A Model to Illustrate the Potential Pairing of Animal Biotelemetry with Individual-Based Modeling

Background. Animal biotelemetry and individual-based modeling (IBM) are natural complements, but there are very few published examples where they are applied together to address fundamental or applied ecological questions. Existing studies are often found in the modeling literature and draw opportunistically on small datasets collected for purposes other than the model application. Animal biotelemetry can provide the robust measurements that capture relevant ecological patterns needed to parameterize, calibrate, and assess hypotheses in IBMs; together they could help meet demand for predictive modeling and decision-support in the face of environmental change. Results. Using an exemplar, simple IBM that uses spatio-temporal movement patterns of 103 acoustic-tagged juvenile yearling Chinook salmon (Oncorhynchus tshawytscha) to quantitatively assess two migratory strategies smolts are hypothesized to use while migrating north through the plume of the Columbia River (United States of America), we nd that animal biotelemetry can provide the robust measurements that capture relevant ecological patterns needed to parameterize, calibrate, and assess hypotheses in IBMs; together they could help meet demand for predictive modeling and decision-support in the face of environmental change. Conclusions. and now mature elds Our hope that and effectively illustrate individual-based perhaps inspire new collaborations between between these elds, as evidenced by the opportunistic use of biotelemetry data in the preceding examples and publication of three of the papers in Ecological Modelling, a specialized journal whose audience may have limited overlap with the biotelemetry research community. In an effort to bridge this gap and highlight the application of animal biotelemetry to individual-based modeling for the biotelemetry community, we present a simple IBM that uses spatio-temporal movement patterns of 103 acoustic tagged juvenile yearling Chinook salmon (smolts) to quantitatively assess two migratory strategies smolts could use while migrating north through the plume of the Columbia River (USA). These strategies include (a) maximizing growth, or (b) selectively using local currents to minimize plume occupancy time, and thus exposure to high predation pressure in the plume region. A quantitative assessment of the patterns in cross-shelf distribution of in-silico smolts against cross-shelf distribution from the telemetered smolts reveal that the strategy of maximizing growth is plausible, but not the selective use of coastal currents to speed passage through a predator-rich region. This model description, and the basic analytical techniques, illustrate individual-based models for the biotelemetry community and can serve as a starting point for collaborations between biotelemetry researchers and individual-based modelers.

nearly every case, small extant datasets were used opportunistically with limited use of quantitative methods for evaluating simulated data against observations. Nabe-Nielsen et al. (13) and Liukkonen et al. (14) use biotelemetry data for IBM parameterization and calibration. Nabe-Nielsen et al., in a modeling framework for predicting the impacts of anthropogenic disturbances on animal movement and tness, used statistical and visual techniques to calibrate a correlated random-walk sub-model with data from a single dead-reckoning tagged North Sea harbor porpoise (Phocoena phocoena), as well as dispersal parameters from 25 satellite-tagged porpoises. Biotelemetry data were not used to corroborate the model. Similarly, Liukkonen et al. used satellite telemetry data to simulate movement patterns and characterize home range formation and spatial ecology of Saimaa ringed seal (Phoca hispida saimensis), with the stated intent to later extend the model to study the effect of changing environmental conditions and different conservation scenarios on population dynamics. They used satellite biotelemetry data from ve seals to parameterize and calibrate 'movement duration' and 'distance from home,' but used the same movement data for model validation (likely due to their small telemetry sample).
Bauduin et al. (15) and Ohashi and Sheng (16) extended the use of biotelemetry data to assessing alternative hypotheses. Bauduin et al. used VHF-based biotelemetry data from 35 Atlantic-Gaspésie caribou (Rangifer tarandus caribou) tracked between 1998 and 2001 to parameterize their IBM model and assess alternative movement hypotheses, including random walk, biased correlated random walk, foray loops to reproduce caribou extra-range movement patterns, and caribou delity during mating season.
Ohashi and Sheng assess ten alternative swimming strategies, including various passive, random, and active directed movements, of juvenile Atlantic salmon (Salmo salmar) migrating in the Gulf of St. Lawrence. They assessed their hypotheses by comparing travel times from juveniles tagged in 2009 and 2010 against travel times of simulated salmon implemented in an ocean circulation model with numerical particle tracking. Although the original dataset included 93 tagged juvenile Atlantic salmon released into the St-Jean River, travel time was measured from only three smolts that were subsequently detected at distant sub-arrays of acoustic receivers in the Gulf of St. Lawrence.
Morrice et al. also extend the use of biotelemtry data to assess alternative hypotheses. They investigate potential migration pathways of yearling and subyearling Chinook salmon (Oncorhynchus tshawytscha) in the Columbia River estuary using alternative movement models; passive drift, random walk, biased correlated random walk (BCRW), and taxis for yearling Chinook and passive drift, random walk, kinesis, and area search for subyearlings. These movement models, and the migration pathways that emerged, were assessed against detections of tagged yearling and subyearling Chinook. Morrice et al.'s model is notable for the large sample of tagged animals used: 3880 yearling Chinook salmon and 4449 subyearling Chinook salmon tagged and released in 2010.
A fuller realization of the promise of IBMs and biotelemetry would include biotelemetry researchers and individual-based modelers working together at the outset of a research project. One reason why biotelemetry data is not more widely used in individual-based modeling may be limited cross-fertilization between these elds, as evidenced by the opportunistic use of biotelemetry data in the preceding examples and publication of three of the papers in Ecological Modelling, a specialized journal whose audience may have limited overlap with the biotelemetry research community.
In an effort to bridge this gap and highlight the application of animal biotelemetry to individual-based modeling for the biotelemetry community, we present a simple IBM that uses spatio-temporal movement patterns of 103 acoustic tagged juvenile yearling Chinook salmon (smolts) to quantitatively assess two migratory strategies smolts could use while migrating north through the plume of the Columbia River (USA). These strategies include (a) maximizing growth, or (b) selectively using local currents to minimize plume occupancy time, and thus exposure to high predation pressure in the plume region. A quantitative assessment of the patterns in cross-shelf distribution of in-silico smolts against cross-shelf distribution from the telemetered smolts reveal that the strategy of maximizing growth is plausible, but not the selective use of coastal currents to speed passage through a predator-rich region. This model description, and the basic analytical techniques, illustrate individual-based models for the biotelemetry community and can serve as a starting point for collaborations between biotelemetry researchers and individualbased modelers.

Model Introduction
The Columbia River basin drains nearly 671,000 square kilometers of land, entering the Paci c Ocean at the border between the USA states of Oregon and Washington. The river creates a dynamic coastal river plume that juvenile yearling Chinook salmon (Oncorhynchus tshawytscha), 'smolts,' must navigate during the earliest stage of their marine migration (17). Previously, the plume was hypothesized to be a refuge from predators and a rich feeding ground (18), but more recent research suggests that the plume is predator-rich and with lower survival in the plume than adjacent estuarine and ocean habitats (19), (20), (21), (22), (23), (24). Due to the perceived importance of the early marine period, there has been a persistent interest in the interaction between the Columbia River plume environment and juvenile salmon migration and survival, particularly as to how migration and survival may be affected by river dynamics (25), (26), (27), (24).
Determining the effects of changing plume conditions on smolt migration and survival requires some understanding of the strategies smolts may employ during this phase of their migration. Two conceivable strategies are (a) to select habitat that maximizes growth or (b) navigate so as to minimize opposing currents during northward migration. The rst strategy would be expected to reduce availability to sizeselective predators while contributing to the growth necessary for them to survive their rst winter at sea. This strategy is also consistent with the critical size, critical period hypothesis which posits that the adult return rate of salmon is in uenced rst by predation-driven mortality at ocean entry, then by starvationdriven mortality during the following winter that results from a failure to build minimum energy reserves (28).
The second strategy, navigating to minimize opposing current while migrating north, would speed northward passage, reducing the period of exposure to predation in the plume region (assuming that survival rates were lower in the plume than farther north). Brosnan et al. (29) demonstrated that survival of groups of yearling Chinook released to migrate through the plume is negatively related to their travel time, so this strategy could be bene cial. This behavior has also been observed in the eld; stereovideographic studies in tributaries of the Fraser River reveal that adult sockeye salmon (Oncorhynchus nerka) migrating upriver selectively travel in lower current (30), and may exploit counter owing eddies (31). Juvenile salmon are believed to similarly exploit turbulent ow during their downstream migration (32), (33).
Individual-based models are a cost-effective means of evaluating which strategy produces results consistent with observed migration patterns. The model presented here was implemented in NetLogo v.5.2, a popular Java-based software for individual-based modeling (34) and uses simulation output (salinity, temperature, and current) from the Virtual Columbia River model developed by the Center for Coastal Margin Observation and Prediction. The IBM doesn't test the effects of different strategies on plume survival, but rather which, if any, of the plausible strategies reproduces migratory patterns observed in acoustic telemetry studies of juvenile yearling Chinook-a rst step in addressing the survival question.

Tagging and Telemetry
The model draws on detections of acoustic tagged juvenile yearling Chinook (Oncorhynchus tshawytscha) and oceanographic simulations from the 2009 spring outmigration. This dataset was chosen because tagged smolts were released from early-April through May, covering a wide range of spring ocean conditions. VEMCO V7-2L acoustic tags were surgically implanted in 1,370 hatchery-raised Columbia River Basin juvenile yearling Chinook smolts. Acoustic receivers were deployed in subarrays across the river at Astoria Bridge and across the continental shelf at Willapa Bay to detect the passage of tagged smolts. Smolts detected on both the Astoria and Willapa Bay sub-arrays (N = 103) were used to calculate plume residence time, and the cross-shelf distribution at Willapa Bay; the relatively small sample size of 103 tracked sh is a consequence of mortality losses and the need for each included smolt to be detected on both subarrays. Greater detail on the tagging procedures and acoustic array used to track tagged smolts can be found in (29) and (35).

Individual-based Model Description
The model was implemented in NetLogo and is described here using the 'Overview, Design concepts, and Details,' or ODD, protocol of Grimm et al. (36).

Purpose
The purpose of the model was to compare two hypothesized strategies smolts may use during their northward migration through the Columbia River plume, i.e., do smolts select habitat to maximize growth, or attempt to minimize opposing currents to more quickly transit through plume region?
Entities, state variables, and scales There were two entities in the model, ocean cells and in-silico smolts. Each ocean cell had six state variables: salinity, temperature, current velocities (x-and y-), water depth, and a representation of feeding conditions. The two types of in-silico smolts are de ned by their migratory strategy, maximizing growth (MaxGro) or minimizing the current opposing their northward migration (MinCur). Each smolt, regardless of strategy, had ve state variables: fork-length, weight, optimal swimming speed, heading, and a binary variable indicating whether the smolt was in estuarine (PSU < 27) or ocean (PSU ≥ 27) waters. Subsequently, each smolt calculated its weight-and temperature-dependent optimal swimming speed, evaluated whether it was in estuarine water (cell salinity < 27 PSU) or marine water (cell salinity ≥ 27 PSU) and moved accordingly (see below). Once all smolts executed their move, the model stepped forward.

Design Concepts
Pearcy (37) is widely credited with the hypothesis that the year class strength of returning salmon is established during the early marine life history, including the period of plume residency. Yearling Chinook departing the Columbia River must travel through the river plume, the dynamics of which are affected by hydropower-regulated river discharge and wind-driven currents and other oceanographic processes (38).
Survival in this region is presumably affected by the migratory strategy adopted by smolts, and insight into the manner that individual smolts interact with the plume environment while migrating, revealed by patterns in their distribution across an acoustic array, could be used to evaluate the impact of changing oceanographic and river conditions (39).
In this model, smolts adapt to changes in themselves and their environment by selecting a weight and temperature-dependent swimming speed and orienting towards habitat consistent with their migratory strategy. Implementing these behaviors requires the following assumptions: 1. Smolts are assumed to employ negative rheotaxis (orientation in the direction of current) to navigate through the model estuary and into the plume. This assumption is grounded in a nding that upregulation of the hormone thyroxine in response to changing light conditions stimulates negative rheotaxis, leading smolts to travel downstream (40).

2.
Smolts are assumed to have a compass sense and the ability to detect gradients in temperature, salinity, current, and prey availability. The speci c mechanisms by which smolts sense these variables are not modeled, but eld and laboratory observations indicate that smolts can sense gradients in the orientation and intensity of magnetic elds (41), ow (30) (42), temperature (43), and salinity (44). Their presence in trawls is also positively correlated with chlorophyll concentration and plankton abundance (45), (46), (47). 3. Smolts are assumed to begin a directed northward migration shortly after reaching the ocean, an assumption that is strongly supported by decades of U.S. and Canadian coastal trawl surveys (48), At every time step, the unique identi cation number, type, size, weight, local ow variables, heading, optimal swimming speed, and model coordinates (including latitude, longitude, and the corresponding grid cell) of each smolt were recorded. All model output analyses were conducted in R (52).

Initialization
The model was initialized with ow environment variables, water depth, and the simpli ed representation of the prey eld speci ed from (53). Initial smolt fork lengths were assigned by drawing randomly from fork lengths of tagged smolts detected at Astoria and Willapa Bay in 2009. The fork length (FL) at tagging was adjusted for growth between release and detection at Astoria by where FL tag is the measured fork length of the tagged smolt at the time of tagging, T riv is the tagged smolts travel time (d) from release to plume entry, and 1.05 (mm d -1 ) is an observed daily growth rate in the Columbia River (49).

Input Data
The ow environment data and depth for each ocean cell at each time step was derived from DB-22, a database of Virtual Columbia River (VCR) model simulations. The VCR was built by the Center for Coastal Margin Observation and Prediction (CMOP) using SELFE, an open-source, community-supported code designed for the effective simulation of 3D baroclinic circulation in the Columbia River estuary and plume that uses semi-implicit nite-element/volume Eulerian-Lagrangian algorithms to solve the Navier-Stokes equations on unstructured triangular grids (54).

Submodel 2: Optimal Swimming Speed
Optimal swimming speed was calculated using the formula described in (55) and parameterized in (56): Where T is temperature and a state variable of the cell and W, weight, is a smolt state variable. ACT is a parameter from the bioenergetics submodel described below.

Submodel 3: Bioenergetics
In bioenergetics models, growth over time is estimated using a simple mass balance equation, growth = consumption -(respiration + egestion + excretion). This submodel uses the Wisconsin bioenergetics equation sets (57); Table 1) for consumption (eqn. 3), respiration (eqn. 1), egestion and excretion (eqn. 2). The equations are parameterized with values from the literature on Paci c salmon bioenergetics ( Table  2). Prey energy density, Q f , is a single global parameter. All smolts use this submodel to grow. The growthseeking model smolts use it to identify and move at a metabolically optimal speed towards habitat (grid cells) where growth opportunities are greatest.
Sensitivity analysis revealed the bioenergetics submodel to be sensitive to two parameters, prey energy density and proportion of maximum consumption, which is consistent with the ndings of similar analysis reported in (55) and (58) (following (59)), as well as (60). When the bioenergetics submodel was validated by simulating migrations in the model world with bioenergetics parameters from Table 2, model smolts grew at rates consistent with a range of reported growth rates for juvenile yearling Chinook observed in the Columbia River plume region (61).

Submodel 4: Prey
A submodel representing declining prey availability with distance from shore was speci ed from Peterson In marine waters, smolts set their heading towards the cell north of their position with maximum growth opportunity or minimum southward owing current that it could reach in one hour at its optimal swimming speed. Smolts moved to the terminal point of a vector that was addition of their movement vector (heading, optimal speed) and the current vector in the cell they occupied. If the smolt's nal position was outside the north, south, or west bounds of the model, the smolt was considered to have permanently emigrated, otherwise it grew according to the bioenergetics submodel. In the estuary, smolts were not permitted to swim east (upriver) of the model boundary.

Submodel 6: Simulated detections
Detections of model smolts on the Willapa Bay subarray receivers were simulated by evaluating smolt paths (Euclidean lines drawn between smolt positions at each model step) to determine if they overlapped any of the detection zones centered on each receiver. The estimated detection range of the acoustic tags was 400 meters. For each smolt path that intersected a receiver's detection radius, the receiver number, smolt number, and time of detection were recorded. Model detections on virtual receivers corresponding to those that were lost during the 2011 season were recorded, but removed prior to analyzing the cross-shelf distributions against observed values.

Model Output Analysis
Model output analyses, including simulated detections, were conducted in R. The cross-array distribution of detections of tagged smolts and model smolts was compared using a modi ed Cramer von-Mises test (64), where the null hypothesis is that there is no difference in the distributions. Syrjala's (64) test was calculated using the R package ecespa, and summary circular statistics using circular.

Model Results
At a signi cance level of 0.05, there was no difference in the distributions of the MaxGro smolts and tagged smolts (p-value = 0.54; Figure 2). There was a signi cant difference in the distributions of the MinCur smolts and tagged smolts (p-value = 0.02; Figure 2). Model smolt tracks ( Figure 3) reveals a more bifurcated pattern in the MinCur types tracks; they traveled close inshore when wind-driven transport was easterly (downwelling), and offshore when westerly (upwelling). MaxGro tracks showed a similar pattern, but their sensitivity to modeled feeding conditions resulted in their being retained on the shelf, and more evenly distributed. The number of smolts crossing the virtual subarray at Willapa Bay was sensitive to the magnitude and direction of wind-driven coastal transport.

Discussion Of The Individual-based Model
The results from the individual-based model suggest that smolts may pursue a strategy of maximizing growth upon beginning their northward migration. Under the critical size, critical period hypothesis, which posits that adult returns are affected rst by predation at marine entry, and then starvation during the rst winter, this strategy could improve the probability of survival if smolts grow to exceed the gape size of predators and attain su cient energy reserves to avoid winter starvation. Additionally, the model indicates that strong westward transport advects smolts off the shelf; a prediction that cannot currently be validated, but could affect survival. Pearcy (37) noted that alongshore or offshore displacement of juveniles immediately after ocean entry could reduce predation and enhance survival. Findings by Brosnan et al. (29) that travel time affects plume survival supports this notion.
However, while it might seem that the selective use of alongshore current by model smolts employing the MinCur strategy would speed passage out of the plume region and enhance survival, this behavior proved to reduce travel time to Willapa Bay by only a few hours relative to the MaxGro strategy. Applying the survival equation used in (29) to estimate survival to the Willapa Bay subarray using median travel times suggests that survival would be virtually indistinguishable between the growth maximizing smolts and current minimizing smolts (0.60 v 0.61, respectively) that reach the array. This suggests there may be little bene t to the active use of favorable current gradients in the plume region.
In addition to the apparently limited survival bene t of selective use of coastal currents in the plume region, there are additional reasons why this pattern may not extend beyond the river. First, the high density of salmon in the river raises the costs of competing for limited food resources and salmon may restrict their feeding and perform the river migration on a limited energy budget (65), (40); but see (66) for evidence that smolt feeding increases in the lower Columbia River. Thus, there may be strong pressure to minimize the cost of migration in the river. Conversely, the early marine environment represents a rich feeding ground where a delay in feeding could result in smolts failing to meet the energy requirements to survive their rst winter at sea (28).
Second, the con ned river environment provides turbulent ow and relative motion cues from the riverbed and similar features that salmon demonstrably respond to (30), (42). Beyond the tidal plume, these cues are weakened. During the spring, ow in the lower Columbia River is approximately 1 m s -1 (and can be much greater) and exceeds 3 m s -1 in the tidal out ow at the river mouth, whereas ambient coastal current velocity is approximately 0.1 m s -1 (67), (25). Shear turbulence is likely reduced in the coastal ocean, and smolts are not positioned to use the seabed as a frame of reference (51). Conceivably, they can detect the ne-scale changes in the intensity of turbulence and/or their motion relative to celestial objects and the earth's magnetic eld. This would allow them to take advantage of small, favorable gradients in coastal current. However, they would require very sensitive absolute and differential thresholds to these cues, which have not been determined experimentally.
The successful transition from river to ocean requires that smolts complete a number of changes (i.e., smolti cation). The precise trigger, or series of triggers, during this complex process that result in the behavioral transition to directed northward migration in the ocean have not been determined. It may be driven by hormonal changes prompted by environmental cues, much as up regulation of thyroxine triggers downstream migration (40). We attempted to capture the transition via a simple thresholddelayed response model, but a mechanistic understanding of the process might explain why smolts are occasionally detected migrating at least as far south as Cascade Head, OR, before turning and swimming north (68).
Incorporating a representation of coastal feeding conditions acted to contain the MaxGro smolts on the shelf, but the fact that nearly 20% of the MaxGro smolts still migrated outside the bounds of the subarray at Willapa Bay is a potentially interesting result. Burke et al. (69) report similar results, although they interpret it as a modeling problem and describe using an Ornstein-Uhlenbeck process that forced model sh to orient towards the historic centers of mass of juvenile salmon sampled across the shelf in an attempt to correct it. Trawl surveys and acoustic receiver subarrays terminate near the shelf break because few or no sh are caught in trawls near the shelf break. Results from this model suggest that there could be a bi-modal distribution of catches/detections that would not be detected under the current sampling regime.
As Pearcy (37) notes, this could have implications for early marine survival. Columbia River basin smolts driven off-shelf at ocean entry might experience reduced predation pressure, particularly from seabirds nesting near the river mouth (19), (24). Their feeding opportunities would likely be reduced (45), (53), but directed northward migration would lead them to quickly regain the shelf environment, which arcs westward north of the Columbia River. Outmigrating yearling Chinook would be placed to take advantage of this offshore transport and potentially reduce their risk of predation since the necessary conditions occur in early spring, when the transition to the upwelling season is beginning and winds are strong.
A second outcome of the model was that negative rheotaxis was su cient to guide smolts out of the estuary. The initial expectation was that the tidally-reversing ow of the Columbia River would signi cantly delay, or prevent, progress towards the ocean, whereas McInerney (70) reasoned that preference for increasing salinity could serve as an orientation mechanism for smolts in the estuary.
However, whether salmon use salinity to orient remains unresolved (12) and the two mechanisms were compared during model development. Both produced similar migratory patterns, although orientation by negative rheotaxis resulted in smolts initially traveling further offshore under west transport conditions than did orientation by salinity. Negative rheotaxis was selected for the nal model because it has a known biochemical basis wherein increasing day length stimulates the upregulation of thyroxine, which in turn causes smolts to orient with the current (40).
Railsback et al. (71) proposed a condition-based movement rule for stream sh in IBM's, termed expected survival, where sh move to habitat where the probability of surviving non-starvation mortality risks, multiplied by the probability of surviving starvation risk over a pre-determined time horizon, is greatest.
Yearling Chinook in the marine environment appear to respond to feeding conditions during northward migration, so applying Railsback et al.'s (71) expected survival, using the rst-winter critical period as the time horizon, may be a fruitful development of the IBM presented here. This would require collaboration among researchers who have collected data on salmon predators, salmon prey, and early marine survival to develop a more accurate representation of predator and prey elds than are currently captured by the IBM; such data are now becoming available (e.g., (24)).

Conclusion
The foregoing model is intended to illustrate the pairing of animal biotelemetry data and individual-based models by highlighting simple ways that biotelemetry data can be used for better informing the development of an IBM. Conditions appear ripe for a productive pairing of individual-based modeling and animal biotelemetry. First, although the body of literature describing IBMs implemented with biotelemetry data is limited and relies largely on repurposing of existing small datasets (but see (11) for re-use of a much larger biotelemetry dataset), it is a promising sign of active interest among individual-based modelers in using biotelemetry data. Second, national and international data assembly, archiving, and access efforts such as the U.S. Animal Tracking Network (atn.ioos.us), the Ocean Tracking Network (oceantrackingnetwork.org), and the International Movebank animal tracking database (www.movebank.org) will make more data discoverable and accessible. This will create opportunities to aggregate datasets su cient for robust IBM parameterization, calibration, and hypothesis testing. Still, the fact that animal biotelemetry data is expensive to collect and requires experience and specialized skills for most elements of array design and deployment, tagging, and data processing and analysis will still require partnerships for prospective projects. Finally, as individual-based modeling and animal biotelemetry have matured, many of the challenges of implementing a successful biotelemetry-supported IBM activity have been identi ed and overcome. Researchers have access to a wealth of methodological information in specialized journals and textbooks (1) (72)

Declarations
Ethics approval and consent to participate: Not applicable. Consent for publication: Not applicable.
Availability of data and materials: Acoustic telemetry data used in this study are available from Kintama Research Services upon request. Virtual Columbia River simulation outputs used in this study are available from IGB upon request and with permission of the Center for Coastal Margin Observation and Prediction.
Competing interests: The authors declare that they have no competing interests.