The research was conducted in the metropolitan Poznań area (western Poland). We selected 20 study sites, located, on average, 1.5 km apart (range: 0.5–4 km; median: 1.4 km) to increase spatial independence among sites (Suppl. material 1: fig. S1). Each site represented a small patch of ruderal vegetation within a suburban landscape, in which Solidago was the dominant (or co-dominant) component (cover 30% to nearly 100%, median 60%). At the site level, goldenrod occurred either as clumps or as a continuous stand. The estimated size of the goldenrod patch - defined as the total contiguous area in the vicinity of the study site where Solidago was present, with no goldenrod observed beyond the patch boundary, ranged from 5 to 1600 m² (median ~100 m²). There was no relationship between patch size and the degree of Solidago cover (Pearson r = 0.242, p = 0.291); however, in patches with low cover, Solidago occurred in clumps, whereas at high cover it formed continuous stands (Mann–Whitney test, p = 0.021). The sites also differed in the proportion of impervious surface area (ISA) within a 250-meter buffer (range: 0.00–52.65). At each study site, we established two 2 m² plots: a goldenrod plot within the Solidago patch and a nearby control plot without goldenrod, dominated by diverse non-Solidago vegetation. Because the paired plots were located within the same local landscape, the control plots should be understood as non-Solidago vegetation patches within landscapes already containing S. canadensis, rather than as controls representing uninvaded landscapes. This spatial configuration, together with the 30% to nearly 100% Solidago cover within focal patches and the continued presence of nearby diverse co-flowering vegetation, is what we refer to as an intermediate stage of invasion. Control plots were located within up to 200 m of the goldenrod plots. In the control plots, the co-flowering vegetation comprised 29 plant taxa (including one taxon identified only to family level, Apiaceae). Based on floristic status, all taxa except two were native or archaeophytes; the remaining two taxa were alien. Achillea millefolium was the most frequently recorded species (16 sites), followed by Cirsium arvense (9 sites) and Tanacetum vulgare (8 sites). Apiaceae (identified to family level) occurred at 7 sites, while Berteroa incana, Medicago falcata, and Senecio vulgaris were each recorded at 6 sites. Chenopodium album and Cichorium intybus occurred at 5 sites, and Hypericum perforatum at 4 sites. All remaining taxa were less frequent (≤3 sites: eg., Convolvulus arvensis, Sisymbrium officinale, Arctium lappa, Artemisia vulgaris, Echium vulgare, Erigeron canadensis, Helichrysum arenarium, Lamium purpureum, Potentilla anserina), including several singletons (e.g., Agrimonia eupatoria, Erigeron annuus, Jasione montana, Leonurus cardiaca, Matricaria chamomilla, Origanum vulgare, Papaver rhoeas, and Trifolium arvense).

The research was conducted in August and September across the goldenrod flowering period over two field seasons (2022–2023). Pollinators were sampled in both the goldenrod and control plots using an entomological net, with 20 minutes of active netting per plot. All surveys were carried out between 9 am and 3 pm under weather conditions suitable for bee activity (warm, sunny, no rainfall, and moderate wind). Study sites were visited in a randomized order so that sampling at each site occurred at different times of day. Each site was surveyed three times per season. Pollinators were assigned to three main groups: wild bees (Apoidea), butterflies (Lepidoptera), and hoverflies (Syrphidae). Easily recognizable species, including most bumblebees, were identified alive in the field, while the remaining specimens were transported to the laboratory for identification. All specimens were identified by the authors. Pollinators were identified to species level whenever possible; however, individuals belonging to the subgenus Bombussensu stricto (four species in Poland: Bombus terrestris, B. lucorum, B. cryptarum, and B. magnus) were pooled and recorded as Bombus s.str. because reliable separation of these cryptic species was not possible using the applied identification approach.

Using the free open-source software QGIS 3.34.11 (QGIS.org 2024), we calculated the mean degree (0–100%) of impervious surface area (ISA) in 250 m-radius buffers around sampling points. Dataset Copernicus Land Monitoring Service High Resolution Layer: Imperviousness for the 2018 reference year (Lot 1 – Imperviousness for the EEA-39 area) was sourced from the Copernicus Land Monitoring Service, available at https://land.copernicus.eu.

To assess the association between Canadian goldenrod dominance and pollinator community composition (bees, butterflies and hoverflies), we performed non-metric multidimensional scaling (NMDS) based on Bray–Curtis dissimilarities, using a site-by-species matrix aggregated at the site level. NMDS was chosen because it makes minimal assumptions about the underlying data distribution and is well suited for community data with many zeros and non-linear relationships in multivariate space. Prior to ordination, we removed species recorded only once across sites to reduce noise from extremely rare taxa, leaving 43 species for the analysis, and square-root transformed the community matrix to down-weight highly abundant species while retaining information on relative differences. Differences in community structure among plot types were tested using permutational multivariate analysis of variance (PERMANOVA; 999 permutations) based on Bray–Curtis distances, and homogeneity of multivariate dispersion among habitat types was evaluated with the betadisper function in the vegan package (Oksanen et al. 2022).

To quantify the effects of invasive Canadian goldenrod (Solidago canadensis), urbanization, and honey bee abundance on wild bee communities, we fitted mixed-effects models in R Core Team 2025 using the glmmTMB package (Brooks et al. 2017). We analysed three response variables for wild bees: abundance, species richness, and alpha diversity (Shannon-Wiener diversity index). Fixed effects included plot type (categorical: goldenrod absent vs. present), the proportion of impervious surface area (ISA) within a 250 m buffer, and Apis mellifera abundance (number of individuals). Each model also included two interaction terms (A. mellifera abundance × plot type; ISA × plot type) to examine whether the relationships between wild bee metrics and honey bee abundance or urbanization differed between goldenrod and control plots. In a separate model, we tested the association of plot type, ISA, and their interaction with A. mellifera abundance.

For butterflies (Lepidoptera) and hoverflies (Syrphidae), we modelled abundance and species richness using the same predictor structure (plot type, ISA, A. mellifera abundance, and the two interaction terms). Shannon diversity was not analysed for these groups because they were sparsely represented, with many samples containing either no species or only a single species, resulting in insufficient variation in Shannon diversity for robust statistical modelling.

To examine whether continuous variation in site-level goldenrod cover was associated with pollinator metrics, we fitted additional mixed-effects models using goldenrod cover (%) as a predictor. For wild bees, butterflies and hoverflies, fixed effects included site-level goldenrod cover, ISA, and A. mellifera abundance. For A. mellifera, we modelled abundance as a function of goldenrod cover and ISA.

To account for spatial autocorrelation among sites, we included the first axis of a principal coordinates of neighbour matrices (PCNM) analysis (PCNM axis 1) as a covariate in each model. We further tested model residuals for spatial autocorrelation using Moran’s I and found no significant effects, indicating that residual spatial structure was adequately controlled (Suppl. material 1: table S7). We used negative binomial generalized linear mixed models (GLMMs) to handle over-dispersed count data for wild bee abundance, species richness, and Apis mellifera abundance, as preliminary checks indicated that Poisson models substantially underestimated the variance. Alpha diversity, which was approximately continuous and symmetrically distributed, was modeled using a linear mixed-effects model (LMM). To reflect the hierarchical structure of the data and site-level heterogeneity, all models included random intercepts for year (two seasons) and site ID. We evaluated model assumptions and diagnostics by testing GLMM overdispersion, assessing LMM residual normality and homoscedasticity using DHARMa package (Hartig 2024), and checking multicollinearity with variance inflation factors (VIFs). We made all visualizations using the ggplot2 package (Wickham 2016).

In total, 1245 insect pollinators were sampled from 20 study sites across two years (40 site-year observations). We collected 1076 individuals representing 45 bee species, 93 individuals from 12 butterfly species, and 76 individuals from 6 hoverfly species. In total, 63 pollinator species (across bees, butterflies, and hoverflies) were recorded. Of these, 45 pollinator species (741 specimens) were recorded in the goldenrod plots, whereas 44 pollinator species (504 specimens) were recorded in the control plots. Apis mellifera Linnaeus, 1758 was the most abundant species (66% of all individuals; 577 individuals in the goldenrod plots, and 254 in the control plots). The next most abundant species were Pieris rapae (Linnaeus, 1758) (4.2% of all individuals) and Eristalis tenax (Linnaeus, 1758) (3.8% of all individuals). Among wild bees, the most abundant taxa were Bombus lapidarius (Linnaeus, 1758) (14% of wild bees), Bombus pascuorum (Scopoli, 1763) (12%), Bombus s.str. (10.5%), and Colletes similis Schenck, 1853 (7%).

Apart from common species, we recorded six taxa listed on the Polish Red List (Banaszak 2002; Buszko and Nowacki 2002): Bombus veteranus (Fabricius, 1793) classified as Vulnerable (VU); Lycaena dispar (Haworth, 1802) as Least Concern (LC); and four species classified as Data Deficient (DD): Colletes similis Schenck, 1853, Hylaeus styriacus Förster, 1871, Sphecodes cristatus Hagens, 1882, and Sphecodes spinulosus Hagens, 1875. We also recorded Halictus scabiosae (Rossi, 1790), a species first reported from Poland in 2020, i.e., shortly before our study. Species composition differed between the two plot types. Twenty-six species occurred in both plot types, whereas 19 were recorded only in the goldenrod plots and 18 only in the control plots. Consistent with these patterns, NMDS ordination based on Bray-Curtis dissimilarities showed a separation in pollinator community composition between goldenrod and control plots (stress = 0.115). This difference was supported by PERMANOVA (χ²-approx. F(1,38) = 9.37, R² = 0.19, p = 0.001; Fig. 1).

To account for potential spatial autocorrelation, we included the first PCNM axis in all models. This term was generally non-significant, indicating limited residual spatial structure in most responses, and is therefore only discussed where it approached or reached significance.

Across pollinator groups, goldenrod-dominated plots were associated with a consistent reorganization of local pollinator use rather than a uniform decline or increase across taxa. Community composition differed between plot types, while responses in abundance and richness were guild-specific: honey bees and hoverflies were more abundant in goldenrod plots, wild bee abundance was lower, although wild bee richness and alpha diversity were unchanged, and butterflies showed lower abundance and richness. ISA showed no consistent main effects across pollinator groups. A significant plot type × ISA interaction was detected only for wild bee richness, whereas the remaining interactions were not statistically significant. Associations between honey bee abundance and wild bee abundance, richness and alpha diversity also differed between plot types, being positive in control plots but negative in goldenrod plots. Finally, site-level goldenrod cover was positively associated only with honey bee abundance and not with the metrics of wild bees, butterflies or hoverflies.

During the flowering peak of Solidago canadensis, pollinator responses differed markedly among guilds between goldenrod-dominated and paired non-Solidago plots. Honey bees were strongly aggregated in goldenrod plots and increased with site-level goldenrod cover, whereas wild bee abundance and both butterfly abundance and richness were lower than in nearby co-flowering control vegetation; hoverflies showed the opposite pattern, with higher abundance in goldenrod plots. Importantly, differences in abundance were not mirrored by comparable differences in species richness: richness was largely unchanged between plot types for wild bees and hoverflies, indicating differences in local activity density and patch use rather than a broad loss or gain of taxa over the sampling period, while butterflies showed lower abundance and richness in goldenrod plots. At the same time, the clear separation in NMDS ordination indicates that pollinator assemblages differed between goldenrod-dominated and nearby non-Solidago vegetation. This suggests that the difference between plot types was not limited to total abundance but also involved a reorganisation of assemblage composition by favouring some taxa while disadvantaging others, leading to species turnover between the two contrasting floral patch types even where overall richness remained similar. Together, these outcomes illustrate a common invasion trade-off in urban systems: mass-flowering invaders may form prominent late-season flowering patches, while their local dominance is associated with reduced floral heterogeneity and guild-specific differences in pollinator assemblage structure (Moroń et al. 2009; Stout and Tiedeken 2017; Kovács-Hostyánszki et al. 2022). However, we did not directly quantify floral resource abundance, nectar production, or pollen quantity and quality in either plot type. Consequently, the observed differences in pollinator visitation cannot be attributed to resource density or quality alone and may also reflect floral diversity, vegetation structure, species-specific preferences, or combinations of these factors. We therefore treat resource concentration and reward quality as plausible explanations rather than mechanisms demonstrated by our data.

Because the paired plots occurred within the same local landscapes and were within the foraging range of the recorded pollinators, differences in captured individuals should primarily be interpreted as differences in local visitation, activity density, and patch use during the sampling period. They do not necessarily indicate corresponding differences in population abundance, demographic performance, or landscape-scale habitat suitability.

Importantly, our control plots represented not only the absence of Solidago, but also the presence of a functionally diverse set of co-flowering ruderal plants that can support different pollinator needs. Several of the most frequent taxa (e.g., Achillea millefolium, Cirsium arvense, Tanacetum vulgare and diverse Apiaceae) are widely visited by adult pollinators and can provide accessible nectar and pollen for multiple groups (Buszko and Masłowski 2008; Bot and Meutter 2023; Nylund 2023). Beyond adult foraging, the control vegetation likely offered larval host resources that are irrelevant to honey bees and most hoverflies but can be crucial for butterflies. This contrast in resource diversity is consistent with the lower use of goldenrod plots by wild bees and butterflies and the greater aggregation of honey bees and hoverflies in mass-flowering patches.

The greater floral and structural heterogeneity of control plots may also have favoured bee taxa associated with particular pollen sources or more diverse floral assemblages, although species-level patterns are discussed in greater detail later in the Discussion and should be interpreted cautiously because many taxa were represented by few individuals.

Overall, the paired comparison contrasted not simply the presence and absence of S. canadensis, but a mass-flowering patch with a more floristically and structurally heterogeneous vegetation patch offering a broader range of potential resources.

The strong association between honey bee abundance and both goldenrod presence and cover is consistent with the resource-concentration effect expected for mass-flowering resources. Honey bees are known to exploit dense flowering stands in a manner comparable to mass-flowering crops (Westphal et al. 2003), and late-season blooms may be especially attractive when most native and cultivated plants have already finished flowering. The high nectar and pollen rewards reported for Solidago flowers (Kabuce 2006) provide a plausible explanation for this aggregation, and goldenrod can form a major component of honey bee forage during this period (Lenda et al. 2021).

However, attraction to goldenrod does not necessarily imply net colony-level benefits. Under laboratory conditions, Lenda et al. (2021) found reduced survival in bees maintained on a goldenrod-honey diet, particularly when bees were unable to defecate and had no access to alternative food sources. More broadly, the same study suggested that goldenrod dominance may reduce the availability of alternative floral resources earlier in the season by displacing more diverse flowering communities. If so, invaded landscapes could generate seasonal forage imbalances, with an abundant late-season resource pulse but reduced resource continuity across the rest of the active beekeeping season. Because our sampling was restricted to the goldenrod flowering period, our study captures only this late-season pulse and cannot assess earlier-season resource displacement or colony-level consequences.

Honey bee abundance also retained a significant spatial structure (PCNM) after accounting for goldenrod cover and ISA, indicating that variation in abundance was not randomly distributed across the study area. Because we did not quantify management-related variables, this pattern may reflect unmeasured drivers typical of managed pollinators, such as apiary distribution, colony density, or the long foraging ranges of honey bees, all of which can weaken relationships between plot-scale habitat characteristics and local abundance (Ropars et al. 2019).

Our results also align with studies showing that late-season mass-flowering resources can attract honey bees disproportionately, while non-crop or more diverse habitats may remain relatively more important for wild bees (Ropars et al. 2019; Riggi et al. 2024). A similar process may occur in invaded urban patches: dense goldenrod stands may concentrate honey bee foraging locally; but whether this increases total resource use, shifts foraging away from alternative flowers, or alters competitive pressure on other pollinators remains unresolved.

Wild bees did visit goldenrod flowers, but their overall abundance was lower in goldenrod plots than in paired non-Solidago vegetation, consistent with earlier reports that wild bees may use goldenrod while remaining less numerous than honey bees in invaded patches (Moroń et al. 2009). In contrast, wild bee richness and alpha diversity showed no clear main effect of plot type, indicating that goldenrod plots supported broadly similar numbers of wild bee taxa, but fewer individuals overall. In other words, rather than causing immediate local species loss, goldenrod may reduce the extent to which wild bees use these floristically simplified late-season patches. More broadly, although late-flowering plants can provide forage during seasonal scarcity, floral quantity does not necessarily translate into resource quality or habitat suitability for wild bees, and conservation outcomes are typically strongest where diverse native plant assemblages are maintained (Scheper et al. 2013).

Species-level patterns provide tentative support for this interpretation. Assemblages in both plot types were dominated by widespread polylectic generalists, but some compositional differences were observed. Species occurring in both goldenrod and control plots were mainly ecologically flexible taxa, whereas species recorded exclusively in S. canadensis plots included several small- to medium-sized, ground-nesting halictids, such as Lasioglossum albipes, L. morio and L. pauxillum, together with several Sphecodes species. As Sphecodes are cleptoparasites of ground-nesting bees, particularly halictids, their presence may reflect local host availability or nesting conditions rather than floral resource use alone. By contrast, control plots included a few additional taxa with relatively narrower floral preference. Osmia spinulosa is broadly oligolectic species, collecting pollen from multiple taxa within the Asteraceae subfamilies Carduoideae (e.g. Carduus, Cirsium) and Asteroideae (e.g. Achillea millefolium, Anthemis) (Wood et al. 2016; Müller 2018). Colletes daviesanus is likewise associated primarily with Asteraceae, showing a strong preference for Tanacetum vulgare, Achillea millefolium, Leucanthemum vulgare, and related taxa, which may shape its local distribution patterns (Müller and Kuhlmann 2008). However, both species were represented by few individuals, so floral preferences, local habitat structure, and stochastic sampling cannot be disentangled.

A key result was that the associations between honey bee abundance and wild bee abundance, richness, and alpha diversity differed between plot types. These relationships were positive in control plots but negative in goldenrod plots, indicating that local co-occurrence patterns varied with floral context. Such plot-dependent reversals are compatible with differential use of contrasting floral patches. Honey bees may aggregate strongly in dense goldenrod stands, whereas wild bees may make greater use of floristically diverse vegetation. The resulting negative association within goldenrod plots does not by itself demonstrate direct interference, exploitative competition, or displacement. However, evidence for A. mellifera - wild bee competition remains mixed, and reported negative effects often involve displacement, that is, changes in where and on which flowers wild bees forage, rather than consistent declines in abundance or richness (Ropars et al. 2019).

Because our data come from flower-based netting during the goldenrod bloom, they most directly reflect visitation patterns and local aggregation. As such, they may not capture longer-term population consequences, which can occur without immediate changes in worker abundance (Westphal et al. 2008; Banaszak et al. 2014; Boyer et al. 2020). Our results are therefore best interpreted as evidence of context-dependent co-occurrence and foraging patterns around a late-season mass-flowering invader, rather than as direct evidence of competition. Alternative explanations - including unmeasured microhabitat differences, variation in the availability of alternative flowers, and spatial structuring linked to apiary placement - cannot be excluded.

Butterflies showed consistently lower local visitation in goldenrod-dominated plots, with both lower abundance and lower species richness than in paired non-Solidago vegetation. The species recorded only in control plots, including Pieris brassicae, P. napi, Pontia edusa, Polyommatus icarus, and Issoria lathonia, are not known to be strict adult nectar specialists (Buszko and Masłowski 2008). Their absence from goldenrod plots should therefore not be interpreted simply as avoidance of Solidago flowers, but may reflect the broader plant diversity and structural heterogeneity of the control vegetation. The control flora also included potential larval host plants for some of these taxa, such as Sisymbrium officinale for pierids and Medicago spp. for P. icarus (Buszko and Masłowski 2008), although our flower-visitor sampling does not demonstrate local reproduction.

Previous studies likewise indicate that butterflies make only limited use of Solidago flowers (Ustinova and Lysenkov 2020). Because butterflies depend on diverse plant communities across life stages, including specific larval host plants and a range of nectar sources for adults, they are especially vulnerable to habitat simplification (Thompson and Pellmyr 1991; Renwick and Chew 1994; Dennis et al. 2004). Compared with the mixed ruderal vegetation in control plots, dense, near-monospecific goldenrod stands may provide fewer of the complementary resources and structural features used by butterflies across different life stages. In addition, such stands may alter vegetation structure and constrain movement through the habitat, acting as partial barriers or creating microhabitat conditions that further reduce butterfly use of goldenrod-dominated patches (Wratten et al. 2003; Moroń et al. 2019). The lower abundance and richness observed in goldenrod plots are therefore consistent with reduced use of floristically and structurally simplified vegetation by butterfly taxa with more complex or specialized ecological requirements. However, because floral resource abundance, host-plant availability, and microhabitat conditions were not quantified directly, these mechanisms remain plausible explanations rather than effects demonstrated by our data.

Hoverflies increased in abundance in goldenrod plots, consistent with earlier studies reporting positive syrphid responses to goldenrod during peak flowering (De Groot et al. 2007; Fenesi et al. 2015). Adult hoverflies are typically flexible foragers (Kaczorowska 2004; Biesmeijer et al. 2006), and many larvae develop as predators, in aquatic habitats, or in decaying organic matter rather than on living plants. This may reduce their dependence on local plant community diversity relative to taxa that rely more directly on host plants or diverse floral assemblages across life stages. Hoverflies are also recognized as important visitors of S. canadensis within its native range (Werner et al. 1980), suggesting that the floral traits of this species are well suited to syrphid foraging (Ustinova and Lysenkov 2020). Behavioural flexibility may further reduce direct overlap with honey bee foraging, as temporal niche partitioning under high honey bee abundance has been reported, with syrphids foraging earlier in the day (Jeavons et al. 2020). Together, these traits provide a plausible explanation for the higher hoverfly abundance observed in goldenrod plots, which is consistent with flexible use of mass-flowering resources. At the same time, hoverfly species richness did not differ between plot types suggesting that the higher abundance in goldenrod plots likely reflected greater representation of a subset of common species rather than the addition of new taxa.

It is also worth noting that we recorded several taxa of conservation interest, although most were represented by single individuals. The Red List butterfly Lycaena dispar (one individual) was captured directly on goldenrod inflorescences. Among bees, Bombus veteranus occurred only in control plots, whereas Hylaeus styriacus, Sphecodes cristatus, and S. spinulosus were each recorded only in goldenrod plots (singletons), and Colletes similis occurred in both plot types. We also recorded one individual of Halictus scabiosae, a species first reported from Poland only two years before our study, likewise collected on goldenrod flowers. Given the very low counts, however, these records are best interpreted as evidence of occasional late-season flower use rather than as an indication that goldenrod supports stable or ecologically meaningful populations of rare taxa.

In our models, impervious surface area (ISA) showed no consistent main effect across pollinator groups. A significant plot type × ISA interaction was detected for wild bee richness, whereas interactions for wild bee abundance and butterfly metrics were not statistically significant. Overall, the present results provide limited evidence that urbanization systematically modified pollinator responses to goldenrod-dominated vegetation, and local floral patch type explained the observed patterns more consistently than ISA.

Previous work suggests that pollinator responses to goldenrod may be nonlinear, with proposed cover thresholds for maintaining pollinator communities and associated services (Moroń et al. 2019). In our study, site-level goldenrod cover did not predict wild bee, butterfly, or hoverfly metrics, whereas honey bee abundance was the only response that increased with goldenrod cover. This pattern may reflect the generally high goldenrod cover at our sites (median: 60%), which compresses the invasion gradient and may limit our ability to detect threshold-like responses. Temporal scope is also likely to matter: Moroń et al. (2019) sampled throughout the growing season, whereas our study focused on the goldenrod flowering period and therefore captured only the late-season phase of pollinator activity. Full-season sampling across a broader gradient of invasion intensity will be needed to determine whether cover thresholds emerge earlier in the season and whether late-season resource pulses translate into net community-level benefits or costs.

Overall, our study shows that local dominance of S. canadensis is associated with shifts in pollinator assemblages in suburban habitats, even when flowering stands were used as prominent late-season flowering patches. Compared with nearby control plots containing more diverse co-flowering vegetation, goldenrod-dominated plots supported higher local abundance of honey bees and hoverflies during the flowering peak, but lower butterfly abundance and richness, lower wild bee abundance, and a distinct shift in pollinator community composition. Together, these patterns indicate that a mass-flowering invader can redistribute local pollinator use of floral patches by concentrating some managed and ecologically flexible visitors, while other guilds are less strongly represented in goldenrod-dominated vegetation. The combination of altered community composition with relatively stable richness in some guilds is consistent with community reassembly occurring before detectable local taxon loss. The context-dependent reversal in honey bee - wild bee associations further suggests that flowering S. canadensis can alter local patterns of pollinator co-occurrence around concentrated floral resources.

Importantly, our study captures an intermediate stage of invasion, in which S. canadensis was already locally dominant within patches but had not yet replaced the surrounding mosaic of other co-flowering vegetation. This highlights that pollinator patch use and assemblage structure may already shift before complete landscape saturation by the invader, while opportunities for coexistence among contrasting floral patches still remain. Because S. canadensis may continue to expand if left unmanaged, retaining non-dominated patches with more diverse late-season vegetation may become increasingly important in suburban landscapes where open habitats are limited. Although our control plots were not true landscape-level controls outside pollinator foraging range from goldenrod, they nevertheless provide a realistic comparison between nearby goldenrod-dominated and non-dominated vegetation. From a management perspective, our findings support limiting the local spread and long-term dominance of S. canadensis and retaining or restoring floristically diverse late-season vegetation. Maintaining this habitat mosaic may help reduce the risk of longer-term biotic homogenization that could arise if heterogeneous vegetation were progressively replaced by dominant invasive stands, while sustaining broader pollinator diversity (McKinney and Lockwood 1999; Olden and Rooney 2006; Carvalheiro et al. 2013).