Introduction

Globally, freshwater ecosystems have undergone massive changes and the increasing loss of freshwater biodiversity is widespread (Palmer & Ruhí 2019; Reid et al. 2019). Floodplain ecosystems of large river are some of the most heavily exploited and valued ecosystems and also among the most modified and degraded (Bayley 1995; Vörösmarty et al. 2010; Castello & Macedo 2016). Human modifications such as river regulation and land reclamation have greatly changed the inundation regimes such as extent, duration, timing and frequency (Poff et al. 1997, 2006; Olden & Naiman 2010; Lu et al. 2018); and have diminished the benefits of hydrological connectivity for vegetation (Guan et al. 2016), fishes (Fontenot et al. 2001; Li et al. 2020), waterbirds (Xia et al. 2017) and many other aquatic organisms (Kupferberg et al. 2012). Thus, floodplain system restoration programs, such as environmental water allocation (Stewart & Harper 2002), reconnecting floodplain with its rivers (Pander et al. 2015) and returning croplands to wetland (Xu et al. 2018), remain a focus of regional biodiversity conservation.
Understanding the underlying abiotic and biotic assembly mechanisms controlling temporal and spatial community structure and patterns is not only a central issue in ecology (Cottenie 2005; Leibold et al. 2004; Magurran et al., 2018 & 2019; McGill 2003; Ning et al. 2019), but also has practical implications for biodiversity conservation (Economo 2011; McGill et al. 2015; Yurkonis et al. 2005). Many mechanisms have been hypothesized as explaining spatial and temporal variations in community structure, i.e. the number of taxa (richness) and their relative abundance (composition). The current advances in metacommunity theory generally suggest that the two broad types of processes, i.e. deterministic and stochastic, work simultaneously in shaping ecological communities (Chase & Myers 2011; Gaston & Chown 2005; Gravel et al. 2006; Daniel et al. 2019). The deterministic assembly generally refers to ecological niche-based mechanisms, including environmental filtering (e.g., climate, pH, nutrients, and salinity) and biological interactions (e.g., competition and facilitation) (Levi et al. 2019; Silvertown 2004). Considering that all species are ecologically equivalent in the probabilistic sense (i.e. individuals have equal chances of reproduction or death, Hubbell 2001; Rosindell et al. 2012), the stochastic processes, however, concerns ecological processes that generate community patterns indistinguishable from neutral simulations through random birth– death events, probabilistic dispersal, and ecological drift (random walk, either to extinction or monodominance) (Hubbell 2001; Gravel et al. 2006; Ning et al. 2019; Zhou et al. 2014). A range of methods have been used to quantify the relative importance of deterministic and stochastic processes in community assembly (Anderson et al. 2011; Baselga 2010), of which the null model approach is the most widely used (Mori et al. 2015; Mori et al. 2013). Through randomizing the empirical community matrix (Gotelli 2000), null models can detect if and by how much the observed variation in community structure (i.e. β-diversity) deviates from the null expectation given a regional species pool (i.e. gamma diversity) (Chase et al. 2011). Large deviations are suggestive of deterministic processes driving community assembly while close to zero deviations are interpreted as neutral community (Gotelli 2000; Tucker et al. 2016).
Research on the temporal dimensions of community ecology is becoming increasingly urgent in the face of accelerating loss of biodiversity in the Anthropocene (Dirzo et al. 2014; Garcia-Moreno et al. 2014; McGill et al. 2015). The study of temporal dynamics can help to enhance our predictions on the responses of communities to natural environmental fluctuations and anthropogenic disruptions (Collins et al. 2000; Kéfi et al. 2019; Rosset et al. 2017) by revealing how species respond to community-level constraints (Musters et al. 2019) and how species interactions change over time (Gonzalez & Loreau 2009; Hallett et al. 2014). This knowledge has strong conservation implications. For example, identifying keystone communities that disproportionally contribute to the maintenance of regional diversity is critical to conservation planning (Ruhí et al. 2017). Many statistical metrics and analytical tools, such as community stability (Tilman 1999), species turnover and community change rate (Collins et al. 2000; Collins et al. 2008), synchrony (Hallett et al. 2016) and temporal beta-diversity index (Legendre 2019) have been developed to quantify and compare the temporal dynamics.
Over the last two decades, studies on spatial variation in community structure (i.e. spatial beta-diversity) have played a central role in understanding how communities are organized along environmental gradients, and how stochastic and deterministic processes influence the functioning of ecosystems (Jia et al. 2020; Heino et al. 2015; Mori et al. 2018). In comparison, far fewer studies have investigated the temporal dynamics of communities (see Korhonen et al. 2010 and Jones et al. 2017 for examples), limiting our understanding on how communities respond to the widespread and rapid changes in the Anthropocene (Dirzo et al. 2014; Ruhí et al. 2017), which may lead to contradictory predictions (see Gonzalez et al. 2016 and Vellend et al. 2017). This limitation is particularly significant for the highly variable ecosystems, such as floodplains, where communities exhibit high intra- and inter- annual variability in community assembly processes (Heino et al. 2015).
Despite decades of study, quantifying the relative importance of deterministic versus stochastic processes in natural community assembly remains a key challenge to ecologists (Dini-Andreote et al. 2015; Tucker et al. 2016), especially in dynamic species-rich landscapes (Ruhí et al. 2017). In this study, we collected monthly samples of benthic fish communities in 12 sites, of which six are located at highly modifiedPopulus nigra plantations and six are from natural Carexsedges and open waters, over three consecutive years in a large floodplain with high intra-annual water level fluctuations. Our main purpose is to quantify the relative importance of deterministic and stochastic processes on the temporal dynamics of fish community, and to investigate if these processes differ between the modified and natural habitats. In natural habitat, inter- and intra- annual variabilities in exogenous physico-chemical factors rising from natural climatic and environmental events, such as fluctuations of water level, pH, salinity and nutrients, are relatively high in comparison with the modified habitat (Bunn & Arthington 2002). Life histories of native biota, especially the exploitative taxa, are finely tuned to capitalize on the specialized temporal niches associated with this variability (Chase et al. 2011). Consequently, temporal diversity can be promoted through an array of mechanisms, such as migration to exploit resources and escape competition (Shaw 2016), synchronous seasonal reproduction, and seasonal fluctuations in abundance (King et al. 2003). Moreover, habitat modification is likely to create dispersal barriers (Li et al. 2020), which may reduce chance colonization by opportunistic species. Thus, our working hypothesis is that the ecological stochastic processes will prevail in the natural sites where dispersal is not constrained and environmental variability is high; and the deterministic processes will dominate the human modified sites, where dispersal is limited by isolation and environment is relatively stable.