Habitat heterogeneity enables spatial and temporal coexistence of native and invasive macrophytes in shallow lake landscapes

Macrophyte invasive alien species (IAS) fitness is often hypothesised to be associated with beneficial environmental conditions (environmental matching) or species‐poor communities. However, positive correlations between macrophyte IAS abundance and native plant richness can also arise, due to habitat heterogeneity (defined here as variation in abiotic and native biotic conditions over space and time). We analysed survey and palaeoecological data for macrophytes in satellite lakes along the Upper Lough Erne (ULE) system (Northern Ireland, UK), covering a gradient of eutrophication and connectivity to partition how environmental conditions, macrophyte diversity and habitat heterogeneity explained the abundance of Elodea canadensis, a widely distributed non‐native macrophyte in Europe. E. canadensis abundance positively correlated with macrophyte richness at both the within‐ and between‐lake scales indicating coexistence of native and invasive species over time. E. canadensis was also more prolific in highly connected and macrophyte‐rich lakes, but sparser in the more eutrophic‐isolated ones. Partial boosted regression trees revealed that in eutrophic‐isolated lakes, E. canadensis abundances correlated with water clarity (negatively), plant diversity (positively), and plant cover (negatively) whereas in diverse‐connected lakes, beta diversity (both positively and negatively) related to most greatly E. canadensis abundance. Dense macrophyte cover and unfavourable environmental conditions thus appear to confer invasibility resistance and sufficient habitat heterogeneity to mask any single effect of native biodiversity or environmental matching in controlling E. canadensis abundance. Therefore, in shallow lake landscapes, habitat heterogeneity variously enables the coexistence of native macrophytes and E. canadensis, reducing the often‐described homogenisation effects of invasive macrophytes.


| INTRODUCTION
Aquatic invasive alien species (IAS), species that have successfully been introduced, established, and spread beyond their native range, are an increasing concern for management and conservation in freshwater ecosystems owing to their potential to cause severe ecological and economic damage (Cuthbert et al., 2021;Strayer, 2010). The ability of aquatic IAS to prosper could depend on resource availability and the physical-chemical environment (Shea & Chesson, 2002;Strayer, 2010). Accordingly, species-and functionally rich communities may limit the abundances of aquatic IAS by reducing access to resources (Levine, Adler, & Yelenik, 2004), whilst beneficial environmental conditions (henceforth referred to as environmental matching) at recipient locations may enhance IAS abundances, irrespective of local native biota (Mack et al., 2000). However, as processes affecting biotic interactions and species distributions are scale-dependent (McGill, 2010), variations in biotic and abiotic conditions within and between sites over space and time (i.e., habitat heterogeneity) may facilitate species coexistence, regardless of their native status (Melbourne et al., 2007). Therefore, habitat heterogeneity could simultaneously increase invasion probability whilst reducing aquatic IAS dominance across a landscape by promoting coexistence in space and time (Clark, Johnston, & Leung, 2013). Landscapes may thus include species that are extirpated at some sites but present at others through spatial and/or temporal storage effects, provided there is sufficient regional connectivity and spatiotemporal habitat heterogeneity (Melbourne et al., 2007).
In lake landscapes, habitat heterogeneity and the distribution of aquatic IAS can be determined by, and correlated to the hydrological network (Salgado et al., 2019a). For example, macrophyte and invertebrate lake communities connected through the hydrological network may be influenced by repeated colonisation events via mass effects, while local environmental factors may dictate community structure through species sorting according to habitat optima in more isolated lakes (Capers, Selsky, & Bugbee, 2010;Padial et al., 2014). Although, there have been attempts to quantify how spatial autocorrelation affects IAS dynamics (Melbourne et al., 2007), disentangling the simultaneous effects of abiotic factors, native biodiversity, and spatially structured dynamics on the abundances of IAS in nature has proved challenging (Nunez-Mir et al., 2017). Available evidence suggests that temporal variation in environmental stress factors and dispersal-related mechanisms promotes co-existence of native and non-native species in freshwater systems (Clark et al., 2013), which could result in a positive relationship between native diversity and aquatic IAS abundance at any one time (Capers, Selsky, Bugbee, & White, 2007). Lake landscapes comprising multiple interconnected lakes that span environmental gradients offer thus, the possibility of more explicitly quantifying how native biodiversity, abiotic factors, and spatial autocorrelation affects aquatic AIS abundances and hence scope to disentangle factors that contribute to regional coexistence.
Among aquatic IAS, macrophyte species are one of the foremost invaders of inland waters across the globe (Bolpagni, 2021). The ecological effects of macrophyte IAS can range from reduced freshwater biodiversity to elevated plant biomass production and altered biogeochemical cycles (Bolpagni, 2021). In Europe, Canadian waterweed (Elodea canadensis Michx.) is considered amongst the most widespread non-native plant species (Hussner, 2012;Nentwig, Bacher, Kumschick, Pyšek, & Vilà, 2018). It was first recorded in Great Britain in 1836 (Simpson, 1984). Thereafter it spread rapidly, reaching the maximum extent of its distribution in Great Britain and Ireland by the middle of the twentieth century (Simpson, 1984). The rapid colonization and spread of this species are commonly attributed to a high capacity for vegetative propagation and tolerance of a broad range of physical-chemical conditions, including low illumination, enabling growth at a wide range of water depths and under eutrophicationinduced shade (Zehnsdorf, Hussner, Eismann, Rönicke, & Melzer, 2015). Once established E. canadensis can quickly replace native submerged macrophytes by forming a dense, closed canopy (Zehnsdorf et al., 2015). Indeed, the propensity for encountering E. canadensis in meso-eutrophic isolated temperate lakes has promoted the view that its spread and dominance across Britain and Ireland is attributable to environmental matching (O'Hare, Gunn, Chapman, Dudley, & Purse, 2012). However, few studies have investigated the role of habitat heterogeneity on E. canadensis abundance variation over space and time, and the extent to which biotic and/or abiotic factors explain its abundances across invaded landscapes.
Here we examine the drivers of E. canadensis abundance in space and time in the Upper Lough Erne (ULE) system, Northern Ireland, a network of interconnected, shallow (<5 m depth), meso-eutrophic (total phosphorus [TP] range = 29-383 μg/L; total nitrogen [TN] range = 0.22-2.25 mg/L), and macrophyte-rich (>40 submerged and floating-leaved species; Table S1) lakes. Present-day and historical data from surveys and sediment core analyses were used to address the following questions: 1. Does habitat heterogeneity (including spatial autocorrelation) per se promote E. canadensis coexistence with native macrophyte communities in space and time (decades-centuries)?
2. Which are the main biotic and abiotic factors that contribute to regional coexistence?
3. To what extent does the variation of biotic and/or abiotic factors contribute to variation in E. canadensis abundance? 2 | METHODS

| Site description
The ULE system is composed of a large (surface area 3,450 ha), generally shallow (mean water depth 2.3 m) central lake (ULE), fed by the River Erne, which is linked by winter floods and various channels and tributaries, to a network of over 50 small (<40 ha), shallow satellite lakes (Loughs; Figure 1). E. canadensis is thought to have colonized the ULE system in the 1880s (Simpson, 1984). Records from the Botanical Society of Britain and Ireland (BSBI) and more recent macrophyte monitoring programmes indicate presence of E. canadensis in ULE and its satellite lakes from the 1950s (Table 1). The zebra mussel (Dreissena polymorpha Pallas) also invaded this system in the 1990s, resulting in strong phytoplankton biomass reductions and increases in water transparency in the central ULE (Minchin, Maguire, & Rosell, 2003).

| Macrophyte sampling
Macrophytes were sampled in four basins within the central ULE and in 13 satellite lakes representing a gradient of nutrient-enrichment, zebra mussel occurrence and hydrological connectivity to the central ULE ( Figure 1). The lakes were grouped into three categories according to Salgado et al. (2019a). Group 1 included the central ULE and lakes directly connected to it through the River Erne (Castle and Derrykerrib) or via tributaries (Doo and Mill). These lakes are mesoeutrophic (TP = 55.1 ± 11.3 μg/L) with clear waters (Secchi depth = 222 ± 47 cm), are macrophyte rich (M = 17.4 ± 2.7), and zebra mussels commonly occur. Group 2 lakes are connected to the central ULE by flows through intermediate lakes and associated tributaries (Killymackan, Cornabrass, and Kilturk). The lakes are eutrophic (TP = 136 ± 54.4 μg/L; Secchi depth = 182.3 ± 55 cm), macrophyte rich (mean = 18.7 ± 4.6), and have sparse occurrences of zebra mussels. Group 3 lakes (Head, Digh, Derryhowlaght, and Gole) are more isolated than Group 2 lakes due to intervening small hills, woodlands, and roads ( Figure 1). These lakes are highly eutrophic (TP = 176.8 ± 89.3 μg/L) with turbid waters (Secchi depth = 113.3 ± 79.3; zebra mussels rarely occur) and macrophyte richness is low (M = 9 ± 3.5).
Macrophyte sampling was undertaken in 1 m 2 units, approximating to the plant neighbourhood scale (i.e., where individual native plants may compete with E. canadensis). The lake percentage volume infested by macrophytes (PVI) method of Canfield and Jones (1984) was used to characterize the distributions and abundances of native  (Table 1) and to assess E. canadensis abundance patterns at the lake scale, percentage of sample occupancy of E. canadensis at each lake was also calculated by dividing the number of sampling points at which E. canadensis was observed by the total number of sampling points within the lake X 100.

| Environmental predictors
Our previous studies of the ULE system demonstrate that macrophyte communities are primarily structured by lake water transparency, which is negatively related to nutrient concentrations (TP and TN) and chlorophyll-a and positively related to zebra mussel occurrence

| Plant macrofossil data
Previously published plant macrofossil abundance data derived from dated sediment cores were used to represent macrophyte community changes over the last c.120 years (Salgado et al., 2019b). Cores were taken from Castle Lough and the Trannish area of ULE (lake Group 1), from Cornabrass and Killymackan (Group 2), and from Gole and Head (Group 3). E. canadensis remains preserve poorly in lake sediments and so we inferred temporal changes in its abundance indirectly from a recent macrophyte study in the ULE system (Salgado et al., 2019a) and from available historical monitoring data (Table 1). Salgado et al. (2019a) showed that macrophyte assemblages now found in the central ULE or closely connected lakes (e.g., Castle and Derrykerrib) are similar to those characterised in sediment cores prior to eutrophication (i.e., pre-1950s). However, macrophytes currently found in the more isolated eutrophic sites (e.g., Gole and Head) resembled those characteristics of sediment cores post-eutrophication (i.e., post-1960).

| Statistical analysis
Two diversity-related measures were previously shown to influence macrophyte IAS fitness (Capers et al., 2007) Feld, Segurado, & Gutiérrez-Cánovas, 2016). BRT constitutes a machine-learning method that combines classical regression tree analysis with boosting (Elith et al., 2008). BRT was ideal for our study as it can accommodate collinear data (e.g., latitude and longitude) and han- To reduce any spatial autocorrelation in the data arising due to the underlying hydrological network and to evaluate whether the importance of diversity and abiotic predictors in explaining E. canadensis abundance shifted with degree of lake connectivity and eutrophication, we ran independent pBRTs for each lake group using the "dismo" (Hijmans, Phillips, Leathwick, Elith, & Hijmans, 2017), and RFAs were then used to assess the extent to which diversity predictors explain E. canadensis abundances through time. Like BRTs, RFA is suited to analysing non-linear relationships by fitting several models (regression trees) to bootstrapped data subsets with the advantage of handling datasets with a low number of observations and predictors, that is, our palaeo-data (Feld, Segurado, & Gutiérrez-Cánovas, 2016). RFAs were run using the function rfsrc of the package "randomForestSRC" (Ishwaran & Kogalur, 2016).

| RESULTS
Except for Gole Lough, E. canadensis was encountered in all study sites (Figure 1). The highest mean sample occupancy of E. canadensis per lake was in Group 1 lakes (48%), followed by Group 2 lakes (32%), and Group 3 lakes (28%). Current native macrophyte species richness and E. canadensis sample occupancies were positively correlated among lakes (r = .44; p = .08) (Figure 2a). This positive relationship became significantly stronger (r = .74; p < .01) after excluding Kilturk Lake, which was identified as an outlier by having 19 native macrophytes species but an E. canadensis occupancy of just 9% (Figures 2a   and S2).
Positive and significant correlations between current native macrophyte species richness and E. canadensis abundance were similarly pBRTs showed that the importance of the pure abiotic fraction in explaining E. canadensis abundance variation declined from 30% in Group 1 to 13% in Group 3 (Figure 4a). Within the pure abiotic fraction, latitude explained almost half of the variation (48%) in Group 1, but just 9% in Group 3 (Figure 4c). Water clarity in Group 3 explained 68% of the abundance variation compared with only 24% in Group 1 (24%). Longitude effects remained relatively constant across the three lake groups, explaining 24% of the abiotic fraction in Group 1, 21% in Group 2, and 23% in Group 3.
The importance of the pure diversity fraction in explaining E. canadensis abundance variation in the pBRTs almost doubled from 17% in Group 1 to 31% in Groups 2 and 3 (Figure 4a). Native beta diversity emerged as the most important predictor, accounting for almost two thirds of the pure diversity fraction in Groups 1 and 2 (61% and 62%, respectively), and 40% of the variation in Group 3 ( Figure 4b). The importance of native Shannon diversity showed an increasing trend from relatively low levels of explained variation in Group 1 (8%) to nearly fourfold higher (31%) in Group 3. The explanatory importance of floating plant cover increased from Group 1 (10%) to Group 3 (16%), whilst overall plant cover was most influential in

| DISCUSSION
The results show that the abundance of E. canadensis in the ULE system is not determined by the single effects of native diversity or beneficial environmental conditions but rather by habitat heterogeneity.
For instance, E. canadensis is commonly reported to dominate over native submerged species once well established, and to exert strong negative ecosystem engineering effects (Zehnsdorf et al., 2015). Conditions considered to favour E. canadensis include nutrients availability, suitable carbon sources, and silty substrates (Zehnsdorf et al., 2015). Such conditions characterised our study sites (Salgado et al., 2019a) and would support an environmental-matching control effect. However, while E. canadensis was present in all but one site, abundances in lakes were generally low to moderate. Moreover, the positive correlation between E. canadensis abundance and native speciose communities over space and time further suggests that native species richness alone does not confer invasibility resistance (Capers et al., 2007). In turn, it indicates sufficient habitat heterogeneity over time within and among-lakes to enable coexistence of native and nonnative macrophytes (Clark et al., 2013) which differ from the oftendescribed macrophyte IAS homogenisation impacts (Muthukrishnan & Larkin, 2020).
Water clarity and a nested spatial dependence between E. canadensis abundance and the location of sampling points in each lake were highlighted as key controlling abiotic factors. Macrophyte species tend to be distributed at certain depths rather than occur across an entire lake (He et al., 2019). Thus, the nested dependence of E. canadensis abundance, likely reflects each lake specific water depth profile and the associated zonation patterns of light availability.
The influence of water clarity on E. canadensis abundance further diminished with lake isolation to the central ULE, which is in line with our previous studies of the ULE system showing that zebra mussels and eutrophication have spread unevenly across the lakes (Salgado et al., 2019a).
Among diversity predictors, much of the spatial and temporal variation in E. canadensis abundance was related to native beta diversity and plant cover; plant community attributes that have been found to better capture macrophyte ecological change in human-dominated landscapes than species richness alone (Fu et al., 2019). The relation of beta diversity and E. canadensis abundance was greatest in the more connected lakes (Group 1), which likely reflects a pronounced influence of source sink dynamics that fosters heterogeneous plant associations under the less stressful environmental conditions (high water clarity and lower nutrients) of this group of lakes (Salgado, Sayer, Brooks, Davidson, Goldsmith, 2018;. In turn, high cover may limit E. canadensis abundance. This pattern was particularly revealed among those eutrophic lakes with a lower degree of connectivity to the central ULE (Groups 2 and 3), and which are dominated by a few  F I G U R E 4 (a) Partial boosted regression trees (pBRT) on the threestudy lake groups showing the pure and shared proportions of explained and unexplained variance in E. canadensis abundance by abiotic metrics and native biodiversity metrics; (b) diversity fraction of the pBRT analysis showing the unique contribution of Shannon diversity, plant cover, submerged plant cover, floating plant cover and beta diversity on E. canadensis abundance variation across the three-study lake groups; (c) abiotic fraction of the pBRT analysis showing the unique contribution of water clarity, longitude and latitude on E. canadensis abundance variation across the threestudy lake groups. Only percentages values of explained variation >10% are presented in the plots [Color figure can be viewed at wileyonlinelibrary.com] submerged and floating species (Salgado, Sayer, Brooks, Davidson, Goldsmith, et al., 2018;. Negative correlations between plant cover and macrophyte IAS were similarly found by Capers et al. (2007) across a series of lakes in Connecticut (USA) and in grassland and dessert plants by Cleland et al. (2004), both suggesting that IAS success is diminished by high resource capture by the resident communities. It is thus likely that different processes may control the establishment   (Clark et al., 2013). However, low water clarity and high native plant cover could lower resource availability for E. canadensis, thereby reducing opportunities to proliferate (Cleland et al., 2004).

| Inferring the history of E. canadensis in the ULE system
Palaeolimnological data reveal that at the time that E. canadensis colonised the ULE system in the late 1800s, macrophyte communities were diverse like those currently observed in Group 1 lakes (Figure 7; Salgado et al., 2019a). Simpson (1984) reported a cycle of local colonization by E. canadensis involving establishment over a three-year period and a subsequent rapid increase in abundance. Given the extensive interconnection by winter flooding E. canadensis probably spread rapidly to many sites. Following its widespread establishment, E. canadensis possibly therefore persisted at moderate abundances for a long period, co-existing with a reported high diversity of other submerged species across the lakes (Salgado, Sayer, Brooks, Davidson, Goldsmith, et al., 2018;. Subsequently, post-1950s, paleoecological data indicate gradual biotic changes associated with more eutrophic conditions that intensified after the 1980s (Battarbee, 1986) but with differential local nutrient concentrations influencing biota (Salgado et al., 2019a).

| Limitations
Reconstructing E. canadensis abundance over time based on survey and sediment core data may have limitations. For instance, some species are likely to have been unrecorded and detection in sediment cores may be biased by preservation issues and under-representation of rare or distantly located macrophyte taxa (Clarke et al., 2014). Our assessments of native macrophyte richness variation over space and time probably favour the more abundant taxa. Unique lake histories could have also introduced some discrepancies between the observed current E. canadensis lake occupancies and the inferred past abundance (Salgado et al., 2019a). Nevertheless, analyses of both palaeo-and Predicting future trajectories of E. canadensis distribution and abundance in the ULE system, and in the United Kingdom in general is, however, challenging. E. canadensis has spread in the British Isles by asexual growth, most likely from male clones (Simpson, 1984). It is therefore possible that conditions (e.g., disease) may eventually challenge the persistence of clonal populations due to the lack of genetic variation. Furthermore, with globalization, unexpected and novel invasion dynamics are more probable (Pyšek et al., 2020). Meanwhile, the ULE system is suggested to be declining through advancing eutrophication, which, if unabated, will eventually override positive regional species storage effects (Salgado, Sayer, Brooks, Davidson, Goldsmith, et al., 2018;. In addition, the sibling invasive species, Elodea nuttallii Planch., is rapidly spreading across the lower part of the ULE system and outcompeting E. canadensis under high nutrient conditions (Kelly, Harrod, Maggs, & Reid, 2015). Quantifying the dynamics of these two invasive species at both landscape and temporal scales is critical, therefore, if invasion processes are to be better understood.

ACKNOWLEDGEMENTS
We thank the Natural History Museum, London, for supporting field- Biological Association. We thank the University of Nottingham and the UKRI-GCRF Living Deltas Hub for supporting J.S. as a postdoctoral researcher and Universidad Cat olica de Colombia for supporting J.S. research. We thank CIRCE under the AU ideas programme for supporting TD contribution. We thank the Lake BESS project (Natural Environment Research Council grant, NE/K015486/1) for funding sediment dating and paleoecological analysis of the Gole lake core and for supporting AB. We thank NIEA for provision of water chemistry data for the central lake (Upper Lough Erne), many landowners for site access and hospitality, Charlotte Hall, Stephen Brooks and Peter Hammond for fieldwork assistance and Laura Petetti for provision of data from the ULET2 core and for fieldwork assistance.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study area openly available