Weather-dependent fluctuations in the abundance of the oak processionary moth, Thaumetopoea processionea (Lepidoptera: Notodontidae)

Population fl uctuations of the well-known oak defoliator, the oak processionary moth (Thaumetopoea processionea L.), were studied using light trap data and basic meteorological parameters (monthly average temperatures, and precipitation) at three locations in Western Hungary over a period of 15 years (1988–2012). The fl uctuations in the numbers caught by the three traps were strongly synchronized. One possible explanation for this synchrony may be similar weather at the three trapping locations. Cyclic Reverse Moving Interval Techniques (CReMIT) were used to defi ne the period of time in a year that most strongly infl uences the catches. For this period, we defi ned a species specifi c aridity index for Thaumetopoea processionea (THAU-index). This index explains 54.8–68.9% of the variation in the yearly catches, which indicates that aridity, particularly in the May–July period was the major determinant of population fl uctuations. Our results predict an increasing future risk of Oak Processionary Moth (OPM) outbreaks and further spread if the frequency of severe spring/summer droughts increases with global warming.


The data used
We used yearly OPM catches of 3 light traps (Fig. 1 and Table 1) over a period of 15 years (1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012) to quantify the population fl uctuations.This period of time is the longest period during which all 3 light traps were operational.These light traps belong to the Forestry Light Trap Network run by the Hungarian Forest Research Institute since the early 1960s.The pairwise distances between the trap locations are as follows: Acsád-Sopron: 41 km; Acsád-Szalafő: 57 km; Sopron-Szalafő: 91 km.The traps are located in old forests, therefore the stand conditions did not change very much during the 15 year study period.The standardized light traps are equipped with 125 W mercury bulbs and operated continuously between 1st March and the end of December.The traps are emptied daily and the species of Macrolepidoptera identifi ed (Hirka et al., 2011).The data collected using light traps might be used for many different purposes, i.e. forecasting outbreaks, faunistic studies, fl ight phenology, diversity trends, etc.Light traps that operate continuously during a whole season at the same place over a long period, are excellent tools for following the population fl uctuations of insects attracted to artifi cial light (Tallós, 1966;Szontagh, 1975Szontagh, , 1968;;Hirka et al., 2011).
Sweden (Lövgren & Dalsved, 2005), two countries where this species is not previously recorded.Of course a contribution of international trade cannot be excluded from these processes.This is the case in the UK, where the species appeared for the fi rst time in 2006 as result of plant trade rather than natural range expansion (Baker et al., 2009;Mindlin et al., 2012).
Greater than the effect on forest health, the spread and outbreaks of OPM increased international awareness of the effect of its urticating hairs on human health (Maier et al., 2004;Gottschling & Meyer, 2006;Green, 2015;Battisti et al., 2017).The human and animal health related importance is well known in Hungary but are normally less frequent and severe than they are likely to be in the Netherlands or Belgium, since OPM populations in Hungary are mainly found in forests rather than in city parks or inhabited areas.
OPM is native to Hungary and the damage it causes has been reported from time to time since the late 19 th century (Szontagh, 1976;Hirka et al., 2011).It is one of the "Top 12" forest pests in Hungary (Hirka et al., 2011), but not classifi ed as a serious pest.The long term  measure of the area damaged per year is 808 hectares countrywide.The average damage for the period 1998-2012 (time span of the present study) is 1,672 ha/year, which is considerably greater than the long term average.The damage caused by OPM peaked in 2004 with a total of 4,270 ha damaged countrywide (source: Forest Damage Database of the Forest Research Institute).OPM's life history in Hungary is similar to its life history in other European countries.The adults normally fl y from late July till mid-September, with a peak in August, but females can sometimes be found as late as early December (Gy.Csóka, unpubl.).Both males and females are capable of fl ying, and are attracted to artifi cial light, therefore can be effi ciently caught and monitored by light traps.
Small and scattered local outbreaks are reported in many different regions of Hungary, where its main host, Turkey oak (Quercus cerris) is abundant.Defoliation is also recorded for other native oaks (Quercus robur, Q. petraea and Q. pubescens).OPM is a thermo-and heliophil species (Szontagh, 1976;Hirka et al., 2011).It does not always behave as a classic outbreak species, as it some places it remains abundant for long periods, neither increasing or collapsing suddenly.Defoliation by this species rarely exceeds 50% at the scale of a stand, unlike the gypsy moth (Lymantria dispar).which can completely defoliate large areas (Mcmanus & Csóka, 2007;Hirka et al., 2011).The high density foci of OPM are rather small and scattered, and do not form a large continuous area.However, there has been a recent expansion of these high density foci at some places.
Understanding the link between climate change and population trends of forest insects and providing reliable forecasts for the future is a major challenge.The main aim of the present study is to quantify the effect of weather on the population fl uctuations of OPM.We hypothesize that the weather conditions (temperature and precipitation) during the larval period of OPM have a signifi cant effect on  Although the daily light trap catches are infl uenced by many environmental factors (temperature, wind rain, clouds, moon phase) (Yela & Holyoak, 1997;Nowinszky et al., 2010;Nowinszky & Puskás, 2017) they are good for monitoring long term population fl uctuations if pooled together for a whole season, since the short term environmental effects are neutralized if collected over a long period of time (Raimondo et al., 2004).
Unfortunately, there are no reliable long term quantitative datasets for the area defoliated by OPM in the vicinity of traps, but the areas with regularly high population densities and moderate defoliation are known (see on Fig. 1).
The meteorological variables used in the analysis were interpolated as described in detail below.We used monthly average temperatures and monthly total precipitation as primary data.Monthly precipitation data was obtained from the hydrological annals, published by the Water Resources Research Centre (VI-TUKI).Additional data was obtained from the Hungarian Meteorological Service (OMSz).The dataset included 608 monthly rain gauge recordings for the years 1997-2012 in Hungary.The number and the location of the rain gauges changed frequently in the given period, thus only stations with a continuous data series were included in the database.To achieve this, a representative radius of 5 km was set for each station.If translocation occurred within this radius, time series were considered continuous.Raw rain gauge records underwent a series of quality tests to identify obvious anomalies and remove false values.Precipitation maps were created using the kriging interpolation method.Kriging is a geostatistical gridding method that has proved useful and been applied extensively for the interpolation of climate data (Dingman et al., 1988;Garen et al., 1994;Hevesi et al., 1992).The reliability of the precipitation maps was assessed using a crossvalidation method.The mean of the deviations from the observed values was 49.4 mm, which is 8.2% of the observed mean annual precipitation in Hungary.
Mean monthly air temperature recorded at a height of 2 meters was obtained from the monthly weather reports published by the Hungarian Meteorological Service (OMSz).The temperature dataset included 31 weather stations in Hungary for the period of 1997-2012.Temperature maps were created using the same kriging interpolation method.The dependence of temperature on altitude was taken into account by applying a monthly vertical gradient.The effect of slope and aspect on air temperature was corrected for by using the solar radiation analysis tool in the Arc-GIS software (Yang et al., 2007).

Analyses
(1) We used the correlation matrix of the yearly OPM catches by the three light traps to compare the temporal patterns in the population fl uctuations at the different locations.We checked whether there is a linear trend in the yearly catches of the three traps for the period 1998-2012 (Student t-test, α = 0.05, DF = 13).
(2) We used CReMIT (Cyclic Reverse Moving Interval Technics) to defi ne the periods within years that had the strongest effect on the catches (Pödör et al., 2014).CReMIT is a data-mining method that helped us test our main hypothesis.CReMIT creates all possible aggregated datasets in a time series to sphere the examination possibilities.The maximum length (ML) and time shifting (TS) of the derived variables are defi ned by the user.The CReMIT method combines the evolutionary and moving interval techniques to create all possible derived time series according to the above mentioned two parameters.For example, if we examine the monthly temperature time series in which ML is 3, TS is 2 and the starting point is 11.2017, than we create, in addition to the basic monthly data, the values for every year: average temperature in June, June-July, June-Aug, July, July-Aug, July-Sep, Aug, Aug-Sep, Aug-Oct, Sep, Sep-Oct, Sep-Nov, Oct, Oct-Nov and Nov. The created time series includes both values for single months and, aggregated time series.
This method provides the average temperatures and sum of precipitation for all the given intervals.These secondary explanatory variables can be correlated with the response variable (light trap catching data) and so expand the scope of the study.The strengths of the correlations with these secondary variables helps in choosing the time windows that have the strongest effects on the response variable.In our analyses all months in the created secondary time periods were given the same weight (1).With different weighting of the different months, the correlations and rvalues could have been greater.However, we wanted to keep the analyses relatively simple in order to avoid speculative explanations of the results.
The lengths of time windows spanned from October to the following July.We decided to exclude both temperature and precipitation for August and September from the analysis.This is because weather conditions in these two months (main fl ight period) may directly infl uence the light trap catches (less fl ight activity, etc.), as demonstrated by Bonsignore & Manti (2013) for Thaumetopoea pityocampa.
In order to decrease the risks of a misleading interpretation, we did not include August and September (the main fl ight period of OPM) in the THAU-index.In addition, we separately checked the direct infl uence of weather conditions (temperature and precipitation) in these two months on the yearly catches.
(3) Based on the results of the CReMIT derived time series analysis, we intended to create a species specifi c variable for OPM.The time windows with the highest explanatory power for all three traps were incorporated in a species specifi c index called "THAU" , which is based on the genus name of OPM.The defi nition of THAU is given in the "Results" section.
(4) We correlated the yearly values of THAU for the three trap locations.We checked whether there is a linear trend in the yearly THAU values calculated for the three trap locations for the period 1998-2012.Then yearly values of the THAU index (as explanatory variable) were correlated with the yearly catches of the light traps (as response variable) using linear regression (Student t-test, α = 0.05, DF = 13).
(5) We also analyzed the delayed effect of the THAU value of the previous year on a given year.In addition, the cumulative effect of a given year's and previous year's THAU was also analyzed.

Correlations and trend analysis of the yearly OPM catches of the three traps
The yearly OPM catches of the three traps are given in supplementary Table S1.There are signifi cant positive correlations between yearly OPM catches of all three traps (rvalues and levels of signifi cance, * = 95%; ** = 99%): Ac- No signifi cant linear trend in yearly catches was found for any of the traps for the given period (1998-2012).

CReMIT analysis -periods in a year when temperature and precipitation had the greatest effect
Average temperature in the period May-July had the greatest explanatory power (Table 2).No strong explanatory power of other time windows was found for temperature.The analysis of precipitation gave a different result.Precipitation in the period from October to July had the greatest explanatory power (Table 3).
There was no signifi cant (p < 0.05) infl uence of weather (monthly precipitation and monthly average temperature) during the fl ight period (August and September separately and also these two months summed together) on the yearly catches.Therefore, we conclude that weather conditions are a far more important in synchronizing remote populations in terms of larval and pupal development than in determining the fl ight period.

Defi nition of THAU
Based on the results presented above we defi ned the THAU as: where T V-VII = average temperature in the period May-July (C°) and P X-VII = summed precipitation between October and the following July (mm).
The explanatory power of THAU is greater than the explanatory power of either temperature or precipitation, which indicates the importance of combining the effect of these two factors.

Correlations and trend analysis of the yearly THAU values of the three traps
The THAU values are highly signifi cantly (p ˂ 0.01) positively correlated among all 3 traps (Acsád-Sopron: 0.80; Acsád-Szalafő: 0.85; Sopron-Szalafő: 0.78).There were no signifi cant linear trends in the yearly catches of any of the traps or the yearly THAU values of any of the trap locations for 1998-2012 period.

Effects of THAU on the yearly catches
The THAU values for the time windows mentioned above were highly signifi cantly (p ˂ 0.01) positively correlated with the yearly catches for all three traps (Table 4).A year's THAU value explained 54.8-68.9% of the variation in the yearly catches, indicating that aridity is one of the major factors infl uencing population fl uctuations.The yearly values of the THAU and the trap catches (after "ln" transformation) are given in Figs 2-4.
The THAU values of the previous year are not significantly correlated with the yearly catches.The average THAU values of a particular year and the previous year were highly signifi cantly (p ˂ 0.01) correlated for two traps (Sopron, Szalafő) and signifi cantly so (0.01 ˂ p ˂ 0.05) for Acsád.

DISCUSSION
The strong positive correlation between the yearly catches of the three traps show evident and strong synchrony in the population fl uctuations at the three spatially rather remote locations.It is similar to the "Moran-effect", when populations of a given species are synchronized by synchronous stochastic effects, for example, weather (Moran, 1953;Liebhold et al., 2004;Hudson et al., 2006;Liebhold , 2006;Allstadt et al., 2015;Chevalier et al., 2015).Similar regional synchrony is demonstrated for T. pityocampa in France by (Li et al., 2015), but in this case low winter temperatures played the key role.
It should be mentioned that caution is needed when studying population fl uctuations/synchrony using light trap catch data, simply because light trap catches are sensitive to weather conditions (Nowinszky & Puskás, 2017).Therefore, the analysis may sometimes detect synchrony between light trap catches rather than synchrony of population fl uctuations, as mentioned by Bonsignore & Manti (2013) for T. pityocampa.This infl uence is much weaker over a long period (several weeks or even months) covering the whole fl ight season.Therefore, only light traps operational for long periods covering the whole fl ight period of a given species provide suitable time series for this kind of population study.
Unfortunately, there are no reliable defoliation data from the close vicinity of the trap locations that would allow us to search for a correlation between the light trap catches and the incidence of defoliation.However, there are signifi cant positive correlations between light trap catches and defoliation for several important oak defoliating Lepidoptera, such as Lymantria dispar (Leskó et al., 1994), Euproctis chrysorrhoea (Linnaeus, 1758) (Leskó et al., 1995), Malacosoma neustria (Linnaeus, 1758) (Leskó et al., 1997) and Operophtera brumata (Linnaeus, 1758) (Leskó et al., 1999).
The infl uence of average temperature in the different time windows recorded in this study differs from that previously reported (Wagenhoff & Veit, 2011;Klapwijk et al., 2013).The highest explanatory power was recorded for the May to July period.Taking April temperature into account only had a signifi cant effect at one of the three locations (Acsád).Therefore, we excluded it, although several papers (Wagenhoff & Veit, 2011;Meurisse et al., 2012;Klapwijk et al., 2013) mention that early spring (March and April) temperatures may infl uence populations due to their direct effect on larval survival and in disrupting the synchrony between budburst and egg hatch.However, the effect of this desynchronization is quite plastic, since young larvae may survive 2-3 weeks of starvation (Wagenhoff et al., 2013) and bud burst occurs earlier following mild winters and in warm early springs (Wagenhoff et al., 2014).The positive effect of temperature in the main larval feeding period is likely to be due to faster development shortening their exposure to both abiotic (i.e.rain) and biotic mortality factors (parasitoids and predators) and providing less favourable conditions for pathogens.
Precipitation (just like temperature) may also have both direct and indirect effects on population fl uctuations.The time window with the greatest explanatory power was that from October to the following July (10 months).The importance of this rather long time window may be due to several different, but interacting factors.The most evident is the effect of spring precipitation on the mortality of the larvae in May and June and of the pupa in June and July.Wagenhoff & Veit (2011) also mention that a wet spring may prevent an outbreak of OPM.In addition, direct mortality caused by rain in May and June, as a high rainfall then may increase the time it takes larvae to complete their development and thus the period they are exposed to other mortality factors (parasitoids, predators, entomopathogenic fungi, etc.).In addition, precipitation summed over a long period (from October to the follow July) has a positive effect on the physiological conditions of the host trees resulting in a more effi cient induced defense against caterpillars.
It should be mentioned that there are considerable discrepancies between the OPM numbers predicted by two traps (Sopron and Szalafő) for the period 2006 to 2012.One possible reason is that in some years several factors, other than weather (parasitoids, pathogens, etc.), may play a signifi cant role in the fl uctuation of OPM populations.
The positive but insignifi cant correlations between the previous year's THAU index and yearly catches indicates there may be a direct effect of weather on population fl uctuations.The average THAU values for a given and previous years explained less variation than the given year values, again supporting a stronger direct (within a year) effect of weather on the population fl uctuations of OPM.
AUTHOR CONTRIBUTION STATEMENT.GyCs and AH developed the idea and wrote the manuscript.LSz identifi ed the moths and contributed to manuscript.NM and ER: interpolation of meteorological data.ZP: CReMIT analysis, statistics.All authors read and approved the manuscript.

Fig. 1 .
Fig. 1.Locations of the three light traps used in this study.(Shaded spots indicate high density populations of OPM.)

Fig. 2 .
Fig. 2. Yearly values of THAU and the OPM catches at Acsád between 1998 and 2012.

Fig. 3 .
Fig. 3. Yearly values of THAU and the OPM catches at Sopron between 1998 and 2012.

Table 1 .
Basic information on the light traps used in this study.

Table 2 .
Signifi cance levels for the correlations between average temperatures in different time windows and yearly catches.

Table 3 .
Signifi cance levels for the correlations between summed precipitation in different time windows and yearly catches.

Table S1 .
Yearly OPM catches in the three light traps.