Application of agent-based modelling for selecting configuration of vertical slot fishway

Physical and 3D numerical hydrodynamic flow models and an agent-based model are developed with the principal objective of analysing fish behaviour in two vertical slot fishway configurations. Fish energy consumption due to swimming represents a crucial criterion for selecting an appropriate fishway configuration. The modelled fish detects ambient flow conditions, makes decisions based on its sensing and cognitive abilities, adapts to the changes in its environment, and moves toward the regions of less turbulent kinetic energy. The results show that fishways with longer pools enable passage of fish at a lower energy consumption.


Introduction
Climate change, environmental degradation, loss of biodiversity, and similar topics, have been increasingly debated in many research papers, various expertise studies, and discussions, as the extremely negative consequences of human-related activities. Accordingly, the interruption of river continuum stands out as one of such unfavourable activities that affect ecosystems related to watercourses. Barrier construction along watercourses (such as hydroelectric power plants, dams, weirs, etc.) has a significant effect on many ecosystem users, especially fish and their migration cycle. An increase in environmental preservation awareness and usage of renewable energy resources has led to a surge in the construction of small hydropower plants. Such facilities have a less negative impact on preservation of river continuum and biodiversity [1]. This paper focuses on the upstream fish migration through a fishway with the smallest possible energy consumption. However, damage to fish and exposure to predators are disregarded [2]. The fishway is a canal designed to obtain such hydraulic flow conditions that fish consider favourable in their upstream or downstream migrations. These engineering facilities are divided into two main groups: technical and natural fishways [3]. This research includes the analysis of a technical vertical slot fishway, as discussed in Section 2 Materials and methods. Fishway design should be adjusted according to the physical characteristics of fish with an emphasis on adapting the flow conditions to the abilities of the weakest swimmer [3]. Hence, fish have to overcome critical fishway sections where maximum flow velocities occur (flow contraction at the location of vertical slots). To swim against the water current in areas of maximum velocities, fish use their burst speed that can be maintained for a few seconds, what requires a certain recovery time afterwards. Burst speed for fish families Salmonidae, Percidae and those belonging to the order of Cypriniformes varies between 10 and 12 body lengths per second (BL/s). Furthermore, a prolonged fish speed which ranges between 3 and 5 (BL/s), needs to be obtained in the other parts of the fishway [3][4][5][6]. Numerous natural and anthropogenic factors that affect fish swimming capability have been discussed in numerous studies. Beamish [5] highlights the water temperature and dissolved oxygen concentration in water as two main factors that negatively affect the prolonged fish speed. Several authors have revealed that the vortex size and turbulence intensity stand out as two hydrodynamic factors that affect fish swimming capability [6][7][8][9][10][11][12]. Although turbulence is usually addressed in a negative context when considering its influence on fish swimming, Liao [7,8] examined the movement pattern used by fish to harness the energy of environmental vortices in turbulent flows in order to minimize their energy consumption. Also, he elaborated on the relationship between the vortex and fish sizes in turbulent flow. If body length is more than an order of magnitude larger than the vortex size, fish swims steadily through it, and vice versa. Hockley et al. [13] state that fish tend to avoid water currents with significant velocity fluctuations, and that their ability to overcome turbulent flow is mainly related to body length, weight, and shape. A more detailed analysis of the relationship between the fish body length, swimming capability, and vertical slot fishway design, is presented by Cai et al. [14]. They examined the behaviour of seven target species that weighed between 10 and 400 [g] with a body length between 5 and 40 [cm]. Their research resulted in recommendations for the flow velocity criteria that should be considered in the vertical slot fishway design. Fish [15] and McKenzie [16] focus on the topic of the energetics of fish swimming. They note that fish tend to use burst velocity even when low velocities with small fluctuations occur, i.e. in flow conditions that could be overcome by using prolonged speed. Furthermore, they present a common equation for calculating the fish drag power. It is obtained from the relation for the hydrodynamic drag force as a function of the fish body size, water density, and flow velocity. Also, the equation is used in this paper for calculating the rate of fish energy loss due to swimming. Fishway configuration plays an important role in ensuring the river continuum. Quality fishway design enables successful fish migrations along watercourse sections where flow discontinuities occur. Flow conditions in fishways are usually examined using physical models in laboratories. In such a way, it is possible to adjust the model geometry relatively fast and obtain results efficiently for various configurations. The Particle Image Velocimetry (PIV) for recording the flow velocity field has been applied successfully in fishways with various slopes, inflows, and widths [17]. Silva et al. [9] observed behaviour of Iberian barbel (Luciobarbus bocagei) in an experimental pool-type fishway with a constant slope, variable inflow, and submerged orifice. Flow velocities were measured continuously with a 3D Acoustic Doppler Velocimeter (ADV) while the fish movement was recorded with a video camera. The conclusion was that Iberian barbel occupies the areas with low turbulent kinetic energy (TKE), i.e. with low turbulence. Adults of this species showed a higher rate of passage success when compared to smallersize individuals. Bermudez et al. [18] analysed movement of brown trout (Salmo trutta), Iberian straight-mouth nase (Pseudochondrostoma polylepis), and Iberian barbel. They presented recirculation zones where low turbulence and velocities occur within vertical slot fishways. These zones are usually designed as resting areas and the study showed that fish stay most frequently behind baffles just downstream of the slot. Furthermore, fish trajectories were defined and presented for the narrow area upstream and downstream of the slot. The authors concluded that fish mostly used the area behind the smaller lateral baffle before transiting to the upstream pool. Li et al. [19] analysed velocity field in two GRAĐEVINAR 73 (2021) 3, 235-247 Application of agent-based modelling for selecting configuration of vertical slot fishway vertical slot fishways for various slopes and baffle shapes. Flow velocity was measured continuously with ADV. Also, a 3D hydrodynamic model was developed and verified based on the observed velocities. It was concluded that baffle shape greatly influences the flow pattern, i.e. TKE, energy dissipation rate, and vorticity. Furthermore, it was concluded that the flow velocity, TKE, and energy dissipation rate are positively correlated with fishway slope. The water level difference between adjacent pools was identified as the main factor affecting maximum velocities at vertical slots. This conclusion was found very useful from the economic point of view as it enables the construction of shorter fishways with steeper slopes can be built. The methodology used in previous studies is also applied in this research paper. A 3D hydrodynamic model of vertical slot fishways was developed, and model simulations were verified based on the observations from the corresponding physical models. The hydrodynamic model is coupled with a fish swimming model, which constitutes a contribution of this research. Fish are complex animals with a highly developed ability for prey capturing, avoiding predators and detecting environmental conditions favourable for their growth and spawning. Therefore, a cognitive aspect plays an important role in the fish swimming model, and should be included in the mathematical formulation of the problem, along with physical elements of the process. The formulation of the behaviour of individuals that is adaptable to environmental conditions represents the core of the Agent-Based Model (ABM) simulations [20,21]. Agent-based modelling is a relatively new approach in investigating the individuals that mutually interact and adapt their behaviour based on changes in their environment [22]. ABM procedures focus on the description of actions and interactions between individuals (agents) and the environment, instead of describing a fully modelled global phenomenon [23]. The purpose of ABM development also defines its complexity. This paper presents an original ABM of fish swimming through fishway developed using the ecological modelling software ECOLab within the MIKE numerical modelling framework (www.dhigroup.com). The modelling aspects that have been considered within this ABM include a physical component involving application of a hydrodynamic model (flow model with the extracted TKE scalar field) coupled with a behavioural component (fish swimming upstream to the zone of increased hormone concentration while avoiding the areas of intensive TKE). Biological components -growth and mortality, and changes in temperature and dissolved oxygen concentration have been neglected due to the short time the fish (agent) spends in the fishway. Lastly, the movement of individual fish is examined for various fishway configurations. The authors of this paper were unsuccessful in finding previous studies related to the coupled hydrodynamic models and ABM of fish swimming through fishway (Eulerian approach for flow field and Lagrangian model for the analysis of fish movement). Earlier studies do not quantify the timedependent fish energy consumption for different fishway configurations. Also, previous studies present the coupled hydrodynamic models with ABMs intending to solve the problem of fish finding the fishway entrance [24][25][26], fish migrations in river estuaries [27], and seasonal migrations of fish in seas [28]. Fish trajectories that follow the agents from their source location in the watercourse to the fishway entrance were extracted in papers [24][25][26] . However, the analysis of fish swimming upstream through the fishway was not conducted. The first part of Section 2 includes a description of datasets used for developing a physical model of the fishway and measurements of flow depth and velocity for two configurations. The development of the numerical flow model of fishways and presentation of the verified model results are given in the second part of the same section. The third part of Section 2 offers a detailed description of the ABM of fish swimming through fishway as well as specifications of ABM simulations. The ABM simulation results are presented in Section 4, and conclusions are given in Section 5.

Numerical model of the fishway
The numerical flow model of the fishway should provide information about key physical processes that affect the cognitive response of fish during their upstream migration through this structure. For instance, it should provide an image of zones with intensive vorticity and/or turbulence that fish can detect and make decisions based on the changes in their environment. A 3D hydrodynamic flow model MIKE 3FM, which is based on the finite volume method, was developed in the scope of this research. Spatial discretization and model mesh is shown graphically below only for the fishway with a pool length of 90 [cm] as the other configuration, i.e. RS-45 is similarly defined with the same density of mesh elements ( Figure 3). The vertical domain of the model is described with 5 σ-layers as such discretization has proven to be the most effective in terms of reducing the computation time with acceptable model accuracy. It is important to note that the sensitivity analysis of model results has also been made. It is based on the variable number of σ-layers. For instance, differences in the modelled TKE field vary less than 3 % for 15, 10 and 5 σ-layers. In the case when the model is vertically discretised with 4 σ-layers and 3 σ-layers, the differences of the same parameter (TKE) increase to 12 % and 35 %, respectively. The turbulence model is defined by the k-e formulation in vertical [29] and the Smagorinsky concept in horizontal direction [30].
The Smagorinsky coefficient is used as one of the calibration parameters and a constant value of 0.1 is adopted when discrepancies of the measured and modelled velocities have been minimized. The calibration procedure and comparison of measurements and simulations are explained in detail in Section 3. Scaling factors for the turbulent kinetic energy (TKE) and its dissipation rate (e) are equal to 1 and 1.3 in the horizontal and vertical directions, respectively. The variable computational time step depends on the mesh element size regulated by the CFL (Courant-Friedrichs-Lévy) parameter limited to the value of 0.8. The velocity and TKE fields, obtained by the hydrodynamic flow model, are used as the basis for the ABM of fish swimming through the fishway. In its upstream movement along the fishway, fish is attracted by a higher ambient hormone concentration, avoiding the zones of high TKE. This cognitive component of fish behaviour represents the essence of the developed ABM.

ABM development and simulations
Numerical simulations included analysis of the movement of fish of an assumed length of 25 [3][4][5][6] and have already been discussed in Section 1 Introduction. Hence, the modelled fish corresponds to the rainbow trout (Salmo gairdneri) [31]. In the text that follows, the term agent denotes an individual fish. ABM simulations were conducted based on the velocity and TKE fields that were extracted from verified results of the hydrodynamic flow model. The basic concept of the developed ABM is presented below.
Fish energy (E) represents the ABM state variable based on the following expression [15,16]: is the target fish speed [m/s] (relative fish speed), i.e. the speed that fish tends to reach considering ambient conditions affecting its movement (flow velocity, turbulence). When swimming collinearly with the water current, fish reduces its relative speed and uses the flow energy for moving upstream. Such movement pattern enables the agent to save energy and use it later for overcoming the areas of highest flow velocities (flow contractions at vertical slots). When swimming against the water current, the agent adaptively uses its continuous or burst swimming speed depending on the ambient flow. The upstream movement of the agent stops when available energy reaches the limit of 0, i.e. the fish is carried away with water current towards downstream fishway sections when condition E = 0 is satisfied. The core of the fish movement motivation is related to the hormone concentration field that monotonously increases in the upstream direction. Also, fish avoids zones of intensive TKE. A detailed description of ABM components for fish movement is provided in the following text. In terms of this research, ABM components are constants, agent sensing functions, and mathematical expressions for describing behavioural and physical fish characteristics. . This approach involving the agent's tendency to move towards lower TKE is derived from previous studies shown in [7,8,13,18]. The second function (θ max,WILL ) is related to the determination of the movement towards higher hormone concentrations based on the ambient 2D hormone field detection within the same search radius. A diagram describing the principle of sensing functions is provided below for the fish movement towards the minimum TKE in its ambient.  Note: The orange symbol represents the current location of the agent, blue triangle the mesh element with minimum TKE, red circle the boundaries within the agent's search radius, crossed element centroids the mesh elements that the agent detects, and circular element centroids the mesh elements that agent is unable to detect during its movement.
In the area of the denser unstructured mesh, the agent detects four nearest elements with the corresponding TKE and hormone concentration values within its search radius. On the contrary, in the zones of the less dense model mesh, the agent detects only one closest element. Arithmetic expressions defined during the development of ABM are described below. The flow area (A) of fish is given by [4]: The agent's relative movement direction (θ agent,rel ), based on the modelled 2D fields of flow direction and TKE, is given by the following formulation: where (j HD,h ) is the direction of the water current vector in horizontal plane that is extracted from the results of the hydrodynamic flow model. A schematic of the agent's decision making regarding its movement direction, based on the ambient hydrodynamic conditions, is provided at Figure 5. The agent's target movement speed (v agent,rel ), based on the modelled 2D fields of flow direction and velocity that are extracted from the hydrodynamic model, is described by the following formulation: where (v HD,h ) represents the flow velocity in horizontal plane obtained from the hydrodynamic model. The agent's decision regarding the swimming mode that will be used (continuous or burst) is defined based on the threshold values of (v HD,h ). These thresholds are derived and adjusted according to results presented in [14]. The dimensionless fish drag coefficient (C b ) is assumed to be dependent on the (v HD,h ), as presented in the expression below: where (C b ) is described by increasing the stepwise function.
Another assumption is that the adopted (C b ) increases with an increase in the flow velocity and the agent's intention to overcome the flow. Considering that the authors are not familiar with the exact and certain values of (C b ), the previous assumption GRAĐEVINAR 73 (2021) 3, 235-247 Application of agent-based modelling for selecting configuration of vertical slot fishway leads to the proportional relationship between (C b ) and the target (relative) movement speed of the agent. These assumptions are considered valid in the light of the fact that similar conclusions are made in [15,16] regarding the relationship between the two described parameters. It is indicated in these studies that (C b ) increases three-to five-fold as a consequence of propulsive movements of the fish. Therefore, if the flow velocity is below 0.2 [m/s], (C b ) is assumed to be 10. If the flow velocity varies between 0.2 and 0.9 [m/s], (C b ) increases three-fold and amounts to 30. For the flow velocity above 0.9 [m/s], (C b ) is equal to 150, which corresponds to the value five times higher than (C b ) for the lower velocity class. The agent's absolute movement direction (θ abs ), which considers both the 2D fields of the outputs from the hydrodynamic model (flow direction and velocity, and TKE) and 2D field of the assumed hormone concentration, is given by the following expression: where (θ abs ) is a function of (v agent,rel ), (θ rel ) and (θ max,WILL ), which determines the final target direction of the agent. The first and second elements given in brackets are related to the magnitude and direction of the first vector, while the third and fourth elements represent the magnitude and direction of the second vector. It is important to note that this function is used to obtain only the agent's resulting movement direction, and so the magnitudes of the two vectors are assumed as equal. In this way, the agent chooses its final movement direction based on hydrodynamic conditions and hormone concentration in its ambient. A schematic for defining the agent's final movement direction is provided in Figure 6. The agent's absolute or resulting movement speed (v agent,abs ) is derived from its target speed (v agent,rel ), its target direction (θ abs ), flow velocity (v HD,h ) and flow direction (θ HD,h ): where (v agent,abs ) is described as a function of (v agent,rel ), (θ abs ), (v HD,h ) i (j HD,h ). The first two parameters in brackets represent the agent's relative (target) movement vector, while the other two are related to the flow vector in the horizontal plane. The agent's resulting speed is obtained in a way that is similar to the schematic shown in Figure 6. (v agent,abs ) differs from the function (θ agent,abs ) in the magnitudes of the two vectors, and in the final result of this function, which is the magnitude expressed in [m/s]. In such a way, the water current defies the agent's target movement, i.e. its target speed and direction described with its flow velocity and direction, which gives the resulting vector of the agent's movement. Both (θ agent,abs ) and (v agent,abs ) are in-built software functions, while (6) and (7) represent generalized equations used for describing the framework of the two functions.
The key parameter to be defined in order to calculate the time variability of the state variable -energy (1), is the agent's power (P) which is given by: where the above equation is derived from the expression for calculation of the fish drag force. The assumption is that (P) is equal to 0 when the agent's relative speed is equal to the flow velocity (v HD,h ). Considering that the agent uses flow velocity when it moves collinearly with water current (4), the energy loss occurs only when it uses continuous or burst swimming speed. The agent's final movement vector is defined by its components in a horizontal plane and vertical direction. The expressions for horizontal movement direction (9) and speed (10) are presented. The final link that is necessary for defining the agent's movement vector involves the vertical speed (v agent,V ), which is given by: (11) where ( The ABM calibration procedure refers to the development of the original ECOLab template that envelopes all the previously described mathematical expressions and functions. The template has been developed exclusively by the authors of this paper while this software is used as a tool for easier conduct of the simulations and focuses on the understanding of crucial processes. Due to the lack of data about fish swimming patterns through the fishways with similar configurations as the two analysed in this research, the ABM was verified based on previous studies that discuss the topic of fish movement in turbulent flows. Following the previous statement, the fish trajectories presented by Bermudez et al. [18] have been used in this paper for ABM verification, i.e. for the establishment of the final ECOLab template. It is important to note that most of the calibration procedure was related to defining the adequate search radius of the sensing functions. This parameter is related to the model spatial discretization where the final adoption of the radius value of 0.018 [m] leads to a rather smooth fish trajectory throughout the fishway. The increase in search radius causes interruptions in the agent's movement with mostly its retention downstream of the fishway baffles. However, the decrease of search radius leads to the unreasonable chaotic movement of the agent. Both increase and decrease of search radius cause higher energy consumption by the agent for covering the same distance, as well as the irregular shape of the trajectory in its upstream movement. Therefore, the search radius stands out as the most sensitive parameter in terms of final model results regarding the agent's trajectory and energy consumption. A total of 6 ABM numerical simulations were conducted (Table 1). A half of these simulations relates to the analysis of a model fish swimming through RS-90, and the other half through RS-45 fishway configuration. The initial location of the agent is assumed to be at the fishway entrance, i.e. on half of the downstream model boundary. The simulations differ for the two configurations in the TKE field obtained from the hydrodynamic model simulations, and the depth of the initial agent's location. Such an approach was applied to show the agent's trajectory during its movement through a particular σ-layer. The computational time step is fixed at 0.05 [s].

Results and discussion
The depth-averaged flow velocities are presented in Figure 7, while the vertical profile of TKE in the fishway is provided in Figure 8. Comparison of the measured and modelled depthaveraged flow velocities for 9 measuring points (see Figure 2) is given in Figure 9 and Figure 10.  In general, RS-45 is characterised by higher velocities as can be seen in the depth-averaged U and V components as well as in the total current speed presented below (Figure 9, Figure 10). The highest TKE is present at the bottom layer. Furthermore, the vertical section of the TKE in the fishway through vertical slots shows higher turbulence within the RS-45 configuration, i.e. TKE is present more than in the RS-90. However, in the mid and surface layers higher TKE is present for the RS-90 configuration (Figure 8, Figure 11- Figure 13).

Figure 10. Comparison of measured and modelled depth-averaged U and V velocity components, and total current speed for RS-45
The model calibration included a comparison of the depthaveraged measured and modelled flow velocities for nine measuring points in both fishway configurations (Figure 2). This analysis included calculation of the two statistical parameters for estimating the level of numerical model certainty -mean absolute error ∆ ABS (12) and coefficient of determination R 2 (13) [33]: where y measured represents measured, and y model modelled U and V components, and the total flow velocity. Parameter i = 1, …, 9 represents the number of measuring points in the middle fishway pool. ∆ ABS and R 2 of depth-averaged U and V components, and the total flow velocity for both fishway configurations, are presented in the following table.
Based on the obtained flow velocity fields (Figure 7), comparison of the depth-averaged velocities (Figure 9, Figure 10), and calculated statistical parameters, it is evident that the RS-45 model provides results with a higher level of certainty than the RS-90 configuration. Both configurations provide more reliable results for U compared to the V flow velocity component. The most significant error is present for the RS-90 model results when measuring points near the left sidewall are revised (points 6-8, see Figure 2). However, it must be noted that significant turbulence is present in the flow, while the calibration was conducted based on the measurements at several points within the fishway. Also, a general overview of the flow pattern needs to be considered in the model calibration. Accordingly, the modelled flow field corresponds to the ones obtained in previous studies for similarly configured fishways [3,12,18,17,34,35].   The ABM simulation results were verified based on the movement patterns, reactions, and camera recorded trajectories of fish in altered flows that occur in vertical slot fishways. One of the studies [18] showed that fish can adapt to the ambient flow conditions, i.e. that their movement is adjusted according to the flow speed and direction. The assumption of the adaptive fish behaviour, as related to the change in movement speed based on the change in its environment, confirms one of the fundamental hypotheses of the cognitive fish aspect within the developed ABM. Fish movement patterns that were investigated earlier in the paper (see Introduction) can be recognized from the ABM simulation results (Figure 11 -Figure 14). The agent avoids the zones with high TKE in its upstream movement, i.e. the fish moves away from turbulent areas. Furthermore, the agent successfully uses burst speed to overcome the zones of high velocities such as flow contraction at vertical slots. On the other side, the agent successfully uses continuous speed during its movement in lowvelocity and stable flow zones. Also, in the areas where the flow is collinear with the agent's preferable movement direction, the agent uses the flow to reduce its energy consumption ( Figure 14). In the zones of rapid change in flow direction and velocity, the agent moves chaotically as can be seen in the RS-90 configuration results, when the agent moves from the middle to the upstream pool (Figure 13-up). The agent's more frequent use of the recirculation zones in RS-90 compared to RS-45 results in lower energy consumption for the first configuration. The agent's residence time in the fishway varies between 7 and 14 [s], as can be seen in energy loss diagrams for the two configurations. Also, the agent's time of stay in the RS-90 fishway is shorter ( Figure 14). The final time step at the two diagrams represents the moment when the agent leaves the model spatial domain, i.e. when it moves through the upstream model boundary. Considering the above, RS-90 represents a more favourable fishway configuration compared to RS-45. The lowest energy loss occurs when the agent is introduced into and moves through the surface, while the highest energy loss occurs for its movement in the bottom σ-layer. Such an energy loss pattern is expected as the highest turbulence occurs near the canal bed in both configurations. Considering that the RS-90 fishway model overestimates flow velocities near the left sidewall (measuring points 6-8, see Figure 2 and Figure 9), it can reasonably be stated that fish could expend even less energy than presented in this paper for such configuration. For the RS-45 configuration, the lowest amount of energy the fish expends in its movement through the surface layer, while swimming in the mid-layer, results in the highest energy consumption. However, it is important to note that the presented ABM simulation results should not be perceived as absolutely correct, but rather useful for comparing different fishway configurations.

Conclusion
Physical and 3D hydrodynamic numerical models were developed for two vertical slot fishway configurations while the numerical component was upgraded with the original ABM of the fish swimming upstream. The calibration and verification of the numerical model were conducted based on the measurements conducted on the physical model (9 measuring points in three layers). The results of past studies that examined fish behaviour in similarly configured fishways were used in the case of verification of the original ABM. The developed ABM considers the fish sensing ability as well as their cognitive aspect of reacting to the changes in their hydrodynamic ambient (flow direction and velocity, turbulence) in order to minimise the agent's energy consumption. Besides considering the abiotic factors, the basic motivation for the agent's movement is the assumed hormone field which is defined as a field monotonously increasing in the upstream direction. The results of six numerical simulations are presented in the paper, with one half relating to RS-90, and the other half to RS-45 configuration. The conducted simulations differ in the depth at which the agent is introduced into the model domain (surface layer, mid-layer, bottom layer). The fish movement pattern is recognized based on the obtained agent's trajectories, from which it can be seen that the fish avoids the zones of high turbulence by adjusting swimming direction and speed depending on the changes in flow direction and velocity. Also, the agent's disorientation in the zones of rapid flow direction and velocity change is successfully modelled with the developed ABM. Such fish behaviour was recognized in previous studies performed on live individuals in altered flows. The results also show that the fishway with a greater distance between pool baffles exhibits more favourable flow conditions in terms of reducing the fish energy consumption. Although such conclusion is rather obvious and expected, the presented results quantitatively confirm this hypothesis. The advantage of longer pools in a fishway lies in the occurrence of recirculation zones in which a reversed flow is present (flow vector directed upstream). It can therefore be concluded that a fishway with a smaller number of longer pools provides a more favourable hydrodynamic environment for fish than the configuration with a greater number of shorter pools. Past ABM studies were mainly related to the modelling of fish individuals as the agents with behavioural abilities who recognize the ambient abiotic factors, but the main objective was to find the fishway entrance. The results presented in this paper are rather useful in terms of improving the previous models and studies by applying the ABM principle in the analysis of fish swimming through differently configured fishways. Therefore, this paper gives an insight into another type of successful application of the ABM for solving real-life problems environmental engineering is concerned with. It is important to note that mutual interaction between fish and flow is not considered within the developed ABM, which is mainly related to neglecting the changes in flow pattern due to the presence of a fish body. This mechanism could be significant in the regions of contracted flow in fishways, such as at baffle openings. In these regions, the flow area is significantly smaller than the cross-section of the recirculation zones. During the fish movement through the openings, the active cross-section is reduced, which results in the change of ambient velocity field and hydrodynamic drag. Therefore, further research should be focused on the improvement of the current ABM with the above-described component. This paper presents a contemporary methodology for modelling typical physical processes on the example of a pilot model. It is certainly advisable to apply such an approach in fishway design procedures as an alternative to the formerly used conventional empirical templates.