Introduction

Disturbances have a major influence on the maintenance or collapse of ecosystems and biodiversity through various biotic and abiotic processes (Sousa 1984; van der Maarel 1993). These are caused not only by natural floods (Pollock et al. 1998), droughts (Fortuna et al. 2006; Fernández et al. 2016), and hurricanes (Spiller and Schoener 2007) but also by human activities, such as burning (Anderson et al. 2014; DeSiervo et al. 2015; Mutz et al. 2017), grazing (Frank 2005; Johansson et al. 2017), mowing (Johst et al. 2006; Uchida and Ushimaru 2014), and logging (Schroeder et al. 2007). Moderate disturbance is thought to reduce the dominance of highly competitive species and allow for the coexistence of diverse species; this is commonly known as the intermediate disturbance hypothesis (Connell 1978; Huston 1979; Buckling et al. 2000; Molino and Sabatier 2001; Cadotte 2007; Uchida and Ushimaru 2014). Most studies have focused on the effects of the frequency or intensity of disturbances (Miller et al. 2011; Hall et al. 2012), although recent studies have paid attention to the impact of disturbance timing on species coexistence and community composition (Crawley 2004; Smith 2006; Hall et al. 2012).

Disturbance timing is also important for population dynamics; experimental studies indicate that the impact of a disturbance depends on the life history stage at which it occurs (Brys et al. 2004; Jantunen et al. 2007; Gignoux et al. 2009). For example, early summer fires increase the number of inflorescences (Fill et al. 2012) and germination (Shepherd et al. 2012) rate more than spring fires, and the coincidence of the timing of mowing and plant blooming reduces genetic diversity (Nakahama et al. 2016) and reproductive success (Jantunen et al. 2007; Nakahama et al. 2016). However, the effects of disturbance timing on consumers, such as insects, remain largely unknown (Humbert et al. 2012; Jakobsson et al. 2018). The impact of disturbance timing may differ between plants and insects because insects have quite different life histories, including ontogenetic niche shifts and behavioral changes.

Grassland biodiversity is in crisis owing to recent management abandonment and intensification (Sirami et al. 2008; Kleijn et al. 2009; Manning et al. 2015; Gossner et al. 2016; Seibold et al. 2019;). Butterflies are good indicators of grassland biodiversity (Oostermeijer and van Swaay 1998; Stefanescu et al. 2005), so identifying the optimal frequency, intensity, and timing of disturbance for butterflies is important for conserving biodiversity in semi-natural grasslands. Previous studies on grassland butterfly populations have indicated that mowing once a year is generally desirable (Johst et al. 2006; Kőrösi et al. 2014; Schwarz and Fartmann 2021), but that mowing during the larval stage greatly reduces survival rates (Courtney and Duggan 1983; Humbert et al. 2010). Nevertheless, the effects of disturbance timing on butterfly population dynamics are poorly known (but see Johst et al. 2006; Kőrösi et al. 2014).

Many species of grassland butterflies live in patchy semi-natural grasslands such as pastures and field margins (Wahlberg et al. 2002; Johst et al. 2006; Hodgson et al. 2009; Kőrösi et al. 2014; Schwarz and Fartmann 2021), where they often form metapopulations (Baguette and Mennechez 2004). Traditionally, patch size and connectivity have been considered the two determinants of metapopulation dynamics (Hanski 1994, 1999; Winfree et al. 2005), but the importance of habitat quality has been suggested as a third determinant (Thomas et al. 2001; Fleishman et al. 2002; Baguette et al. 2011). Habitat quality includes various biotic and abiotic processes, such as bottom-up effects of food resources (Dennis et al. 2003), top-down effects of natural enemies (Yamanaka et al. 2009; Altermatt et al. 2012; Senft et al. 2017), mutualistic interactions (Senft et al. 2017), and physical abiotic micro-environments (Fortuna et al. 2006; Altermatt et al. 2012). Recent studies have examined the effects of disturbance on metapopulation dynamics (Schroeder et al. 2007; Hodgson et al. 2009; Fedrowitz et al. 2012; Johansson et al. 2017; Mutz et al. 2017). However, few studies have focused on disturbance timing at the patch scale in grassland metapopulations (Wahlberg et al. 2002; Mutz et al. 2017; Popović and Nowicki 2023).

The present study aims to clarify the effects of frequency and timing of disturbance on subpopulation dynamics in a metapopulation of the butterfly Plebejus argyrognomon in agricultural landscapes. This species inhabits grassland patches at farmland margins maintained by periodic mowing; the annual mowing frequency and timing vary between habitat patches. Because this butterfly species is multivoltine, with three generations per year (Kawazoe and Wakabayashi 1976), the optimal timing of disturbance may differ between generations.

This study employed a two-step analysis to separately evaluate the effects of disturbance frequency and timing on subpopulations (Table 1). The first step was to identify the appropriate disturbance frequency based on the relationship between subpopulation size and mowing frequency on an annual basis, with habitat connectivity and patch area treated as covariates. Habitat quality was not included in the analysis because mowing frequency and habitat quality were generally correlated (Johst et al. 2006; Noordijk et al. 2010; Wynhoff et al. 2011; Heuss et al. 2019). The second step was to explore the disturbance timing within a year that maximizes the subpopulation size under the appropriate mowing frequency identified in the first analysis. Since the effect of timing is estimated while the frequency is kept constant, it may reflect direct effects on butterflies themselves, rather than indirect effects mediated by the change in habitat quality. Therefore, we used both the disturbance timing and habitat quality as explanatory variables in the second-step analysis. Here, we focused on the abundance of the larval stage for the above analyses, because the larval stage appears to be the most susceptible to mowing (Courtney and Duggan 1983; Humbert et al. 2010).

Table 1 The two-step analysis used to evaluate the effects of mowing frequency and timing on subpopulation size of Plebejus argyrognomon
Full size table

This study addressed the following three questions: (1) Is larval abundance greatest in habitats that are mown once during the developmental period? (2) At which life history stage does mowing maximize larval abundance? (3) Is larval abundance larger in patches with higher connectivity with other patches?

Materials and methods

Study sites and species

This study was conducted in Iijima Town, Nagano Prefecture, central Japan (35°50′N, 137°55′E), with annual precipitation of 2007 mm and average temperature of 10.9 °C (Fig. 1A). This area is an agricultural landscape consisting of farmlands, small forests, residential areas, and roads. P. argyrognomon praeterinsularis (endemic subspecies of Japan) is now listed as an endangered (EN) species in Japan (Ministry of the Environment, Government of Japan, 2020), and this area has the highest number of remaining habitats for it in Japan (Miyashita et al. 2021). European subspecies are common in central Europe but is one of the most endangered butterflies in Sweden and Norway which, like Japan, are located at the edge of its global distribution. In the study area, P. argyrognomon has three generations per year; the 1st, 2nd, and 3rd adult periods are from late May to mid-June, mid-July to early August, and early September to mid-October, respectively. This species is a specialist herbivore that feeds only on the low shrub Indigofera pseudotinctoria (Fabaceae) in the larval stage and has a facultative mutualistic relationship with ant species (such as Camponotus japonicus and Formica japonica) during the larval and pupal stages. The species overwinters as eggs laid near the base of the only host plants.

Fig. 1
figure 1

The location of the study area in Japan (A), the spatial distribution of the 159 habitat patches (B), and the photo showing an example of host-plant patches and habitat patches (C aerial, D ground photo)

Full size image

The host plant I. pseudotinctoria is found along farmland edges and roadsides. We comprehensively examined the locations of host-plant patches throughout the study area before this study. There are several subpopulations beyond the north of the range shown in Fig. 1B, but they are more than 800 m apart, with forests and gorges in between. Moreover, no other subpopulations are found south of this range. Therefore, we regarded this group of habitat patches as a metapopulation and all host-plant patches within a 20 m range were considered to belong to the same “habitat patch” for P. argyrognomon; otherwise, they belonged to another habitat patch (Fig. 1C) (Thomas and Harrison 1992; Ojanen et al. 2013). This range was determined by observing the area over which the butterfly flies during a short period of time. We surveyed 143 habitat patches (286 host-plant patches) in 2018 and 159 habitat patches (339 host-plant patches), including newly found sites, in 2019 (Fig. 1B).

Butterfly abundance

We monitored P. argyrognomon adults and larvae for two years in 2018 and 2019. Adults were examined every two weeks in all habitat patches and larvae were examined at the peak of each generation in all host-plant patches. All adults found were plotted on a map by walking throughout the habitat patch. As habitat patches are linear and not large in area (median: 1079 m2), this survey appears to provide good estimates of local abundance. Actually, our three repeated surveys conducted at nine habitat patches in June 2020 recorded a similar number of individuals for all surveys in a given patch (r = 0.906 for 1st vs. 2nd surveys, r = 0.954 for 2nd vs. 3rd surveys). Because larvae make characteristic feeding marks on their leaves (Fig. 2), all the host plants were examined for the presence of these marks throughout the host-plant patch. If larvae were found, the accompanying ants were also recorded. When only feeding marks were left, those marks within a range of 30–40 cm were regarded as a single larva. Larval abundance in each host-plant patch was calculated by combining the number of larvae actually found and the number of larvae inferred from the feeding marks (Wahlberg et al. 2002). Larval surveys were conducted on all except rainy days, and adult surveys were conducted from 9:00 a.m. to 4:00 p.m. on sunny, light-windy days.

Fig. 2
figure 2

Feeding marks made by a P. argyrognomon larva on a leaf of I. pseudotinctoria

Full size image

Vegetation survey

We placed 2 × 2 m quadrats in each host-plant patch and recorded the coverage and height of host plants within the quadrats, along with the overall vegetation coverage and height. The number of quadrats differed according to the patch area (range: 1–12). The total number of quadrats placed was 520 in 2018 and 591 in 2019. Surveys were conducted in June, July, and September 2018 and 2019.

Ant abundance

To investigate the relative abundance of the ant species that could accompany the larvae, a pitfall trap was set at the center of each quadrat. A cylindrical container with an inner diameter of 5 cm and a depth of 4 cm was used for the pitfall trap. This was filled with 10 ml of a threefold diluted honey solution, a small piece of fish sausage, a few drops of detergent, and a small amount of pepper powder to prevent disturbance by animals. All traps were set on 24-h rain-free days in June and August both, 2018 and 2019 both.

Mowing frequency and timing

We visited all host-plant patches at least once a week from May to October and recorded whether the vegetation in the quadrats was mown or not. Because the frequency of mowing within the host-plant patch was not spatially identical, the mowing frequency of a host-plant patch was defined as the number of mowing averaged over all quadrats within the host-plant patch. The mowing date was defined as the day when more than half of the quadrats in the habitat patch were mown.

Data analysed

As described in the introduction, we used a two-step analysis to evaluate separately the effects of mowing frequency and timing (Table 1). In the first step, the appropriate mowing frequency was explored. Subpopulation size was defined as the number of larvae found in each host-plant patch. The mowing frequency used for the analysis was the number of mowing from the emergence of 1st generation larvae to the disappearance of the 3rd generation (from early May to the end of August). In the second step, the mowing day that maximized the subpopulation size under a single mowing was estimated (see Sect. “Results” for the selection of single mowing).

Effects of disturbance frequency

To clarify the relationship between larval population size and mowing frequency, generalized additive mixed models (GAMMs), which can account for nonlinearity, were used for the analyses. GAMMs were performed in 2018 and 2019 separately.

The statistical model is as follows.

$${\mathrm{log}(L}_{i})={\beta }_{0}{+\beta }_{1}s\left({F}_{i}\right)+{\beta }_{2}{Area}_{i}+{\beta }_{3}{S}_{i}+ID+\varepsilon$$
(1)

where \({L}_{i}\) is the mean larval abundance across all generations (from 1st to 3rd) at host-plant patch i, F is the mowing frequency from May to August, Area is the area of the host-plant patch, S is the connectivity index of habitat patch i, and ID is a random variable representing the habitat patch ID. Non-linearity was considered only for the mowing frequency F (s() represents the smoothing spline). \(\beta\) is the coefficient and \(\varepsilon\) is the error term that follows a zero-inflated negative binomial distribution. S is expressed as follows (Hanski 1999):

$${S}_{i}={\sum }_{i\ne j}{N}_{j}exp\left(-\alpha {d}_{i,j}\right)$$
(2)

where \({d}_{i,j}\) is the distance between habitat patch i and neighboring habitat patch j (measured between patch edges), \({N}_{j}\) is the mean adult abundance (both males and females) across all generations in habitat patch j, and α represents the dispersal capacity (1/average dispersal distance in km) of adult P. argyrognomon.

To explore the dispersal capacity that led to the highest model performance, we performed GAMMs for models with different dispersal capacities (α = 1, 2, 4, and 10, corresponding to the mean dispersal ability of 1000, 500, 250, and 100 m, respectively), based on the dispersal capacity estimated in an earlier study (Zhang and Miyashita 2018). We then selected the model with the best performance using the Watanabe–Akaike information criterion (WAIC)(Watanabe 2010), and evaluated the effect of each explanatory variable.

All explanatory variables were scaled. The analysis was performed using the {brms} package (Bürkner 2017) for GAMM with Bayesian methods and the {MuMin} package (Barton 2019) for model selection using WAIC in R 4.1.3 (R Development Core Team 2022). We ran five MCMC chains for 5000 iterations for estimation after the removal of the first 1000 iterations (burn-in). Convergence was assessed using the Gelman-Rubin diagnostic (\(\widehat{R}\)), and all parameters had \(\widehat{R}\) < 1.05. Non-informative prior distributions were used for almost all the parameters (Table S1). All the explanatory variables were standardized.

Effects of disturbance timing

The results of the first analysis indicated that the subpopulation size decreased with mowing frequency, but mowing less than once resulted in large estimation errors due to the small sample size (Fig. 3) (see Sect. “Results”). Therefore, the mowing timing that could maintain the largest subpopulation size in each generation was explored, using host-plant patches with single mowing, excluding those that were partially mowed at different times. The 2nd and 3rd generations were used for the analysis because very few patches are mown in May when 1st generation larvae are present. Analyses were performed using GAMMs in which the following model was used:

$${\mathrm{log}(L}_{i,t})={\gamma }_{0}+{\gamma }_{1}s\left({D}_{i}\right)+{{\gamma }_{2}Host}_{i,t}+{{\gamma }_{3}Ant}_{i,t}+{\gamma }_{4}{Area}_{i}+{{\gamma }_{5}N}_{ID,t-1}+{{\gamma }_{6}S}_{ID,t-1}+ID+\varepsilon$$
(3)

where \({L}_{i,t}\) is larval abundance at host-plant patch i in generation t, D is the mowing date (the number of days from May 1 when host plants begin to sprout), Host is the host plant coverage, Ant is presence or absence of two ant species (Camponotus japonicus and Formica japonica) captured by traps, with which P. argyrognomon is assumed to have a symbiotic relationship (Fig. S3), Area is the area of the host-plant patch, \({N}_{t-1}\) is the peak abundance of adults in the parent generation (a measure of temporal autocorrelation), and \({S}_{t-1}\) is the connectivity index. Non-linearity was considered only for the mowing date D (s() represents the smoothing function). \(\gamma\) is the coefficient and \(\varepsilon\) is the error term that follows a zero-inflated negative binomial distribution. \({S}_{i,t-1}\) can be expressed as follows (Sutherland et al. 2014; Johansson et al. 2017):

Fig. 3
figure 3

Average annual larval abundance in relation to mowing frequency in 2018 and 2019. Thick lines represent means and thin broken lines represent the 95% confidence limits

Full size image
$${S}_{i,t-1}={\sum }_{i\ne j}{N}_{j,t-1}exp\left(-\alpha {d}_{i,j}\right)$$
(4)

This index represents connectivity, expressed as potential adult immigrations in the parental generation. The \(\alpha\) was fixed to the value that had been determined in the first analysis.

Note that host plant coverage and symbiotic ant abundance were only weakly correlated with the mowing date, indicating no serious collinearity. All explanatory variables were scaled. The number of chains and iterations, burn-in, and the convergence assessment were the same as in the first analysis. Non-informative prior distributions were used for almost all parameters (Table S2). All explanatory variables were standardized.

Results

Effects of disturbance frequency

Due to lack of vegetation or ant-trapping data, the number of host-plant patches used in the present study was 274 (132 habitat patches) in 2018 and 334 (159 habitat patches) in 2019. The total number of larvae found in two years was 3775.

First, we examined the influence of patch connectivity on local population size of P. argyrognomon. The model with an α-value of 2 showed the lowest WAIC, and the model without connectivity had a much higher WAIC in both 2018 and 2019 (ΔWAIC = 8.07 in 2018, and 15.31 in 2019, Fig. S1), indicating the significance of connectivity.

Next, we evaluated the effect of explanatory variables by fixing the connectivity index to \(\alpha\) = 2. Mowing frequency, area, and connectivity showed strong effects on larval abundance in both years, and the effect of connectivity was larger in 2019 than in 2018 (Table 2). Larval abundance decreased with increasing mowing frequency in both years, with mowing twice or more having a particularly strong detrimental effect (Fig. 3).

Table 2 Estimates of the posterior mean, and 75%, 95% credible intervals (CI) for models of average annual larval abundance (1st analysis)
Full size table

Effects of disturbance timing

As this analysis used data from host-plant patches that had been mown only once, the number of host-plant patches used here was 71 (46 habitat patches) in 2018 and 106 (70 habitat patches) in 2019.

The posterior distribution of the nonlinear term for the mowing date had a 95% credible interval that did not cross zero for any year or generation (Table 3). In 3rd generation larvae, a hump-shaped relationship was observed between the mowing date and abundance in both years, with a peak at approximately 70–80 days (July 10–20) (Fig. 4). In other words, larval abundance in the 3rd generation showed a clear peak when the previous generation had been disturbed at the adult stage. The advantage of this timing of mowing was considerable; the abundance was several times that of other timings. In the 2nd generation, a weak hump-shaped relationship was observed in 2019, with a peak around day 50 (June 20) in 2019, but no clear pattern in 2018 (Fig. 4).

Table 3 Estimates of the posterior mean, and 75%, 95% credible intervals (CI) for models of for larval abundance in each generation (2nd analysis)
Full size table
Fig. 4
figure 4

Larval abundance in 2nd and 3rd generations in relation to mowing date in 2018 and 2019. Thick lines represent means and thin broken lines represent the 95% confidence limits. The vertical lines on the x-axis represent the rug distribution. The stage and generation of P. argyrognomon are indicated below the figure. E egg, L larva, A Adult

Full size image

Larval abundance was associated with adult abundance in the preceding generation (Table 3). Ant presence showed a clear positive coefficient, but host plant coverage did not. Connectivity indexes showed a larger positive coefficient in the 2nd generation than the 3rd in both years.

Discussion

Studies on the conservation management of grasslands to halt the rapid decline of grassland species have frequently emphasized the frequency and intensity of grassland disturbances (Tälle et al. 2016; Jakobsson et al. 2018; Filazzola et al. 2020), and sometimes the disturbance timing (Jantunen et al. 2007; Humbert et al. 2012; Nakahama et al. 2016; Tian et al. 2021). However, few studies have simultaneously assessed the appropriate disturbance frequency and timing for the conservation of grassland species (Johst et al. 2006; Kőrösi et al. 2014). First, the present study showed that less frequent mowing during the season would increase subpopulation size (Fig. 3). This may be related to an increase in habitat quality due to an increase in host plant coverage and symbiotic ant density, as revealed by the additional analysis (Fig. S4). This is plausible because host-plant availability generally reinforces the bottom-up effect, whereas ant density mitigates top-down effects of predators and parasitoids (Weeks 2003; Hill et al. 2022). Second, the mowing in mid-July (July 10–20), which corresponds to the adult period of the 2nd generation, was suggested as suitable for the larval population of the 3rd generation (Fig. 4). Although much less clear, mowing in mid-June (June 20, 2019) appeared suitable for the 2nd generation larvae, which also corresponds to the adult period in the 1st generation (Fig. 4). Taken together, mowing during the adult stage appears to maximize the larval abundance of the next generation. There are three possible reasons for this. One is that disturbance is fatal in the larval stage due to food deprivation, injuring and trampling of larvae (Courtney and Duggan 1983; Humbert et al. 2010), whereas the effect of disturbance in the adult stage would be much weaker, because adults can avoid the detrimental effects by flying. The second reason is that females may prefer to lay eggs in habitats immediately after recovery from disturbance, resulting in an increase in offspring generation (Haan and Landis 2019; Knight et al. 2019). The host plant I. pseudotinctoria is a low shrub with a ground-crawling form, allowing females to lay eggs on the remaining leaves and branches, even immediately after mowing. For the third reason, habitats after disturbance are expected to have higher food quality of host plants (Martinsen et al. 1998; Takagi and Miyashita 2012) and fewer natural enemies (Haan and Landis 2019; Yeh et al. 2021), so females may prefer to lay eggs in such habitats (preference-performance linkage). In contrast, mowing during the adult stage has been reported to be disadvantageous owing to emigration (Feber et al. 1996; Aviron et al. 2007; Popović and Nowicki 2023). However, emigration appears to be ephemeral, as most habitat patches are surrounded by non-hostile matrix, such as grasslands of field margin and road verges, as well as crop vegetation. Although these grasslands lack host plants and thus are not habitat patches for P. argyrognomon, they are likely to function as temporal refuges. It is likely that individuals that moved to nearby vegetation could soon return to their original habitat patches and might not have received strong detrimental effect by mowing in the adult stage.

Disturbance timing has received attention as an element of disturbance regimes (Crawley 2004; Miller et al. 2012), but earlier studies have focused on plant populations or communities, with little attention paid to consumer populations. The results of the present study suggest that the effects of disturbance on consumer population dynamics may vary greatly depending on which life-history stage of a target organism coincides with the timing of the disturbance. This is almost consistent with a simulation study that mowing in the young larval stage is unsuitable for the population dynamics of two Maculinea butterflies, in which only young larvae stay on aboveground host plants (Johst et al. 2006). Future research is needed to clarify the mechanisms by which matching affects consumer population dynamics.

The above results have important implications for current grassland management practices in the study area. Many host-plant patches are mown twice or more, and even patches that are mown once are rarely mown in mid-July (Fig. S2), so the current mowing regime is mostly not beneficial for P. argyrognomon. By interviewing farmers about the reasons for the high frequency of mowing, we found that frequent mowing is thought to be necessary to maintain a good landscape appearance and to reduce rice plant pests. Since the main insect pest, the stink bug, invades rice paddies from grasslands during rice flowering, mowing in mid-July (before rice flowering) is suggested to be effective in controlling pests (Teramoto 2003; Ono et al. 2007). As mowing once in July is compatible with rice pest control, it seems highly possible to establish this practice in habitat patches adjacent to rice paddies if farmers understand the conservation value of P. argyrognomon.

Habitat connectivity had a positive effect on subpopulation size of P. argyrognomon. However, the effect of connectivity was variable, being more evident in 2019 and more prominent in the 2nd than in the 3rd generation. Although the reason is unknown as to why connectivity was more evident in 2019, the between-generation difference may be due to the difference in habitat quality affecting dispersal and colonization of butterflies (Fleishman et al. 2002; Matter et al. 2009; Baguette et al. 2011; Akeboshi et al. 2015). The most likely reason is the difference in the availability of nectar resources for adults. As the inflorescences of the host plant I. pseudotinctoria are so abundant with full bloom in July when 2nd generation adults emerge (Fig. S5), high patch quality might have reduced emigration (Baguette et al. 2011), weakening the effect of patch connectivity for 3rd generation larvae.

The influence of disturbance on metapopulations is well-supported by theoretical studies (Ellner and Fussmann 2003; Johst and Drechsler 2003; Wilcox et al. 2006; Reigada et al. 2015). Field studies have also suggested that metapopulation dynamics are driven not only by the intensity and frequency of disturbance (Thomas et al. 2001; Hodgson et al. 2009; Johansson et al. 2017), but also by the years since disturbance (patch age) (Wahlberg et al. 2002; Schroeder et al. 2007; Caruso et al. 2010; Mutz et al. 2017). Although the present study did not directly address metapopulation dynamics, disturbance timing was shown to have a significant effect on subpopulations independently of the effect of patch connectivity. It seems natural that this effect could propagate to the metapopulation scale. It should be noted, however, that mowing many patches simultaneously may not produce desirable results for metapopulation dynamics. Laboratory and mesocosm experiments have indicated that asynchronous, rather than synchronous, disturbances lead to the stability of metapopulation persistence (Matthews and Gonzalez 2007; Vasseur and Fox 2009; Fox et al. 2011; Steiner et al. 2013; Duncan et al. 2015). Moreover, variations in the optimal timing of disturbance among habitat patches could be associated with subtle differences in the phenology of butterflies, host plants, and natural enemies among patches. Experimental manipulation of the degree of mowing synchronicity in multiple metapopulations is a promising approach to explore truly appropriate habitat management in heterogeneous landscapes.

We have demonstrated that mowing during the adult period (mid-June or mid-July) can increase the subpopulation size of P. argyrognomon several times more than mowing at other timings. However, this was derived from correlational analysis and the causality is still unclear. It is therefore necessary to conduct a long-term mowing experiment, including mowing in late fall and early spring when P. argyrognomon is inactive. Furthermore, uniform application of locally appropriate management across the landscape can make undesirable results for the metapopulation level. Previous metapopulation studies suggest that synchronous disturbance can undermine metapopulation stability and persistence (Matthews and Gonzalez 2007; Vasseur and Fox 2009; Fox et al. 2011; Steiner et al. 2013; Duncan et al. 2015). Therefore, a slight shift in mowing timing among adjacent habitat patches may promote adult re-immigration and help maintain the metapopulation of P. argyrognomon in agricultural landscapes.