Extensive green roofs (EGR) are a type of urban ecosystem that offer a range of ecosystem services (Oberndorfer et al., 2007; Kowarik, 2011). One of the ecosystem services provided by EGRs is the support of urban native biodiversity, through the implementation of diversified native plant communities based on the analogous habitat hypothesis (Lundholm and Richardson, 2010; Rivière et al., 2022). The analogous habitats hypothesis proposes that species which are adapted to similar abiotic conditions in natural habitats can also be successful on green roofs (Lundholm, 2006) and can serve as a starting point for selecting species. Green roofs provide solutions to develop urban grassland and herbaceous ecosystem networks in cities (Sutton et al., 2012; Williams et al., 2014; Thuring and Grant, 2015). However, harsh environmental conditions on extensive green roofs (e.g., high level of solar radiation, shallow substrate, poor water retention capacity) associated with future climate evolution can threaten the success of these native plant communities, and associated ecosystem services, when the functioning and dynamic of EGR ecosystems are not well-understood.

Green roof plant communities are dynamic. When initiated by seeding on bare substrate, the EGR plant community is marked by changes in plant composition over time as this situation relates to primary ecological succession conditions (Brown and Lundholm, 2015). The seeded plant community evolves through time in accordance to seeded species response to environmental conditions at the roof scale: roof height, slope, exposure, substrate depth and composition (Madre et al., 2014; Brown and Lundholm, 2015). EGR are also characterized by local microenvironmental variations impacting the composition and dynamic of the plant community (Heim and Lundholm, 2014; Roulston et al., 2020; Bradbury, 2021; Rivière et al., 2022). In addition to the response of seeded species to environmental conditions, a key driver of EGR plant community evolution is spontaneous plant colonization with non-seeded plant species (Thuring and Dunnett, 2019; Vanstockem et al., 2019). Weedy or unwanted species colonizing spontaneously EGR are identified as a limit to EGR acceptance and maintenance (Vanstockem et al., 2019). Spontaneous colonization by plant species is often linked to presence of bare soil or gaps in the vegetation cover due to the extreme abiotic growing conditions on EGR (Köhler, 2006; Sutton and Lambrinos, 2015).

Soil seedbank is a key component of plant community dynamics (Thompson et al., 1997; Vandvik et al., 2016). It influences the vegetation dynamic in the short and long term (Basto et al., 2018). The seedbank can provide information on species that may have formerly existed in the vegetation, species that represent the pool for the vegetation succession, and, represents the source of regenerative potential for the restoration of altered plant communities. Seedbanks on EGR [being biodiversity reservoirs (Vandvik et al., 2016)] may be crucial as an insurance for maintaining plant communities during periods of extreme conditions (e.g., extreme drought or frost) and can potentially offer a basis for restoration for vegetation gaps. Despite its importance in understanding vegetation response to extreme EGR conditions, the study of seedbank from extensive green roof (Vanstockem et al., 2018) or even urban areas (Albrecht et al., 2011) has received little attention. In an explorative study of seedbanks on 109 EGRs in temperate Europe, Vanstockem et al. (2018) confirm that EGRs do develop seedbanks, dominated by weedy plant species and exhibiting a strong discrepancy with vegetation composition.

To promote native biodiversity, EGR should be considered analogous to natural habitats (Lundholm, 2006). The harsh environmental constraints of EGR are similar to abiotic conditions encountered on species rich dry grasslands and rocky environments in temperate Western Europe (high solar radiation, low soil depth (<20 cm) and high variability in soil water retention capacity). Western Europeans dry grasslands are composed of isolated climax herbaceous habitats on most harsher environmental conditions and, for the largest surfaces, seral semi-natural habitats originating from centuries of extensive grazing and mowing. Rivière et al. (2022), demonstrated experimentally that extensive green roofs can act as analogous habitats for native flora of European dry calcareous grasslands (71% of native seeded grassland species established on an experimental green roof). The most successful species were common dry grassland species such as grass species, succulent species (sedum sp.), Echium vulgare L. and Papaver argemone L. However, success of native grassland plant species, plant community composition and structure were demonstrated to vary through time and to depend on microenvironmental conditions at the roof scale, with large surface of bare soils left after 3 years in the most constrained conditions (low substrate depth and high solar exposition). Decrease in plant cover was also measured in years with severe drought. The contribution of spontaneous ruderal species to the plant communities increased over time. Studies of seedbank in natural situations have shown that calcareous grassland species are mainly perennial stress tolerant species presenting mostly transient seedbank that stay viable in the soil for a very limited period of time and/or are absent from the seedbank (Dutoit and Alard, 1996; Willems and Bik, 1998; Bisteau and Mahy, 2005). A lack of well-developed permanent seedbank of target seeded species can be a threat for long term maintenance of analogous dry grassland vegetation of EGR facing colonization by spontaneous ruderal species.

In this study, we analyzed the seedbank contribution to the plant community of an extensive green roof analog to dry calcareous grassland after 4 years of development in order to understand the potential contribution of seedbank to the resilience of vegetation to harsh EGR conditions and to colonization by spontaneous species. The specific questions asked are: (i) what are the main seedbank characteristics and composition of an analogous urban grassland developed on an extensive green roof, (ii) what is the level of similarity between seedbank and vegetation, (iii) how is the seedbank and abundance of spontaneous species related to vegetation gap and microenvironmental variations on EGR.

Increasing attention has been paid to plant community dynamics on extensive green roof in ecological literature (Roulston et al., 2020; Bradbury, 2021; Rivière et al., 2022). Change in plant community over time is key to EGR design and their acceptability (Thuring and Dunnett, 2019; Vanstockem et al., 2019). It is also a major determinant of EGR capacity to sustain native biodiversity through development of analog habitat (Lundholm, 2006; Rivière et al., 2022). In this study, we examined the development of seedbank in relation to the installation of an analogous dry grassland on an experimental EGR. In contrast to our recent knowledge increase on above ground vegetation patterns and processes on green roofs (Madre et al., 2014; Brown and Lundholm, 2015), seedbank development and its role for green roof vegetation has been seldom explored directly (Vanstockem et al., 2018). The study of seedbank is important to understand its contribution to the resilience of vegetation to harsh EGR conditions. Specifically, a lack of well-developed permanent seedbank of target seeded species can be a threat for long term maintenance of analogous dry grassland vegetation of EGR facing colonization by spontaneous ruderal species.

Our study of the standing plant community evolution on an experimental EGR indicates that green roofs work as an analogous habitat supporting seeded species from native dry grasslands (Rivière et al., 2022). However, the evolution over time of plant community on the EGR diverges from the initial seeding by the absence of a set of seeded grassland species and the increase in spontaneous species cover over time. The species richness of spontaneous species tends to increase to the same level as species richness of seeded species (12 vs. 16) in 4 years. Development of a high richness of spontaneous species on experimental green roofs over a short time period has been repeatedly reported. For example, Dunnett et al. (2008) in Sheffield, UK, identified 35 colonizing species for 15 planted species, Olly et al. (2011) on experimental plots in Birmingham, UK, identified 29 spontaneous species for 17 seeded species and Bates et al. (2013) identified 31 and 37 colonizing species on two brown roofs in the same city. All the spontaneous species observed on the studied EGR are anemochorous except C. album, Geranium sp, P. oleracea and Vicia sativa L. Kleyer et al. (2008). Those latter species may have been imported to the roof by birds or by people visiting the roof. This suggests that most spontaneous species reach our EGR by wind, a pattern reported by Madre et al. (2014). We cannot exclude that some species might have been introduced via seed-contaminated substrate, however, we think that this is negligible as there was very low spontaneous species development on the first year of the experience.

Using a “space for time” approach, Vanstockem et al. (2018) demonstrated the presence of seedbanks on established extensive green roofs aged of 19 years. In the current study, for the first time, we experimentally demonstrate the accumulation of a seedbank on an experimental green roof over a short period of time (4 years) from its settlement from bare substrate. A total of 18 species were found in the seedbank. Mean seedbank SLI indicated a transient global seedbank, but seed accumulation in the deepest substrate layers suggested ongoing constitution of a permanent seedbank, mainly due to spontaneous species with permanent seedbanks, such as C. album and P. oleracea.

Despite a set of common species between the seedbank and the vegetation, and a trend for plant community to become more similar to the seedbank through time, in depth comparison reveal divergence between both, with spontaneous ruderal species dominating the seedbank (mainly P. oleracea) when considering seedbank density, and seeded grassland species dominating the vegetation when considering the species cover. High dissimilarity between the composition of species of the seedbank and the vegetation has been measured in established EGR and the majority of species in EGR seedbanks are spontaneous colonizers (Vanstockem et al., 2018).

Only a small proportion (9 out of 24) of seeded grassland species are detected in the seedbank with mostly lower seed densities as compared to the original seeding, indicating that a seed reserve has not accumulated over time. Limited development of grassland seedbank is due to the non-development of some seeded species (eight species) and to the absence of seedbank for a set of species (eight) present in the vegetation.

Comparison with the seedbank of (semi-)natural systems with comparable environmental characteristics [habitat template sensu Lundholm (2006)], such as calcareous grasslands on shallow soils, can provide useful insights on EGR seedbanks. Many authors showed that species dominating calcareous grassland vegetation were mostly absent from the seedbank, and when present were not abundant (Davies and Waite, 1998; Bisteau and Mahy, 2005; Bossuyt et al., 2006). Almost all calcareous grassland species produce short-lived (transient) seeds that stay viable in the soil for a limited time period. Grassland species invest more resources in individual survival than in seeds production (Eriksson and Eriksson, 1997).

Spontaneous species observed on the experimental EGR are mainly ruderal and for the majority (60%) have accumulated a seedbank despite low cover in the above ground vegetation, to the exception of P. oleracea, which was one of the dominant species of the vegetation. Vanstockem et al. (2018) also reported a dominance of ruderal species in seedbank of established EGR. Ruderal species allocate more resources to producing a large number of seeds rather than developing their cover (Grime, 1988; Šerá and Šerý, 2004). Ruderal spontaneous species are also present in seedbank composition of natural dry grassland but have a limited development in the vegetation (Bisteau and Mahy, 2005). In contrast to Vanstockem et al. (2018) who observed limited presence in the seedbank species pool in comparison to the vegetation, the two most abundant phanerophyte species in the seedbank of our EGR (Buddleja davidii Franch., Betula sp) were almost absent from the vegetation, indicating that seed rain may be an agent responsible for seed accumulation in the seedbank (Davies and Waite, 1998). The presence of B. davidii in the seed bank may pose a potential problem for future vegetation development on the roof as it has a very high invasive potential in disturbed, open and dry environments (Tallent-Halsell and Watt, 2009; Monty et al., 2019), even though it is currently absent from the established vegetation in permanent plots.

We demonstrate that after 4 years of development, both standing vegetation and seedbank species composition were influenced by the variation of environmental factors: substrate depth and sun exposure. The substrate depth is known to be an important factor structuring green roof vegetation both for seeded/planted species and for colonizing species (Durhman et al., 2007; Köhler and Poll, 2010; Madre et al., 2014). In more stressful environment on our EGR (low substrate depth, high sun exposure) vegetation cover was limited with large proportion of bare substrate (Rivière et al., 2022). Vegetation gaps are an important component of seeded EGR, during the vegetation development phase and along the life-cycle of EGR when extreme environmental conditions and abiotic conditions can locally affect vegetation (Vanstockem et al., 2019). Colonizing plants can develop on green roofs because there is available space due to changes in the planted populations, with a decrease in some of the planted species (Dunnett et al., 2008). Gaps in vegetation can impact seedbank characteristics and the development of vegetation by providing an opportunity for colonization by spontaneous and ruderal species (Vanstockem et al., 2018). However, in our study, we do not find a relationship between space left vacant by grassland vegetation and spontaneous species development. This can result from extreme environmental conditions affecting in the same way the two categories of species and preventing ruderal spontaneous species to germinate and colonize bare substrate. This lack of gap colonization potential may be reinforced by the fact that seedbank density was correlated to dominant species cover in the vegetation, indicating, that seedbank accumulates in the vicinity of mature individuals growing in the most favorable environment of the green roof. Also, the role of the seedbank in gap regeneration depends on the severity of disturbance which should relocated buried seed to the surface. In undisturbed vegetation, seedlings very rarely originate from the seedbank (Milberg and Persson, 1994; Kalamees and Zobel, 2002).

In conclusion, in contrast to Vandvik et al. (2016) and Vanstockem et al. (2019) suggestion that seedbanks on EGR could serve as a diversity insurance during periods of extreme conditions (e.g., extreme drought or frost) and can therefore potentially offer a basis for restoration of vegetation gaps, we experimentally measured that the maintenance of a diversified analogous grassland on an extensive green roof cannot rely mainly on an accumulated seedbank. Grassland seeded species did not contribute significantly to the seedbank. Seeded community is therefore unstable and vulnerable to sudden and profound changes under extreme hydric stress conditions. In this case, standing vegetation will be dominated by spontaneous species and the habitat developed will no longer be analogous to dry grassland. Maintenance of target grassland vegetation, in response to disturbance by extreme events, will depend mainly on the perennial habit of grassland species and their high tolerance to stress. Punctual seeding intervention may be needed in case of extreme climatic events affecting grassland species survival.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

GM and LR carried out the conception, design of the study, and wrote the first draft of the manuscript. LR and AS carried out the experiment and performed the statistical analysis. All authors contributed to the conception of the manuscript, manuscript revision, read, and approved the submitted version.