Behavioral Assumptions in Models of Fish Movement and Their Influence on Population Dynamics

—This study investigates the movement and growth of cohorts in a coastal fish stock by simulating animal responses to spatial heterogeneity of biotic and abiotic conditions in a dynamic marine landscape. A coastal bay is modeled using spatial and temporal data on prey distribution, benthic habitat, depth, and salinity. Prey abundance and salinity vary daily through an annual cycle to create a spatiotemporally dynamic environment with seasonal fluctuations in the quality and distribution of habitats favoring growth. Three movement behaviors—random walk, kinesis, and gradient response via restricted-area search—simulate fish cohort movements in relation to environmental characteristics. A bioenergetic growth model is used to describe somatic growth by comparing spatiotemporally variable prey consumption rates and metabolic requirements. This facilitates evaluation of the way in which movement behavior influences the ability of cohorts to locate and occupy favorable habitats in a heterogeneous environment. Random movement behavior proved inefficient for locating preferable habitats and resulted in the lowest cohort growth trajectory and stock biomass per recruit. Kinesis and restricted-area search behaviors resulted in similar spatial distributions and characteristics of stock biomass when cohorts were initially distributed at random. However, the results from the restricted-area search simulations were highly sensitive to the initial positions of cohorts. The restricted-area search simulations also resulted in high variation in growth rates among cohorts, reflecting complex interactions between behavioral mechanisms and the structure of local heterogeneity. The results show that movement models reflecting similar density patterns can differ in their influence on cohort growth and mortality. In particular, the presence of local optima can bias the results of movement models employing directional responses to a gradient structure. These results underscore the importance of sound theoretical assumptions in movement model construction and suggest that minimalism be adopted in the absence of empirical support for behavioral assumptions concerning animal responses to environmental cues. Models that attempt to capture the spatiotemporal dynamics of fish stocks must include some specification of movement behavior to reproduce the distributions of individuals within stocks. The difficulties of directly observing fish movements and precisely measuring stock distributions complicate selection of an appropriate movement model. Phenomenological models, which recreate ob* Corresponding author: humstonr@vmi.edu 1 Present address: Virginia Military Institute, Department of Biology, Lexington, Virginia 24450, USA. Received February 18, 2003; accepted April 20, 2004 served spatiotemporal stock distributions without employing a mechanistic response to environmental conditions, offer little utility for predicting the responses of stocks to potential environmental changes. Statistical models of fish–habitat associations (see Guisan and Zimmermann 2000 for a review) provide some predictive utility when one is considering habitat alterations (Olden et al. 2002). However, assuming that a stock distribution is solely a function of habitat affinities excludes the influence of social interactions as well as aspects of habitat structure (patchiness, distance between patches, etc.) that combine with movement 1305 BEHAVIORAL MODELS OF FISH MOVEMENTS behavior to influence patterns of aggregation (Flierl et al. 1999). Models that explicitly simulate animal–habitat interactions through a hypothesized behavioral response (e.g., Dagorn et al. 2000; Humston et al. 2000; Huse 2001) are often difficult to parameterize and validate in addition to presenting the basic challenge of developing mechanistic models from principles of behavioral ecology (Kirby 2001). Modern programming and computational capabilities encourage development of complex behavioral models of movement to replicate observed distributions of stocks. However, the assumptions underlying the hypothesized behavior can be difficult to justify empirically, especially in a rule-based model (Humston et al. 2000; Kirby 2001). Models are often selected based on their ability to produce spatial patterns matching those that are expected or observed, with less consideration as to how the underlying behavioral mechanisms may influence the dynamics of stock production. In a simulation in which the spatial structure of the bio-physical environment is modeled explicitly, the choice of the models used to simulate the movement of organisms is paramount in maintaining a realistic simulation (Tyler and Rose 1994; Giske et al. 1998); improper model design can render simulation results inapplicable. Model structure dictates simulated behavior, in that movement behavior must function logically within the spatiotemporal scales of data resolution. Models should ideally be selected based on their mechanistic similarities to observed movement behavior rather than their ability to recreate spatial patterns, thus avoiding the pitfalls of inferring spatial processes from spatial patterns (Parrish and EdelsteinKeshet 1999). Large-scale movements and coarsescale distributions can be considered the composite results of many fine-scale behavioral ‘‘decisions’’ in which fish react to ambient conditions with respect to their environmental and physiological requirements (Flierl et al. 1999; Grunbaum 1999; Kirby 2001). It follows that even large-scale movements are modeled most realistically through an understanding of the proximate mechanisms determining movement behaviors at fine scales. Developing technologies in tagging and tracking allow study of fish movement behavior across many spatiotemporal scales. However, the functional limitations of tags and receivers can preclude their application in some systems or for certain species. Analysis of telemetry data to quantify behavioral response mechanisms is a burgeoning area of study (e.g., Newlands 2002; Newlands et al. 2004), but it is still limited in application. Such analyses require that data be collected on the biophysical environment experienced by free-living fish at a corresponding resolution during tracking (Kirby 2001), which can exceed technological and field observation capabilities. Development of movement models is therefore confronted with a lack of data describing fine-scale behavior. This does not imply that attempts at structuring firstorder movement models should not proceed. However, in such circumstances modeling approaches should employ proximate mechanisms that do not assume undocumented cognition or behavioral sensitivity (e.g., memory or spatial learning capabilities). Approaches should balance acceptable model performance and efficiency with the appropriate detail of underlying behavioral mechanisms. We designed spatiotemporal simulations to examine how differences in behavioral mechanisms can influence predicted stock size structure given the movement and growth dynamics. The simulation environment incorporated georeferenced data on benthic habitat, salinity, depth, and forage distribution in Biscayne Bay, Florida. The growth of individual cohorts of bonefish Albula vulpes was simulated following the principles of bioenergetics and predator–prey functional responses outlined in a previous study by Ault et al. (1999b). The choice of study area and species was motivated by concurrent field studies on the biology of bonefish; however, the representation of bonefish herein is as a ‘‘straw man’’ species. The movements of multiple cohorts were modeled as behavioral responses to biotic and abiotic habitat conditions, specifically those conditions that affect the growth rate in the bioenergetic model (forage density and the physical conditions influencing metabolism). The spatial patterns of forage distribution and salinity variation were dynamic in time; therefore, the growth of individual cohorts in the simulation depended on their ability to consistently locate suitable environmental conditions. Three movement models were employed to compare the stock growth trajectories predicted under different assumptions of behavioral movement. One model simulates random movement, with no behavioral response to environmental stimuli, thereby providing a baseline or control scenario. The other two are mechanistic models, each of which assumes a different suite of behavioral responses yet can produce comparable aggregations in favorable areas. As a control comparison, we also modeled stock growth under the assumption of no movement of cohorts. In simu-