Introduction

High altitude plants in the Himalayan region play a vital role in medicines used to maintain human health. They are also used for other purposes such as construction, dyeing, fodder, food, fuel, incense, soap and sources of fibres (Manandhar 2002; Rokaya et al. 2010). For all of these purposes, plants are generally collected in the wild (Hamilton 2004; Ghimire et al. 2008). Due to improper harvesting techniques, these species are often subjected to over-exploitation and face the risk of extinction (Gaoue and Ticktin 2007; Schmidt et al. 2011). It is thus important to understand their detailed biology in order to predict the future of these species and establish efficient strategies for their conservation. Identification of sustainable harvesting strategies, which use natural resources without seriously depleting them (Martínez-Ballesté et al. 2005), helps to both conserve species and maintain the livelihoods of many rural people, who obtain a significant portion of their subsistence needs and incomes from plant resources (Hamilton 2004; Schmidt et al. 2011).

Studies from Nepal, in the central Himalayan region, have shown that the medicinal plant trade has a positive impact on the livelihoods of local people (Olsen and Bhattarai 2005; Olsen 2005). For trade, medicinal plants are generally harvested from the wild (Rai et al. 2000; Hamilton 2004; Kandari et al. 2012), which causes a serious reduction in plant populations (Sheng-Ji 2001). It is thus necessary to determine which plants are endangered, vulnerable, or should be prioritized for conservation (Kala et al. 2004). To identify the target plant species for conservation, different indices (use value index, sensitivity index and importance value index) have been used in India (Dhar et al. 2000), and rapid vulnerability assessment, which identifies plants that are vulnerable to over-exploitation through a relatively rapid filtering, has been used in Nepal (Lama et al. 2001). Specific studies on two medicinal plant species in Nepal, Nardostachys grandiflora and Neopicrorhiza scrophulariiflora, have shown that a large volume of rhizomes is collected in the wild (Olsen 2005; Ghimire et al. 2008). These studies indicated that sustainable management plans should include types of harvesting patterns, plant life forms and growth patterns (Ghimire et al. 2008) and should also include cyclical harvesting to protect plant species against potential extinction (Mariot et al. 2014).

Studies of population dynamics are an important baseline for understanding the future of plant populations and enable prediction of the effects of different harvesting techniques on the performance of plant populations. Among the available techniques, transition matrix models are commonly used techniques for describing population dynamics of plants and predicting their viability in the future (Caswell 2000a; Crone et al. 2013). These techniques can be applied when devising sustainable and profitable management plans for harvesting various plants species (Ghimire et al. 2008; Schmidt et al. 2011; Mondragón Chaparro and Ticktin 2011; Klimas et al. 2012). In a review, Schmidt et al. (2011) found 46 studies that used matrix models to assess the effect of harvesting on non-timber forest products. Most of the studies included palm and tree species (33 sp. div. + 1 cycad sp. div.), and only twelve focused on herb species. However, only two papers have focused on harvesting of medicinal plants (Nantel et al. 1996; Ghimire et al. 2008), so further studies are clearly needed.

Although many studies have demonstrated large variation between years and habitats in the life histories of perennial plants (e.g. Jongejans and De Kroon 2005; García et al. 2008; Jongejans et al. 2010; Heinken-Šmídová and Münzbergová 2012), studies on sustainable harvesting of medicinal plants generally only last for two to four years. This is quite a short time period to reveal possible temporal variability in population dynamics and its effect on population persistence (Crone et al. 2011). Moreover, with the exception of a study by Ghimire et al. (2008), the effect of harvesting on medicinal plants has only been explored in one habitat type. In a study on the harvesting of rhizomes of a perennial plant from Nepal, Ghimire et al. (2008) showed a large difference in sustainable harvesting between populations in the outcrops and in the meadows. The study stressed the importance of long-term research in multiple habitats in addition to studies over variable time periods.

To develop sustainable harvesting strategies for highly used perennial medicinal plants from the genus Rheum that grow in the high altitudes of the Himalayan region, we collected data on population dynamics from different populations (two habitat types for Rheum acuminatum and one habitat type for Rheum australe) over a six-year period in central Nepal and asked the following questions: (1) What are the population dynamics of two Himalayan rhubarb species (Rheum acuminatum and R. australe) in Nepal in different habitat types? (2) What is the effect of harvesting on population dynamics of the two species, and how does it differ between the different habitat types?

Material and methods

Study species

We studied two species of rhubarb belonging to the Polygonaceae family that grow in the high altitudes of the Himalayan range. Rheum acuminatum is a long-lived, perennial more than 1 m tall. It is found in high altitude habitats (2,800–4,000 m a.s.l.) on slopes, in forests or open rocky areas. Its leaves are generally less than 20 cm long and are wide, apex acuminate or long acuminate and rarely obtuse. Its flowers are dark purple and bloom in multiple clusters. Rheum australe is also a long-lived, perennial herb more than 2 m tall. It is found on grassy slopes, along streams, near bodies of water, in shady and rocky areas and at elevations of 3,200–5,200 m a.s.l. Its leaves are over 20 cm long and are apex obtuse. Its flowers are similar to those of R. acuminatum (Wu et al. 2003). Both species bloom from June to late July and fruit from late July to September. Seeds are dispersed from the end of October to November by wind, water, gravity or animals. They germinate in March/April of the following year.

All the below ground parts (the stems and roots) of both species of rhubarb are widely used in traditional folk, Ayurvedic and Tibetan medicines in Nepal (Lama et al. 2001; Baral and Kurmi 2006; Rokaya et al. 2010, 2012) and are often harvested in the wild in large amounts for trade. People in the buffer zone of the Langtang National Park depend on natural resources, and harvesting of medicinal plants is often a major source of income. The trade of R. australe in Nepal increased from 915 kg of belowground organs in 2008/2009 (GoN/MoFSC 2009) to 4,385 kg in 2014/15 (GoN/MoFSC 2015). Moreover, it is assumed that much unreported trade occurs as well. The trade of R. acuminatum is not well documented, but it is often mixed with R. australe and is commonly called ’Padamchal’ in the Nepali language. Rheum acuminatum is also used as substitute for R. australe in traditional or Ayurvedic medicine (Rokaya et al. 2012).

In Nepal, there are generally three groups of medicinal plant users (Ghimire et al. 2005, 2008): (i) commercial collectors, who harvest plants in large volumes for trade; (ii) traditional healers, who collect plants in small amounts for use in traditional medicine; and (iii) local non-specialists who collect plants for their own uses (dyeing, condiments, self-medication, veterinary use, etc.). To identify methods of harvesting of the Rheum species, we carried out interviews with local people, commercial collectors (outside the park area) and traditional healers. We also observed the actual harvesting practices in the field by asking one traditional healer (a local person) and two local non-specialists (who were supposed to reflect harvesting patterns of local as well as commercial collectors) to collect plants in August 2007. Harvesting for trade requires collection of a large volume of plants; thus, collectors haphazardly uproot plants everywhere, irrespective of plant stages. The traditional healers select mature plants for uprooting based on the concept that mature plants are more potent than young plants. Analogously, local non-specialists, such as commercial collectors, randomly uproot plants in places where they are abundantly available. The frequency of harvesting (every year, once every two or three years, etc.) and amount of harvesting depend on the availability of resources in nature as well as the type of plant user. Although no useful information about the total decrease in number and size of the rhubarb population is available, there are multiple examples of severely destroyed populations due to harvesting. Regions with the most affected populations include central and mid-western Nepal (GoN/MoFSC 2015).

Plant populations and sampling designs

A demographic study was performed in four populations of R. acuminatum and two populations of R. australe in the Gosaikunda region of Langtang National Park (central Nepal). Two populations of R. acuminatum were located under thickets or forest (Chandan Bari and Cholang Pati, ~ 3,500 m a.s.l.), hereafter called ’forest’. Two others were in open grasslands or rocky areas (Phulung Ghyang, ~ 3,450 m a.s.l. and Pongjo Kharka1, ~ 3,550 m a.s.l.), hereafter called ’open’. The two studied populations of R. australe were located in open grasslands and rocky areas (Phongjo Kharka2, ~ 3,550 m a.s.l. and Mirki Bhir, ~ 3,450 m a.s.l.). More precise locations are not provided for the purpose of conservation. The study populations were located with the help of local people (animal herders and traditional healers) in the region. Populations were located far (> 6 km) from villages and faced low human pressure with no harvesting regime.

In August 2007, we selected approximately 100 plant individuals per population (usually all plants within the population), permanently tagged them and followed them every year in August until 2012, i.e. for five subsequent years. When possible, we ensured that individuals of different stages (as defined below) were marked in similar proportions (see Münzbergová and Ehrlén 2005). Every year, small new vegetative plants found within the locality were also tagged. These plants were found in areas without a previous occurrence of rhubarb plants, and there were no old leaves from large plants in the same spot. Moreover, they were not connected to surrounding plants. Thus, all small new vegetative plants were very likely germinated from seeds. Therefore, we only considered generative and not clonal reproduction in our study. In each census, we recorded the number of leaves, total leaf length and petiole length of the longest leaf in each plant individual. For flowering plants, we recorded the length of the flowering stem. Mortality of the plants was also recorded in the following year.

Using regression analyses, we looked for thresholds in recorded plant traits that differentiated vegetative plants with low and high probability of flowering in the next year. Petiole length was ultimately the best predictor of plant size and flowering in the next year. We thus classified individuals of both rhubarb species into three stage classes based on petiole length of the longest leaf as well as flowering: small vegetative (petiole length < 20 cm for R. acuminatum and < 25 cm for R. australe and not flowering), large vegetative (petiole length > 20 for R. acuminatum and > 25 cm for R. australe and not flowering) and flowering individuals. Small vegetative plants were categorized to include small plants with zero or very low probability of flowering in the next year and to contain a proportion of approximately one-third of tagged plants. Small vegetative plants can grow to large vegetative plants in the next year; in rare cases, they can remain in the stage of small vegetative plants for one more year. Most of the large vegetative plants stay in this stage for more than one year, and when they gather enough resources, they flower and produce seeds. The plant usually dies after flowering; however, it is not strictly monocarpic because it can become vegetative after flowering again. In very rare cases, it can flower in two consecutive years. We thus consider both studied species as perennial nonclonal with polycyclic shoots (Klimešová et al. 2016).

Seed production and seedling recruitment

After the maturation of seeds in October of each year, we randomly sampled seeds from 10 flowering individuals from each population to calculate the total seed production per plant. To assess the seedling recruitment rate, we sowed the seeds in the field immediately after seed collection. We established three sowing plots (1 × 1 m) per year and locality, and 100 seeds were sown into each plot. The plots of following years were laid adjacent to plots of previous years. However, probably due to intensive seed herbivory, cold temperatures and rotting of seeds due to easy absorption of water from the environment, we recorded no natural seedling recruitment in the field. In subsequent models and analyses, we thus used seedling recruitment rates of 0.5, 0.75, 1.0, 3.0 and 5.0% as well as greenhouse seedling recruitment (30%) and compared the results with changes in the real population size during the six years of the study. The best match was found with the recruitment rate of 0.75%, which was therefore used in the subsequent analysis for all populations and species; only results with this recruitment rate are presented.

To estimate survival of seeds in the seed bank, we buried 10 nylon bags per locality, each containing 100 seeds, in autumn of 2007 and 2008. We excavated the seed bags in the following two years (spring of 2008, 2009 and 2010), but all of the seeds were decayed. We thus concluded that our study species do not form a persistent seed bank.

Data analysis

Comparison of vital rates

We compared differences in the probability of stasis, growth, retrogression, flowering and mortality between R. acuminatum in forest and open habitats and R. australe in open habitats (these three types will hereafter be called population types). For this comparison, we used logistic regression with stage in the previous year, year, species in population type, locality nested within population type and their interactions as independent variables. Stasis was defined as the plant surviving and remaining in the same stage. Growth was defined as plant growth to a larger stage, i.e. growth from a small vegetative to a large vegetative plant or flowering and of large vegetative to flowering plants. Retrogression was defined as shrinkage from a larger stage to a smaller stage, i.e. shrinkage of a large vegetative or flowering plant to a small vegetative plant or shrinkage of a flowering plant to a large vegetative plant.

The differences in number of seeds per plant were compared by generalized linear models with Gamma distribution. Year, population type, locality nested within population type, and their interactions were used as independent variables. In these analyses, all factors were fixed. Analyses of vital rates were carried out in R version 3.0.2 (R Development Core Team 2013).

Population dynamics

Data on recruitment, growth and survival of individuals classified by size were gathered over five transition periods (2007–2012). Stage-based population projection matrix models were used to estimate demographic parameters (Caswell 2000a). Models were in the form of

$$ {n}_{\left( t + 1\right)}= A{n}_{(t)}, $$

where n is a column vector containing the number of individuals in each stage at time t or t + 1, and A is a square matrix with the matrix elements representing transition probabilities among stages. There are three stages in our study system: small vegetative (a1), large vegetative (a2) and flowering individuals (a3). a ij indicates transitions in the matrix A from stage j to stage i in one-year time intervals. From each transition matrix constructed for each population and year, we consequently calculated lambda (λ), which is known as the population growth rate (Caswell 2000a), and elasticity, which is usually used as a measure of the contribution of a matrix element to fitness (de Kroon et al. 2000). Elasticity analyses can be used to identify potential management targets because changes in vital rates with high elasticity will produce larger changes in λ (Caswell 2000b).

Both the 95% confidence intervals (CI) of λ and elasticity values of each transition matrix were estimated by the bootstrap percentile-interval method (Alvarez-Buylla and Slatkin 1994). Bootstrapping the original data used to derive the transition matrices was performed 105 times (see Dostálek and Münzbergová 2013 for details). For this purpose, a MATLAB script developed for a previous study was used (Münzbergová 2006).

To summarize information on λ and elasticity for population types, populations and years, we used the stochastic simulation approach as suggested by Caswell (2000a) and Rydgren et al. (2007). Using this approach, we incorporated environmental and temporal stochasticity within population type into the model and into estimates of λ and elasticity values. We calculated two types of stochastic population growth rates – λS (population growth rate based on the stochastic simulation approach). First, λS for every year and population type was based on a combination of data from two populations. There were five such λS values for each population type based on years 2007–2012. By combining data from two populations, we were able to compare variation in λ among years. Second, overall λS were based on a combination of data from two populations and all years 2007–2012 for each population type. There was thus one overall stochastic population growth rate value for each population type. This approach enabled us to compare the population growth of all three population types in the long run. Simulations were performed using a MATLAB script developed in a previous study (Münzbergová 2005). We ran the same stochastic simulations for the bootstrapped matrices described above and were therefore able to construct a 95% confidence interval of the stochastic population growth rate and elasticity (Černá and Münzbergová 2013).

A life-table response experiment (LTRE) (Caswell 2000a) with a fixed factorial design was conducted to examine the contribution of each transition to the observed variation in λ between species in the two habitat types (see Appendix  1 for details). Compared to the elasticity analysis, the LTRE analysis quantifies the observed effects of single matrix elements on observed variation in the population growth rate, as opposed to expected effects. Potential best management actions identified by elasticity analysis are sometimes not realized, because it is difficult to change vital rates with the highest elasticity values. Then, LTRE analyses enable identification of variable matrix elements with the highest positive or negative contributions to the population growth rate. To visualize the LTRE values and compare the three population types, we also summed the positive and negative contributions separately for survival (including transitions a11, a12, a22, a23, a33), as well as fecundity (a13) and growth (a21, a31, a32), and plotted these values (Jongejans and De Kroon 2005; Černá and Münzbergová 2013; Münzbergová 2013).

Harvesting simulation

To reveal the effect of harvesting on plant survival and growth, we performed a harvesting experiment in the field. We selected a population of approximately 50 R. acuminatum individuals next to a population at Phulung Ghyang, marked the plants and asked a local to harvest the plants so that the actual local harvesting technique would be applied. We followed these plants (observing survival and possible resprouting) in subsequent years. No plants were found in this place the next year or following years. We assume that none of the harvested plants was able to successfully resprout because all the belowground organs were harvested during the experimental process.

To assess the effects of harvesting on performance of the species, we conducted projections of population size over 30 years under four different harvesting scenarios (see below). To simulate environmental stochasticity, one of the 10 matrices (2 populations × 5 transition intervals) available for each population type was randomly selected in each simulation step. In each step, the resulting population vector (number of plants in each stage) was replaced by values drawn from a Poisson distribution with the appropriate mean, which is a standard procedure to simulate demographic stochasticity (Caswell 2000a). This projection was repeated 1,000 times for each population type (Münzbergová 2005). The harvesting was simulated by removing 0, 25, 50 and 75% of the natural population every 1, 3, 5 and 10 years. This was carried out in different scenarios, including removal of (i) only plants in the large vegetative and flowering stages, (ii) only large vegetative plants or (iii) only flowering plants. Small vegetative plants are not intentionally harvested for belowground organs. However, they are usually destroyed when commercial collectors randomly dig up the belowground organs of mature plants, leading to the destruction of all plants in a population. Therefore, in simulation tests, we also used a scenario (iv) in which all of the plants (small and large vegetative and flowering plants) were removed. For the initial population vector for the harvesting simulations, we used 100 plant individuals divided according to stable stage distribution (Caswell 2000a) and averaged all populations of both species, i.e. 40 small vegetative, 52 large vegetative and 8 flowering plants. The stable stage distribution was very similar between the population types. By using the mean stable stage distribution, we ensured that populations of all population types were under exactly the same starting conditions in our simulations. The MATLAB script used during the analysis had been developed previously for different studies (Münzbergová 2005,2006).

All analyses were performed using Matlab version 6.0.0.88 (The MathWorks, Inc., Natick, Massachusetts, USA).

Results

Comparison of vital rates

All tested vital rates (probability of stasis, growth, retrogression, flowering and mortality) significantly varied among years and population types. The differences between populations within species in different habitat types were much smaller in comparison to the differences caused by variability among years and population types. Individuals of R. australe in open habitats survived longer in the same stage (stasis), less often shrunk to smaller stages and had lower mortality in comparison to R. acuminatum in both forest and open habitats. On the other hand, R. acuminatum in forest habitats flowered earlier in comparison to the other population types. R. acuminatum in open habitats was in between the strategies of the other two population types and had lower probabilities of both stasis and flowering as well as a higher probability of mortality (Table 1 , Fig.  1 ).

Table 1 Effects of stage in previous year, year, population type, locality nested within population type and their interactions on probability of stasis (survival and remaining in the same stage), growth (growth to larger stage), retrogression (shrinkage from larger stage to smaller stage), flowering and mortality in the next year. Tests were done using logistic regression. Significant results (P < 0.05) are in bold. Residual degrees of freedom – 2,828, d.f – degrees of freedom; Expl. dev. – proportion of explained deviance out of total deviance in the model.
Full size table
Fig. 1
figure 1

Comparison of vital rates among three population types (R. acuminatum in forest and open habitats and R. australe in open habitats). Vital rates are the probability of stasis (survival and remaining in the same stage), growth (growth to a larger stage), retrogression (shrinkage from a larger stage to a smaller stage), flowering and mortality in the next year. Boxes show means, standard errors and 1.96 standard errors (defining 95% confidence interval of the mean) of percentage of individuals undergoing given transition. Different letters indicate significant differences between population types within one vital rate (Tukey’s post hoc test, P < 0.05).

Full size image

Rheum acuminatum produced more seeds in the forest habitats (297 ± 19, mean ± SE, seeds per plant) in comparison to open habitats (262 ± 13 seeds per plant). However, R. australe produced many more seeds (480 ± 21 seeds per plant). Seed production also strongly varied between years (Table 2 ).

Table 2 Effect of year, population type, locality nested within population type and their interactions on seed numbers of R. acuminatum and R. australe. Tests were done using generalized linear models. Significant results (P < 0.05) are in bold. Residual degrees of freedom – 232.
Full size table

Population growth rates

We recorded significant variation in the population growth rate between populations and years of all population types. The stochastic population growth rates (λS) based on a combination of data from two populations within one year varied from 1.01–1.10 for R. acuminatum growing in forest habitats, 0.90–1.07 for R. acuminatum in open habitats and 0.96–1.20 for R. australe growing in open habitats during five transition intervals (Fig.  2 ). Overall stochastic population growth rates (λS) based on a combination of data from 2007–2012 showed that R. acuminatum growing in open habitats had the lowest population growth rate (λS = 0.97), followed by R. acuminatum growing in forests (λS = 1.07) and R. australe in open habitats (λS = 1.09; see Fig.  2 ).

Fig. 2
figure 2

The stochastic population growth rates (λS) and their 95% confidence intervals for three population types. The last values in each population type (indicated with asterisks) show the results for all transition intervals combined (2007–2012). The dashed line (λ = 1) indicates a stable population.

Full size image

Elasticity

The transition with the highest elasticity value in all population types was stasis in the stage of large vegetative plants (a22). Conditions leading to the damage of these large vegetative plants are most destructive to the Rheum populations and have the greatest effect on the population growth rate. Other important transitions were associated with growth (a21 and a32) and fecundity (a13). We recorded no significant differences in elasticity values between population types (Fig.  S1 ), indicating that similar management action should be taken to support all three population types.

Life table response experiment

According to LTRE analyses, population growth of R. acuminatum from forest habitats is positively affected by growth-related transitions (mainly the high probability of growth of large vegetative to flowering plants). For R. australe in open habitats, it is positively affected by fecundity and survival-related transitions (mainly the high production of small vegetative plants from flowering plants and high survival of plants in the large vegetative stage). In contrast, populations of R. acuminatum in open habitats contributed negatively to variation in the overall population growth rate by transitions related to survival, fecundity and growth (mainly due to low production of small vegetative plants from flowering plants and low growth from small to large vegetative plants; see Table 3 , Fig.  3 , Fig.  S2 ).

Table 3 Results of LTRE analyses comparing the contribution of single matrix elements to observed variation in overall population growth rate between population types.
Full size table
Fig. 3
figure 3

Result of LTRE analysis showing positive and negative contributions of three population types to the observed variation in overall population growth rate grouped by life history components – survival (S), i.e. stasis and retrogression, fecundity (F) and growth (G).

Full size image

Effect of harvesting on population size

Harvesting 25% of all plant stages (small and large vegetative and flowering individuals) every year led to a significant decline in population size and increased the probabilities of extinction within 30 years in all three population types (Fig.  4 , 5 , Table S1 ). The extinction probability was highest in R. acuminatum in open habitats. The negative effect of harvesting was partly reduced when only large vegetative and flowering plants were harvested. The lowest extinction probabilities were achieved when harvesting only flowering plants (Fig.  4 , 5 ).

Fig. 4
figure 4

Extinction probabilities of three population types in 30 years with harvesting intensity of 25% of individuals in all (All stages), only large vegetative and flowering (Large veg. & flow.), only large vegetative (Large vegetative) or only flowering (Flowering) stages. 1, 3, 5, 10 on the X-axis indicates the interval of harvesting in years.

Full size image
Fig. 5
figure 5

Extinction probabilities of Rheum species in three population types in 30 years with harvesting frequency every three years. Individuals in all, only large vegetative and flowering, only large vegetative or only flowering stages are harvested in proportion of 0, 25, 50 and 75% of all plants in the population as indicated on the X-axis.

Full size image

Harvesting could be considered sustainable (the extinction probability in 30 years was close to zero) when harvesting 25% of all plant stages in a rotation of 3–5 years for R. australe and R. acuminatum growing in forest habitats. R. acuminatum growing in open habitats was extremely vulnerable to harvesting; even when 25% of the vegetative or flowering plants were removed one time in 10 years, there was an approximately 15% probability of population extinction within 30 years (Fig.  4 , 5 , Table S1 ).

Discussion

Although the population dynamics among the studied Rheum species in the two habitat types and among different years of the study were relatively stable, there were significant differences in sensitivity to harvesting. Sustainable harvesting of the populations is possible using long rotational periods (at least five years) or when harvesting is restricted to certain stages. The optimal harvesting strategies, however, differ between the species and between the two habitat types. This result suggests that design of optimal harvesting strategies must be both species- and habitat-specific.

Very few studies concerning the impacts of harvesting non-timber forest products on population dynamics have lasted for more than two to three years (review by Schmidt et al. 2011). Thus, few studies have been able to assess temporal variability and its effects on long-term population dynamics, despite the fact that temporal variability can play a crucial role in the overall dynamics of a species (Oostermeijer et al. 1996; Fréville et al. 2004; Marrero-Gómez et al. 2007; Jongejans et al. 2010; Crone et al. 2011; Bucharová et al. 2012). In particular, the studies aimed at providing recommendations for sustainable harvesting should last longer in order to achieve results that are not biased by their disregard of temporal variability. In our study, variability in survival, growth, flowering and seed set resulted in variability in the population growth rate between years and populations of the two Rheum species. Different climatic factors, such as the timing of snow melting, rainfall and temperature during the growing season, probably contributed to this variability (Marrero-Gómez et al. 2007; Nicolè et al. 2011). Weather conditions could also affect seed production and seedling recruitment; an example is seen in the study by Fréville et al. (2004). Moreover, the seeds of both species are large and easily visible and thus represent an important source of food for some local predators. Specifically, a frequent predator of rhubarb seeds is Royle’s Pika (Ochotona roylei) (Khanal 2007), which can consume approximately 10–15% of the total seed production every year (M. Rokaya, personal observation). Variable seed production and herbivory together with very low seedling recruitment is another important source of temporal variability (Münzbergová 2005; Maron and Crone 2006; Russell et al. 2010; Rokaya and Münzbergová 2012). The strong intensity of herbivory may also explain why we detected no natural germination in the field sowing experiments. High sowing intensity in the case of large Rheum seeds may attract seed predators and lead to a greater reduction in available seeds than would be expected when seeds are dispersed naturally. Furthermore, the survival of seeds in open habitats throughout the winter is low; mature seeds are easy prey or are subjected to desiccation until their death due to sunlight (before winter) and cold temperatures during winter.

Population growth rates of both Rheum species in the two habitat types in our study were close to 1, which corresponds to the stable population growth rate of many perennial plant species (García et al. 2008; Ramula et al. 2008). García et al. (2008) also showed that the longer the species lifespan is, the more stable the population dynamics of the species. This is also the case for the Rheum species studied here, which live a relatively long time (usually five or more years). Perennials are usually better able to cope with changing weather conditions because they have storage in their belowground organs. However, this can also make them very sensitive to harvesting (Ghimire et al. 2008). The effect of harvesting on population dynamics depends on the stage and plant part that is harvested. Because survival of large vegetative individuals had the highest elasticity, harvesting of these plants with the potential to flower in subsequent years could severely decrease population growth rates (Law 2007; Ghimire et al. 2008). The harvesting of underground parts (in the case of Rheum species) also contributes more to the decline in plant populations compared to harvesting of aboveground parts because the underground parts of plants usually die after harvesting (Ticktin 2004; Rock et al. 2004; Lázaro-Zermeño et al. 2011). However, either the selective harvesting of different plant stages or long-term rotational harvesting should reduce the risk of population decline. The results of our study suggest that harvesting of both species can be considered sustainable (the probabilities of extinction within 30 years are very low) under certain conditions. The best strategy for sustainable harvesting of R. acuminatum and R. australe in forest habitats is to harvest only flowering plants after they produce mature seeds because flowering plants usually die after seed production. The population is also not as endangered by removing flowering plants because harvesting of reproductive structures, including propagules and flowers, usually has a low impact on the population dynamics of perennial species (Ticktin 2004; Emanuel et al. 2005; Schmidt and Ticktin 2012; Castro et al. 2015). The main problem associated with Rheum plant species is that there are not many flowering plants (approximately 8% of plants were flowering in the studied populations). According to our models, if plants of all stages are harvested even at low harvest intensity (25% every three years), a decline in projected population sizes is expected for both species growing in any habitat type (R. acuminatum in the forest habitat as well as R. australe in the open habitat). This finding is also comparable to findings from different parts of the world for other non-timber forest product species whose belowground organs or whole plants were harvested (Nault and Gagnon 1993; Nantel et al. 1996; Raimondo and Donaldson 2003; Freckleton et al. 2003; Rock et al. 2004; Hernández-Apolinar et al. 2006; Ghimire et al. 2008; Lázaro-Zermeño et al. 2011). Thus, commercial collectors or local non-specialists who collect plants haphazardly and destroy small vegetative or seedling individuals are the most important contributors to population decline, as found in the case of N. grandiflora growing in Nepal (Ghimire et al. 2008). However, the traditional healers, who collect plants in small amounts for traditional medical uses, endanger plants less. The data analysis in our study species revealed that harvesting plants in specific stages or maintaining a certain time period for rotation of the harvest has almost no negative impact on population sizes. However, this is not true for R. acuminatum in open habitats, which is strongly advised not to be harvested at all. Although both species in the two habitat types have relatively similar population growth rates, harvesting even a few individuals over a very long time period can have serious consequences for the existence of R. acuminatum in open habitats. This finding is similar to the results shown for the economically important species N. grandiflora, found in west Nepal (Ghimire et al. 2008). While populations of N. grandiflora in the outcrops were stable and harvesting even a very small proportion of the population could lead to extinction, populations in the meadows were performing well and a much larger proportion could be harvested. Together with the results of our study, this emphasizes that harvesting recommendations should be specific to not only species but also habitats.

Conclusion

Our study demonstrated that the random uprooting typical of commercial harvesting has a strongly negative impact on the populations of both Rheum species, and there is a chance of extermination of the whole population as the reproductive capabilities of plants are reduced. Thus, to protect these Himalayan endemic plants from extinction, the following management strategies should be adopted: (i) Local harvesting regulations are required for different populations, as the sensitivity of populations to harvesting differs between species in different habitat types; (ii) The transplantation or cultivation of plants through belowground organs/seeds is required either in situ or ex situ; and (iii) Rotational harvesting or selective harvesting at particular plant stages is necessary to protect natural populations.