Volume 63, Issue 3 p. 260-271
Open Access

Climate change shifts population structure and demographics of an alpine herb, Anemone narcissiflora ssp. sachalinensis (Ranunculaceae), along a snowmelt gradient

Yuka Kawai

Yuka Kawai

Faculty of Environmental Earth Science, Hokkaido University, Sapporo, Hokkaido, Japan

Search for more papers by this author
Gaku Kudo

Corresponding Author

Gaku Kudo

Faculty of Environmental Earth Science, Hokkaido University, Sapporo, Hokkaido, Japan


Gaku Kudo, Faculty of Environmental Earth Science, Hokkaido University, Sapporo, Hokkaido 060-0810, Japan.

Email: [email protected]

Search for more papers by this author
First published: 31 May 2021
Citations: 1

Funding information: JSPS KAKENHI, Grant/Award Numbers: 24570015, 21370005; Environmental Research and Technology Development Fund, Grant/Award Number: D-0904

Gaku Kudo is the recipient of the 2022 Young Author Award.

The copyright line for this article was changed on 23 August 2022 after original online publication.


Alpine ecosystems, characterized by cold climates and short growing seasons, are thought to be most vulnerable to climate change. Warmer temperatures and earlier snowmelt extend the growing season length and increase drought stress for alpine plants, resulting in changes to their distribution. Anemone narcissiflora ssp. sachalinensis is a perennial herb that grows in the alpine snow-meadows of northern Japan. In the last few decades, its distribution has shifted toward later snowmelt habitat in the Taisetsu Mountains of Hokkaido. We recorded demographic data for this species at early, middle and late snowmelt habitats over four years (2009–2012), and constructed transition matrix models to evaluate how demographic parameters and population growth rate vary between local habitats along a snowmelt gradient. The proportion of reproductive plants was low and seed production was limited in the early snowmelt habitat, with drier soil conditions, in comparison to the middle and late snowmelt habitats, with moist soil conditions. Evidence of the transition from small plants to those in the reproductive stage was limited in the early snowmelt habitat, suggesting that growth was inhibited; the local population in this habitat was estimated to be sustained by seed migration from later snowmelt habitats. These results indicate that advancing snowmelt under climate change may decrease the reproductive activity and population growth rate of snow-meadow plants if seed migration from later snowmelt populations is limited, resulting in the extinction of local populations.


Alpine ecosystems, which are characterized by cold climates, short growing seasons, and heterogeneous snow distributions reflecting geographical undulation, are particularly vulnerable to climate change (Ernakovich et al., 2014; Grabherr, Gottfried, & Pauli, 1994; Theurillat & Guisan, 2001). Warmer temperatures modify the accumulation of snow during the winter and advance snowmelt in the spring, resulting in an extension of the snow-free period and a decrease in the water available to alpine plants during the summer (IPCC, 2014). Longer growing seasons combined with drought stress affect the distribution pattern and species diversity of alpine plant communities (Barros, Thuiller, & Münkemüller, 2018; Ernakovich et al., 2014; Theurillat & Guisan, 2001; Winkler, Butz, Germino, Reinhardt, & Kueppers, 2018). For many plant species, there is ample evidence of distribution range shifts toward higher elevations (Matteodo, Wipf, Stöckli, Rixen, & Vittoz, 2013; Moritz & Agudo, 2013; Pauli, Gottfried, Reiter, Klettner, & Grabherr, 2007); the species compositions of alpine plant communities have changed in many mountain regions around the world (Amagai, Kudo, & Sato, 2018; Gritsch, Dirnböck, & Dullinger, 2016; Pickering, Green, Barros, & Venn, 2014; Zorio, Williams, & Aho, 2016).

The distributions of alpine plant species are primarily determined by the spatial heterogeneity of environmental conditions, such as snowmelt time, soil moisture, land surface stability, nutritional conditions and geological properties (Nagy & Grabherr, 2009). In particular, the spatial heterogeneity of snowmelt time, which determines the growing season length, is a crucial factor that affects the distribution pattern (Hülber, Bardy, & Dullinger, 2011; Kudo & Ito, 1992; Litaor, Williams, & Seastedt, 2008), flowering phenology (CaraDonna, Iler, & Inouye, 2014; Kudo, 2016), leaf physiology (Choler, 2005; Kudo, Nordenhäll, & Molau, 1999; Wheeler et al., 2014), reproductive output (Gezon, Inouye, & Irwin, 2016; Kudo & Hirao, 2006; Lluent, Anadon-Rosell, Ninot, Grau, & Carrillo, 2013; Moriwaki, Takyu, & Kameyama, 2020) and demographic parameters of alpine plants (Campbell, 2019; Hülber et al., 2011; Kawai & Kudo, 2018). Therefore, snowmelt time advancement due to global warming may lead to changes in population dynamics, distribution shifts and/or the local extinction of alpine plants.

Various demographic parameters affect the persistence of local populations, such as the growth, mortality and fecundity of plants (Silvertown & Lovett-Doust, 1993). The demographic parameters of alpine plants may vary among local populations in response to snowmelt conditions (Kawai & Kudo, 2018). Consequently, the predicted response of local alpine plant populations to advancing snowmelt times is a crucial issue in global change biology. Although many studies have compared the reproductive activity and seed-set success of local populations under different environmental conditions (Galen & Stanton, 1991; Gezon et al., 2016; Kudo & Hirao, 2006; Lluent et al., 2013; Moriwaki et al., 2020; Price & Waser, 1998; Wipf, 2010), only a few studies have evaluated the effects of reproductive output on population dynamics (Campbell, 2019; Iler et al., 2019). Comparisons of local population demographic characteristics under different snowmelt conditions may provide useful insights for the impact of climatic change on the population dynamics and persistence of alpine plants. Additionally, seed migration from other populations plays an important role in the persistence of local populations under heterogeneous environmental conditions (Ferrer, Montaña, & Franco, 2015). Because alpine ecosystems are composed of highly localized habitats associated with various snowmelt conditions, the habitat-specific demographic characteristics of local populations and seed migration between local populations should shape the overall population dynamics, that is, metapopulation structure (Eriksson, 1996). The persistence of local populations under climate change should therefore take into account the contributions of seed migration from external populations, originating from different habitats.

Based on the ecological importance of the snowmelt regime in alpine ecosystems, we compared the reproductive performance and demographics of a perennial alpine herb, Anemone narcissiflora ssp. sachalinensis, in three habitats along a snowmelt gradient in the Taisetsu Mountains, northern Japan. We observed that a large population of this species in a snow-meadow in this mountain region had disappeared in only 4 to 5 years in the time since 1990 (see Figure S1). The rapid extinction of this local population is suspected to be caused by climate change, that is, advancing snowmelt time, warmer temperatures and/or drier soil conditions; however, the mechanism for this local extinction had not yet been confirmed. To understand local population degradation, it is crucial that population dynamics under different snowmelt conditions be monitored.

The transition matrix model of population dynamics is a useful and powerful tool for understanding and managing plant populations; it can be used to characterize the dynamics of a population and identify the key life-history stages that contribute to population growth (Caswell, 2000; Silvertown & Lovett-Doust, 1993). In this study, we postulate that earlier snowmelt decreases the reproductive performance of A. narcissiflora due to drought stress, and that this decrease in reproductive activity may degrade population growth. As a result, the distribution of A. narcissiflora will shift to a later snowmelt habitat along the snowmelt gradient. Specifically, we attempted to address the following questions: (1) Does reproductive performance decrease in the early snowmelt habitat along the distribution range? (2) How do demographic parameters differ between local habitats along the snowmelt gradient? (3) Do the important life-history stages affecting population dynamics differ between the early and late snowmelt habitats? Based on these questions, we discuss the ecological impacts of climate change on alpine ecosystems in terms of the population dynamics of snow-meadow plants.


2.1 Study plant

A. narcissiflora ssp. sachalinensis (Ranunculaceae) is a deciduous perennial herb that grows in the alpine meadows of northern Japan. Rhizomes of this species grow vertically on which several radical leaves are produced as a fascicle. Sometimes, additional ramets are produced from the rhizomes of large plants in the vicinity of main ramets. Aboveground plant parts emerge soon after the disappearance of snow, then flowering occurs within a few weeks; the achenes mature 1 month after the anthesis. Major flowering periods are from mid to late July. Reproductive plants commonly produce a single erect scape with one to five whitish flowers. Individual flowers contain 6 to 30 pistils that develop to ovate achenes. This species is weakly self-compatible and pollinators are required to achieve fruit set. In our preliminary experiment, seed set via artificial outcrossing was 63%, whereas seed set via a bagging treatment was only 13% (Y. Kawai & G. Kudo, unpublished data).

2.2 Study site

This study was conducted in the central part of the Taisetsu Mountains, Hokkaido, northern Japan (43°33′N, 142°51′E). Because of heavy snowfall and prevailing northwest winds during the winter, deep snowdrifts form on the southeastern-facing slopes, which cause a clear gradient in snowmelt time over a local area. Snowfall usually begins in late September. Summer temperatures at 1700 m elevation are 8.2°C in June, 12.3°C in July, 12.5°C in August and 7.3°C in September. Monthly precipitation is 147 mm in June, 260 mm in July, 360 mm in August and 234 mm in September (G. Kudo, unpublished data). Annual temperature has increased at the pace of 0.33°C per decade during the last 30 years and snowmelt has advanced 4.1 days per decade during the last 25 years in the study area (Kudo, 2014).

In 1988, three 20 × 20 m plots (A–C) were established to record the vegetation pattern along the snowmelt gradient (Kudo & Ito, 1992; Figure S2a). These plots were located within a distance of 100 m; the distance of plot A to B was about 70 m and the distance of plot B to C was about 30 m. The average time for snow disappearance over the 25 years from 1988 to 2012 was early June, late June and early July in plots A, B and C, respectively. Dwarf shrubs (e.g., Sieversia pentapetala, Rhododendron aureum, Phyllodoce aleutica) and herbaceous plants (e.g., Primula cuneifolia, Potentilla matusmurae, Peucedanum multivittatum) are common in late snowmelt locations in this region. In the 1988 survey, A. narcissiflora was present only in plots A and B (Kudo & Ito, 1992); however, in a preliminary survey in 2008, we detected that A. narcissiflora had expanded its distribution to plot C (Y. Kawai & G. Kudo, unpublished data), indicating a shift in its range toward late snowmelt locations. In our preliminary survey, we detected that soil moisture during the summer season increased along the snowmelt gradient from plot A to plot C (Figure S2b). In the early snowmelt plot A, soil moisture often decreased below 40%, whereas the soil moisture in plot C remained above 50% throughout the summer.

2.3 Monitoring of demographic characteristics

In 2009, we established three fixed quadrats (1 × 1 m) in each plot (A, B and C), and tagged all A. narcissiflora individuals (including ramets) for identification. During the survey period from 2009 to 2012, the demographic features (i.e., death, growth or stasis) of all plants were recorded, and new recruits were tagged every year. During the pre-flowering period (from early to mid-July), we recorded the leaf number and reproductive status (i.e., nonflowering or flowering) of all plants within the quadrats. For single-leaved young plants, plant height was also measured to determine their size class. For flowering plants, the total number of flowers and developed achenes were recorded during the flowering and fruiting periods. Demographic data for the three quadrats within each site were pooled for analysis. In total, 957 individuals were observed across all plots over the whole survey period.

2.4 Analysis of reproductive properties

For each reproductive function, that is, flowering occurrence, flower production and fruit (achene) production, the effects of leaf number (representative of plant size) and plots (growing habitat along the snowmelt gradient) on reproductive performance were analyzed. We used generalized linear mixed models (GLMM), postulating a binomial error distribution with a logit-link function for the flowering probability and a Poisson error distribution with a log-link function for both the number of flowers and fruits per plant. Full models included the number of leaves per plant, the plot (A, B or C) and their interaction as explanatory variables. In the GLMM for flowering probability, year (2009–2012) was included as a random factor. In the GLMMs for flower and fruit production, the individual reproductive plant was included as a random factor across years. To find the best fit model for the data, full models were compared with all possible subset models using the Akaike's information criterion (AIC) corrected for finite sample size. For the model selection analyses, we used the “lme4” package in R 2.5.0 (R Development Core Team 2018, http://www.r-project.org/).

2.5 Matrix model analysis

We constructed a stage-classified transition matrix model for each plot by pooling the census data over the 4 years (Figure 1). Individual plants were assigned to one of five size classes based on the leaf number, as follows: Class S: single-leaved plant shorter than 3 cm in height; Class L1: single-leaved plant taller than 3 cm in height; Class L2: two-leaved plant; Class L3: three-leaved plant; and Class L4: greater than or equal to four-leaved plant. Class S plants were ascribed as seedling recruitments. Flowering was commonly observed in Class L4 plants, but only occasionally in Class L3 plants.

Details are in the caption following the image
The basic structure of the transition matrix model. S: current-year seedling, L1: single-leaved class, L2: two-leaved class, L3: three-leaved class, L4: not less than four-leaved class. Sexual reproduction occurs mostly at L4, but rarely at L3, and vegetative ramet production occurs only at L4 [Color figure can be viewed at wileyonlinelibrary.com]

The transition matrix element for seedling recruitment (Fseedling) was derived from Class L3 (rarely) and L4 plants (mostly), in which the total number of newly emerged seedlings was allocated to Class L3 and L4 proportionate to their flower production. The transition matrix element for ramet recruitment by vegetative propagation (Framet) was obtained as the average number of newly emerged ramets per L4 plants, that is, the number of new ramets in each class was divided by the number of L4 plants, in which we postulated that vegetative ramet production occurred in large-sized plants based on field observations. We calculated a population growth rate (λ) from the dominant eigenvalue of the matrix model.

Elasticities of λ with respect to individual elements, that is, the relative contribution of individual elements to the population growth rate, were calculated using the obtained matrixes. The elasticity of a given matrix element is defined as the ratio of the proportional change in λ to that in the matrix element (aij). The elasticity matrix is calculated as:
where Sij is the sensitivity matrix. The elements of any elasticity matrix satisfy the conservation law, whereby the sum of the elasticities over all elements is equal to 1 (De Kroon, Plaisier, van Groenendael, & Caswell, 1986):

We classified the matrix elements into five categories, that is, seedling recruitment, ramet recruitment, retrogression (downward change in the stage), stasis (no change in the stage) and progression (upward change in the stage). Then, we defined five types of elasticities of λ: Eseedling, Eramet, Eretrogression, Estatis and Eprogression. For instance, Eseedling is the sum of the elasticities of λ to the matrix element of seedling recruitment (Fseedling). Under these definitions, due to the aforementioned theory, the sum of the five elasticities is equal to 1.

Furthermore, we analyzed the effects of seedling recruitment on population growth using the obtained transition matrix model for each plot. In this analysis, we simulated changes in λ in response to changes in Fseedling values from 0 to the values obtained in each plot. All simulations were conducted across 50 times (years) starting from the present size distribution in each plot for the calculation of λ values.

2.6 Estimation of seed dispersal between subpopulations

Because the three plots (i.e., subpopulations) were located adjacently within 100 m (Figure S2a), we need to evaluate the contribution of seed migration between the plots to seedling establishment in each subpopulation. We therefore estimated the number of seedlings that originated from within or outside the plot. The average seed germination rate was obtained as the ratio of the total number of seedlings divided by the total number of achenes across the three plots. For each subpopulation, the number of seedlings originating from that plot was estimated by multiplying the germination rate by the total seed number in that plot (across the three quadrats in each plot). Finally, the number of seedlings originating from migrated seeds was determined by subtracting the estimated number of seedlings originating from that subpopulation from the total seedling number observed in that subpopulation.


3.1 Population structure and reproductive activity

The population density of A. narcissiflora was highest in plot B (126.9 plants/m2), intermediate in plot A (71.4 plants/m2) and lowest in plot C (49.3 plants/m2). Figure 2 shows the size distribution (based on leaf number) of individual plots over 4 years (2009–2012). There was significant variation in plant size across plots (p < 0.0001, Kruskal–Wallis test). Large-sized plants with four or more leaves (Class L4) were abundant in plot B (44.3%) and plot C (39.0%), whereas the proportion of large-sized plants was small in plot A (12.6%), where small-sized plants with a single leaf (Class L1) were most common (27.6%).

Details are in the caption following the image
Size distribution of individual plots according to leaf number (a: plot A, b: plot B, c: plot C). One-leaf plants are classified as current-year seedlings (gray) and older than one-year plants (open). Reproductive plants are displayed in closed area. The cumulative number of plants in the three quadrats and across years (2009–2012) is shown in each plot in which breakdowns of seedlings (leaf number = 1) and reproductive plants are shown in parentheses. The kernel regression line is shown for each plot [Color figure can be viewed at wileyonlinelibrary.com]

Reproductive plants were most abundant in plot B (10.7 plants/m2), followed by plot C (5.6 plants/m2) and lowest in plot A (2.7 plants/m2). The proportion of reproductive plants increased from the early to late snowmelt plots; this proportion was 3.7% in plot A, 8.3% in plot B and 11.5% in plot C. Flowering started to occur once plants reached the four-leaved size (Figure 2). In the GLMM for flowering probability, plant size (p < 0.0001), plot B (p = 0.013) and plot C (p < 0.001) had positive effects, whereas the interaction between leaf and plots had a negative effect (p < 0.001; Table S1); this was because large-sized plants were rare in plot A in comparison to in plots B and C. The average leaf number for flowering plants was 6.8, 7.3 and 7.8 leaves in plots A, B and C, respectively.

Reproductive plants had between 1 and 10 flowers (average = 3.3 flowers), and this increased with leaf number (z = 6.46, p < 0.0001), but there were no differences in flower production between the plots (the effect of plot was excluded according to the AIC for model selection). Fruit (achene) number also increased with leaf number in reproductive plants (z = 12.07, p < 0.0001), but there was no difference in fruit production between the plots (the effect of plot was excluded according to the AIC for model selection). These results indicate that reproductive performance was size-dependent across the plots (Figure 3).

Details are in the caption following the image
Relationship between plant size, expressed as leaf number and flower number (a) or fruit number (b), across plots (A–C). Each line was obtained by a generalized linear mixed model (GLMM) [Color figure can be viewed at wileyonlinelibrary.com]

3.2 Demographic analysis

The projection matrix was obtained by the sum of transition rates (retrogression, progression and stasis), seedling recruitment by seed production and ramet recruitment by vegetative propagation (Table 1). The population growth rate (λ) was larger than 1 in every population (ranging from 1.017 to 1.048), indicating an increasing number of plants from year to year. The highest λ value was found in plot B, where the population is projected to double after 14.7 years (1.04814.7 = 2) under stable environmental conditions. The projection matrixes showed that survival and remaining in the same size class were the most common events for all size classes for all populations (Table 1).

TABLE 1. Transition matrixes, population growth rate (λ), survival rate of each size classa and stable size distribution for local populations (plots A–C)
(a) Plot A (early snowmelt site) λ = 1.039
Size class (t + 1) Size class (t)
S L1 L2 L3 L4
S 0.580 0.008 0.584
L1 0.248 0.684 0.216 0.062 0.095
L2 0.018 0.257 0.519 0.320 0.041
L3 0.007 0.191 0.474 0.108
L4 0.007 0.031 0.124 0.865
Survival rate 0.84 0.95 0.96 0.98 0.99
Stable distribution 0.18 0.33 0.25 0.11 0.14
(b) Plot B (middle snowmelt site) λ = 1.048
Size class (t + 1) Size class (t)
S L1 L2 L3 L4
S 0.486 0.308
L1 0.243 0.831 0.088 0.009 0.057
L2 0.069 0.578 0.093 0.012
L3 0.006 0.245 0.454 0.059
L4 0.006 0.078 0.444 0.939
Survival rate 0.73 0.91 0.99 1.00 1.00
Stable distribution 0.19 0.33 0.07 0.07 0.34
(c) Plot C (late snowmelt site) λ = 1.017
Size class (t + 1) Size class (t)
S L1 L2 L3 L4
S 0.520 0.041 0.347
L1 0.252 0.780 0.273 0.000 0.023
L2 0.008 0.122 0.545 0.130 0.012
L3 0.121 0.522 0.040
L4 0.348 0.948
Survival rate 0.78 0.92 0.94 1.00 0.99
Stable distribution 0.18 0.39 0.12 0.05 0.26
  • a Size class was defined as follows: S: single-leaved plant shorter than 3 cm in height; L1: single-leaved plant taller than 3 cm in height; L2: two-leaved plant; L3: three-leaved plant; and L4: greater than or equal to four-leaved plant.

The seedling survival rate was relatively low (0.73–0.84) compared to that of larger size classes (0.91–1.00) in every population (Table 1). The stable size distributions, obtained from the right eigenvector of the projection matrixes, indicate that the proportions of small-sized plants (Class S and L1) were similar across the plots (51–57%), but the proportions of larger plants (Class L2 to L4) varied highly between plots (Table 1). In plot A, Class L2 plants were most common (25%), whereas Class L4 plants occupied only 14% of the population. In contrast, Class L4 plants were most common in plots B (34%) and C (26%), whereas the proportions of Class L2 and L3 plants were relatively small in these plots (5–12%). These differences reflected the larger growth probability from L3 to L4 in plots B (0.44) and C (0.35) in comparison with that in plot A (0.12), and the larger retreat probability from L3 to L2 in plot A (0.32) in comparison with that of plots B (0.09) and C (0.13).

3.3 Elasticity analysis

The elasticity matrix indicated that the probability of surviving and staying in L4 class was the most important element contributing to λ; this tendency was stronger in plots B (0.55) and C (0.57) than in plot A (0.30; Figure 4; Table S2). The elasticity matrix of plot A showed the relatively higher importance of the survival probability of small-sized plants (Class S to L2; 0.51) in comparison with that in plots B (0.28) and C (0.28). The recruitment of seedlings (sexual reproduction) and ramets (vegetative propagation) had smaller effects on λ than the progression and stasis of existing plants. The relative contribution of seedling recruitment to λ was highest in plot A (0.031).

Details are in the caption following the image
Results of an elasticity analysis for population growth rate (λ), conducted for each plot across size class (a: plot A, b: plot B, c: plot C). Refer to Figure 1 for the components of the transition matrix model. Size class was defined as following five categories: small: single-leaved plant shorter than 3 cm in height; L1: single-leaved plant taller than 3 cm in height; L2: two-leaved plant; L3: three-leaved plant, and L4: greater than or equal to four-leaved plant [Color figure can be viewed at wileyonlinelibrary.com]

The contributions of sexual reproduction (i.e., seedling establishment) to λ were simulated using the projection matrix model for individual plots (Table 1), in which Fseedling values were changed from 0 to the values obtained in the projection matrix model (i.e., 0.584, 0.308 and 0.347 in plots A, B and C, respectively), then λ values were calculated (Figure 5). There were positive relationships between Fseedling and λ in every plot. When Fseedling values decreased close to 0, λ values decreased below 1 in plots A and C, indicating a decrease in population size. In contrast, the λ value in plot B remained above 1 even without the recruitment of seedlings.

Details are in the caption following the image
Effect of seedling recruitment (Fseedling in Figure 1) on population growth rate (λ) of each plot [Color figure can be viewed at wileyonlinelibrary.com]

3.4 Estimation of seed migration

On average, A. narcissiflora dispersed 73.3 seeds and established 9.5 seedlings per quadrat (1 m2) across plots A to C every year, which indicates a germination rate of 0.13%. Seed production varied highly between the plots (i.e., 18.2, 133.1 and 66.6 seeds/m2 in plots A, B and C, respectively). Thus, the estimated number of seedlings was 2.4, 17.3 and 8.7 seedlings/m2 in plots A, B and C, respectively, with no seed dispersal between plots. The average number of seedlings across all study years was 4.9, 16.8 and 6.8 seedlings in plots A, B and C, respectively. Therefore, the proportion of seedlings originating from migrated seeds was largest in plot A (2.5 of 4.9 seedlings were estimated to be migrators). In contrast, the observed seedling numbers in plots B and C were smaller than the expected number of seedlings. This indicates that local populations acted as a source in plots B and C, whereas the local population in plot A acted as a sink, assuming that seed dispersal commonly occurs between the plots.


Our analyses of the size distribution and demographic parameters of A. narcissiflora along the snowmelt gradient suggest that the distribution shift toward later snowmelt locations is related to habitat-specific population dynamics. In the early snowmelt habitat (plot A), the proportion of reproductive plants was small, and the population growth rate (λ) relied on the recruitment of small-sized plants that were maintained by seed migration from other habitats. In contrast, subpopulations in the middle (plot B) and late snowmelt habitats (plot C) were composed of many reproductive plants, in which the seed source for seedling recruitment was sustained within each habitat. These results indicate at least two ecological aspects to the population dynamics of alpine plants responding to variation in snowmelt time: the ecological significance of the snowmelt gradient as a determinant of the distribution range of each species, and the impact of climate change on the distribution pattern of alpine plants.

4.1 Population dynamics along the snowmelt gradient

Important life-history stages affecting the population dynamics of plants vary depending on the growing habitat. For instance, growth and/or survival at a specific stage is an important determinant of the population growth rate (λ) for long-lived plants inhabiting a stable habitat, whereas seed production is a key factor affecting the population dynamics of short-lived plants inhabiting an unstable habitat (Silvertown, Franco, Pisanty, & Mendoza, 1993). Habitat-specific variations in demographic properties are detectable not only for interspecific comparisons, but also for intraspecific comparisons between populations under different environmental conditions (Fowler & Antonovics, 1981; Horvitz & Schemske, 1995). Although our study was conducted in a small area, the steep snowmelt gradient covers the whole distribution range of A. narcissiflora in terms of the snow conditions of the alpine ecosystem. Therefore, information about habitat-specific demographic parameters provides a clue for the prediction of population dynamics and distribution shift along the snowmelt gradient. The transition matrix models revealed that the population growth rate (λ) was larger than 1 in every subpopulation, indicating that these local populations are sustainable at the present time. Elasticity analysis demonstrated that the maintenance of individuals in a reproductive stage was the most important factor for population growth in every habitat. In the early snowmelt habitat, in which reproductive individuals were scarce, the persistence of small-sized individuals contributed to population growth more intensively than in the other habitats. Because seedling establishment within the early snowmelt habitat was predicted to be limited, self-sustainability of the subpopulation in this habitat might be difficult without seed migration from other habitats.

Flower and fruit production strongly depended on plant size, as represented by leaf number. The small number of reproductive individuals in the early snowmelt habitat might be related to the drier soil conditions in this location (Figure S2). Generally, snow-meadow herbs are sensitive to soil moisture and drier conditions often suppress their growth and reproductive performance (Campbell, 2019; Iler et al., 2019; Lluent et al., 2013). The transition matrix in the early snowmelt habitat was characterized by a larger retrogression and smaller progression of L3 plants, in comparison with the other habitats (Table 1). This indicates that the transition of smaller individuals to the reproductive stage was restricted, resulting in lower seed productivity in the early snowmelt habitat. Nevertheless, the proportion of seedlings in the early snowmelt habitat (18%) was comparable to that in the middle (19%) and late snowmelt habitats (18%); therefore, stable seedling establishment in the early snowmelt habitat might be maintained by seed migration from the later snowmelt habitats.

The subpopulation in the middle snowmelt habitat showed the highest population growth rate; population density and the number of reproductive individuals were higher here than in the other habits. Interestingly, this subpopulation was predicted to be maintained even without the recruitment of seedlings because of the high survival rate of large-sized individuals and vegetative ramet production by L4 individuals. In contrast, the survival rate of seedlings was lowest in this habitat (0.73), probably because of shading stress due to the high population density (Hülber et al., 2011). Large seed production in the middle snowmelt habitat was modeled to be a major seed source to the neighboring subpopulations.

The subpopulation in the late snowmelt habitat was established in the past 20 years; it is where λ was the lowest (1.02), and is maintained by seedling establishments, like the subpopulation in the early snowmelt habitat. Although population density was the lowest in the late snowmelt subpopulation, the proportion of reproductive individuals was comparable to that in the middle snowmelt subpopulation, probably due to the moist soil conditions owing to the late snowmelt. The seed productivity of this subpopulation (67 seeds/m2) was estimated to be much larger than that of the early snowmelt subpopulation (18 seeds/m2). A short, snow-free period might limit population growth in late snowmelt locations, however, if a short growing period suppresses the establishment of seedlings (Galen & Stanton, 1999; Scherff, Galen, & Stanton, 1994). The survival rates of seedlings and small-sized individuals (L1) were smaller (Table 1), and the contribution of seedling establishment to population growth rate was lower than that of the early snowmelt habitat (Figure 5). These results suggest that a late snowmelt may be beneficial for large-sized reproductive individuals, but disadvantageous for seedling establishment.

Spatial heterogeneity, even at a relatively small scale, can produce a similar level of demographic variation within a continuous population as that between populations (Fowler & Antonovics, 1981; Horvitz & Schemske, 1995). Local scale demographic variation has been documented along the environmental gradient within populations for several plant species (Kawai & Kudo, 2018; Vega & Montaña, 2004). Local snowmelt gradients in alpine ecosystems are a crucial environmental factor responsible for diverse plant communities at the local scale (Hülber et al., 2011; Kudo & Ito, 1992). The present study successfully explains the mechanism determining the distribution range of a single species along a snowmelt gradient. The distribution in early snowmelt locations might be limited by the restriction of growth to the reproductive stage due to drought stress, whereas the distribution in late snowmelt locations might be limited by the difficulty of seedling establishment due to the short growing period.

4.2 Climate change impacts on alpine vegetation

Whether alpine plants can keep up with ongoing climate change is crucial to predicting the impacts of climate change on alpine ecosystems. Although there are a few examples of the rapid adaptation of phenological traits (Franks, Sim, & Weis, 2007), the progress of climate change often outpaces the possible adaptation of life-history traits, even in short-lived organisms (Wilczek, Cooper, Korves, & Schmitt, 2014). Some studies have demonstrated that the extension of the growing season length and enhanced drought stress due to advanced snowmelt times decreases the survival and reproductive success of snow-meadow plants, resulting in a declining population growth rate (Campbell, 2019; Iler et al., 2019). Increasing drought stress, caused by advanced snowmelt and warm summer temperatures, may cause the local extinction of alpine plant populations (Gritsch et al., 2016; Rumpf et al., 2019).

The present study explains the mechanism for the local extinction of alpine snow-meadow plants under climate change. In the early snowmelt habitat, A. narcissiflora relied on seed migration from later snowmelt habitats for the maintenance of the subpopulation. Generally, the seed dispersal distance of alpine plants is small, especially for species with no inherent dispersal syndrome (Morgan & Venn, 2017; Scherff et al., 1994). Because A. narcissiflora seeds are dispersed by gravity, later snowmelt populations acting as a source of seed dispersal should be located close to the early snowmelt habitat. Because our study site was located on a steep snowmelt gradient, the distance between the early and middle snowmelt subpopulations was 60 to 70 m (Figure S2). As previously noted, we observed that a large A. narcissiflora population disappeared during the 1990s in a snow-meadow located about 2.5 km east of our study site, where snowmelt progressed quickly over a gentle slope and flowering occurred simultaneously across a few hundred meters (Figure S1). This indicates that there were no late snowmelt populations around this area. With increasing drought stress caused by early snowmelt from the 1980s to 1990s, it is hypothesized that seed production decreased and this population degraded without seed migration from later snowmelt populations. This scenario indicates the importance of source–sink relationships among local populations for the stable existence of alpine plant populations under climate change (Dias, 1996; Eriksson, 1996; Ferrer et al., 2015).

Alpine vegetation diversity is composed of a mosaic of plant communities, reflecting the heterospecific environmental conditions at the local scale, where the existence of a snowmelt gradient is an important component (Björk & Molau, 2007; Hülber et al., 2011; Kudo & Ito, 1992). This indicates that modification of the snowmelt pattern could affect the vegetation structure of alpine ecosystems. Our comparisons of snow-meadow vegetation in the Taisetsu Mountains revealed that the mosaic structure of plant communities has been obscured over the last 40 years because of plant species invasions in later snowmelt locations and the decline of some herbaceous species in the snow-meadow communities, including A. narcissiflora (Amagai et al., 2018). These results highlight the vulnerability of alpine plant communities to climate change, especially in late snowmelt locations (Björk & Molau, 2007; Hülber et al., 2011; Matteodo, Ammann, Verrecchia, & Vottoz, 2016). In the snow-meadow, where a local A. narcissiflora population was extinct (Figure S1), graminoid species, such as Agrostis flaccida and Calamagrostis langsdorffii, increased the plant cover (Amagai et al., 2018). Such a transition of species composition might be caused by the modification of edaphic conditions under climate change although the existence of interspecific competition between decreasing and increasing species is unknown.

The present study demonstrated the importance of heterogeneous micro-habitats for the stability of local populations in alpine ecosystems. Although evidence for changes to the vegetation structures and species compositions of alpine plant communities have accumulated on a global scale, our understanding of the mechanisms of the local extinction and distribution shift of alpine plants is limited. Population biologists are facing the challenge of understanding the contribution of local population dynamics to the overall distribution patterns of individual species under climate change.


We sincerely thank to Takenori Takada for his kind instruction for the application of transition matrix model. This study was supported by Environmental Research and Technology Development Fund (D-0904) and JSPS KAKENHI (21370005, 24570015).


    The authors declared no potential conflicts of interest.