Decadal changes in masting behaviour of oak trees with rising temperature

Decadal changes in masting behaviour—directional changes in seed production with fluctuations on a decadal time‐scale—are attracting widespread attention in the context of global climate change. However, our mechanistic understanding of the effects of climate on seed production on a decadal scale is unsatisfactory, partly because of the insufficient statistical analyses of long‐term data on masting. We detected decadal changes in masting behaviour in the Japanese oak Quercus crispula based on long‐term data (38 years: 1980–2017) from the Kitakami Mountains of Japan. The moving average of seed production in a 20‐year sliding window increased, whereas the coefficient of variation decreased. A wavelet power spectrum, as well as a second‐order log‐linear autoregressive (AR) model showed that masting intervals shortened from 3‐ or 4‐year cycle to a 2‐year cycle. The moving average of seed production increased linearly as the moving average of temperature increased. Temporal variations of the two AR model coefficients as a function of temperature were well described by concave curves. Synthesis. By conducting the statistical analyses of a long‐term seed production dataset, we obtained significant evidence of decadal changes in the masting behaviour of the Japanese oak and showed that the shortening of the masting interval was associated with rising temperature. A resource allocation shift and an environmental veto were discussed as possible mechanisms underlying the decadal change.

regeneration of trees, their population dynamics and the evolution of seed consumers, and it eventually affects the dynamics of the entire community (Jones et al., 1998;Koenig, Knops, Carmen, & Pearse, 2015). Masting even affects the disease risk to humans. For example, acorn masting increases the risks of people contracting hantavirus (Reil, Imholt, Eccard, & Jacob, 2015) or Lyme disease (Bogdziewicz & Szymkowiak, 2016) through increases in the populations of rodent vectors. For a deeper understanding of forest ecosystems, and for sustainable forest management planning, information about the determinants and underlying mechanisms of masting is essential (Pearse, Koenig, & Kelly, 2016;Vacchiano et al., 2018).
Decadal changes in masting behaviour-directional changes in seed production with fluctuations on a decadal timescale-are attracting widespread attention in the context of global-scale environmental change (Ascoli et al., 2017) and impacts on ecological interactions (McKone, Kelly, & Lee, 1998). Seed production of various temperate species is increasing (Allen, Hurst, Portier, & Richardson, 2014;Ascoli et al., 2017;Buechling et al., 2016;Caignard et al., 2017;Richardson et al., 2005), although decreases in seed production have also been reported (Redmond, Forcella, & Barger, 2012). A shortening of masting intervals has been observed in beeches in northern Europe (Müller-Haubold, Hertel, & Leuschner, 2015;Övergaard, Gemmel, & Karlsson, 2007), whereas a meta-analysis of world-wide datasets has shown an increase in inter-annual variation in seed production (Pearse, LaMontagne, & Koenig, 2017). Suggested causes of decadal changes in masting behaviour include global changes in, for example, climate and nitrogen deposition (Caignard et al., 2017;Övergaard et al., 2007), but Kelly et al. (2013) have reported that masting may be insensitive to gradual increases in temperature. These discrepancies in findings on masting behaviour and our understanding of its mechanism can be attributed not only to interspecific differences in the response to climate change  but also to an insufficient understanding of how climate and resources drive seed production (Pearse et al., , 2017. The effect of weather as a proximate cause of masting is of interest from at least two perspectives-resource dynamics and pollination (Allen, Millard, & Richardson, 2017;Koenig et al., 2015;Pearse, Koenig, & Knops, 2014;Richardson et al., 2005). Many studies examining resource dynamics (Bogdziewicz, Steele, Marino, & Crone, 2018;Bogdziewicz et al., 2019;Pesendorfer, Koenig, Pearse, Knops, & Funk, 2016;Satake & Bjørnstad, 2008;Schermer et al., 2019;Venner et al., 2016) have adopted the resource budget model, which assumes that plants accumulate resources each year and produce seeds once the stored resources exceed a threshold (Isagi, Sugimura, Sumida, & Ito, 1997). Recovery from resource depletion by reproduction may be faster in higher productivity environments, enabling plants to reproduce more frequently (Kelly & Sork, 2002;Satake & Bjørnstad, 2008).
Moreover, Satake and Bjørnstad (2008) have shown that the input of surplus resources (net production per year) can be a determinant of local masting interval variations. This idea of a relationship between surplus resources and local masting intervals may help explain decadal changes in masting behaviour. Weather factors-particularly temperature and precipitation-are considered to be the drivers of resource priming for reproduction because of their effects on photosynthesis and productivity in trees (e.g. Allen et al., 2014Allen et al., , 2017Müller-Haubold et al., 2015;Piovesan & Adams, 2001;Richardson et al., 2005;Smaill, Clinton, Allen, & Davis, 2011). Therefore, we predicted that, in response to favourable temperatures and precipitation, trees would more frequently shift their resource allocation pattern from storage to reproduction, thus shortening the masting interval. Furthermore, an enhanced resource supply may lead to an increase in the total mass of seed produced.
However, resource dynamics alone cannot explain the population-level synchrony of seed production, which is an essential feature of masting (Crone & Rapp, 2014;Isagi et al., 1997). Flowering or pollination may be a key to that synchrony. One proposed the mechanism of synchronization is pollination efficiency, which is often referred to as pollen coupling (Isagi et al., 1997;Satake & Iwasa, 2000). However, the pollination Moran effect  may be more applicable as a mechanism of synchronization in oaks (Bogdziewicz et al., 2017). In support of a pollination Moran effect,  showed that pollination failures caused by environmental vetoes could synchronize the dynamics of seed production among individual trees.
The environmental veto hypothesis, according to which specific weather conditions can prevent individual trees from flowering or being pollinated over a substantial area, may thus be a plausible mechanism linking masting (seed production synchrony) with climate Pearse et al., 2016). Koenig et al. (2015) showed that low springtime temperatures caused pollination failure in the valley oak Quercus lobata. In turn, poor seed production due to pollination failure in a low-temperature year may contribute to the tuning of the resource dynamics of individual trees and thus to synchronization of mast years . Therefore, we can predict that low temperatures in spring sharpen masting in oak species.
A variety of statistical techniques have been developed for analysing the time series of cyclic fluctuations or spatial synchrony in animal populations (e.g. Bjørnstad, Falck, & Stenseth, 1995;Cazelles et al., 2008;Haydon, Stenseth, Boyce, & Greenwood, 2001;Royama, 1992), and some of these techniques have been adopted in recent studies on masting (Ascoli et al., 2017;Chen, Brockway, & Guo, 2018). Although the mast year frequency (or masting interval) has been measured as a key component of temporal variability in seed crops (Abrahamson & Layne, 2003;Allen, Mason, Richardson, & Platt, 2012;Kasprzyk, Ortyl, & Dulska-Jeż, 2014), frequency analyses are prone to bias as a result of the subjective categorization of years into mast or non-mast years.
In contrast, a wavelet analysis can describe the decadal changes in population fluctuations avoiding the bias associated with a dichotomous categorization (Cazelles et al., 2008), and Ascoli et al. (2017) have introduced the wavelet analysis as a technique for investigating long-term changes in masting patterns. A second-order log-linear autoregressive (AR) model has also been used to describe the decadal changes in wildlife populations (e.g. Cornulier et al., 2013;Ims, Henden, & Killengreen, 2008). The AR model can be used to investigate how the current population density reflects population densities during the previous 1 and 2 years. Seed production in a certain year may be determined by the interaction between the input (photosynthesis) and output (seed production) of resources in previous years; thus, the AR model may be useful for analysing inter-annual fluctuations of seed crops.
The aim of this study was to investigate the decadal changes in the masting behaviour of the Japanese oak Quercus crispula in relation to climate change. This deciduous tree species is a principal component of cool-temperate forests in Japan. In recent decades, atmospheric temperature and precipitation in the study area have been increasing. We therefore performed time-series analyses by using the wavelet power spectrum and an AR model for seed production in Q. crispula in the Kitakami Mountains of Japan, where seed production by this species has been observed for 38 years . Our results indicate that the masting interval shortened over this 38-year period. In addition, average number of seed production and the mass of individual seeds increased, whereas the coefficient of variation (CV) of seed production decreased.
Variations in weather factors-in particular temperature-showed a good agreement with AR model results, variations of average number and CV of seed production. Then, as possible mechanisms underlying the observed decadal changes in masting behaviour, we examined both a shift in the allocation of resources and an environmental veto.

| Study species
The Japanese oak Q. crispula (a synonym of Q. mongolica var. grosseserrata) is a broad-leaved deciduous tree species in the family Fagaceae. It is native to East Asia and is distributed widely in Japan (Yagihashi, Matsui, Nakaya, Taoda, & Tanaka, 2003 (Figure S1). Along with Fagus crenata, Q. crispula is often dominant in cool-temperate forests, though it occasionally forms monospecific stands (Ohba, 2006). Quercus crispula is monoecious and anemophilous. The seeds develop slowly for about 2 months after the completion of flowering and more rapidly thereafter (Nakajima et al., 2012).

| Study site and field research
The field survey was performed in a Q. crispula stand at the Nakaimura research site (39°48′32″N, 141°33′27″E, 900 m a.s.l; Figure S1), which is located in a national forest in the central region of the Kitakami Mountains. Although the stand history is not known in detail, the stand age is estimated to be 135 years (Sanriku-Hokubu District Forest Office, 2016).
Since 1980, annual seed production has been monitored in a plot (35 m × 35 m; 0.12 ha) in this research site ( Figure S1); it contained a total of 56 stems (467 stems/ha)-mainly Q. crispula-with a DBH of 28.7 ± 9.6 cm (M ± SD). A grid was superimposed on the plot, and 16 seed traps were set at the grid nodes. All reproductive organs, including both seeds and cupules, that fell into the seed traps each year between mid-July and early November were collected over the

| Annual fluctuation of seed production
The change in the masting interval was analysed by wavelet analysis, using the WaveletComp package 1.1 (Roesch & Schmidbauer, 2018) in the R statistical environment (R Development Core Team, 2013). This analysis is appropriate for data with irregular timeseries characteristics and is superior for quantifying the dynamics of populations with non-stationary dynamics (Cazelles et al., 2008). The significance of the wavelet power spectrum, which shows time-series changes at time t in the analysis period, was tested by bootstrapping with 10,000 replications. A calculation process based on Markov process-based resampling of observed values was adopted.
The AR model was also used to describe the temporal changes in the annual fluctuation of seed production. The equation for the AR model, which was originally used to describe the change in population density between year t and year t − 1 (i.e. x t − x t−1 ) as a function of density, was transformed into the following simple one (Royama, 1992): where x t represents the natural logarithm of annual seed production (the number of seeds) in year t, x t = log e (N t ). The coefficients [1 + a 1 ] and a 2 represent the effects of seed production during the previous 1 and 2 years, respectively, on seed production in the current year t, and ε t is an error term for density-independent effects in year t (independent random numbers with mean = 0).  Figure 3c (see also Bjørnstad et al., 1995). Temporal changes in the masting interval cause a point plotted in a 2 − [1 + a 1 ] space to be relocated over time ( Figure 3c). Therefore, masting pattern changes can be traced by observing how plots of the AR model coefficients move over time.
State-space model was used to estimate the number of mature seeds and the two coefficients of the AR model. We used JAGS ver. 4.3.0 (Plummer, 2017) and the Markov Chain Monte Carlo (MCMC) method, taking measurement errors into consideration, to estimate the number of seeds per 1.0 m 2 in the 16 seed traps and the two coefficients of the AR model ([1 + a 1 ] and a 2 ) governing all 16 time series. The numbers of chains and iterations were set at 3 and 50,000 respectively; the initial burn-in was set at 10,000 and thinning was set at 40 (see the JAGS code, Appendixes S3 and S5 and seed-fall data, Appendix S4). Good convergence (R < 1.02) of MCMC was confirmed for the 608 estimates of the number of seeds (16 traps × 38 years; Figure S2).
Although in general a shorter time frame provides a more sensitive estimation of the variation of [1 + a 1 ] and a 2 , the estimates based on short time frames may not be robust because of the small sample size (i.e. the number of years). Therefore, to explore the characteristic length scale at which the underlying dynamics could be most clearly observed (Habeeb, Trebilco, Wotherspoon, & Johnson, 2005), we used the sliding window method and compared the results from various time frames. For example, when the window size (i.e. time frame) was set at 11 years, a 1 and a 2 were estimated for each of 28 periods (1980-1990, 1981-1991, …, 2006-2016, 2007-2017). We compared the convergence (R), stability and sensitivity of the a 1 and a 2 estimates among 15 window sizes (11-25 years). Estimates of a 1 and a 2 obtained with smaller window sizes (≤13 years) showed poor convergence (R > 1.1, Figure S3), whereas convergence was satisfactory with larger windows (>13 years, R ≤ 1.006, Figure S3). In accordance with the reduction in erratic behaviour ( Figure S3), the stability, which was assessed based on the change rate of the coefficients between neighbouring years, became higher as the size of window increased ( Figure S4). The stability levelled off when the window exceeded 20 years. In contrast, the sensitivity, which was represented by the range of the coefficients, decreased with the increase in the window size in general ( Figure S4). However, it was stable at intermediate values when the window size ranged between 15 and 21 years ( Figure S4). Considering these results, we adopted a 20-year time frame as the characteristic length scale.
In addition to [1 + a 1 ] and a 2 , the average and CV of seed production were calculated by using a 20-year sliding window. Furthermore, seed weight was compared among those years in which data were available by using Tukey-Kramer's HSD test.

| Weather effects on seed production
The effects of climate change on masting behaviour were analysed by examining the temperature and precipitation during the growing and flowering seasons. At the study site, Q. crispula blooms and develops new leaves from late May to early June, and mature seeds and leaves fall from mid-October to early November. In a cool-temperate forest located 500 km south-west of our study site, where the annual mean temperature is 7.2°C, Q. crispula actively photosynthesize between June and September (Muraoka & Koizumi, 2005). The annual mean temperature at our study site is comparable to that during the 38-year study period, and it ranged between 4.4 and 7.1°C. Therefore, at our study site, we considered the growing season of Q. crispula to be between June and September and the flowering season to be from late May to early June.
We used meteorological data recorded at the Yabukawa Meteorological Station (39°47′00″N, 141°19′42″E, 680 m a.s.l), which is 20 km south-west of the study site. The data were downloaded from the website of the Japan Meteorological Agency (2019).
The 20-year moving averages of mean temperature and precipitation during the growing season were analysed as proxies for productivity. Similarly, the 20-year moving averages of mean temperature and precipitation during the flowering season were used as those for environmental vetoes.
To explore the influence of weather factors on masting behaviour, we compared the results of three regression models (linear, quadratic and multivariate). The explanatory variables were the 20-year moving averages of temperature or precipitation, or both, in each season, and the objective variable was average seed production, its CV or one of the two AR model coefficients ([1 + a 1 ] or a 2 ) in the 20-year sliding windows.

| Annual fluctuation of seed production
The average of seed production of the 16 seed traps fluctuated widely, from 0.5 in 1988 to 117.7 in 1987 ( Figure 1a); the average throughout the study period was 28.3, with a large standard deviation (31.9). During the study period, high seed production never occurred in two successive years, but successive years of low abundance were observed several times between 1980 and 2000. The interval between peak years-where a peak year is one in which seed fall was higher than in adjacent years-ranged from 2 to 4 years until  (Figure 3a, b). The net value of a 1 was always negative, ranging from −1.782 to −1.374 ( Figure 3a) and its variance was small (0.016). Because none of the 95% credible intervals included zero, the effect of a 1 was always significant. In contrast, a 2 exhibited large variation (variance = 0.040) and ranged between negative and positive values ( Figure 3b). Assessment of the significance of a 2 on the basis of the 95% credible interval revealed that its effect was significantly negative in nine periods (in the 4th and from the 6th to 13th period), significantly positive in two periods (18th and 19th) and non-significant in eight periods (1st to 3rd, 5th and 14th to 17th). The changes in both coefficients [1 + a 1 ] and a 2 were well described by a concave approximation against period (Figure 3a When we plotted the values of each coefficient in each period in a 2 − [1 + a 1 ] space, the first period (1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999) was located on the contour indicating a 3-year cycle (Figure 3c, d). The positions of the plotted points moved downward along the contour, indicating a 3-year cycle as the effect of a 2 became stronger (i.e. as a 2 became more negative).

| Weather effects on seed production
In the growing season (June-September), monthly mean temperature showed a gradual upward trend, with high inter-annual fluctuation ( Figure S6). Its 20-year moving averages increased monotonically with period number in all months ( Figure S6; Table S1). Although monthly precipitation showed a gradual increasing trend in July and August, its moving average fluctuated more than that of temperature ( Figure S6; Table S1).
The 20-year moving average of seed production showed good agreement with that of mean temperature in the growing season ( Figure 4a). In the linear regression model, but not in the quadratic model, temperature had a significant positive effect on average seed production (Table 1). Similarly, precipitation had a significant positive effect on the moving average of seed production in the linear model but not in the quadratic model (Table 1). The multivariate model with two explanatory variables (temperature and precipitation) showed a significant contribution of temperature and not of precipitation (Table 1). Moreover, the moving averages of temperature and precipitation were highly correlated with each other (r p = .847, t = 6.566, p < .0001). Therefore, we focused on the effects of temperature to the CV of seed production and masting interval. Temperature had a significant negative effect on the CV of seed production in the linear model but not on the quadratic model ( Table 2). The regression of coefficient [1 + a 1 ] against temperature was significant only in the quadratic model (  Figure 4b). Although the regression of a 2 was significant in both the linear and quadratic models, the quadratic model showed higher adjusted coefficient of determination than the linear model (  Figure 4b).
In the flowering season (late May to early June), the moving average of temperature increased monotonically, whereas that of precipitation showed a decreasing trend ( Figure S6; Table S1). Temperature had a significant positive effect on seed production in the linear model but not in the quadratic model (Table S2). Similarly, precipitation had a significant negative effect on the seed production in the linear model but not in the quadratic model (Table S2). The multivariate model with the two explanatory variables (temperature and precipitation) revealed a significant contribution of temperature but not of precipitation (Table   S2). Therefore, we focused on the effects of temperature even in the flowering season. Temperature also had a significant negative effect on the CV of seed production in both the linear and quadratic models (Table S3). The regression of coefficient [1 + a 1 ] against temperature was significant in the quadratic model (Table S3), whereas the regression of coefficient a 2 against temperature was significant in both the linear and quadratic models (Table S3).
In addition to the masting interval, average seed production may show temporal changes. During our study period, we observed an increasing trend in seed production (Figure 2), which was consistent with trends reported for two Quercus species in Europe (Caignard et al., 2017) and one Nothofagus species in New Zealand (Allen et al., 2014;Richardson et al., 2005) but opposite to the trend of the pinyon pine (Pinus edulis), which has shown a decrease in cone production in North America (Redmond et al., 2012).
Our temporal variation results are basically opposite to the seed production trends found in a meta-analysis conducted by Pearse et al. (2017). The meta-analysis showed an increase in the temporal variation represented by the CV, whereas in this study the temporal variation represented by the CV decreased. There may be an interaction between the temporal variation of seed production (CV) and the masting interval; in the studied forest, the increasing frequency of the output of stored resources (i.e. the shortening of the masting interval) during the study period may have reduced the CV of seed production by lowering the peak seed production level (Figure 1).
TA B L E 1 Summary of regression model results for the relationships between seed production and weather factors in the growing season. The objective variable was the 20-year moving average of seed production. Explanatory variables were the 20-year moving average of temperature, precipitation or temperature and precipitation from June to September. Thus, a clearer understanding of the relationship between the seed production amount and its temporal variation may be gained by considering the effect of the masting interval.

| Mechanisms underlying the decadal changes in masting behaviour
What In the flowering season, temperature showed a rising trend, whereas precipitation decreased ( Figure S6; Table S1). Although these changes in the weather conditions were not large, Schermer et al. (2019) pointed out that even a subtle change in weather conditions at the time of pollen release and aerial diffusion can, by making pollen limitation more likely, significantly affect reproduction of oak trees. The regression models showed that average seed production increased with rising temperature and decreasing precipitation (Tables S2, S3) and that the CV of seed production decreased with rising temperature. These results imply that warmer springtime temperatures accompanied by less precipitation may have reduced the frequency of an environmental veto of the pollination process and contributed to the higher seed production and its lower variability in the second half of the study period.
TA B L E 2 Summary of regression model results for the relationships between temperature in the growing season and the coefficient of variation (CV) of seed production and each of the two AR model coefficients. The objective variable was CV of seed production, [ The regression models showed that the effect of temperature on average seed production in the growing season was similar to that in the flowering season (Tables 1; Table S2). In addition, in the growing season regression models, temperature well explained the CV of seed production and the variation of the AR model coefficients (Table 2; Table S3). These results suggest that changes in the resource dynamics during the growing season may exert influence on the decadal changes in masting.
Rising temperatures and a consequent surplus of resources may cause oak trees to shift their allocation of resources from storage to more frequent use. In Q. crispula, rising temperatures encourages photosynthesis in the canopy (Hikosaka, Nabeshima, & Hiura, 2007). Moreover, Nabeshima, Kubo, Yasue, Hiura, and Funada (2015) reported that the number and total area of earlywood vessels in tree stems of this species had increased since 1970s. These vessel changes may improve water transport efficiency and encourage the production of photosynthates and thus surplus resources. These findings support our inference that temperature in the growing season may be an indicator of surplus resources. The variations in the moving average of seed production and the AR model coefficients [1 + a 1 ] and a 2 were well explained by that of temperature in the growing season ( Figure 4). Thus, an increase in surplus resources can reasonably be interpreted as having a causal connection with the observed decadal changes in masting.
The effect of surplus resources on masting may not be straightforward. The concave curve describing the relationship between the AR model coefficient for a delayed effect (a 2 ) and temperature ( Figure 4b) was an unexpected result. If the masting level is constant, the delayed effect of a 2 should simply become weaker (i.e. a 2 should increase) as surplus resources increase, because the stored resources consumed (converted to output) would likely be recovered quickly. In point of fact, after the 8th period (1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006), the values of a 2 increased linearly ( Figure 3b). However, a 2 was relatively high in the earlier periods, even though the level of surplus resources may have been low. The behaviour of a 2 may reflect the interplay between the storage and output of surplus resources. The masting level varied in the study period, particularly in the earlier periods.
As the resource depletion by seed production in the previous year is considered to be a fundamental determinant of seed production in the current year, we predicted that the effects of [1 + a 1 ] would always be negative and stable. Although the observed values of a 1 were always significantly negative (Figure 3a), the curve describing the relationship of a 1 with temperature was also concave ( Figure 4b). However, in comparison with a 2 , the variation of a 1 was small and the change in masting behaviour was determined mainly by the variation of a 2 . In another deciduous oak Quercus serrata, the total time of carbon accumulation in seeds has been estimated to be 1.38 years (Ichie et al., 2013); therefore, carbon stored for more than one year might be used for seed production. This finding is consistent with our result that [1 + a 1 ] was a fundamental component that always influenced seed production, and a 2 was the component that made masting behaviour variable.
Differing from Ichie et al., (2013), several studies showed that reproduction in masting species did not depend on stored carbon reserves (Hoch, Siegwolf, Keel, Körner, & Han, 2013;Igarashi, Shibata, Masaki, Tayasu, & Ichie, 2019). Reproduction in beech does not deplete stored carbohydrates, but it does change the amount of nitrogen stored (Han & Kabeya, 2017; see also Han, Kabeya, Iio, Inagaki, & Kakubari, 2013;Miyazaki et al., 2014), and masting in F. crenata is well explained by the interplay between stored nitrogen and climatic cues (Abe et al., 2016). However, nutrient accumulation may not have direct effects on the masting of Q. crispula-the target species of this study-because the seeds of this species contain only small amounts of protein (Shimada, Saitoh, Sasaki, Nishitani, & Osawa, 2006). Nitrogen may have indirect effects on seed production through bio-assimilation in Q. crispula, because in Quercus species, photosynthetic activity increases as the leaf nitrogen content increases (Takashima, Hikosaka, & Hirose, 2004). Our limited knowledge about carbohydrate-nitrogen interactions (Allen et al., 2017;Han & Kabeya, 2017) and the variation of carbon storage in relation to masting request further researches on the resource dynamics of masting.
The two possible mechanisms underlying decadal changes in masting that we examined, namely a resource allocation shift and an environmental veto, are not mutually exclusive. Inter-annual fluctuations of seed production have various aspects, including the number and mass of seeds; its variability represented by the CV; the masting interval; the peak size; and the interactions among them. Each aspect may have a different mechanism. From these aspects, testing the effects of a resource allocation shift and an environmental veto on the decadal change in masting behaviour provides us a good opportunity to identify the determinants and underlying mechanism of masting itself, because identified covariates explaining decadal changes would also be the candidate determinants of masting. To generalize our findings, it is necessary to investigate whether similar changes in masting behaviour occur in various places with different climatic features. Physiological and phenological data on seeding are also needed to further evaluate these two possible mechanisms of decadal changes in masting.
The decadal changes in masting behaviour may provoke various changes in the forest ecosystem. Solbreck and Knape (2017) showed that highly fluctuating biennial seed production can lead to predator satiation. However, biennial seed production in Q. crispula may reduce predator satiation. One of the major seed consumers of Q. crispula is curculionid weevils, which generally have a prolonged diapause of 2 years (Maeto & Ozaki, 2003). Therefore, the life cycle of the weevils might synchronize with seed production on a 2-year cycle. The decrease in the CV of seed production may also reduce the predator satiation effect. Thus, it would be interesting to investigate how the decadal changes in masting behaviour influence seed consumers and its feedback to trees (Bogdziewicz, Marino, Bonal, Zwolak, & Steele, 2018).
Because the Japanese wood mouse Apodemus speciosus increases in abundance after the masting year of Q. crispula (Onodera et al., 2017;Saitoh et al., 2007), the shortening of the masting interval may increase the frequency of high-density years of the wood mouse. Rodents are a significant vector of tick-borne pathogens, and the short masting interval may increase the risk for zoonoses.
However, several vertebrate species may be involved in the population dynamics of ticks (Ostfeld, Levi, Keesing, Oggenfuss, & Canham, 2018;Takumi, Sprong, & Hofmeester, 2019). Thus, to establish a comprehensive forest management program that includes the management of zoonosis risks, studies of decadal changes in masting behaviour in the context of community ecology should be conducted.

ACK N OWLED G EM ENTS
We are grateful to: A. Satake for helpful advice; H. Qingmin for many discussions and critical reading of the first manuscript; S. Sakurai for initiating the research at this study site; W. Suzuki, K. Osumi and T.
Yagi for field assistance; and K. Sekimura, R. Sato and H. Sarudate