the more total Olea pollen will be produced. The effect of the accumulated daily maximum temperature seems to be different according to the period: a negative correlation on days 21–40 and a positive correlation on days 41–50. In addition, the accumulated mean temperatures from days 61–80 (March) have a negative correlation with the pollen index (r = −0.52). It seems that the mean temperature on days 301–330 (October–November) does influence the pollen index of the Olea pollen season (r = −0.50). The pollen index has a negative correlation (r = −0.46) with the accumulated rainfall amount on days 141–160, during the pollination period. This could be interpreted as the wash out of airborne pollen by rain. However, the accumulated rainfall in January (days 1–30) has a positive correlation with the pollen index and more particularly on days 1–10 (r = 0.52). Rainfall during the vegetative period has a positive effect on pollen production. In order to forecast the annual production of the Olea pollen, the rainfall variable on days 141–160 were not integrated in the model. The pollen quantities depend not only on meteorological conditions before pollen release, but also maybe on those prevailing during pollen release (e.g. pollen wash-out by rainfall, stopping of pollen grain release in the occurrence of frost, etc.). The model selected to predict the pollen index of the Olea pollen season achieved 83.2% of explanation in 1995 and 67.6% in 2009 (Table V).

Duration

The accumulated daily minimum temperature on days 341–360 (December) gave the best explanations for the length of the season, which was the same variable used to predict the start date (r = −0.59). In addition, the accumulated minimum temperature during the last ten days of March were positively correlated with the length of the pollen season. If the accumulated daily mean temperature increased during this period, the duration of the Olea pollen decreased. The accumulated rainfall from days 31–60 (February) has a positive correlation with the pollen season length (r = 0.54). The same variable was used to predict the end date. The minimum temperature and the amount of rain in the first three months prior to the Olea pollination play an important role determining the length of the Olea pollen season. The retained model enabled the forecast of the duration of the season with 86.8% accuracy in 1995 and 70.0% accuracy in 2009 (Table V).

Peak date

The variables with the highest coefficient correlation are accumulated daily maximum temperature on days 121–150 during the pollination period (r = −0.74), and more precisely, on days 141–150. This period has not been integrated in the model. The accumulated daily maximum temperature on days 31–40 (first ten days of February) and on days 91–120 (April) has a negative coefficient of 0.70 and 0.64, respectively. It seems that the accumulated rainfall in February (r = 0.43) does slightly influence the start day of peak count. The global radiation at the end of December, influences the peak date in plants with a spring flowering. Temperature was the best variable for forecasting the maximum value date. The selected model enabled the forecast of the start of the peak count with 98.1% accuracy in 1995 and 95.9% in 2009 (Table V). The difference between the predicted and actual date of the start of the peak period for Olea was three days in 1995 and six days in 2009.

Peak value

The variables with the highest coefficient correlation are the accumulated daily maximum temperature on days 21–40, which was the same variable used to predict the pollen index (r = −0.60) and on days 101–120, the same variable used to predict the end date of the season. In addition, the accumulated daily mean temperature on days 41–50 is correlated with the peak value as well as the pollen index. It seems that the accumulated global radiation in December (days 341–360) does slightly influence the peak value (Table III). The retained model enabled the forecast of the peak value for the Olea pollen season with 94.2% accuracy in 1995 and 91.8% in 2009 (Table V).

Pre-peak pollen index

The variables with the highest coefficient correlation with the PPI are the accumulated daily maximum and mean temperature on days 41–50 (r = −0.52 and r = −0.59, respectively) and the accumulated rainfall in January, more particular, on days 1–10 (r = 0.52). These environmental parameters were the same variables used to predict the pollen index of the Olea pollen season. The accumulated daily maximum temperatures from days 21–40 have a negative correlation with the PPI (Table III). The effect of the maximum temperature seems to be different according to the period. This observation has been made in the pollen index study. In the present case, the selected model enabled the forecast of the PPI with 41.4% in 1995 accuracy and 94.4 % in 2009 (Method 1, Table V). We obtained low accuracy from this method. However, the PPI represented about 57.6 ± 15% of the annual pollen count. Then, we can calculate the PPI from the predicted pollen index, obtained from the previous model. This process was able to predict the PPI of the Olea pollen season successfully with an accuracy of 86.6% in 1995 and 96.4% in 2009 (Method 2, Table V).

Discussion

Long-term trends in Nice

The advantage of the non-parametric statistical tests over the parametric tests is their better suitablility for non-normally distributed, missing data and extreme values, which are frequently encountered in environmental time series. These tests can be applied to a small number of observations. These statistical tests are little affected by errors within the data values and are robust because insensitive to the ‘extreme’, missing values and values below a detection limit. In the future, subjective forecasts by linear extrapolations could be made with the rates of annual change with a confidence interval. The Mann–Kendall test does not account for the magnitude of the data. In this study, results are little affected because data pollen present a maximum jump lower than one order of magnitude.

Trends in Olea pollen

In the last decade, a delay in the start of the Olea pollen season has been observed. Between 1990 and 2009, a delay in the start (0.54 days/year) and final dates could be identified (Table VI). An increase of the mean pollen index and the annual PPI was obtained. Globally, we observed an upward trend in annual pollen production and a stability of the mean duration of the Olea pollen season (Table VI). Olive pollen trends reflected a flowering delay, particularly in the last years (1998–2001), linked to adding behaviours of winter-chilling and spring-warming climates (Orlandi et al., 2004).

Table VI. Main feature trends (Method 1) of the Olea pollen season and confidence interval CI 95% (Min–Max 95%) obtained by the Mann–Kendall test in Nice over the period 1990–2009

	Average	Trends	CI 95%	Trends (%/year)
Pollen index (grain/m^3)	2502 ± 1248	+40 grain/year	(−76; +127)	+2.4
Start date (days)	115 ± 11	+0.54 days/year	(−0.70; +1.46)
End date (days)	197 ± 18	+0.60 days/year	(−1.46; +2.33)
Length (days)	82 ± 22	+0.06 days/year	(−2.39; +1.80)
Peak value (grain/m^3)	320 ± 193	+14 grain/year	(−3; +31)	+8.2
Peak date (days)	151 ± 8	−0.46 days/year	(−1.13; +0.17)
Pre-peak pollen index (grain/m^3)	1504 ± 876	+40 grain/year	(−63; +119)	+4.0
Note: significance levels α > 0.1.

Frenguelli (2002) reported an advance in the start date of Olea pollen season (0.4 days/year) in Perugia (central Italy) over a period of 20 years. Galán et al. (2005) carried out a study in three southern Spanish cities (Málaga, Granada and Priego). For each area, an advance of the onset of the Olea pollen season was observed over the 1992–2001 period (1.5, 0.7 and 1.8 days/year, respectively).

Meteorological trends

Some authors argued that meteorological conditions could influence the pollen season timing as well as its quantitative importance (Frenguelli, 2002; Frenguelli et al., 1991; Emberlin et al., 1997, 2002; Menzel, 2002). Weather conditions directly influence pollen counts by influencing pollen production, the time of the start of flowering, the abundance of vegetation and by controlling the amount of pollen released daily into the air. The present study was able to show that the pollen season depends not only on meteorological conditions before pollen release, but possibly also on those prevailing during pollen release. The study of the pollen-season timing of species allows us to explain changes in ecology related to the possible global climatic change (Frenguelli et al., 1991; Emberlin et al., 1997, 2002). To evaluate the effect of the main meteorological parameter changes on pollination, a trend analysis was studied between 1990 and 2009.

For Nice, we obtained by using the Mann–Kendall test, starting from annual averaged meteorological data (Table VII), a decrease for the precipitation amounts and for the relative humidity over the 1990–2009 period. A significant upward trend was observed for the daily minimum, mean and maximum temperatures, between 0.047 and 0.059 °C/year. The daily global radiation increased 5.71 J/cm²/year over the study period. By using the Seasonal Kendall test (Table VIII), decreasing trends for the precipitation amounts and the relative humidity during the warm and cold seasons were found. For the minimum temperatures, the study related a significant increase during the warm season and a downward trend in winter. Globally, the maximum and mean temperatures and the global radiation increased during the warm and cold seasons. These temperatures only decreased in winter but significantly increased in spring. The global radiation increased in winter. The average mean annual temperature in Nice has increased over the last 20 years, revealing a clear and significant linear trend. There is, therefore, a significant warming during the spring and summer whereas we observed a slightly decrease of temperatures during the winter.

Table VII. Annual trends of the main meteorological parameters and confidence interval CI 95% (Min–Max 95%) obtained by the Mann–Kendall test in Nice over the 1990–2009 period

	Average	Trends	CI 95%
Precipitation amounts R (mm)	772 ± 232	−15.4 mm/year (α > 0.1)	(−34.5; +10.4)
Minimum temperature T n (°C)	12.5 ± 0.4	+0.050 °C/year (α = 0.01)	(+0.016; +0.079)
Maximum temperature T x (°C)	19.8 ± 0.5	+0.047 °C/year (α = 0.1)	(−0.001; +0.092)
^(1) Mean temperature T mean (°C)	15.7 ± 0.5	+0.059 °C/year (α = 0.01)	(+0.025; +0.087)
Global radiation GR (J/cm^2)	1525 ± 52	+5.71 J/cm^2/year (α = 0.05)	(+0.79; +8.81)
Relative humidity RH (%)	69.8 ± 3.4	−0.44 %/year (α = 0.001)	(−0.62; −0.22)
Note: Significance levels α.

Table VIII. Seasonal trends of the main meteorological parameters obtained by the Seasonal Kendall test in Nice over the 1990–2009 period

Meteorological parameters		Cold period 22 September− 20 March	Warm period 21 March–21 September	Winter	Spring	Summer	Autumn
Precipitation amounts R (mm)	Average	504±193	268±99	147±95	142±53	126±76	357±173
	Trends (mm/year)	−8.7	−4.0	+0.86	−0.24	−4.12	−8.40
Minimum temperature T n (°C)	Average	8.2±0.5	16.7±0.7	6.6±0.7	14.1±0.9	19.3±0.8	9.7±0.7
	Trends (°C/year)	+0.018	+0.074**	−0.023	+0.103**	+0.047	+0.054
Maximum temperature T x (°C)	Average	15.7±0.5	23.8±0.7	14.3±0.8	21.0±0.9	26.6±0.7	17.2±0.7
	Trends (°C/year)	+0.023	+0.067**	−0.020	+0.093*	+0.030	+0.048 ^+
Mean temperature T mean (°C)	Average	11.2±0.5	20.2±0.7	9.8±0.8	17.6±0.9	22.8±0.7	12.6±0.7
	Trends (°C/year)	+0.038^+	+0.078***	−0.015	+0.098**	+0.038 ^+	+0.064**
Global radiation GR (J/cm^2)	Average	888±49	2155±67	1021±73	2182±74	2129±77	755±58
	Trends (J/cm^2/year)	+2.9	+7.5**	+2.2	+7.9**	+7.3**	+4.2^+
Relative humidity RH (%)	Average	68.6±3.1	71.0±4.3	67.1±4.1	72.3±3.5	69.7±5.5	70.2±4.2
	Trends (%/year)	−0.31**	−0.52***	−0.31^+	−0.36**	−0.63***	−0.40**
Significance level α = > 0.1; ^+significance level α = 0.1; *significance level α = 0.05; **significance level α = 0.01; ***significance level α = 0.001.

The seasonal trends for the Olea pollen and the main meteorological parameters, the most influencing pollen features, are in agreement. The observed increase in Olea pollen abundance coincides with a rise in air temperature, which is the meteorological factor to have experienced a sustained and significant change over the same period. The air temperature is without doubt the parameter, which is able to determine inter-annual changes in the Olea pollination in Nice.

This study underlines how changes in temperature can have a great influence on the pollination of many plants with consequences in various sectors such as the allergological (Emberlin et al., 1997, 2002; Chuine, 2000). The data obtained here give indirect evidence of the flowering phenophase that occurs in the olive trees. These species are very sensitive to climatic changes, which can be used to demonstrate climatic variations. The olive as a typical Mediterranean species is sensitive to low temperatures and resistant to water shortage. In this manner, the northern and the southern limits of olive cultivation in the Mediterranean basin are conditioned by low temperature and low rainfall (Orlandi et al., 2005).

Conclusions

The olive tree, Olea europaea, is very common in south-eastern France and is recognised as one of the most important allergenic pollen in the Mediterranean. This study allowed the development of predictive models for calculating the main characteristics of the Olea pollen season over the last 20 years.

A clear relationship between the characteristics of the Olea pollen season in Nice and the temperature recorded during the months before the flowering period (January–February) has been demonstrated in determining reproductive growth and anthesis. Other environmental factors, such as global radiation and rainfall, may be of great influence in determining the pollination. Rainfall has a great influence in determining the final date of pollination. A negative correlation with the accumulated rainfall amount during the pollination period was obtained and can be interpreted as the wash out of airborne pollen. However, rainfall during the vegetative period has a positive effect on pollen production. The pollen quantities depend not only on meteorological conditions before pollen release, but also maybe on those prevailing during pollen release.

With respect to the analysis of global trends, we observed an upward trend in annual pollen production and a stability of the mean duration of the Olea pollen season. The study allowed us to explain changes in ecology related to the possible climatic change. The observed increase in Olea pollen abundance coincides with a rise in air temperature in Nice within the last 20 years. The air temperature is the parameter, which is predominantly useful to determine inter-annual changes in Olea pollination. From a climatic point of view, the suitable area for olive cultivation could be enlarged due to the changes in temperature and precipitation patterns. In this way, a spatial increase and a northward shift of the potential olive cultivation area is expected under the condition of climate warming (Orlandi et al., 2005).

Acknowledgements

This work was made possible by the technical contribution of the Aeroallergen Monitoring Network (RNSA) for data acquisition, analysis and interpretations.