Fish (Spinibarbus Hollandi) Dynamics in Relation to Changing Hydrological Conditions: Physical Modelling, Individual-Based Numerical Modelling, and Case Study

ECOHYDROLOGY Ecohydrol. 6, 586–597 (2013) Published online 16 April 2013 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/eco.1388

Fish (Spinibarbus hollandi) dynamics in relation to changing hydrological conditions: physical modelling, individual-based numerical modelling, and case study

Rui Han,1 Qiuwen Chen,1,2* Koen Blanckaert,1,3 Weiming Li2 and Ruonan Li1 1 RCEES Chinese Academy of Sciences, Beijing, 100085, China 2 China Three Gorges University, Yichang, 443002, China 3 Department of Limnology of Shallow Lakes and Lowland Rivers, Leibniz-Institute of Freshwater Ecology and Inland Fisheries (IGB), Berlin, Germany

ABSTRACT The paper reports the development of an individual-based fish dynamics model, of which the key components are the rules for the movement of individual fish and the definition of the habitat suitability. The distribution of the fish mainly depends on the flow conditions (velocity, depth, substrate) and life cycle of the fish. A major contribution is the refinement of the rules for fish movement, based on laboratory experiments under volitional swimming conditions, which also provided the ranges of preferential velocities and substrate size for the target fish, Spinibarbus hollandi. Moreover, they provided data on the fish trajectories and distribution patterns that allowed for validation of the movement rules. The validated fish dynamics model was applied to investigate the effect of discharge increase during the dry season by means of reservoir operation in the Lijiang River, which was the subject of field investigations in 2007 and 2008. The model results indicated that reservoir operation leads to an increase of fish biomass. According to the fish movement rules, fish cannot always escape from riverbed regions that dry during decreasing discharge events, which causes them to be trapped and die. Reservoir operation decreases the area of dry riverbed and reduces the travel distance for fish to escape from dry regions. Critical advantages of the individual fish model over global models defined on the population level are that they can account for the time that the fish needs to reach a region of suitable habitat and for the spatial pattern of suitable zones and their connectivity. Copyright © 2013 John Wiley & Sons, Ltd.

KEY WORDS ﬁsh behaviour; ﬁsh dynamics; laboratory scaled model; numerical simulation; individual-based model; vector- based movement Received 1 February 2012; Revised 16 February 2013; Accepted 14 March 2013

INTRODUCTION dynamics in relation to physical and environmental parameters (substrate, flow depth, flow velocity, food Fish are an essential component in the river ecosystem and may resources, dissolved oxygen, etc.) predicted by a hydro- be an important economic resource. Therefore, predictions of logical model. Most models do not represent individual fish changes in the fish population in response to natural or behaviour but use global aggregated parameters such as anthropogenic causes, such as climate change, floods, population abundance or biomass. Such models parameterize droughts or hydrological changes induced by reservoirs are individual interactions in a simplified way (Ekeberg 1993, important (Brookshier et al., 1995). The total biomass, spatial Steel et al., 2001, Li et al., 2011a, 2011b). These global distribution and dynamics of the fish are determined by the models merely focus on equilibrium conditions (Crowder fish behaviour (Railsback et al., 1999), which depends on et al., 1992) and are not appropriate for investigating the multiple interrelated physical, physiological and environ- fish dynamics in response to changes in physical or mental processes, as well as interactions between individuals. environmental parameters. These global models cannot, Recently, the progress in computational capacity and forexample,representwhetherornotfish is able to escape spatial data collection (Chen et al., 2010) has facilitated the to shelters during important flood events or avoid being development of a variety of numerical models for fish trapped on dry riverbed during droughts. Moreover, dynamics (Rose et al., 1996). These models predict the fish validation of global models is complicated by the difficulty to measure experimentally the parameters, which are defined at the population level. Most species have mainly been investigated at the individual level (Reed, 1983), *Correspondence to: Qiuwen Chen, RCEES Chinese Academy of Sciences, which favours the use of individual-based models that Beijing, 100085, China. E-mail: [email protected] directly parameterize the behaviour of individual fish

(DeAngelis and Cushman, 1990; Bian, 2003; Humston et al., 3. To apply the validated model for the prediction of the 2004; Li et al., 2010). Calibration and validation of such effect of flow regulation through reservoir operation on models, however, critically rely on experimental data on the Lijiang River in China, in which Spinibarbus individual fish behaviour (DeAngelis and Gross, 1992). At hollandi is the dominant fish species. present, there is a paucity of such experimental data. Experimental studies on fish behaviour, including SPINIBARBUS HOLLANDI swimming orientation and speed, flow cues for migration, preferences for velocity and flow depth, have attracted Spinibarbus hollandi (Osteichthyes, Cypriniformes, increasing interest (Coutant, 2000). Bailey and Batty Cyprinidae, Spinibarbus) is largely present in southern (1983) examined predating behaviour by Aurelia aurita Asia. It is an important aquaculture species in South on early first-feeding stage larvae of the herring Clupea Chinese rivers, such as the Yangtze River, the Pearl River harengus. Webb (2002) investigated the posture, depth and and the Lijiang River. swimming trajectories of various fishes under different Spinibarbus hollandi (S. hollandi) (Figure 1) is charac- conditions of flow perturbation and turbulence. Burrows terized by a blunt and protractile snout, small eyes on the (2001) and Gibson et al. (2002) studied juvenile plaice upper side of the head, a slightly oblique mouth, a posterior Pleuronectes platessa behaviour in relation to depth end of the upper the jaw that reaches the anterior margin of changes. Bégout Anras and Lagardère (2004) investigated the eyes, two pairs of barbels, with the maxillary barbels swimming behaviour of rainbow trout as a function of longer than the mandibular barbels at the corner of the stocking density. Pavlov et al. (1994) and Skorobogatov mouth, an elongated cylindrical body, large and cycloid et al. (1996) investigated effects of turbulence on fish scales, and a complete lateral line. Its dorsal fin origin is in behaviour. These investigations were all performed in front of its pelvic fin origin, its pectoral fin ends are distant laboratory flumes or tanks of simple geometry, which ignore from its pelvic fin origin, its pectoral and pelvic fins are the heterogeneity in physical and environmental conditions situated at the lower side of its body, and it has a forked found in natural rivers. Moreover, the fish behaviour in these caudal fin. experiments was investigated under conditions of forced The life cycle of S. hollandi can be roughly divided into six swimming, which are not representative of conditions stages (Yin, 1998): embryo (1–2 days), larva (3–4days), encountered in natural rivers. Only recently, fish behaviour juvenile (30 days), young (1–2 years), adult (3.8–6years)and has been investigated under volitional swimming conditions. senility (7–8 years). During its life cycle, young S. hollandi’s Castro-Santos (2005) analysed the volitional swimming paired fins are orange and turn to grayish when fish grows up. behaviour of migratory fishes when traversing velocity ThespawningperiodforS. hollandi lasts from early May to barriers. Li et al. (2011b) studied the behaviour of the end of August. During this period, fish spawn in water Spinibarbus hollandi in a laboratory physical model. with slow velocity, and eggs are attached to gravel (Cai et al., The present paper has three objectives: 2007). S. hollandi does not have a migrating behaviour during its life cycle and spawns in its living area. 1. To improve the parameterization of the behaviour of the Most previous research on S. hollandi focussed on its food fish species Spinibarbus hollandi (Figure 1) and to habits and optimal conditions for spawning, while little is acquire data for validation of models for fish dynamics known at present about the fish’s preference with respect to by means of dedicated laboratory experiments under hydrological parameters. S. hollandi is an omnivorous fish volitional swimming conditions, with a broad feeding ability, including algae crustacean and 2. To develop and validate an individual-based model for aquatic insects. S. hollandi is also known to have a preference the dynamics of Spinibarbus hollandi, which is a further for running water with rocky substrate at the middle and refinement of the model reported by Li et al. (2010), bottom layers of rivers.

Figure 1. Spinibarbus hollandi.

Figure 2. D-ended recirculating flume with flat bottom and vertical baffle. Flow depth in the experiments was 1.0 m.

Field observations on the Lijiang River have indicated a length of 43 m and a width that varies from 5 to 13 m. that the flow velocity and the substrate size are the dominant The flume was characterized by a natural heterogeneous hydrological parameters with respect to the behaviour and bathymetry installed in mortar. The flow conditions were dynamics of S. hollandi. Hence, the fish preference in created and controlled by a pump, a line of valves and a relation to these two parameters has been investigated in rectangular weir (Figure 3). two dedicated laboratory experiments under volitional In both flumes, velocities were measured with a SonTek swimming conditions. Acoustic Doppler Velocimeter at the centreline every 0.5 m in streamwise direction. Measurements were made in points situated at the bottom, at mid-depth and at the water surface in LABORATORY EXPERIMENTS the first flume, and at mid-depth and the water surface in the second flume. In each point, the measuring time was 30 s. Methods Ninety healthy two-year-old S. hollandi were selected Laboratory experiments on preferential flow velocity, from the fish captured in the Lijiang River and transported preferential substrate size, and swimming velocities and to the laboratory. Their weight and length were in the range patterns were performed in two laboratory flumes at the 0.300–0.350 kg and 0.26–0.28 m, respectively. The fish Three Gorges University. were then raised with artificial food for 1 month in a The first flume (Figure 2) was a 17.5-m-long and 3-m-wide 17.5 m 3m 1.5 m flume in the laboratory, supplied D-ended fibreglass tank with flat bottom. To create a velocity with fully aerated running water to ensure the water quality gradient under laboratory condition, an artificial vertical conditions. Before experiments, all fish were randomly baffle was attached at the bottom at an angle of 5 with respect divided into three groups, with 30 individuals each. During to the sidewalls. Thus, the flow section S varied around the the experiments, water temperature was controlled around flume, leading to a longitudinal gradient in cross-sectional 21 C, pH was from 6.9 to 7.2 and the dissolved oxygen averaged velocity, U = Q/S,whereQ is the applied discharge was from 5.5 to 6.5 mg/L, similar to the water quality in the and S the area of the cross-section. Flow was recirculating Lijiang River. Prior to experiments, the fish were through the flume. acclimated to the appropriate water environment for 7 days. The second flume (Figure 3) was a 1/60 scale model of a In both flumes, fish behaviour and movement were 2.6-km-long reach of the Jinsha River in China, leading to continuously observed and recorded with underwater

Inflow 13 Flow direction 5 5 m m Substrates Distribution Area 2

4 3 Outflow 1

43m

Figure 3. Physical scale model of a reach of the Jinsha River. Heterogeneous bathymetry installed in mortar, with flow depths varying from 0.5 to 1.2 m. A 6-m-long zone near the flume entrance has been installed with five sequential 1.2-m-long stripes of increasing substrate size. Points 1 to 5 indicate locations where experimental data have been acquired for model validation.

5 Residence time ratio (%)

0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 Velocity (m s-1)

Figure 4. Preferential velocity curve for S. hollandi based on observed residence times in the experiment in the D-ended ﬂume (Figure 2).

Table I. Substrate material and accumulated residence time of S. hollandi in experiments under still-water conditions performed in the physical scale model (Figure 3).

Substrate material according to Lane (1947) Particle diameter (mm) Accumulated residence time (h) coarse sand 1 0.3 Very ﬁne gravel 3 0.4 Very coarse gravel 40 1.0 Very coarse gravel 50 0.8 Mixture of very coarse sand and very coarse gravel 2 (10%) and 45 (90%) 2.2 Smooth mortar – 19.3

infrared cameras. To avoid disturbing the fish, observations with an optimal water depth (Li et al., 2011b) that varies in the were made with short-range telescopes. model between 0.5 and 1.2 m. Near the flume entrance, five transversal stripes of 1.2 m length were installed with Preference to flow velocities different sediment size d,whichcanbeclassified as follows d fi Spinibarbus hollandi’s preferential flow velocities were (Lane 1947): coarse sand ( = 1 mm), very ne gravel (d = 3 mm), very coarse gravel (d = 40 mm), very coarse investigated in the first flume (Figure 2). The water depth fl gravel (d = 50 mm), a mixture of 10% very coarse sand in the ume was set at an optimal depth of 0.5 m (Li et al., (d = 2 mm) and 90% very coarse gravel (d =45mm).Note 2011b), and the recirculating discharge was set at 3 1 that the substrate material in the Lijiang River is heteroge- 0.3 m s , yielding a cross-sectional averaged velocity in neous and consists of grass, sand, gravel and cobbles. The the flume that varied from less than 0.2 m s1 to more than fl bottom in this 6-m-long zone with substrate was attened, 2ms 1, thus providing volitional swimming conditions. and the water depth was 1.0 m. Experiments were performed using the first group of 30 Experiments were performed using the second group of individuals, which could freely swim around the flume. 30 individuals, which could freely swim around the flume. The duration of the experiment was 24 h. The duration of the experiment was 24 h. The underwater A preferential velocity curve for S. hollandi (Figure 4) was cameras continuously recorded the location of the fish around established on the basis of the observed residence time in each the flume, leading to the residence times in each substrate location of the flume of the 30 investigated individuals. S. zone that is summarized in Table I. Fish spend a total of 4.7 h hollandi preferred staying in velocities ranging from 0.3 to in the zone where the bed was covered with sediment, which 0.6 m s 1 (Figure 3). When velocities were higher than corresponds to a time fraction of about 20%. Because the zone 0.7 m s 1, fish residence time declined rapidly. Residence where the bed is covered with sediment only corresponds to became negligible for velocities higher than 1.5 m s 1,and about 12.5% of the total area of the physical model, this very few individuals were able to exceed this limiting velocity. indicates that the fishhaveapreferenceforabottomcovered with sediment. Furthermore, the results clearly indicate Preference to substrate size S. hollandi’s preference for coarser sediment, and mainly Preference to substrate size was investigated in the laboratory the mixture of very coarse sand and very coarse gravel. This physical scale model (Figure 3) under still water conditions, result was confirmed by nonparametric Kruskal–Wallis

analyses (PKW < 0.01). It should be noted that the preference downstream of the sand island, point 4 was close to the for substrate size may depend on the life stage of the downstream boundary, and point 5 was in the middle of the fish; therefore, the aforementioned results only apply for channel at the left side of the sand island. young 2-year-old S. hollandi. Table II reports the magnitude of the velocity measured near the water surface in these four points, as well as the Swimming rates, swimming patterns, fish behaviour and relative residence time in each of these points. residence time To assess and validate the fish model, experiments on swimming rates, swimming patterns, fish behaviour and FISH DYNAMICS MODEL fl residence time were performed in both umes. As aforemen- The fish dynamics are largely driven by the river fi fl tioned, sh were continuously tracked in both umes by hydrology, and primarily by the substrate type, the flow means of underwater cameras, which allows for determining depth and the flow velocities. These parameters are fi fi the sh trajectories, the sh behaviour, the swimming rate and determined by means of a flow model. The fish dynamics fl the residence time in each point of the umes. It should be in response to these hydrological parameters, as well as noted that the swimming rate does not correspond to a real environmental parameters and other important processes in fi instantaneous swimming speed of the sh, but rather to the the lifecycle of the fish are described by the fish dynamics swimming speed averaged over a period. model. The fish do not affect the hydrological parameters. The laboratory experiments in the first flume did provide a range of swimming rates, with an average value of 0.02 m Flow model s1. Obviously, the swimming rate is determined by the environment, and hence differs in laboratory flumes and The hydrological parameters were calculated with the natural rivers. Zhao (1982) and Yin (1998) report swimming open-source modelling program SELFE (Zhang and rates of Cyprinidae, including S. hollandi in natural rivers of Baptista, 2008). The model solves the depth-averaged 30 to 50 m per day. At present, no experimental data is shallow-water equations on an orthogonal curvilinear grid fi available that would allow differentiating the swimming rate by means of a nite-element method and a semi-implicit according to the fish age. Hence, all fish have the same numerical scheme. The model requires discharge at the fl fl swimming rate. in ow boundary and water depth at the out ow boundary In the next section, the swimming rates observed in the as boundary conditions. Moreover, it requires bathymetry fl first laboratory flume will be used for the assessment and and bed roughness as input parameters. The ow model fl fl validation of the fish movement rules by means of provides the ow velocity and ow depth. simulation of the experiment in the second laboratory flume, whereas the swimming rates observed in natural Fish model rivers will be used in the application of the model to the The individual-based fish model represents each individual Lijiang River. fish (Coombs, 1999), as a discrete particle, which position An experiment with the third group of 30 individuals was is tracked in a Lagrangian way. The position is determined performed in the second flume to acquire experimental data by the movement of the fish relative to the flow. The fish for independent model validation. The discharge in this movement and behaviour depend on the life cycle of the experiment was 0.7 m3 s1, and the flow depth was in the fish (Coombs, 1999), the hydrological parameters and range 0.5 to 1.2 m because of the heterogeneous natural environmental parameters, such as the availability of food bathymetry. Attention was focussed on the five representative resources. These parameters have to be provided as input to points in the flume indicated in Figure 3: point 1 was near the the fish model. The model provides the fish distribution right bank, points 2 and 3 were in the middle of the model pattern by means of a time ratio of fish presence in each

Table II. Data in five points in the physical scale model indicated in Figure 3. The relative residence times in the five points add up to 100% by definition.

Velocity magnitude at Relative residence time Relative residence time the water surface U – experimental value – model prediction Monitored point (m/s) (%) (%)

1 1.42 19.35 20.34 2 0.21 22.58 26.44 3 0.29 25.81 22.04 4 0.26 22.58 22.71 5 0.51 9.68 8.47

1200 Original Curve (days), t0 (days) is the time when the fish mass equals to zero, 1000 Corrected Curve which corresponds to the time when larvae transform into fi 1 800 juvenile sh, and the growth parameter k (days )isa characteristic of the fish species. 600 For each life stage, a mortality rate is defined. For the eggs

Weight (g) 400 and larvae, it expresses the percentage of individuals that die

200 during the time required for the transformation into larvae and juvenile fish, respectively. For the fish, the mortality rate 0 0 500 1000 1500 2000 2500 3000 expresses the percentage of individuals that die during 1 year. Time (day) Fish also die if their minimal living requirements are not met. Figure 5. Von Bertalanffy growth function (Equation (1)). Original curve In the present case, for example, the drying of part of the (Yin 1998) for the family of Cyprinidae, to which S. hollandi belongs, and riverbed during low flow can cause death of fish. corrected curve for S. hollandi based on fish data collected in 2007–2008 on the Lijiang River. Representation of fish and initial fish distribution A number of eggs, larvae and fish are initially randomly distributed in the investigated domain. An initial age distribution of the fish is defined. The age is the principal characteristic of individual fish. Its weight is then obtained by means of Equation (1). With the field surveys on the Lijiang River in 2007 and 2008, the following relation between the weight W (g) and the body length L (cm) has been determined, with an R2 value of 0.996 (Figure 6):

W ¼ 0:0173L3:0638 (2)

Figure 6. Relationship between body length and weight of S. hollandi Suitability map based on field surveys in the Lijiang River in 2007 and 2008. To increase or maintain its fitness (Railsback et al., 1999), fi point of the investigated domain and the total fish biomass. sh will try to move to more suitable places, whereby the The main components of the fish model will first be suitability depends on hydrological and environmental fl fl described immediately hereafter in a general way, and the parameters, such as the ow velocity, ow depth, substrate specific model configurations adopted in the simulation of size, availability of food resources, dissolved oxygen and fi the laboratory experiments and the case study will water quality. Therefore, it is essential to de ne the pattern subsequently be detailed. of the suitability in the investigated domain. In the present paper, the suitability is defined in a simplified way and only fl fl Fish life cycle takes the ow velocity, the ow depth and the available food resources into account. The velocity is taken into fi Alifecycleof sh is implemented in the model that starts with account by means of maps of the preferential velocities, as fi eggs. The process of spawning is represented in a simpli ed derived in the laboratory experiments (Figure 4). Flow fi waybydening the number of spawned eggs during the depth is indirectly taken into account: it is always suitable, spawning period. In the present model set-up, the number of except when the riverbed becomes dry during low- fi spawned eggs is not related to the number of adult sh in the discharge events, which can trap fish and cause their reach. The eggs subsequently transform into larvae after a death. The food resources are also taken into account in an fi fi de ned period and subsequently into juvenile sh after indirect way: the available food resources at a certain fi another de ned period. The weight of eggs and larvae is location and the daily food consumption of the fish fi negligible and not accounted for in the total sh biomass. The determine how many fish can be sustained at that location. fi fi fi juvenile sh subsequently develop into young sh, adult sh As long as this maximum number of fish is not reached, fi and senile sh according to the von Bertalanffy growth food resources are available and do not affect the function (von Bertalanffy 1938, Figure 5): suitability, which remains uniquely based on the preferen- hi fl fi ðÞ 3 tial ow velocity. Once the maximum number of sh is ¼ kt t0 Wt W1 1 e (1) reached, the location becomes unsuitable, even if the fish preferential velocities occur. This treatment of the effect of the W1 (g) is the average ultimate weight of an adult fish, food resources availability also accounts in a simplified Wt (g) is the average weight of an individual fish at time t way for competition between individual fishes. Substrate

Table III. Parameters adopted in the ﬁsh dynamics model for the model validation by means of the experiment in the second laboratory ﬂume (Figure 3, second column) and the model application to the reach of the Lijiang River (Figure 7, third column).

Model application Parameter Model validation – second laboratory ﬂume – Lijiang River

Initial number 90 1000 of individuals Life cycle Young (2 years old) Juvenile: 30 days Young: 1–2 years Adult: 3–6 years Senility: 7–8 years Age structure – 30-day-old juveniles 60%, 1-year-old young fish 20%, 3-year-old adult fish 10%, 7-year-old senile fish 10% Accumulated mortality – 1% Food resource(mg/L) – 1.02 Time increment in fish model 100s 24 h Fish swimming rate 0.02 m s1 50 m day1 Poisson coefficient (l) – 2 0:003ðÞtt 3 Von Bertalanffy – Wt ¼ c 1004 1 e 0 growth equation The correction factor c based on field surveys is c = 0.8 for juvenile fish, c = 0.5 for young fish, c = 1.0 for adult fish and c = 0.9 for senile fish Preferential velocity 0.3–0.6 m s1

preference is not taken into account in the model interdependent or independent (Marsh and Jones, 1988; applications reported hereafter, because the bed in the Wu et al., 2000). In the reported model, independent second laboratory flume consisted of mortar and insuffi- distance and direction are implemented. Hence, the cient substrate data were available for the Lijiang River. movement due to fish swimming can be written as follows: Fish movement (swimming rate, direction) ! ! ! L ¼ L a L (4) Eggs and larvae do not move. The movement of fish towards fish swimming fish swimming fish swimming more suitable places determines the spatial distribution of the fi sh in the river; therefore, the movement rules are critical where the first term represents the distance and the second to accurate model simulations (Railsback et al., 1999). term the direction. The maximum distance that a fish can fi The re nement of the movement rules based on the swim during the time increment Δt is given by its fl experiments in the two laboratory umes (Figures 2 and 3), swimming rate (Table III) multiplied by the time increment. as compared with that in Li et al. (2010), is one of the major The real distance that the fish will swim, however, is contributions of the present paper. typically shorter and determined by the movement rules. fi The movement of an individual sh within the period of Also, the direction of fish swimming is determined by the Δt one time! increment can be represented by the displacement movement rules. fl vector Lfish, which is the resultant of drifting by the ow and These movement rules are a set of hierarchical rules that fi displacement due to sh swimming: express the behaviour of the fish. Fish will preferentially migrate to the closest suitable location within reach that is ! ! ! situated in the direction against the flow. If no suitable L ¼ L þ L fish flow fish swimming (3) location against the flow direction is within reach, fish will migrate to the most suitable location that is located upstream of its present location at random direction. If The displacement due to drifting by the flow is defined suitability is homogenous, fish will migrate towards the fl ! as the product of the ow velocity vector vflow and the time location with the lowest living requirements, which, in the increment Δt. The flow velocity vector at the fish location present model implementation, amounts to the location is interpolated from the flow model grid. with the highest available food resources. Finally, in a Movement due to fish swimming can be classified by homogeneous environment, fish movement will be whether swimming distance and direction are random.

These fish movement rules constitute the most critical Elevation component of the fish dynamics model and largely determine the model results. To assess and validate the China fish movement rules, the fish dynamics model was applied to simulate the experiment in the second flume (Figure 3). The choice of model parameters is summarized in Table III and will now be justified. The steady flow in the second flume was computed on a grid consisting of about 14 000 grid cells, providing the velocity distribution to the fish dynamics model. The Guilin velocities in the flume ranged from about 0 to about 1 m s 1. Qingshitan Reservoir The simulated experiment lasted 1 day, which allows discarding the effects of the life cycle and the availability of food resources, and accentuating the role of the movement rules. Whereas the flow patterns of 30 fish individuals were experimentally investigated, the numerical simulations were performed for 90 individuals aged 2 years, to have less Yangshuo uncertainty in the statistics of the residence time ratios in each 00.51.0 km point of the flume. With the experiments in the first flume (Figure 2), the preferential velocities were defined in the range from 0.3 to 0.6 m s1 (Figure 4), and the swim rate was set at Figure 7. Situation of the investigated 7000-m-long Daxu reach on the 1 Lijiang River between the cities of Guilin and Yangshuo in Southeast China. 0.02 m s .Nofish migrated in or out of the experimental The colour code in the right figure indicates bed morphology, whereby the flume. A time step of 1 s was adopted in the fish model, which zero level corresponds to the lowest level in the investigated reach. is quite typical for investigations in rather small spatial zones. Goodwin et al. (2006), for example, also adopted a time step of 1 s to simulate migration in a fish pass. Migration through inflow and outflow boundaries The model was validated with data from the experiments in fl Fish migration occurs through the upstream and downstream the second laboratory ume (Figure 2). Because it was fi fi cross-sections that delimit the investigated reach. Because dif cult to measure accurately the sh residence time in every fl migration of fish in and out of the investigated reach occurs point of the laboratory ume, the validation focussed on the fi fi randomly and independently, the number of fish migrating sh residence time in ve representative points indicated in fl through the inflow and outflow boundaries in each time step Figure 3. Comparison of observations in the laboratory ume fi was assumed to have a Poisson probability density: and model predictions (Table III) indicates that the sh model simulates satisfactorily the relative residence time ratios in all n five points, which lends credit to the fish movement rules. l l PΔ ðÞ¼n e (5) t n! where l parameterizes the Poisson probability density and PΔt (n)is the probability that a number of n > 0 individuals will cross the boundary during the time step Δt. As a characteristic Flow quantity In Guilin station of the Poisson probability density, the number of fish that will Effluent of Qingshitan Reservoir cross the boundary during a time step Δt is on the average ) l fi ) -1 equal to . Incoming sh were introduced at the most suitable -1 s s 3 location in the cross-section, and the age structure of the 3 incoming fish population was specified. On the average, the number of outgoing fish matches the number of incoming fish. Fish that were located close to the outgoing boundary Discharge(m and lived in the most unsuitable conditions were selected to Discharge(m migrate out of the investigated reach. 0 60 120 180 240 300 360 Time (day) Assessment and validation of fish movement rules fi Figure 8. Hydrographs recorded in 2004 at the Guilin hydrological station A major originality of the present paper is the re nement of (left axis) at the upstream boundary of the investigated reach and at the the fish movement rules, as compared with Li et al. (2010). outflow of the Qingshitan reservoir (right axis).

Residence time ratio (%) MODEL APPLICATION -3 With regulation 1.3×10

Study site Without regulation The fish dynamics model was applied to the Daxu reach of the Lijiang River, which is situated in Southwest China (Figure 7). River cruises on the Lijiang River between the 0 ’ cities of Guilin and Yangshuo are one of China s major Figure 9. Accumulated fish presence frequency predicted by the fish tourist attractions and are of fundamental importance for dynamics model in the year 2004. Flow is from left to right. the region’s economy. The Lijiang River also supports fi extensive shery, particularly for S. hollandi. provided as output the daily-averaged water levels and the The investigated reach has a compound-channel morphology flow velocities. The flow model was calibrated by means of and a length of 7000 m. Measurement of the bathymetry the measured flow data until the relative error between fl and the ow velocities were performed with a Sontek/YSI measurements and simulations was smaller than 5%. The fi Rivercat acoustic Doppler pro ler (San Diego, California, flow model calibration is reported in detail by Li et al. USA, Xylem Inc.). Over 70 cross-sections were measured (2010, 2011a). along the 7000-m-long reach. A description of the The main parameters of the fish model are summarized bathymetry in the entire reach was obtained from bilinear in Table III and explained hereafter. interpolation of the measured cross-sections. The bed Initially, 30 000 eggs, 500 larvae and 1000 fish in- material is heterogeneous and includes zones of grass, dividuals were randomly distributed in the model. The eggs sand, gravel and cobbles. transform into larvae within a period of 2 days with a Because of the subtropical monsoon climate and the mortality of 99%, and the larvae transform into juvenile ’ karst geology, the Lijiang s hydrology is characterized by fish within 4 days with a mortality of 50%. The initial strong seasonal fluctuations: the discharge varies from 12 3 1 number of eggs and larvae and the mortality rates were to 12 000 m s during extreme floods, with an annual fi 3 1 3 1 chosen to yield 400 juvenile sh on a population of 1000 average of 120 m s . Discharges lower than 30 m s at fish, on the basis of information provided by local the Guilin hydrologic station do not allow for navigation of fishermen. The age structure of the fish was 600 juveniles the cruise ships and impose threats to the ecosystem, (30 days old), 200 young fish (1 year old), 100 adult fish because large parts of the riverbed become dry. To reduce (3 years old) and 100 senile fish (7 years old), and the flow fluctuations during flood events, to maintain a mortality of fish was set at one per thousand, on the basis of minimum discharge for cruise ship navigation and to fish data obtained from field experiments in 2007 and improve ecological conditions during the dry season, a 2008, information from the local fishermen and previous number of reservoirs already operate on the upstream work on the Lijiang River by Li et al. (2010). branches of the river, and more reservoirs are planned. It is During the simulated 1-year period, biomass of the fish important to assess the effect of the altered hydrological population changes. Fish biomass decreases as a result of regime on the river’s ecosystem. The present paper fish mortality and increases as a result of growth of the fish. investigates the effect of Qingshitan reservoir, which is In the present simulation, the natural mortality rate of fish operational since 1957, on the population of S. hollandi.It has been set at one per thousand, on the basis of previous is complementary to investigations on the effect on riparian vegetation dynamics (Chen and Ye, 2008; Ye et al., 2010) and fish habitat (Li et al., 2010, 2011a). 0.6 Without Mortality 0.5 With Regulation fi Without Regulation Model con guration 0.4 Simulations of the fish population were performed for 0.3 scenarios with a natural and an altered hydrological regime for the year 2004, which was a rather dry year (Figure 8). weight (t) 0.2

The discharge increase due to Qingshitan reservoir is most 0.1 important during the dry season (October to December). The 7000-m-long investigated reach was discretized on a 0 0 60 120 180 240 300 360 grid containing 240 grid cells in the longitudinal direction Time(day) and 10 grid cells in the transverse direction, with an average mesh area of about 220 m2. The measured daily averaged Figure 10. Evolution of the total fish biomass predicted by the fish dynamics model in the year 2004. The reference line labelled ‘without discharges and water levels were used as the upstream and mortality’ represents the fish biomass evolution due fish growth without downstream boundary conditions, respectively. The model any mortality.

Copyright © 2013 John Wiley & Sons, Ltd. Ecohydrol. 6, 586–597 (2013) FISH DYNAMICS IN RELATION TO CHANGING HYDROLOGICAL CONDITIONS 595 work on the Lijiang River by Li et al. (2010). This show large spatial and temporal variations from nearly zero mortality rate, however, only marginally influences the minimum velocities to maximum velocities of more than results of the present 1-year investigation. The main cause 2ms1. The swim rate was set at 50 m per day (Zhao 1982, of mortality, which causes a considerable decrease of Yin 1998). A time step of 1 day was adopted in the fish biomass in the Lijiang, is the trapping of fish in riverbed model, which is coherent with the time step of the flow areas that dry during low-discharge conditions. model and appropriate for investigation fish dynamics in a Yin (1998) provides the following parameters in the von relatively large river reach over a total duration of 1 year. Bertalanffy (1938) growth function for the family of Time steps of 30 days have been reported in literature for Cyprinidae, to which S. hollandi belongs: W1 = 1004 g and investigations in a larger domain and on a longer scale k = 0.003 days1 (Equation (1)). With the fish data (Railsback et al. 1999). The value of l =2 ofthePoisson collected in 2007–2008 on the Lijiang River for S. distribution was estimated by the maximum-likelihood hollandi, however, this growth function has been multi- method using data collected from measurements, which plied by a correction factor c, with c = 0.8 for juvenile fish, consisted in counting the number fish captured by fishing c = 0.5 for young fish, c = 1.0 for adult fish and c = 0.9 for nets that spanned the cross-section. This value does, senile fish. Both the original and corrected growth however, only weakly influence the model results. The functions are shown in Figure 5. age of incoming fish was defined as 2 years. The available food resources in the Lijiang River are available from field measurements of the plankton concen- Model results tration, which in the average amounts to 1.02 mg l 1. S. The simulations mainly focus on the spatial distribution hollandi consume daily 2% to 5% of their body weight, and biomass of S. hollandi in the investigated reach and which amounts to less than 10 g per day. For the application how they are affected by the operation of Qinshitan of the model to the Lijiang River, the total plankton mass in reservoir. Food resources were found to be sufficiently one computational cell can sustain at least 28 fish species. available in the Lijiang River and never constituted a As aforementioned, the suitability in the present application constraint that affected the fish distribution. is defined in a simplified way and mainly takes the velocity Figure 9 shows the patterns of the accumulated fish into account. On the basis of the laboratory experiments, a presence frequency for the scenarios with and without flow range of suitable velocities were defined as 0.3 to 0.6 m s1. regulation, which amounts to the percentage of time that It should be noted that the velocities in the investigated reach fish spent in each point of the domain. The spatial patterns are quite similar for the scenario with and without flow regulation: fish are mainly found near both bends in the reach. This can tentatively be attributed to the higher variability of the flow depth and the flow velocity in the bends as compared with straight reaches. The riverbed is typically inclined in transverse direction in a bend, with flow depths that increase from the inner bank to the outer bank, and the velocities show important spatial gradients throughout the bend (Blanckaert 2010). This implies that channel bends will provide suitable conditions for the fish fl under a much broader range of discharge conditions than Figure 11. Water depth computed with the ow model during the fl maximum (top) and minimum (bottom) discharge in 2004 for the scenario straight reaches, where the ow depth and velocities show with flow regulation. The lower figure shows that large parts of the less heterogeneity. The higher heterogeneity of the flow riverbed fall completely dry during the dry season. depth in both bends as compared with the straight reaches is illustrated in Figure 7. fl Residence time ratio (%) The model results indicate that the increase of the ow 5.2×10-2 discharge during the dry season leads to an increase in size of the regions where fish frequently occur, that is, an increase in size of the zones of suitable habitat, as well as a With regulation reduction of the distance between these suitable zones. These favourable effects for the fish population are well illustrated by the evolution of the total biomass of the fish 0 during the year 2004 (Figure 10). Figure 12. Historical spawning grounds (top) Accumulated fish presence When omitting fish mortality, the total biomass would frequency predicted by the fish dynamics model during the spawning period (May–August) in 2004 for the scenario with flow regulation. Flow gradually increase according to the von Bertalanffy growth is from left to right. function (Equation (1)) as illustrated in Figure 10. The real

Copyright © 2013 John Wiley & Sons, Ltd. Ecohydrol. 6, 586–597 (2013) 596 R. HAN et al. growth of the total biomass is smaller because of fish in biomass evolution between scenarios with and without mortality. For both scenarios with and without flow reservoir operation. This mortality rate due to drying cannot regulation, the growth of the biomass is very similar be accounted for in global models. In the simulation of the during the wet season (May to August, Figure 8), but the second laboratory flume, a global model without movement biomass grows faster during the dry season (September to rules would predict the highest relative residence time to November, Figure 8) for the scenario with flow regulation occur in point 5, where the most suitable velocities occur, and (Figure 10). This difference is mainly due to a higher the lowest relative residence time in point 1, where unsuitably mortality of fish in the scenario without flow regulation. high velocities occur (Figure 3 and Table II). The dynamic During the dry season, large parts of the riverbed fall fish model includes fish movement rules that express if and completely dry, leading to the trapping of fish and causing how the fish can reach the most suitable regions. The relative their death. isolation of the region of suitable velocities around point 5 Figure 11 shows the simulated flow depth during the impedes fish of reaching this region. This implies that the maximum and minimum discharges in the year 2004. Even spatial patterns of the suitability, and especially the for the illustrated scenario with flow regulation, large parts connectivity between suitable regions, play a major role, of the riverbed fall completely dry. The flow regulation which cannot be accounted for in global models. Rapid floods does, however, increase the discharge during the dry season are another example where these advantages are critical. and thereby reduce the riverbed area that dries, which leads Contrary to global models, individual-based models can take to a decrease in fish mortality. into account whether or not fishcanreachshelters Both results in Figures 9 and 11 indicate that the flow characterized by suitable conditions for surviving. regulation has favourable effects on the fish population. Individual based fish models are important tools for a These model results should merely be seen as indications, variety of practical applications. As illustrated in the however, because the paucity of fishery data does not allow present paper, they allow assessing the influence of natural direct model validation. To further assess the plausibility of or anthropogenic modifications to the river system, for the model results, Figure 12 focuses on the accumulated example, modifications to the hydrology or river training presence frequency during the spawning period from May works. They can also be an important guide in the design of to August. The spawning period was chosen because of its river restoration schemes and allow verifying that the importance in the life cycle of S. hollandi and because connectivity between suitable regions remains guaranteed historical information is available on the location of the during important events such as floods and droughts. spawning grounds. Surveys have shown that the Daxu Furthermore, individual-based fish models can provide section has three traditional spawning grounds (Figure 12). guidance in defining sustainable fishing practice. The Regions with high presence frequency (Figure 12) simulation of the temporal evolution of the biomass would predicted by the model coincide reasonable well with the allow defining how much fish may be captured per life historical spawning grounds, which lends further credit to stage to maintain the fish population at a sound and the model results. sustainable level. At present, there is, for example, overfishing in the Lijiang River. There is an urgent need for field data, including spatial fish DISCUSSION distributions, mortality rates and their dependence on age and season, migration and swimming rates. At present, the The results convincingly demonstrate that individual-based available data from field campaigns on the Lijiang River did fish dynamics models that include realistic rules for the fish only allow confirming the plausibility of the model results but behaviour and movement are capable of predicting the did not allow a direct validation. spatial distribution of the fish and the evolution of the There is also a need to further refine and develop individual- biomass. based fish dynamics models. The modular architecture of the These individual-based models have important advantages reported model makes extensions, model refinements or over models that are globally defined on the population level applications to other fish species rather straightforward. and typically are only based on the habitat suitability without Extensions could include the inclusion of additional including rules for the fish movement and behaviour. This processes that are important in the life cycle of the fish, means that global models cannot account for the time that the such as migration for spawning, or interactions with fish needs to reach a region of suitable habitat, which was predators and preys. Refinements could concern the found to be of dominant importance in the reported model definition of the suitability parameters, which may vary applications. In the application to the Lijiang River, fish could according to the life stage of the fish. Additional laboratory not always escape from riverbed regions that dry during experiments on the preferential velocity range, for decreasing discharge episodes. The mortality of fish trapped example, could be performed for fish of different ages. in dry regions was found to be the major cause of differences Suitability could also account for turbulence, which

DeAngelis DL, Cushman RM. 1990. Potential applications of models in strongly affects the transport of nutrients and oxygen. fi fl forecasting the effects of climate change on sheries. Transactions of Especially, ow regions of considerably increased turbu- the American Fisheries Society 119: 224–239. lence are known to be relevant, such as occurring behind DeAngelis DL, Gross LJ. 1992. Individual-based Models and Approaches in Ecology. Chapman and Hall: New York large boulders and hydraulic structures such as bridge piers fi fl Ekeberg Ö 1993. A combined neuronal and mechanical model of sh and groynes, or mixing zones where ows from different swimming. Biological Cybernetics 69: 363–374. rivers meet. Gibson RN, Robb L, Wennhage H, Burrows MT. 2002. Ontogenetic changes in depth distribution of juvenile flatfishes in relation to predation fish and temperature on a shallow-water nursery ground. Marine Ecology Progress Series 229: 233–244. ACKNOWLEDGEMENTS Goodwin RA, Nestler JM, Anderson JJ, Weber LJ, Loucks DP. 2006. fi – The Authors are grateful to the financial support of National Forecasting 3-D sh movement behavior using a Eulerian Lagrangian- agent method (ELAM). Ecological Modelling 192: 197–223. Basic Research Program 973 (2010CB429004), National Humston R, Olson DB, Ault JS. 2004. Behavioral assumptions in Nature Science Foundation of China (51279196), ‘100 Talent models of fish movement and their influence on population Program of Chinese Academy of Sciences (A1049)’, and the dynamics. Transactions of the American Fisheries Society 133(6): 1304–1328. Chutian Scholarship (KJ2010B002). The fourth author was Lane, EW 1947. Report of the subcommittee on sediment terminology. partially funded by the Chinese Academy of Sciences Visiting Transactions, Americal Geophysical Union, 28(6): 936–938. Professorship for Senior International Scientists, grant Li R, Chen Q, Ye F. 2011a. Modelling the impacts of reservoir operations on the downstream riparian vegetation and fish habitats in the Lijiang number 2011T2Z24, and by the Sino-Swiss Science and River. Journal of Hydroinformatics 13(2): 229–244. Technology Cooperation for the Institutional Partnership, Li W, Chen Q, Huang Y. 2011b. Laboratory study on ethology of – grant number IP13_092911. Spinibarbus hollandi. Acta Ecological Sinica 31(5): 1404 1411. (in chinese) Li W, Han R, Chen Q, Qu S, Cheng Z. 2010. Individual-based modelling of fish population dynamics in the river downstream under flow regulation. Ecological Informatics 5: 115–123. REFERENCES Marsh, LM, Jones, RE, 1988. The form and consequences of random walk Bailey KM, Batty RS. 1983. A laboratory study of predation by Aurelia movement models. Journal of Theory of Biology 133: 113–131. aurita on larval herring (Clupea harengus): experimental observations Pavlov DS, Lupandin AI, Skorobogatov MA. 1994. Influence of flow compared with model predictions. Marine Biology 72: 295–301. turbulence on critical flow velocity for gudgeon (Gobio gobio). Bégout Anras ML, Lagardère JP. 2004. Measuring cultured fish swimming Doklady Biological Sciences 336: 215–217. behaviour: first results on rainbow trout using acoustic telemetry in Railsback SF, Lamberson RH, Harvey BC, Duffy WE. 1999. Movement tanks. Aquaculture 240: 175–186. rules for individual-based models of stream fish. Ecological Modeling Bian L 2003. The representation of the environment in the context of 123:73–89. individual-based modeling. Ecological Modelling 159: 279–296. Reed M 1983. A multidimensional continuum model of fish behavior. Blanckaert K 2010. Topographic steering, flow recirculation, velocity Ecological Modelling 20(4): 311–322. redistribution, and bed topography in sharp meander bends. Water Rose KA, Tyler JA, Singhdermot D, Rutherford ES. 1996. Multispecies Resources Research 46, W09506, doi:10.1029/2009WR008303. modeling of fish populations. Computers in Fisheries Research, Brookshier PA, Flynn JV, Loose RR. 1995. 21st Century Advanced Chapman and Hall: London; 194–222. Hydropower Turbine System. American Society of Civil Engineers: Skorobogatov MA, Pavlov DS, Lupandin AI. 1996. Effect of New York. current velocity and turbulence intensity on the distribution of the Burrows MT. 2001. Depth selection behavior during activity cycles of juvenile roach Rutilus rutilus in a water stream. Journal of Ichthyology 36: plaice on simulated beach slope. JournalofFishBiology59: 116–125. 654–658. Cai Z, Lin G, Ni J, Ye Z, Tang Z, Zhang S. 2007. Study on the Steel EA, Guttorp P, Anderson JJ, Caccia DC. 2001. Modeling juvenile reproductive biology of Spinibabus calddwellis. Guangxi Agricultural migration using a simple Markov chain. Journal of Agriculture Sciences 2:24–28. Biological Environment Station 6(1): 80–88. Castro-Santos T 2005. Optimal swim speeds for traversing velocity Von Bertalanffy L. 1938. Quantitative theory of organic growth. Human barriers: an analysis of volitional high-speed swimming behavior of Biology 10(2): 181–213. migratory fishes. Journal of Experimental Biology 208: 421–432. Webb PW. 2002. Control of posture, depth, and swimming trajectories of Chen Q, Han R, Ye F, Li W. 2010. Spatio-temporal ecological models. fishes. Integrative and Comparative Biology 42(1): 94–101. Ecological Informatics 6(1): 37–43. Wu H, Li BL, Springer TA, Neill WH. 2000. Modelling animal movement Chen Q, Ye F. 2008. Unstructured cellular automata and the application to as a persistent random walk in two dimensions: expected magnitude of model river riparian vegetation dynamics. Lecture Notes in Computer net displacement. Ecological Modelling 132: 115–124. Science 5191: 337–344. Ye F, Chen Q, Li R. 2010. Modeling the riparian vegetation evolution due Coombs S 1999. Signal detection theory, lateral-line excitation patterns and to flow regulation of Lijiang River by unstructured cellular automata. prey capture behavior of mottled sculpin. Animal Behavior 58:421–430. Ecological Informatics 5(2): 108–114. Coutant, CC, Whitney R, 2000. Fish behavior in relation to passage Yin M. 1998. Fish Ecology. Higher Education Press. Beijing, China; through hydropower turbines: a review. Transations of the American 51–58. (in chinese) Fisheries Society, 129(2), 351–380. Zhao C. 1982. Fish way. Electronic Industry Press. Beijing, China; 97–102. Crowder LB, Rice JA, Miller TJ, Marschall EA, 1992. Empirical and (in chinese) theoretical approaches to size-based interactions and recruitment Zhang Y, Baptista AM. 2008. SELFE: a semi-implicit Eulerian– variability in fishes. Individual-based Models and Approaches in Lagrangian finite-element model for cross-scale ocean circulation. Ecology. Chapman and Hall: New York; 237–255. Ocean Modelling 333:26–52.