Long- term population dynamics: Theory and reality in a peatland ecosystem

1. Population dynamics is a field rich in theory and poor in long-term observational data. Finding sources of long-term data is critical as ecosystems around the globe continue to change in ways that current theories and models have failed to predict. Here we show how long-term ecological data can improve our understanding about palaeo-population change in response to external environmental factors, anteced ent conditions and community diversity. 2. We examined a radiometrically dated sediment core from the Didachara Mire in the mountains of south-western Georgia (Caucasus) and analysed multiple biological proxies (pollen, fern spores, non-pollen palynomorphs, charcoal, diatoms, chryso phyte cysts, midges, mites and testate amoebae). Numerical techniques, including multivariate ordination, rarefaction, independent splitting and trait analysis, were used to assess the major drivers of changes in community diversity and population stability. Integrated multi-proxy analyses are very rare in the Caucasus, making this a unique record of long-term ecological change in a global biodiversity hotspot. 3. Synthesis. Population changes in the terrestrial community coincided primarily with external environmental changes, while populations within the peatland community were affected by both internal and external drivers at different times. In general, our observations accord with theoretical predictions that population increases lead to greater stability and declines lead to instability. Random variation and interspe cific competition explain population dynamics that diverged from predictions. Population change and diversity trends were positively correlated in all taxonomic groups, suggesting that population-level instability is greater in more diverse We used statistical techniques that detect major changes in population size and stability for each taxo nomic entity without the need for analogue- based bias corrections (Birks et al., 2012; Walker & Pittelkow, 1981; Walker & Wilson, 1978). Population changes were compared to independent climatic records (regional- scale conditions) and changes in sediment composition within the site (local- scale conditions). Based on ecological theory, we pre dicted that populations achieve greater stability following an increase in population size (Hamilton et al., 2009; Lande, 1993), and that taxonomic groups of higher diversity (richness) exhibit greater community- level sta bility and population- level instability compared to low- diversity groups (Gross et al., 2014; Loreau & de Mazancourt, 2013; Tilman, 1996). in the (or Caucasus Minor) Mountains of though There is a to theory with long-term to test the predictive success of theoretical frameworks, thereby improving their ability to predict future change.


| INTRODUCTION
Improving knowledge of ecological processes is increasingly important as anthropogenic impacts affect global ecosystems in potentially irreversible ways (Cahill et al., 2013;Kubisch, Holt, Poethke, & Fronhofer, 2014;Parmesan, 2006). In a changing environment, extinction risk is an ever-present concern (Crone, 2016). Population theory predicts that extinction risk is reduced in populations that are stable or increasing (Channell & Lomolino, 2000). Growing populations are considered more resilient to environmental change and demographic disruptions than shrinking ones (Hamilton et al., 2009;Lande, 1993). In reality, however, populations and communities change in ways that current theory cannot always predict (Moritz & Agudo, 2013;Zhang, Yan, Krebs, & Stenseth, 2015).
Population dynamics and species diversity are closely intertwined.
At the community level, high-diversity ecosystems tend to be more stable and resilient to environmental changes than less diverse systems (Cardinale et al., 2012;Tilman & Downing, 1994;cf. May, 1972).
At the population level, on the other hand, empirical data and modelling suggest that individual populations in diverse communities should experience greater instability than those in less diverse communities (Loreau & de Mazancourt, 2013;Tilman, 1996). Little scientific attention has been devoted to examining this relationship temporally using palaeo-data. Knowledge of the factors that drive diversity on centennial to millennial scales is limited geographically, temporally and taxonomically (Birks, Felde, Bjune, et al., 2016;Birks, Felde, & Seddon, 2016;Feurdean et al., 2013).
In this paper, we use high-quality palaeoecological data from a peatland to address the following questions: 1. Do palaeo-population dynamics correspond temporally to environmental changes (extrinsic vs. intrinsic) and, if so, does this relationship differ for different groups of organisms?

2.
To what extent do palaeo-population changes conform to the predictions of population theory, especially in relation to diversity?

| Research design
We adopted a palaeoecological approach to address our research questions. Biological fossils from various taxonomic groups were analysed at comparable sampling intervals in a single core to provide a long temporal sequence of palaeoecological changes in a peatland ecosystem. Palaeopopulation dynamics were reconstructed based on fossil accumulation rates (Flenley, 2003;Jeffers et al., 2015). We used statistical techniques that detect major changes in population size and stability for each taxonomic entity without the need for analogue-based bias corrections (Birks et al., 2012;Walker & Pittelkow, 1981;Walker & Wilson, 1978).
Population changes were compared to independent climatic records (regional-scale conditions) and changes in sediment composition within the site (local-scale conditions). Based on ecological theory, we predicted that populations achieve greater stability following an increase in population size (Hamilton et al., 2009;Lande, 1993), and that taxonomic groups of higher diversity (richness) exhibit greater community-level stability and population-level instability compared to low-diversity groups (Gross et al., 2014;Loreau & de Mazancourt, 2013;Tilman, 1996).

| Study site
Our study site is a poor fen, Didachara Mire, located near the upper forest limit in the Lesser Caucasus (or Caucasus Minor) Mountains of communities, even though diverse communities are themselves more stable. There is a continuing need to confront population theory with long-term data to test the predictive success of theoretical frameworks, thereby improving their ability to predict future change.

K E Y W O R D S
chironomids, diatoms, diversity, Georgia, palaeoecology and land-use history, pollen, population dynamics, testate amoebae Georgia (41°41′02″N, 42°29′49″E; see Figure S1). It sits at the foot of Mt Tsvinta (2,423 m) on the NW flank of the Arsiani Range at an elevation of 2,000 m a.s.l. The peatland forms part of the headwaters of the Acharistskali River, a tributary of the Chorokhi (Çoruh) River.
Didachara is an optimal site for studying environmental change as it is situated at the intersection between two global biodiversity hotspots (Caucasus and Irano-Turanian: Mittermeier et al., 2005), two distinct vegetation belts (Colchic mixed coniferous forests and subalpine vegetation: Nakhutsrishvili, 2013) and two major climatic zones (maritime and continental : Javakhishvili, 1964; see Figure S1).
The peatland occupies a small, closed basin ( Figure S2). Its predominant vegetation is a relatively species-rich Sphagnum-Carex associ-  (Margalitadze, 1982;nomenclature after Gagnidze, 2005). Subalpine meadows of the West Lesser Caucasian type (Nakhutsrishvili, 2013) extend from the site's elevation to the highest elevations of the nearby mountains (2,452 m). Margalitadze (1982) previously reconstructed the Holocene vegetation history of the site on the basis of undated pollen assemblages in two sediment cores. The present study greatly improves on that earlier reconstruction and is among the first of its kind in the Caucasus region to integrate dated palaeoecological data from various proxies (De Klerk et al., 2009;Moiseenko et al., 2012).

| Field and laboratory methods
In August 2003, three parallel sediment cores were collected from the centre of the peatland using a Streif-Livingstone corer ( Figure S2; Merkt & Streif, 1970). Core C, the most complete sequence, was selected for detailed multi-proxy palaeoecological analyses, loss-onignition and radiocarbon dating. In the laboratory, samples were taken at regular intervals to extract fossilised pollen, spores and non-pollen palynomorphs (0-910 cm), midges and mites (2-798 cm), diatoms and chrysophytes (610-810 cm) and testate amoebae (70-630 cm); these depth ranges were selected based on preliminary analyses and sediment characteristics. Statistically valid numbers of diatoms were only recovered in lake sediments. Testate amoebae were mostly found in peat.
Chironomids are sensitive indicators of changing freshwater environments, including temperature variations (Walker, 2001).
Chironomids and other invertebrate remains (biting midges and oribatid mites) were quantified in 48 samples. Sediment samples were processed following standard techniques (Brooks, Langdon, & Heiri, 2007;Walker, 2001). Wet sediments (1-3 g) were initially washed through a 100μm mesh sieve. Invertebrate remains were sorted and picked out from the sieve residue in a Bogorov counting tray under a stereomicroscope at 25-40× magnification and mounted on microscope slides using Euparal ® . Chironomids were identified at 200-400× magnification following Brooks et al. (2007) and Andersen, Ekrem, and Cranston (2013a). Ceratopogonidae (biting midges) were separated into morphotypes following Walker (2001). Only chironomids were used for analyses requiring percentage data (e.g. zonation and ordination). Accumulation rates for all invertebrate taxa were calculated by wet sediment weight (remains g WW −1 year −1 ). At least 55 remains were identified per sample (mean: 700).
In freshwater environments, diatoms and chrysophytes may be used to trace changes in nutrient status, salinity and pH (Smol, Birks, & Last, 2001). Diatoms and chrysophyte cysts were enumerated in 23 lake sediment samples. Samples were digested in H 2 O 2 and HCl, and mounted in Naphrax following Renberg (1990). Above a depth of 735 cm at least 250 diatom frustules were identified per sample, but below this depth, 110-215 valves were counted per sample due to low concentrations (mean valve count for the core: 435). Diatom accumulation rates were not estimated.
Eleven samples of wood and other identifiable macrofossils were submitted for AMS radiocarbon dating. Radiocarbon dates were calibrated using the IntCal13 calibration curve (Reimer et al., 2013) and a calendar age of −53 bp (2003 ad-collection year) was assigned to the core top. We produced an age-depth model ( Figure S3) using Bacon 2.2 (Blaauw & Christen, 2011), a Bayesian approach that considers prior information on sedimentation rates. Stratigraphic diagrams showing the relative abundance of each taxon by sample depth and modelled age were produced using Tilia 2.0.32 (Grimm, 2013).

| Numerical analyses
To analyse palaeo-community interactions and factors influencing ecosystem development, we used assemblage zones and multivariate ordination scores as the basis for comparison (Ammann et al., 2000). Assemblage zones were determined through binary splitting, the number of zones being assessed statistically by the broken stick model (Bennett, 1996). Non-metric multidimensional scaling with Bray-Curtis dissimilarity measure was applied to each of the five proxy datasets (upland pollen, wetland pollen and spores, chironomids, diatoms and chrysophytes, and testate amoebae). Non-metric multidimensional scaling was selected over other techniques because it makes no assumptions about response models and is therefore applicable to datasets of different variances. Data were square-root transformed prior to analysis in PAST (Hammer, 2015).
To assess whether individual palaeo-populations have changed in consistent ways, we used independent splitting, a valuable and underutilised tool for analysing multi-proxy data (Birks & Birks, 2006;Birks et al., 2012;Lotter, Birks, & Zolitschka, 1995). Independent splitting treats each palaeoecological taxon independently, which is only possible when accumulation-rate data and a robust age-depth model are available. Changes in accumulation-rate data through time are divided (split) into sections with homogenous mean and standard deviation statistics. These statistics are used to assess relative changes in palaeo-population size and stability, without the need for quantitative population estimates (Walker & Wilson, 1978).
Independent splitting was implemented in Psimpoll (Bennett, 2005). Pollen taxa with <10 occurrences were excluded, as were testate amoeba, midge and mite taxa with <5 occurrences. At each statistically significant population change (Walker & Wilson, 1978), we observed whether stability increased or decreased according to the ratio of mean to standard deviation (M/SD; Walker & Pittelkow, 1981). Results were compared to the null model of Blaauw, Bennett, and Christen (2010), with 450 random proxy walks of 200 samples at population sizes of 500 and 2,000 (representative of local proxies and terrestrial pollen respectively: Blaauw et al., 2010). Observed population changes in fossil data were also compared to species traits by associating fossil taxa with traits from representative or similar species (Fournier, Lara, Jassey, & Mitchell, 2015;Kleyer et al., 2008).

| RESULTS
Palaeoecological data from Didachara show significant long-term variations in all the taxonomic groups analysed at decadal to multicentennial scale ( Figure 1). A total of 208 terrestrial pollen taxa, 32 wetland taxa (local pollen and non-pollen palynomorphs), 52 diatom taxa and chrysophyte cysts, 61 testate amoeba taxa, 28 midge taxa and oribatid mites were identified ( Figure S4). In interpreting the data, an important distinction must be made between the larger spatial scale represented by upland pollen compared to the other proxies which represent communities within the site itself. Pollen from major pollenproducing trees in mountainous areas is readily transported upslope by wind (Kvavadze, 1993). Upland pollen therefore comprises a mixed signal derived from source plants at varying distances from the site (typically 1-10 km: Mariani, Connor, Theuerkauf, Kuneš, & Fletcher, 2016;Markgraf, 1980). Other proxies (wetland plants, diatoms, testate amoebae, midges and mites) are largely derived from local, in situ populations. Hence, upland pollen provides a terrestrial community background against which to interpret local changes in the lake/peatland community. In the following section, major palaeoecological transitions are interpreted with reference to environmental conditions.

| Terrestrial community development since 13,000 cal. bp
The earliest upland palaeovegetation zone (Poaceae-Artemisia; Figure 1) indicates a largely treeless landscape of grassy steppe, with moderate burning and deposition of minerogenic sediments, typical of Late-glacial environments across the region (Messager et al., 2013(Messager et al., , 2017Wick, Lemke, & Sturm, 2003). Increasing temperature and precipitation at the beginning of the Holocene (c. 11,700 cal. bp; Figure 2) allowed the lake to fill, meadow and marsh vegetation to expand (Poaceae-Potentilla zone), and fire to increase as fuel limitations decreased. Meadow vegetation was replaced by deciduous woodland species (Ulmus-Fraxinus zone) around 9,900 cal. bp, likely favoured by a decreased impact of fire (limit zones 2-3). Today Ulmus and Fraxinus rarely occur at the upper forest limit, only achieving dominance in low-middle elevation forests (Ketskhoveli, 1971).
F I G U R E 1 Major taxa from the various taxonomic groups at Didachara, southern Georgia. Data are presented in stratigraphical order in both percentages (filled curves) and accumulation rates (bars); x-axes are scaled independently for accumulation-rate data (lower scales: remains cm −2 year −1 ). Ages, depths, lithology and loss-on-ignition (LOI) results are given on the left side; assemblage zones are given on the right. *monolete fern spores; **5. Trinema (dom.); ***Hyalosphenia-Pseudodifflugia. Further details in Table S1 and Figures S3 and S4 0  50  100  150  200  250  300  350  400  450  500  550  600  650  700  750  800  850  900  Neither tree is a major pollen producer (Connor, 2011), so it seems certain that they were canopy dominants in Early Holocene vegetation around Didachara, facilitated by increased temperatures and a lack of competition. Fraxinus was probably out-competed by the cold-and shade-tolerant F. orientalis as it expanded its range around 8,200 cal. bp (Fagus-Ulmus zone). Mixed coniferous-deciduous forests followed around 6,300 cal. bp, with P. orientalis and A. nordmanniana increasingly important (Picea-Abies zone). This association is typical of the mountain forests of Colchis (Western Caucasus) today (Nakhutsrishvili, 2013), although the timing of its Holocene establishment varies from place to place (Connor & Kvavadze, 2008).
Fire activity in the Picea-Abies zone peaked at ca. 4,000, 3,000 and 1,800 cal. bp, indicating disturbances in the forest. A brief episode of deforestation followed at 950 cal. bp (Poaceae-Fagus zone), mainly affecting coniferous forest and accompanied by an increase in anthropogenic pollen indicators (Behre, 1986). Human populations increased in the highlands during the medieval period (Burney & Lang, 1971), with widespread ecological impacts (Connor, 2011).
Since 300 cal. bp coniferous forests have returned. Abies populations have not recovered as successfully as Picea, being more sensitive to human activities (Connor, 2011).
Overall, significant changes in the terrestrial community's structure (assemblage zone boundaries) are closely associated with regionalscale climatic changes recorded in isotopic records (Figure 2).

| Lake/peatland community development since 13,000 cal. bp
Within Didachara's aquatic and wetland vegetation, sedges (Cyperaceae) were dominant from 13,000 to 9,900 cal. bp (Figure 1), suggesting a shallow or fluctuating water-table. Spores of dunginhabiting fungi (i.e. Cercophora and Sporormiella) suggest herbivores were present, perhaps visiting the site as a waterhole. Around 9,900 cal.
Diatom and chrysophyte palaeo-assemblages are recorded from 11,200 to 7,700 cal. bp. An initial succession of benthic taxa (e.g. epiphytic Gomphonema olivaceum and epipelic Fragilariforma nitzschioides) suggests an initially low lake level, corresponding to the final stages of steppe vegetation in the surrounding landscape. A peak in chrysophyte cysts constitutes the next phase (10,960-10,800 cal. bp), perhaps influenced by increased nutrient inputs as the lake filled (Smol, 1985), a supposition corroborated by rapid fluctuations in mineral content ( Figure 1). Aulacoseira nygaardii, a species typical of oligotrophic lakes, peaks shortly after (10,800-10,640 cal. bp), followed by a largely benthic diatom community including aerophilic species (Orthoseira roeseana) and indicative of nutrient enriched, shallow water with limited thermal stratification (Pinnularia-Orthoseira zone). After 9,970 cal. bp, tychoplanktonic A. nygaardii again becomes dominant, with short-lived peaks in Pinnularia microstauron and chrysophytes around 9,300 and 8,600 cal.
bp. This Aulacoseira-Pinnularia zone indicates a higher lake level and corresponds to the Cyperaceae-Pediastrum zone in the wetland vegetation and the expansion of Ulmus-Fraxinus woods around the study site.
Compared to other proxies, midge and mite palaeo-assemblages exhibit more consistency over the period analysed (11,000 cal. bp onwards). Limnophyes, a chironomid often found in semi-aquatic habitats among mosses and macrophytes (Andersen, Saether, Cranston, & Epler, 2013b), is common in all zones except in the uppermost Tanytarsus-Psectrocladius zone (Figure 1). In the earliest zone, cool conditions are inferred until 10,100 cal. bp, based on the abundance of Krenopelopia, Micropsectra radialis-type and other cold-stenothermic taxa (Cranston & Epler, 2013;Pankratova, 1977). This is followed by a phase with warmth-adapted taxa (e.g. Corynoneura scutellata-type, Procladius) that Overall, significant changes in the lake/peatland community prior to 6,000 cal. bp (lake phase) are associated with changes in the terrestrial community (Figure 2). After this time, significant changes relate to shifts within the peatland community that are largely independent of terrestrial community changes.
F I G U R E 3 Testing population dynamics theory with palaeo-population data. The graphs indicate the percentage of major population changes that show an increase or decrease in stability following a change in palaeo-population size. "Predicted" responses accord with population dynamics theory, in contrast to "unpredicted" responses. Hollow histograms show results for the more abundant taxa only (number of observations n given in parentheses). Box plots indicate random walk simulations with population sizes of 500 (representative of local proxies) and 2,000 (representative of upland pollen: Blaauw et al., 2010). Asterisks denote observations that exceed 95% confidence intervals [Colour figure can be viewed at wileyonlinelibrary.com]

| Intrinsic and extrinsic drivers of palaeopopulation dynamics
In the Didachara record, significant population changes were not evenly distributed through time (Figure 2). The most prominent changes in terrestrial populations are associated with rapid changes in precipitation, temperature and seasonality at the Pleistocene-Holocene boundary, c. 11,700 cal. bp (Göktürk et al., 2011;Messager et al., 2013Messager et al., , 2017Wick et al., 2003;Wright et al., 2003). Regional climatic shifts are "slow/large" environmental changes that theory predicts would reduce ecosystem resilience (Seddon, Froyd, Leng, Milne, & Willis, 2011;Seddon et al., 2015). Our observations suggest that these changes impacted upon both terrestrial and lake/peatland communities ( Figure 2). Population-level reactions were particularly strong within the terrestrial community. This is perhaps due to the larger spatial scale represented by the upland pollen proxy and its greater sensitivity to regional-scale climate change.
The close correspondence between terrestrial population changes and regional climatic trends supports the idea that environmental changes are primary drivers of long-term community development Simulations and empirical evidence show that biotic interactions (such as competition) may override climatic controls on species abundance (Brooker, 2006;Miller et al., 2008).
Terrestrialisation (or "macrosuccession": van der Valk, 2012) had major impacts on lake/peatland population dynamics. Populations of midges, mites, testate amoebae and other wetland taxa increased at the transition from shallow lake to peatland, 4,500-4,000 cal. bp ( Figure 2). Chironomid population dynamics subsequently tracked temperature changes and testate amoebae population dynamics followed increasing oligotrophy ( Figure S6). Community dynamics also changed during the mid-late Holocene, with lake/peatland community shifts generally occurring out-of-phase with terrestrial shifts (Figure 2).
Terrestrialisation was related primarily to intrinsic processes of organic matter accumulation within the peatland. Its timing, however, was influenced by extrinsic climatic warming, indicated by increases in thermophilous chironomid assemblages at Didachara (Figure S6), as well as isotopically and palynologically inferred temperatures across the region (Connor, 2011;Wick et al., 2003). Terrestrialisation was therefore intrinsically driven and climatically mediated, a finding that agrees with research into peatland systems elsewhere (Belyea & Baird, 2006;Lavoie et al., 2013).
As in other parts of Eurasia (Seddon et al., 2015), we suggest that "slow/large" processes diminished in the Caucasus during the mid-Holocene ( Figure 2) and that greater resilience developed in the peatland community (i.e. macrophytes, testate amoebae and invertebrates).
This resilience weakened around 1,000 cal. bp as anthropogenic deforestation impacted on the terrestrial community, creating feedbacks onto water-table depth (via tree removal), macrophyte cover (sunlight availability) and trophic status (acidification). Responses across various trophic levels in and around the site are evidence of a trophic cascade triggered by deforestation (Ireland & Booth, 2012).

| Antecedent conditions
A key question in ecology is whether populations, distributions or ranges change in predictable ways (Brooker, 2006). Population dynamics theory predicts that growing populations should subsequently achieve stability, while declining populations will experience greater instability (Channell & Lomolino, 2000). If this were true in every instance, a great number of ecological problems could be instantly solved. Allee effects and other concepts are used to describe exceptions to this theory (Kuparinen et al., 2014;Oliver et al., 2015).
Two-thirds of our observations support the key theoretical predictions of population theory (Figure 3a). The observations are remarkably uniform given that they derive from mostly unrelated taxonomic groups. Random walk simulations produce similar results to the fossil data (Figure 3a), casting doubt on whether the observed differences in palaeo-population response are real or artefactual. The null model has no inherent consideration of species interaction or environmental forcings (Blaauw et al., 2010), factors expected to influence palaeopopulation dynamics at Didachara. While our results support the predictions of population theory, they call for caution in attaching ecological explanations to patterns that may be caused by random variation.
Certain fossil observations fall beyond the bounds of the random simulations (asterisked in Figure 3a). This is particularly true for testate amoebae and wetland pollen, taxonomic groups at Didachara whose histories are closely intertwined. It raises the possibility that community-level interaction influences population dynamics, although taphonomy, life history, niche specialisation, environmental change and interspecific interaction may be equally important.
Several limitations apply to our palaeo-population analyses.
Although population splits are significant, stability changes in the fossil data were not significantly different to those generated by the null model. Generation and response times of testate amoebae and chironomids are many times faster than terrestrial trees. Sampling intervals for short-lived taxa may be inadequate to detect rapid changes in population size and stability. Even generation times of terrestrial trees vary by several orders of magnitude (Walker & Chen, 1987).
It is unlikely that any feasible sampling interval can adequately represent all palaeo-population changes for every taxon. It may be necessary to model empirical data using Bayesian statistics (Clark & Bjørnstad, 2004) to target particular taxa or groups. Such datasets could be used to identify taxa whose histories exhibit early warning F I G U R E 5 Relationship between community-level richness ( Figure 4) and population-level instability (number of independent splits: Figure 2). Timeslices as in Figure (Scheffer et al., 2009). Independent splitting results hint at population instability predicting subsequent population decline among longer lived taxa ( Figure S7), but sampling intervals may be inadequate to address this using the current dataset.
While there is considerable replication within the Didachara record (337 palaeo-population changes), they solely relate to a single core from a single site and are dominated by terrestrial pollen changes.
We encourage replication in other palaeo-records to determine the generalisability of these findings. Questions also remain regarding the degree to which population dynamics can be overridden by random variation, competition and environmental change.
We used trait analysis to gain further insight into population dynamics that diverged from theoretical predictions. Tall canopy trees are disproportionately represented among taxa that conflict with theory-they increased and subsequently experienced greater variability ( Figure S5). Competition between canopy dominants seems the most likely explanation for this pattern (Bennett, 1986), as P. orientalis, A. nordmanniana, Pinus kochiana and F. orientalis all inhabit the same bioclimatic niche (Nakhutsrishvili, 2013).
Low-growing taxa are most frequent among species whose populations have decreased, which reflects the replacement of low-statured Late-glacial steppes and Early Holocene meadows by taller forest vegetation. Species that do not follow the predictions of population dynamics theory, however, tend to be taller ( Figure S5). Additional height may confer competitive advantages (Brooker, 2006), particularly in an increasingly forested environment.
Testate amoebae with smaller apertures are more prevalent among the populations that conflict with theoretical predictions (Figure 3), including abundant taxa such as Difflugia and Hyalosphenia species.
Our results suggest that organism size affects population dynamics (Brooker, 2006;Pelletier, Clutton-Brock, Pemberton, Tuljapurkar, & Coulson, 2007) and that biotic interactions such as competition may play a stronger role in ecosystem dynamics than is generally acknowledged (Jeffers et al., 2015). Wider application of independent splitting to other suitable palaeoecological records is likely to uncover consistent ecological responses for taxa whose life cycles are too long or cryptic to monitor using alternative methods.

| Diversity and stability
Ecological theory states that greater diversity lends stability to communities (Cardinale et al., 2012;Tilman, Isbell, & Cowles, 2014;cf. May, 1972). Upland pollen has the highest overall richness in rarefaction results (Figure 4). Over the long term, upland pollen tends to exhibit relatively gradual transitions between different assemblages ( Figure S6). Chironomids have the lowest overall richness and exhibit more pronounced fluctuations in assemblages ( Figure S6), although fewer significant shifts in assemblage structure (Figure 2). Seen through ordination results, our observations suggest that more diverse communities are inclined towards greater long-term stability, but seen through numbers of statistically significant assemblage shifts, the inverse is true, raising questions about how best to assess long-term stability (see also Loreau & de Mazancourt, 2012). Life histories are also critical: it may be expected that upland vegetation composed of long-lived trees may have a greater degree of ecological inertia than short-lived chironomid communities (Smith, 1965;Von Holle, Delcourt, & Simberloff, 2003).
Richness in the various taxonomic groups has not remained stable through time (Figure 4). Peaks in richness in the terrestrial community are associated with periods in which grasses (Poaceae) played a prominent role (zones 2 and 6; see Figure 1). The lowest richness is associated with forested periods (zones 5 and 7) because of the dominance of competitive, late-successional tree species (e.g. Fagus and Picea: Nakhutsrishvili, 2013). The grassland zones have greater palynological evenness, a characteristic that has been shown to correspond to greater habitat diversity (Feurdean et al., 2013;Matthias, Semmler, & Giesecke, 2015). These zones were also more affected by fires (Figure 1), which are a key disturbance agent contributing to greater landscape diversity and taxonomic richness in Europe (Colombaroli, Beckmann, van der Knaap, Curdy, & Tinner, 2013;Giesecke, Wolters, Jahns, & Brande, 2012) and the Caucasus (Connor, 2011).
Within the peatland community, long-term changes in diversity are apparent in all taxonomic groups. Richness among wetland vegetation indicators rises abruptly around 4,000 cal. bp, coinciding with the onset of terrestrialisation, which likely created a diversity of habitat patches of peatland surface and open water. Diatom richness was highest in the zone with predominantly benthic species (Pinnularia-Orthoseira zone) and corresponds to more minerotrophic conditions and the absence of competing Pediastrum. Among midges (chironomids) and mites, the greatest richness occurs in the most recent zone (Tanytarsus-Psectrocladius), which is likely related to increased temperatures and greater effective moisture. For testate amoebae, constant-sum and accumulation-rate rarefaction results are in close agreement, indicating a steady rise in richness through the last 8,000 years that tracks the inferred expansion of oligotrophic conditions ( Figure 4 and Figure S6).
Population-level instability is apparently linked to higher community diversity, as suggested by positive correlations between diversity estimates and population changes ( Figure 5). Correlations are strongest for testate amoebae (r 2 : 0.38) and wetland vegetation (r 2 : 0.29). These long-term observations appear to support theoretical predictions and previous empirical observations from shorter term ecological studies (Gross et al., 2014;Loreau & de Mazancourt, 2013;Tilman, 1996).
Highly diverse systems experience greater population-level variation after a disturbance due to interspecific competition. The effects of this competition on overall community structure and biomass are to lend stability, either because there is a greater probability that disturbanceresistant species will be present compared to a low-diversity system (Tilman, 1996) or because species responses to disturbance occur at different times and rates (Loreau & de Mazancourt, 2013).
Surprisingly, random walk simulations produce comparable relationships between richness and population change ( Figure 5). The relationship is stronger (r 2 : 0.16) for a simulated population size of 500 compared to 2,000 (r 2 : 0.05), replicating correlations observed in the fossil communities. In communities with a low overall population, increases in richness are likely to be accompanied by significant changes in population sizes as new species migrate into the community. This effect is diluted in larger, more diverse communities in which many populations are already present and interspecific interactions may be more critical.
Our results, while representing conditional support for the richness and stability relationships observed in short-term studies, suggest that there is greater scope for testing ecological theories using long-term ecological data, model simulations and alternative indices of diversity and stability.

| CONCLUSIONS
Population dynamics theory makes important predictions that are notoriously difficult to test using short-term datasets. We analysed palaeoecological data from a peatland ecosystem to provide a long-term perspective on population dynamics. In accordance with predictions, populations of taxa that lived within the peatland tended to respond concurrently to internal changes, especially terrestrialisation. This pattern was not consistent through time, however, with both peatland and terrestrial taxa being impacted by external environmental changes such as major climatic shifts and human impacts at various times. The data show that antecedent population increases led to future population stability in two-thirds of cases. Random variation and inter-species competition emerged as likely explanations for population changes that did not conform to this theoretical prediction. This result highlights the need to integrate intracommunity interactions into population models and confront these with null models. As predicted, taxonomic groups of higher diversity tended to have greater community-level stability and population-level instability compared to low-diversity groups, although questions remain about how to best assess community stability. Future research should aim to expand knowledge of long-term population changes beyond a single site, aiming at replication on a regional scale.