Rheoreaction impacts dispersal of fish larvae in restored rivers

Abstract Connectivity of nurseries and spawning habitats for young of the year life stage is essential for successful recruitment of fish populations and therefore provides a key indicator for river restoration measures. Models for dispersal offer the potential to draw conclusions regarding restoration scenarios and to fill knowledge gaps about possible implications for fish populations. A newly developed rheoreaction‐based correlated random walk model (RCRW), in combination with a three‐dimensional numerical model and a non‐steady‐state particle tracing model, was applied for nase carp larvae (Chondrostoma nasus) before and after a restoration project on the river Danube, Austria. Spatio‐temporal patterns of dispersal of virtual larvae, attached with rheoreactive behaviour, were analysed within both scenarios. In comparison to the heavily modified river reach, the restored reach enabled a greater amount of upstream movement from the release site and showed a generally higher variability of spatio‐temporal distribution patterns. In contrast, estimated total settlement of rheoreactive larvae was substantially higher for the situation prior to the restoration measure. By comparing model results with a previously field experiment it was found that model simulations including rheoreaction as a single behaviour for navigation could not explain the whole pattern of larval dispersal. Therefore it is highly recommended for future studies to develop larval dispersal models by considering other factors (i.e., behaviour, bio‐energetics and environmental factors) of existing and future individual‐based models, which could serve as a tool to analyse the effect of restoration measures for recruitment of riverine fish populations.

Rheoreaction, defined as the fish's behaviour induced by the current (Pavlov, Kostin, Zvezdin, & Ponomareva, 2010), is known to be a driving factor for larval dispersal in rivers. Behavioural aspects of fish larvae were studied recently for riverine (Humphries & King, 2003;Lechner et al., 2014;Schludermann, Tritthart, Humphries, & Keckeis, 2012;Zens, Glas, Tritthart, Habersack, & Keckeis, 2018) and marine YOY species (Leis, 2006;Leis & Carson-Ewart, 1997), showing that they considerably effect habitat connectivity. Lechner et al. (2014) encountered enhanced drift, settlement and habitat connectivity of YOY fish for a near-natural shoreline, when compared to a heavily modified one. Schludermann et al. (2012) identified a significantly higher amount of recaptured settled larvae remaining upstream of a release site compared to the proximate areas downstream. Reichard, Jurajda, and Smith (2004) found larval drift maxima close to the shoreline but nevertheless drift was recorded in the main channel as well. Keckeis, Frankiewicz, and Schiemer (1996) and Loisl, Singer, and Keckeis (2014) determined the highest fish densities on gravel bars, representative for a typical natural hydro-morphological condition, during the reproductive period. Lorenz, Stoll, Sundermann, and Haase (2013) detected significantly higher densities of YOY fish due to restoration measures. On the other hand Keckeis (2014) found contradictory short-term effects of a restoration measure in the Danube River on the sublittoral assemblage as well as on the inshore areas (larvae). Although the abundance of larvae decreased the abundance of the sublittoral fish community increased immediately after implementing the restoration measure. In this context, White, Gerken, Paukert, and Makinster (2010) and Eick and Thiel (2013) suggest that river engineering structures may improve habitats for YOY fish, but a broader view considering the whole life cycle of fish is needed.
The nase carp (Chondrostoma nasus), a rheophilic cyprinid widespread in European rivers (Lelek, 1987), is often used as a target species in fishecology studies (Keckeis et al., 1996;Ovidio & Philippart, 2008;Pichon, Tales, Gorges, Baudry, & Boët, 2016). The females spawn demersal eggs which developed within the gravel substrate. After hatching the larvae exhibit a benthic lifestyle before they swim up and disperse upstream or downstream aiming at finding suitable nursery habitats.
In contrast to marine areas, analyses of the connectivity between hatcheries and nurseries of YOY fish within inshore habitats in rivers by means of analytical or numerical tools are lacking so far. Due to the complexity of spatio-temporal dispersal processes, such tools provide advantages to separate the factors and cues and to draw conclusions on various scenarios of restoration measures prior to their implementation in the field. In this study, a rheoreaction-based larval model, developed in an experimental flume (Glas et al., 2017), was compared to dispersal patterns observed in the field (Lechner et al., 2014). Moreover, the influence of rheoreaction on dispersal of early life stages was investigated for two different scenarios, a heavily modified shore and a restructured, near natural gravel bar (before-after comparison).

| Study site
The study was conducted in the national park "Nationalpark Donauauen, http://www.donauauen.at" at a river reach within the free flowing section of the Austrian Danube east of Vienna-near the village Witzelsdorf-between river-km 1891.0 and 1894.0 ( Figure 1).
Mean discharge at the study site is Q = 1930 m 3 /s and low discharge (94% probability of exceedance) is Q = 980 m 3 /s. In the scope of an integrative restoration measure, the hydro-geomorphological characteristics were adjusted at the left shore to improve ecological and navigational demands (Klasz, Schmalfuß, & Schlögl, 2008). The scenarios before (2007) and after (2011) the restoration measure, conducted from 2009 to 2010, were considered in this study ( Figure 1). The groyne layout was adapted from a former orthogonal layout to an attracting layout. Groyne heights, ranging between a crest height equals to the water level at mean flow conditions, were reduced to a crest elevation equal to low flow water levels (+0.30 m). Groyne spacing of approximately one times the average groyne length was increased to a spacing twice the average groyne length. In order to establish a bypass flow along the shoreline at low flow, groyne roots were lowered to a water level below the low flow water levels. This adaption, in combination with a lowered crest elevation of an upstream guiding wall, increased the proportion of discharge in the groyne field and along the shoreline during mean flow by 6% (Habersack, 2013). The rip rap along the shoreline was removed and thus a near-natural gravel bar was formed as a consequence of the bank erosion, the increased flow along the shoreline and the new attracting groyne layout.  Details of the particle tracing model are presented in Tritthart, Liedermann, and Habersack (2009). In terms of the unsteady particle tracing model, flow fields were calculated for a number of steadystate conditions sufficient to cover the desired hydrograph in a first step. Then, while running the particle tracing model, representative flow fields along the hydrograph were taken into account to calculate particle trajectories. Further information on the unsteady particle tracing approach is available in Tritthart, Gmeiner, Liedermann, and Habersack (2019).

| RCRW model based on Glas et al.
This 2D model approach was developed in the framework of a flume study, considering rheoreaction as a driving factor for dispersal of nase carp larvae. With the help of a raster based analysis, observations were categorized after Pavlov (1994) to derive the following movement patterns: "active downstream" (orientation equal to the flow and speed was greater than the flow velocity), "active-passive" (orientation against the flow and speed was less than the flow velocity) and "passive" (random orientation and speed was equal to flow velocity). A fourth one ("active upstream": orientation against the flow and speed is higher than flow), introduced by Zens et al. (2018), was considered as well. The RCRW model generates those movement patterns as well as durations within certain movement patterns (as derived from observations as well) over the simulation time. Horizontal swimming, based on the biased and correlated random walk model (Codling et al., 2004), was altered in terms of the inherited bias.
To reproduce Pavlov's types of movement patterns, the preferred movement direction in the RCRW model was changed to be in accordance with the local flow direction. An additional user-defined angle was introduced to shift the preferred movement direction towards the shoreline. Furthermore, horizontal swimming speed was derived from mean flow velocity and movement pattern. Further details of the model approach are given in Glas et al. (2017).
In this study, the field experiment was reconstructed numerically

| Restored geomorphological situation (2011)
Trajectories of the RCRW model (Figure 3(a) and (b)) were assigned to one of four typical path types (Table 1) With respect to temporal variability, it was found that dispersing virtual larvae reached the exit of the observed area 1.5 days after release ( Figure 4). Therefore, larvae can be related to types A, B and C, as outlined in the distribution of RR (%) in GF2 and GF3 within the same period (Figure 4). Repeated circular movements of virtual larvae at the shoreline in GF1-increasing retention temporarily-were observed for types B and C, comprising upstream movements along the shore and downstream dispersal in areas with higher flow velocities further away from the shoreline. This was in accordance with the flume experiment (Glas et al., 2017). Furthermore, higher fish densities on gravel bars (Keckeis et al., 1996;Loisl et al., 2014) were likely due to circular movements. Passive particles left the observed area substantially later than virtual larvae (path type B or C), as local flow velocities along the passive trajectories were notably smaller. Peaks of exit events occurred 1.6 and 3.7 days after release (Figure 4). It was obvious that the peaks were related to the instantaneous increase of the hydrograph. and GF3 (Figure 4) suggests that most of the larva were drifting passively. However, 0.3% of larvae were located at the shorelines along GF3 (Figure 3(b)) 7 days after release. Finally, 8 days after release, 99.6% of released virtual larvae were washed out of the study area.
Sparse settlement of larvae could be detected along the shoreline of the island facing the main channel.
As a consequence of increased lateral dispersal of virtual rheoreactive larvae, the chance for them reaching areas with higher flow velocities and thus being dispersed to the exit downstream, were increased. The peak in passive wash-out of virtual larvae during increased discharge (6-8 days after release) was not reflected by the findings of Lechner et al. (2018) in the field, where drift rates seemed to be insensitive to discharge on the gravel bar. This contradiction might be due to the assumption of "passive" transport in the RCRW model, for flow situations when flow velocity exceeded U ≥ 0.225 m/s (as no data from the flume experiments was available for this case).

| Situation prior to restoration (2007)
In 2007, virtual larvae showed low dispersal distances (11.6 ± 0.2 m) during the first day after release (low flow, Q = 1,024 m 3 /s). Dispersal activity was restricted to a small pool, shaped by the higher crest elevations of groynes, which was eventually connected to the river at ris-

| Rheoreactive virtual larvae versus observed larvae (2011)
In terms of the reconstruction of the mark-recapture experiment, the modelled drift rate of rheoreactive larvae was overestimated at nets applied during the first day (Table 2) when compared to the observed data. The opposite holds true for the following days, as no larvae were found in any of the virtual nets. Furthermore, modelled drift was observed at SS1 and SS3 (1 day after release) only, whereas observed drift occurred at all SS. Moreover, recaptured larvae from PAS (Lechner et al., 2014), representing larval settlement at inshore habitats, showed larvae at inshore areas along the whole stretch of the

| CONCLUSIONS
Rheoreaction as a key behaviour for larval dispersal (Pavlov et al., 2010) was investigated for an integrative restoration project on the river Danube by means of a recently developed and validated RCRW model. Analytical methods for assessing the complex spatiotemporal processes of numerical larval drift and dispersal were developed and presented. In comparison to the heavily modified river reach, the restored, near natural reach enabled a higher amount of upstream dispersal and a generally higher variability in terms of the complex spatio-temporal distributions of larvae. In contrast, total settlement of rheoreactive larvae was substantially higher at the situation prior to the restoration measure. A possible explanation can be found in the theory of river engineering structures improving habitat conditions for YOY fish (White et al., 2010) for specific hydrographs. However, this may also raise questions in terms of habitat connectivity on heavily modified rivers. Also Keckeis (2014) found a contradictory short-term effect of a restoration measure for nase carp larvae in the river Danube. However, a comparison of numerical larval drift and settlement with field-based observations (Lechner et al., 2014) suggests that rheoreaction is important but not the only behaviour determining dispersal of larvae. Obviously, virtual larvae attached with rheoreaction behaviour were able to withstand the flow and increase inshore retention, but larvae were dispersed rapidly, when flow conditions locally increased. In comparison to passive particles, virtual rheoreactive larvae frequently changed the lateral position in relation to the shoreline and therefore were more often exposed to higher local flow velocities. Therefore, aiming for the development of an IBM T A B L E 2 Averaged number of larval drift (L4, inshore release) per SS and day after release (RR Ind./ 1000m 3 filtered water) for observed (Lechner et al., 2014)  for riverine fish larvae for future investigations, the following is suggested: (a) inclusion of a vertical swimming behaviour (Vikebø et al., 2007)  (d) implementation of variable instead of constant model parameters with regard to specific biotic or abiotic aspects; (e) inclusion of other settlement cues, as suggested by Lechner, Keckeis, and Humphries (2016), for example, acoustic or visual signals (Codling et al., 2004;Staaterman et al., 2012) or hydraulic gradients, indicating shore proximity; (f) determination and inclusion of the mode of drift entrance (Lechner et al., 2016), characterized by "active," "coincidental" and "passive" modes; (g) investigation and inclusion of a potential effect of larvae hiding in proximity to the gravel (Holzapfel & Hauer, 2013) or the boundary layer (Nikora, 2010); (h) inclusion of further aspects affecting dispersal (e.g., growth, feeding, mortality, predation, vessel-induced waves).
As a consequence, RCRW and IBM have the potential to serve as an important tool to assess river restoration measures in terms of habitat connectivity of YOY fishes, aiming to improve plannings of future restoration measures and to increase knowledge relating to some hydro-morphological conditions.

DATA AVAILABILITY STATEMENT
Data sharing is not applicable to this article as no new data were created or analyzed in this study.