Generation cycles in experimental populations of a multivoltine insect

Abstract

1. Some animals exhibit large‐amplitude fluctuations with periods close to their generation length. These fluctuations can be caused by seasonal environmental influences, scheduled pesticide applications, or ecological factors such as intraspecific regulation and consumer–resource interactions. While theory provides various mechanisms for how environmental and ecological factors might generate generational fluctuations, there has never been a field experiment testing the relative contributions of seasonal demographic synchronisation and intraspecific regulation to generation cycles in natural populations.

2. The smaller tea tortrix, Adoxophyes honmai, is a serious pest of tea plants and a temperate multivoltine insect that undergoes 3–5 large‐amplitude generation cycles each year under natural conditions in Japan. Theory suggests that these fluctuations may represent limit cycles driven by asymmetric intraspecific interference, where older larvae directly affect younger ones. However, in the field, these populations also experience strong seasonality and periodic insecticide applications that are thought to either generate or modify these fluctuations.

3. We conducted a replicated field‐cage experiment on the tortrix populations to manipulate the initial degree of stage synchrony and the timing of introduction. The experiment included four treatments contrasting pulsed versus continuous age structures at the onset of spring, along with two introduction timings separated by 20 days (approximately half a generation time). We minimised artificial interventions, such as harvesting and insecticide application, as well as the effect of natural enemies while allowing meteorological influences during the season.

4. To compare the field‐cage experiment results with theoretical predictions, we constructed an age‐structured population model featuring asymmetric larval interference. We executed simulations using the same introduction scenarios as in the field‐cage experiment.

5. We observed the emergence of clear generational cycles in all treatments of the field‐cage even in the absence of any initial demographic synchrony. This suggests an internal mechanism regulating population cycles, possibly intraspecific interference. However, the generational cycles in the field‐cage were synchronised across treatments and with outer field populations. The results from the field‐cage experiment and simulation analyses indicate that external environmental factors, such as precipitation, acted as a pacemaker for the generational cycles created by the internal regulatory mechanism.

Although theory suggests various mechanisms by which environmental and ecological factors may drive generational fluctuations, our field‐cage experiment is the first to demonstrate how internal dynamics and external disturbances jointly produce synchronised, large‐scale outbreak cycles.

1. INTRODUCTION

Oscillatory population cycles indicate biotic regulation. For example, Nicholson's blowfly data, widely cited in numerous introductory ecology textbooks, demonstrated clear cycles in adult population size, so‐called ‘generation cycles’ (Nicholson, 1935). Populations exhibiting these generation cycles consist of individuals at the same developmental stage, and the cycle periods approximately correspond to the generational duration from egg to adult. However, it is reasonable to assume that patterns of adult emergence become diffuse over time, resulting in overlapping generations with multiple life stages, given variation in developmental rates among individuals and extended oviposition periods (Bjørnstad et al., 2016). A substantial body of theoretical research has shown that intraspecific competition for resources is one of the major causes of generation cycles (Gurney et al., 1980; May, 1973; Nisbet, 1996).

Generation cycles in insect populations have primarily been studied in laboratories or tropical environments, where external conditions can be assumed to be relatively constant (Bjørnstad et al., 1998; Briggs et al., 2000; Godfray & Chan, 1990; Nakamura et al., 2004). Recently, however, such cycles have also been documented in field populations of the smaller tea tortrix, Adoxophyes honmai, in temperate regions (Yamanaka et al., 2012). A. honmai exhibits regular outbreaks throughout its distribution in Japan, ranging from four generations in the north to five or six in the south (Smith et al., 2025). These generation cycles were typically regarded as transient phenomena following seasonal developmental synchronisation during winter, similar to monitoring records of other pest insects. However, previous theoretical analyses have suggested that generation cycles in A. honmai are alternatively driven by intraspecific interference among larvae (Bjørnstad et al., 2016; Nelson et al., 2013).

Simpson et al. (2018) observed frequent aggressive and cannibalistic behaviours among larvae in laboratory populations. Subsequently, Nelson et al. (2019) quantified the strength of direct offensive behaviours, including cannibalism, and found that older larvae exhibited stronger offensive capabilities than younger ones. Furthermore, they incorporated the estimated age‐specific competitive parameters into an age‐structured model, which successfully reproduced generation cycles in simulations. It has been reported that cannibalistic behaviour can also create generation cycles in other systems (Rosenheim & Schreiber, 2022). Briggs et al. (2000) found that the generation cycles in Indian meal moths in the laboratory were well aligned with the simulation results involving egg cannibalism and asymmetric interference among larvae. Nakamura et al. (2004) reported that a ladybird in tropical crop fields showed clear generation cycles, and these cycles could be simulated by a simple mathematical model incorporating egg cannibalism. Therefore, the generation cycles in A. homai may also have been caused by their cannibalistic behaviour.

However, we still cannot eliminate alternative explanations for the cause of generation cycles in A. honmai. Powell et al. (2000) asserted that seasonality itself can directly cause generation cycles. A. homai overwinters only in the larval stage, and the onset of winter induce uniform development at the beginning of each growing season (Honma, 1966), meaning the generation cycles could be merely a transient dynamics after winter. Godfray and Chan (1990) showed that periodic pesticide applications can lead to generational cycles in tropical pests. Since A. homai is an important pest in tea plantations, regular pesticide applications might be an alternative explanation for the generation cycles. Additionally, prey—predator interactions can create generation cycles, especially when the lifespans of predators and preys differ (Bonsall & Hassell, 2010; Knell, 1998).

In our study, we conducted a series of experimental field introductions of A. honmai using enclosed field‐cages to evaluate the effect of asymmetric larval interference while minimising other factors. Specifically, young larvae were introduced into the field cage either via one single introduction of all larvae (pulsed introduction), or via multiple introductions of smaller numbers of larvae (continuous introduction) at the start of spring to assess the impact of winter synchronisation. If the pulsed introduction creates generation cycles while the continuous one does not, the generation cycles in A. honmai can be assumed to be transient dynamics. We also implemented treatments in which larvae were introduced into the field cages 20 days later, that is, a half‐generation delay, after both the pulsed and continuous introductions. If asymmetric larval interference alone drives generation cycles, these cycles should be out of phase with those generated by the earlier introductions. The field cages were enclosed with fine mesh to prevent the migration of wild populations and reduce the influence of outside predators. We did not apply any insecticides or harvest tea leaves but allowed meteorological influences during the experiment.

We constructed an age‐structured model with asymmetric larval interference based on Nelson et al. (2019) and compared the simulation results to field‐cage experiments. Since we could not directly observe the fate of each individual in the cage experiment, our simulation analyses are expected to enhance understanding of the mechanisms that create generation cycles.

In addition, we monitored the dynamics in uncaged populations where plucking and pesticide applications are routinely carried out, and in nearby abandoned tea orchards where no pesticides or plucking were applied. Comparing the population dynamics in the field‐cages with those in the station (managed) and the abandoned orchards (unmanaged) allows us to evaluate the impact of management methods on generation cycles in this pest insect.

2. MATERIALS AND METHODS

2.1. Field‐cage experiment

Sixteen field‐cage chambers were built at the NIPP/NARO Kanaya station (34°48′ N, 138°8′ E) (see Appendix S1). Each field‐cage consisted of four interconnected greenhouse frames, completely covered with fine mesh netting (1.25 mm mesh size), and each chamber was separated by the same netting. The mesh allows for the penetration of rain and wind but prevents the immigration of A. honmai adults and most of their predators, including parasitoid wasps. The chambers measured 5.4 m in width, approximately 6 m in length, with a maximum height of 3.4 m and a minimum height of 1.8 m. Each chamber contained two rows of tea plants, each 4.5 m in length and 1.8 m in width, covering approximately 16.2 m² per chamber.

We set up two crossed treatments to test the difference between pulsed and continuous age‐structure at the starting point and to investigate the effect of introduction timing (Table 1). The first treatment varies the degree of synchrony with either a pulsed introduction (treatments (a) and (b)), where all insects were released at once, or a continuous introduction (treatments (c) and (d)), where insects were released over ten occasions for 40 days. The second treatment manipulates the alignment of introduction timing with expected phenology, either in alignment (treatments (a) and (c)) or 20 days later, which is nearly equal to half a generation length (treatments (b) and (d)). Each treatment has four replicate chambers, for a total of 16 chambers. In the pulsed treatments, approximately 2000 eggs were introduced at four locations in each chamber, simultaneously (8000 individuals in total per chamber), while the continuous treatments involved approximately 200 eggs at four locations, transferred into the chamber ten times every 4 days (8000 individuals in total per chamber). The introduction of 8000 individuals was approximately four times higher than the field density of older larvae during outbreak periods (Kodomari, 1987). This large number was chosen to account for larval mortality, particularly during the early instars. The experiment started on 6th May 2019 for treatments (a) and (c) at the same time as the first cycles in the field populations and on 26th May 2019 for treatments (b) and (d), half a generation length later.

TABLE 1.

Two crossed treatments for 16 field‐cage experiments.

Treatment	Cage ID a	Date of introduction
Pulsed introduction
Treatment (a)	a1, a2, a3, a4	May 6th
Treatment (b)	b1, b2, b3, b4	May 26th (a half‐generation shift from treatment (a))
Continuous introduction
Treatment (c)	c1, c2, c3, c4	May 6th, 10th, 14th, 18th, 22nd, 26th, 30th, June 3rd, 7th, 11th
Treatment (d)	d1, d2, d3, d4	May 26th, 30th, June 3rd, 7th, 11th, 15th, 19th, 23rd, 27th, July 1st (a half‐generation shift from treatment (c))

^a The positions and inside photos of the cages can be found in Figure S1.

To adjust the age of introduced larvae, we prepared egg masses, instead of the young larvae, that were ready for hatching in the exact numbers needed for each date of introductions across the four treatments (Table 1). We placed the egg masses in metal mesh cases to prevent unintended predation by predators such as slugs (10 cm in diameter and 8.5 cm in height, see Appendix S1, right middle panel). The mesh on the metal cases was coarse enough to allow the first instar larvae to escape but fine enough to prevent their parasitoid wasps. They were positioned among the rows of tea plants on the morning of the introduction date. The metal mesh cases were inspected in the afternoon to determine whether all larvae had hatched and escaped to the tea plants. The laboratory populations for the introductions were originally established from wild populations around Kanaya station (Kawasaki et al., 1983). They have been maintained for over 40 years under 21.0°C and 15L:9D light condition. We created a detailed schedule to ensure that a sufficient number of adults could be obtained for copulation a week before the introduction date; thus, the rearing conditions for the larvae were adjusted among groups from 15 to 24°C to align with the timing of pupation.

The imago population in each chamber was monitored daily by counting the male moths captured by funnel traps (Sumitomo Chemical Co., Ltd., Tokyo, Japan) with synthetic female sex‐pheromone lures (SE lure for the smaller tea tortrix, Shin‐Etsu Chem. Co. Ltd., Tokyo, Japan). Captured males were counted and released every morning.

Prior to the experiments, each chamber was treated with the short‐acting pesticide, emamectin benzoate (Affirm®, Syngenta Japan, Tokyo, Japan), which generally retains its residual effect for only a few days, to remove all resident A. honmai larvae and associated natural enemies, including parasitoids. Each chamber was subsequently treated with mating disruption using rope‐type dispensers of Hamaki‐con N® (Shin‐Etsu Chem. Co. Ltd., Tokyo, Japan) placed along the rows of tea plants from 16 April, in order to prevent mating by wild moths not experimentally introduced. The mating disruption treatments continued until 4th June 2019, well before the first emergence of adults from the introduced populations. Field examination confirmed that no males were present between 7th May and 4th June 2019, ensuring that all chamber populations originated from the introduced individuals. To minimise any artificial intervention, pesticides and fungicides were not applied during the experiments. However, twigs were pruned once on 10th July to maintain the health of the tea plants. The pruned twigs were left adjacent to the tea plants to allow the living larvae move back to the tea plants.

For statistical analyses, daily catches were recorded and smoothed using 5‐day averaging to reduce short‐term noise and emphasise periodic patterns if they exist. The analyses were conducted on the moth catches, which were log‐transformed after adding 0.5 to stabilise the variance in population size and also to reduce the impact of outliers (Keitt, 2008; Yamamura, 1999). Calendar dates were transformed to the physiological timescale ($\varphi$) according to Yamanaka et al. (2012).

$$\varphi \left(d\right)={\int }_0^d\frac{1}{1+e^{-\gamma ·\left(T\left(\xi \right)-ϵ\right)}}\mathit{d\xi },$$ (1)

where T(d) is the temperature of the day‐d. The parameter $\gamma$ controls the steepness of the sigmoid development function, and $ϵ$ is the temperature at which the development rate function reaches half the asymptotic value. Mean daily temperatures (T(d)) were recorded by the weather monitoring system at Kanaya station (https://www.naro.affrc.go.jp/org/vegetea/mms‐kanaya/Menu.html, accessed August 2025). Insects grow fast in summer and slow in early spring and late autumn. Equation (1) is used to quantify the growth schedule of the insects based on heat calories, eliminating seasonal effects on growth rates. The function is a nonlinear analogue to the day‐degree calculation used in pest management (see Appendix S2).

2.2. Simulation analyses

To compare the results of the field‐cage experiments with the theoretical predictions, we constructed an age‐structured population model based on Nelson et al. (2019). Specifically, realistic variation in each stage was modelled as a gamma‐chain model, in which each stage is divided into $N_{\mathrm{sub}}$ (=70) substages. Each substage assumes an exponential waiting time to progress to the next substage in a chain manner. After combining all the substages, this algorithm creates a gamma‐distributed delay with a specific variation to pass through each life stage, from egg to adult. Detailed formulations can be found in Appendix S3. Temperature‐dependent growth of eggs, larvae, pupae, and adults was modelled using the logistic function of Equation (1). The daily birth rate of female moths and the death rate at all stages were formulated using the same baseline logistic function. The daily mean temperature in 2019 at the Kanaya station was incorporated into the simulations.

All parameters related to growth speed, birth and death rate were estimated from previous studies in the laboratory and field (Tables S1 and S3 in Appendix S3). However, parameters for asymmetric larval interference were taken from the laboratory experiment of Nelson et al. (2019) and rescaled to the physiological scale of our study. No additional mortalities, such as winter hardiness, natural enemies, or pesticides, were assumed. We executed simulations with the same introduction scenarios as in the field‐cage experiment. Introductions in the simulations were modelled as a gamma function, where larval neonates are quickly recruited into the modelled population within a day (Equation S10 in Appendix S3).

In addition to the simulations above, we employed the same age‐structured/gamma‐chain model to explore the effect of an external perturbation, namely, mortality caused by rain, on synchrony among treatments. We conducted simulations with the four introduction scenarios, adding mortality caused by rain only during the egg period. It should be noted that this implicitly assumes the mortality of neonates just hatched from eggs and also the prevention of new egg recruitment by mated females, in addition to the direct mortality during the egg period (Equation S11 in Appendix S3.

2.3. Field population census in managed and abandoned tea orchards

The number of wild adults outside the field cages was monitored daily using pheromone traps with a synthetic female sex‐pheromone lure (SE lure) and with a 1‐m diameter water pan at the Kanaya station. The pheromone traps were placed at two locations (SE1 and SE2), at least 50 m apart from each other and from any field cage to avoid mutual interference. At the Kanaya station, scheduled pesticide spraying and pruning to maintain the rows of tea plants were regularly conducted, but away from the field cage.

In addition, three pheromone traps, equipped with sticky sheets (Sankei chemical, Kagoshima, Japan) and SE lures, were placed in the abandoned tea orchards, located 1.5 km south of the Kanaya station. The abandoned tea orchards measured approximately 200 m in length and 50 m in width in total, and it was confirmed that there were no pesticide applications, no plucking of tea leaves for harvesting, nor pruning of twigs to maintain the tea rows. The three traps, designated SE3, SE4 and SE5 from north to south, were placed at least 30 m apart, and the number of male catches was counted daily.

2.4. Wavelet analyses

To examine the changing generation cycles over time in both the field‐cage experiments and the managed and abandoned tea orchards, wavelet analyses were conducted (Carmona et al., 1998; Cazelles et al., 2014; Keitt, 2008). Wavelet analysis offers a powerful tool to visualise and quantify periodic patterns across different time scales, making it particularly useful when dominant cycles vary over time. The wavelet transformation is defined as

$$w_k\left(t,s\right)=s^{-1}\sum _i\psi \left(\frac{t-v_i}{s}\right)x_k\left(v_i\right),$$ (2)

where $\psi \left(\tau \right)$ is the Morlet base function. $x_k\left(v_i\right)$ is the population size at time‐ $v_i$. s is the period scales examined. The modulus of the wavelet transformation $\left|w_k\left(t,s\right)\right|$ is the wavelet power spectrum, which represents the strength of periodicity and can be mapped onto a two‐dimensional coordinate system with a time axis (t) and scale axis (s).

Before the analyses, we combined the data across treatments since the trap catch trends were similar within the treatments. Specifically, we summed the trap captures from the four replicates in the four treatments in the field‐cage experiment, as well as those from SE1 and SE2 at the Kanaya station, and from SE3, SE4 and SE5 in the abandoned orchard, respectively. These summed values were then smoothed, log‐transformed after adding 0.5, and standardised with a mean of 0 and a standard deviation of 0. The calendar dates of the moth catches were transformed to the physiological timescale ($\varphi$), resulting in irregularly sampled data. We calculated the wavelet power spectrum using the above processed time series records in $\varphi$‐scale with the mvcwt R‐library, as mvcwt is suitable for irregularly sampled data (Keitt, 2008; R Development Core Team, 2025).

The significance of dynamical changes in cyclic dynamics was evaluated by performing randomisation tests. We created 500 surrogate records by randomising the phases of a Fourier transform, preserving the power spectra of the original data (Cazelles et al., 2014). The Ebisuzaki method in the rEDM R‐library was employed to generate surrogate data (Sugihara et al., 2012). We calculated the wavelet power spectra of the 500 surrogates and estimated one‐sided 95% confidence limit for the null hypothesis, assuming that the cyclic behaviour of the population dynamics had not changed over time.

We also calculated the localised wavelet modulus ratio among the pheromone trap catches at the station, in the abandoned tea orchard, and in the field‐cage experiments to evaluate inter‐site synchrony. The wavelet modulus ratio is widely used to detect population synchrony and to visualise temporal changes in dominant cycles and their variability across different time scales (Keitt, 2008; Lasky et al., 2016; Vasseur et al., 2014). It is calculated as,

$$\rho \left(t,s\right)=\frac{{\Lambda }_{t,s}\left(\left|\sum _k{W}_k\left(t,s\right)\right|\right)}{{\Lambda }_{t,s}\left(\sum _k\left|W_k\left(t,s\right)\right|\right)},$$ (3)

where ${\Lambda }_{t,s}\left(·\right)={\int }_{-\infty }^{\infty }{e}^{-\frac{1}{2}{\left(\frac{t-\tau }{s}\right)}^2}\left(·\right)\mathit{d\tau }$ and $\left|·\right|$ executes modulus calculation. The numerator measures the total variation among populations, while the denominator quantifies the combined variation of all the populations. If the populations fluctuate asynchronously over time, the modulus ratio tends to zero. Conversely, it approaches one if the variations among populations become synchronous. To test the significance of inter‐site synchrony, we performed a randomisation test, creating 500 surrogate records using the Ebisuzaki method for all the station, abandoned orchard, and field‐cage experiments, respectively, as described in the previous section. We also calculated the global modulus ratio (the mean over time) and the mean modulus ratio during the generational periods (29.7 ≤ ϕ ≤ 33.5) throughout the time series.

2.5. Ethical approval

This study did not require ethical approval or permission as they were carried out in areas where such permissions are not mandated.

3. RESULTS

3.1. Field‐cage experiment

The results of the field‐cage experiments for the four introduction treatments, with all four replicates, are shown in Figure 1. The population dynamics in all treatments displayed generation cycles around the generational period (ϕ = 30.7), especially in the latter part of the season. For the pulsed introduction of treatments (a) and (b), the first pulse of adults appeared at the appropriate time after the growth period of the introduced larvae. However, the second pulse showed a slight delay in treatment (a) and a slight acceleration in treatment (b). Finally, the conspicuous and synchronous cycles emerged in the third and fourth pulses in the latter part of the season, regardless of the introduction timing. The emergence of the generation cycles was significant according to the randomisation test. Although there were no clear generation cycles at the beginning of the experiment, the cycles became more distinct in the latter part of the season.

FIGURE 1.

Male moth catches in the field‐cage experiments (left panels) and their wavelet power spectrums (right panels). Larvae were introduced in one time in (a and b) while continuously in (c and d). The introduction dates were shifted half a generation later in (b and d). The arrows indicate the timing of introduction. Heat maps in the wavelet power spectrums show the gradation from low (navy blue) to high (red) power, and the results inside the dotted lines are significant. Blue tick lines in the wavelet power spectrums represent the 5% significance levels based on randomisation test with 500 iterations. The time schedule was transformed into the physiological timescale (ϕ), but ticks on the horizontal axes are located at every generational time and labelled with the calendar date. Lines at ϕ = 30.7 represent the generational time of A. honmai.

For the continuous treatments (c) and (d), adults appeared repeatedly before the completion of the introduction procedure (Figure 1). These individuals were those introduced at the earlier time points. Since then, adults have appeared continuously and intermittently in treatment (c) until the large pulse in mid‐September. In contrast, there was a small but conspicuous peak of adult emergence around August 7th in treatment (d), as seen in treatment (b). The noticeable and synchronous cycles emerged in the latter part of the season around the generational period (ϕ = 30.7), even for the continuous introduction. These distinct cycles in the latter part of the season also appeared synchronous across all treatments. This will be tested using the wavelet modulus ratio in Section 3.3.

3.2. Simulation analyses

The simulation results with the same introduction scenarios in the field cage experiment are shown in Figure 2. The simulated dynamics differed from those in the actual field‐cage experiment, displaying clear regular generation cycles and maintaining a consistent pace around the generational period (ϕ = 30.7). The cycles depended on the timing of the introduction. Since the cyclic behaviours were relatively constant compared to those in the field‐cage experiments, changes in the cycles were not significant around the generational period compared to those in the field‐cage experiment (Figure 2, right panels).

FIGURE 2.

Adults in the simulations (left panels) and their wavelet power spectrums (right panels). Larvae were introduced in one time in (a and b) and continuously in (c and d). The introduction dates were shifted a half generation later in (b and d). The arrows indicate the timing of introduction. Heat maps in the wavelet power spectrum show the gradation from low (navy blue) to high (red) power, and results within the dotted lines are significant. Blue tick lines in the wavelet power spectrums represent the 5% significance levels based on randomisation test with 500 iterations. Lines at ϕ = 30.7 represent the generational time of A. honmai.

The timings of the first and second peaks of adults in the pulsed introduction were somewhat concordant with the simulation results (Figures 1a,b and 2a,b). However, the third and fourth peaks of adults emerged in an asynchronous manner in the simulation, while those in the field cage were synchronised regardless of the introduction timings. This suggests that other factors may have affected the field‐cage populations beyond larval interference.

In the simulations of the continuous introduction, there were substantial blank periods before the first adult emergence (Figure 2c,d). Due to strong larval interference in the model assumption, only individuals from the final introduction could grow to adulthood in the simulations. In contrast, a small number of individuals emerged in the field‐cage experiment during the introduction procedure (Figure 1c,d). We suspect that some larvae may have escaped from intraspecific interference because the chambers in the field cage likely provided more space for propagation than in the pinpoint simulations. The difference between the simulations and the field‐cage experiment became more apparent after August: the adult emergence patterns in the simulations maintained their generation period, whereas those in the field‐cage experiments changed and became synchronised with each other.

3.3. Trap catches in the managed and abandoned tea orchards

There were five clear generational cycles in the managed tea orchard (Figure 3a), while the third and fourth peaks were not evident in the abandoned orchards (Figure 3d). Unlike the results in the field‐cage experiment, the adult catch dynamics in the managed and abandoned tea orchards were regular and consistent, somewhat similar to the simulation of treatment (a) (Figure 2a). Since the cyclic behaviours in both the managed and abandoned tea orchards were relatively constant but clearer in the early part of the season than in the latter part after August, changes in the cycles were significant in May (Figure 3, right panels).

FIGURE 3.

Male moth catches in the managed tea orchard (a) and in the abandoned tea orchard (b). Corresponding wavelet power spectrums are shown in the right panels. Heat maps in the wavelet power spectrum show the gradation from low (navy blue) to high (red) power, and results within the dotted lines are significant. Blue tick lines in the wavelet power spectrums represent the 5% significance levels based on randomisation test with 500 iterations. The time schedule was transformed into the physiological timescale (ϕ), but ticks on the horizontal axes are located at every generational time and labelled with the calendar date. Lines at ϕ = 30.7 represent the generational time of A. honmai.

Although the generation cycles in the managed and abandoned tea orchards resembled the simulation of treatment (a), they also appeared synchronous with those in the field‐cage experiment (Figures 1 and 3). This will be tested by the wavelet modulus ratio in Section 3.3 below. It should be noted that the trap catches in the managed and abandoned tea orchards were 6.4 to 21.3 times larger than those in the chambers of the field‐cage experiment. This was because the traps in the station collected male moths in the open field, while the traps in the chamber only caught males in a small area.

3.4. Wavelet modulus ratio among populations and simulations

To estimate inter‐populational synchrony, we calculated the localised wavelet modulus ratio among trap catches at the station, in the abandoned tea orchard, and in the four treatments of the field‐cage experiment (Figure 4a). The modulus ratio was generally high ($\approx$1.0) throughout the period, especially in the latter part of the season. The global modulus ratio around the generational period (ϕ = 30.9) was significantly coherent based on the randomisation test (Figure 4b). Additionally, we observed an increase in synchrony among populations at the scale of the generational period ($29.7\le \varphi \le 33.5$) in Figure 4c. It should be noted that significant synchrony was detected from the end of June, while the peaks in the field‐cage experiment became more pronounced later in September and October (Figure 1). Although it was difficult to confirm visually from the results of trap catches in the field experiment, some synchronised emergence of adults may have started from the end of June. Furthermore, we observe small peaks of adult emergence around 3rd August before the clearer peaks in September and October across all four treatments (Figure 1).

FIGURE 4.

Modulus ratio among all pheromone trap records from four treatments of the field‐cage experiment, as well as in both managed and abandoned tea orchards in (a). The global modulus ratio (the summation over time) is shown in (b). The summation of the modulus ratio over the generation time ($29.2\le \varphi \le 34.0$) throughout the time series is also shown in (c). Blue tick lines in (a) represent the 5% significance levels (the area of the significant cycles covered by a large region coloured in red). Red dotted lines in (b) and (c) indicate the 95% confidence limit based on randomisation test with 500 iterations; mean modulus ratio values above this line represent significant synchrony among populations.

The wavelet modulus ratio among simulation results of the four treatments, including trap catches in the managed and abandoned tea orchards, was calculated to compare with the field‐cage experiment (Figure 5a). Since the adult emergence patterns in the simulations were asynchronous among the treatments, the wavelet modulus ratio was generally low throughout the period, especially in the latter part of the season. Both the global modulus ratio and the modulus ratio around the generational period were not significant throughout the period (Figure 5b,c).

FIGURE 5.

Modulus ratio from simulation results and in both managed and abandoned tea orchards in (a). The global modulus ratio (the summation over time) is shown in (b). The summation of modulus ratio over the generation time ($29.2\le \varphi \le 34.0$) throughout the time series is also shown in (c). Blue tick lines in (a) represent the 5% significance levels, with significant cyclic dynamics found in the bottom and left parts (the opposite side to those in Figure 4). Red dotted lines in (b and c) indicate the 95% confidence limit based on randomisation test with 500 iterations; mean modulus ratio values above this line represent significant synchrony among populations.

3.5. Simulation with rain mortality

An additional external factor, rain mortality, substantially altered the timing of the simulated generation cycles (Figure 6). Although we did not assume growth delay at any stage, asymmetric larval interference with rain mortality for eggs generated synchronous dynamics among the four treatments in the latter part of the season. The synchronous dynamics across the four introduction scenarios can be observed over a wide range of parameters if the mortality is strong enough to significantly reduce egg numbers (Appendix S4). Rain mortality should not be restricted to the egg stage; mortality during larval period also creates inter‐populational synchrony (Appendix S5). While rain mortality should be substantial, if it exceeds a certain threshold, some populations may go extinct during the simulation, depending on the timing of rainfall.

FIGURE 6.

Simulated population dynamics with the four introduction scenarios in the field‐cage experiment, which includes egg mortality caused by rain. Parameters and model structure can be found in Appendix S3.

It is important to note that the synchronous cycles observed in the simulations were not immediately influenced by rainfall; adult populations did not exhibit local fluctuations at the time of rain. The simulated populations continued to display clear generational cycles, but they gradually became synchronised toward the end of the season (Figure 6). Rain‐induced mortality reduces egg and early‐larval densities but does not directly affect older larvae or adults. Under low‐density conditions, weakened intraspecific interference allows most older larvae to survive, often leading to outbreaks after rainfall. Thus, rainfall likely acted as a pacemaker.

4. DISCUSSION

We found that the introduced populations in the field‐cage exhibited clear generation cycles late in the season, both in the pulsed and continuous introduction treatments (Figure 1). Although only adult males could be monitored using pheromone traps, we assumed that the abundance of adult males accurately tracked overall adult population dynamics, including females, because field records from pheromone and light traps were highly concordant in A. honmai (Smith et al., 2025). Therefore, we conclude that generation cycles, where individuals at the same developmental stage exhibit cycles approximately equal to the generation length from egg to adult, did occur in our field‐cage experiments. The simulation results further supported this interpretation, showing that adults and larvae emerged in a cyclic manner (Figure 2).

The field‐cage treatments were designed to directly test whether the cycles resulted from over‐winter demographic synchronisation. Since disrupting synchronisation still yielded strong generation cycles, we can conclude that the generation cycles are not caused by transient dynamics after winter. Considering that the field‐cage populations, as well as those in the abandoned tea orchard, were not influenced by pesticides or plucking, the generation cycles of A. honmai were not caused by human intervention. Therefore, some biotic regulation must be generating the generation cycles.

Several field reports of generational cycles indicated that intra‐ or interspecific interactions could be a cause of these cycles. For example, Godfray and Hassell (1989) postulated that the cycles of the coconut moth, Levuana iridescens, were caused by the host‐parasitoid interaction, with the difference in lifespan between host and parasitoid being key in generating these cycles. In our field‐cage experiment, the fine mesh net prevented the majority of natural enemies, but Trichogramma spp. could have penetrated the chambers due to their small size (Ishijima et al., 2009). We cannot eliminate the possibility of this host‐parasitoid interaction as a cause of generation cycles in A. honmai. However, we introduced egg masses that were about to hatch in metal cases. The metal mesh was fine enough to prevent Trichogramma spp., and there was little chance of parasitism, at least at the beginning of the experiments. Moreover, Takagi (1974) reported that Trichogramma spp. are generalist parasitoids and are not the main natural enemy of A. honmai. We suspect that Trichogramma spp. did not influence the cyclic behaviour of A. honmai in the field‐cage experiment.

Alternatively, we hypothesise that intraspecific interference, including cannibalism among larvae, was the main cause of the generation cycles, as several previous works reported frequent cannibalistic behaviour among larvae of A. honmai (Nelson et al., 2019; Simpson et al., 2018). Intraspecific interference appeared to be more severe in the latter part of the season than in the early period following introduction, as the generation cycle became more conspicuous. In fact, Indian meal moths in laboratory settings and Indonesian ladybeetles in tropical crop fields demonstrated clear generation cycles due to intraspecific cannibalism (Briggs et al., 2000; Nakamura et al., 2004). Our simulation model also predicted clear generation cycles based on the same parameters as our field‐cage experiment (Figure 2). It should be noted that the simultaneous introduction of 8000 individuals in the pulsed introduction treatments (a) and (b), where the population size was four times larger than the field density of older larvae, may have caused temporary intense interference among larvae. However, the subsequent adult emergence from those initial larvae did not reveal clear generation cycles as observed later in the season (Figure 1), which means that synchronisation emerged in later generations. We suspect that the population size declined to a certain level due to asymmetric interference and background mortality and that the intensity of larval interference was milder than in the later seasons.

Unexpectedly, the population dynamics in the cage experiment eventually became aligned across treatments, even though the populations were introduced with a half‐generation lag in treatments (b) and (d) (Figure 1). Moreover, they synchronised with those in the managed and abandoned tea orchards (Figures 3 and 4). We suspect that this synchronisation was caused by the ‘Moran effect,’ a well‐known phenomenon where populations experiencing synchronous environmental fluctuations tend to synchronise, especially if their dynamics are cyclic (Bjørnstad, 2000; Moran, 1953; Reuman et al., 2025). Numerous empirical studies highlight the role of external factors in driving spatial synchrony among populations. For example, isolated sheep populations on separate islands can show clear synchronisation if they are internally regulated and exposed to strong, frequent external perturbations (Grenfell et al., 1998). In addition, Reuman et al. (2025) pointed out that populations with cyclic dynamics tend to synchronise easily. For example, spongy moth defoliation is well‐known to have cyclic population dynamics, and the precipitation shared by regional populations acts as a ‘pacemaker’ synchronising the populations (Haynes et al., 2019). A. honmai exhibits pronounced generational cycles, which may be attributed to intraspecific cannibalism. This characteristic might facilitate inter‐populational synchrony more readily than in other animal species.

Though we could not directly observe how external causal factors influenced synchronisation in the cage experiment, we suspect that heavy rains during June and July 2019 might have caused mortality in the young larval stages or mating suppression in adults. Moths generally experience high mortality during their early instars (Kamata & Igarashi, 1994; Rahman et al., 2019), and their mating behaviour is known to be suppressed by rainfall (Paulen & Kobolka, 2018; Pitcairn et al., 1990). Typically, there is a long and continuous rainy season from mid‐June to late July across wide areas of Japan. At Kanaya station in 2019, the rainy season started on 7th June, with heavy intermittent rain occurring particularly from 4th to 6th, 14th to 17th, 22nd and 26th July (Japan Meteorological Agency, https://www.jma.go.jp, accessed November 2023). In our simulation with rain mortality at the egg stage, we found synchronised generation cycles among the different treatments (Figure 6). Although we still do not know how the rain affected mortality at each stage, and other meteorological factors may have also influenced the population, some external factors, possibly rain mortality, may have acted as a pacemaker for the generation cycles.

Previous research showed that synchrony among populations is also induced by dispersal (Fox et al., 2011; Kendall et al., 2000). The chambers in our field‐cage were intended to be completely separate, but the stuffing materials (the polyethylene sponge tubes) fell off, resulting in small gaps at the lower part of the cage roofs between most pairs of chambers. This likely occurred when a typhoon hit the area in early August. These gaps were small (approximately 2 cm × 48 cm at maximum) and located at a high position, considering the flight activity of A. honmai (1.8 m in height). We can assume that female moths tend to be sedentary, especially when in a suitable habitat (Shirai & Kosugi, 2000). Therefore, the majority of females may not have migrated among cages and did not substantially affect population growth.

In conclusion, our findings indicate that a biotic regulation mechanism, potentially involving asymmetric larval interference and the Moran effect, such as rain mortality, acted in concert to produce the clear and synchronised generational cycles of A. honmai, a multivoltine insect inhabiting temperate zones. The mechanism of winter synchronisation is unlikely to contribute to strong cycles in this pest insect. These cycles have often been perceived merely as transient dynamics after winter, particularly by many field practitioners. Although several research reports have elaborated on the mechanisms of these generation cycles through theoretical models or statistical analyses based on long‐term records, our designed field‐cage experiment provides new insights into how internal mechanisms and external perturbations generate synchronised large‐scale generation cycles.

AUTHOR CONTRIBUTIONS

Takehiko Yamanaka., Ottar N. Bjørnstad. and William A. Nelson. designed the study. Yasushi Sato conducted experiments and compiled the data. Takehiko Yamanaka and William A. Nelson executed statistical analyses.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

STATEMENT ON INCLUSION

All authors contributed to the writing. Our study was an internationally collaborative work and includes all the scientists who carried out the experiment as the authors.

Supporting information

Appendix S1. Details of field‐cage experiments.

Appendix S2. Physiological time‐scale (ϕ) transformation.

Appendix S3. Age‐structured model.

Appendix S4. Simulation results incorporating rain mortality at the egg stage.

Appendix S5. Simulation results incorporating rain mortality at the young larval stage.

Yamanaka, T. , Sato, Y. , Nelson, W. A. , & Bjørnstad, O. N. (2026). Generation cycles in experimental populations of a multivoltine insect. Journal of Animal Ecology, 95, 782–794. 10.1111/1365-2656.70229

DATA AVAILABILITY STATEMENT

The data and scripts that reproduce the results of this study are provided and can be accessed via an OSF link (https://doi.org/10.17605/OSF.IO/JKZ3A; Yamanaka et al., 2026).