NEW MODELS OF ANIMAL MOVEMENT
By
ANDREW M. HEIN
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA 2013 ⃝c 2013 Andrew M. Hein
2 To my parents, brothers, and sister
3 ACKNOWLEDGMENTS I want to begin by thanking my committee chair, Jamie Gillooly, for his guidance, encouragement, and his infectious enthusiasm for ideas. I will continue to strive to emulate his willingness to consider any scientific question without being intimidated by paradigm. I also want to thank my committee co-chair, Scott McKinley, for his constant willingness to collaborate and for his commitment to rigorous logic in science.
The afternoons spent at his chalk board have been among my most educational and enjoyable experiences as a graduate student.
The work presented in this dissertation benefitted greatly from discussions with my committee members Doug Levey, Bob Holt, and Jose Principe, and also with Mary Christman and Ben Bolker. Individual chapters were greatly improved by comments from S. P. Vogel, T. Bohrmann, A. P. Allen, and J. H. Brown, J. Casas, M. Vergassola,
I. Couzin, A. Brockmeier, E. Kriminger, and many others. I am very grateful for funding from a University of Florida Alumni Fellowship, a National Science Foundation Graduate Research Fellowship under Grant No. DGE-0802270, and the National Science
Foundation under Grant 0801544 in the Quantitative Spatial Ecology, Evolution and
Environment Program at the University of Florida.
I could not have completed this work without the encouragement and support of my family and friends. I especially want to thank my brother, Luke. I also owe special thanks to Gabriela Blohm, who spent many long hours discussing ideas with me and exhibited a saintly patience when I had a new idea or discovery that I could not help but share with someone. Finally, I want to thank my parents: my father, for encouraging my philosophical tendencies, and my mother for always reminding me of the right to pursue my curiosity.
4 TABLE OF CONTENTS page
ACKNOWLEDGMENTS ...... 4
LIST OF TABLES ...... 8
LIST OF FIGURES ...... 9 ABSTRACT ...... 10
CHAPTER
1 INTRODUCTION ...... 12
1.1 New Models of Animal Movement: Constraints of Physics, Constraints of Information ...... 13 1.2 Biomechanics, Energetics, and Animal Migration ...... 14 1.3 Sensory Information and Models of Animal Movement ...... 15 1.4 Linking Movement Behavior and Encounter Rates of Interacting Species . 16
2 ENERGETIC AND BIOMECHANICAL CONSTRAINTS ON ANIMAL MIGRATION DISTANCE ...... 17
2.1 Model Development ...... 18 2.1.1 Parameterizing Model for Walking, Swimming, and Flying Migrants 19 2.1.2 Model Predictions ...... 22 2.2 Materials and Methods ...... 22 2.3 Results ...... 24 2.4 Discussion ...... 26
3 SENSING AND DECISION-MAKING IN RANDOM SEARCH ...... 33
3.1 Model Development ...... 34 3.1.1 Searching Without Olfactory Data ...... 36 3.1.2 Incorporating Olfactory Data to Make Search Decisions ...... 37 3.1.3 Interpreting Scent Signals ...... 38 3.2 Materials and Methods ...... 39 3.2.1 Scent Propagation ...... 39 3.2.2 Simulation Details ...... 40 3.3 Results ...... 40 3.3.1 Visual-Olfactory Predators Find Targets Faster and More Reliably Than Visual Predators ...... 40 3.3.2 Visual-Olfactory Predators Learn From No-Signal Events ...... 42 3.3.3 Visual-Olfactory Predators Concentrate Search Effort Near Targets 43 3.4 Discussion ...... 43
5 4 SENSORY INFORMATION AND ENCOUNTER RATES OF INTERACTING SPECIES ...... 50
4.1 Materials and Methods ...... 51 4.1.1 Encounter Rate and Search Behavior: Some Definitions ...... 51 4.1.2 Framework for Modeling Movement Decisions ...... 52 4.1.2.1 Sensory signals and search behavior ...... 52 4.1.2.2 Perfect sensing and response ...... 53 4.1.2.3 Purely random search ...... 54 4.1.2.4 Imperfect sensing and response ...... 55 4.1.3 Encounter Rate Simulations ...... 56 4.1.4 Estimation of Scaling Regimes and Exponents ...... 57 4.2 Results ...... 58 4.2.1 Encounter Rates of Purely Random Predators are Near-linear in Prey Density ...... 58 4.2.2 Encounter Rates of Signal-modulated Predators Change Nonlinearly with Prey Density ...... 59 4.2.3 Sensory Response Allows Predators to Encounter Nearby Targets more Frequently ...... 60 4.3 Discussion ...... 60
5 CONCLUSIONS ...... 67
APPENDIX
A MIGRATION MODEL DERIVATION, SENSITIVITY, AND STATISTICAL ANALYSES ...... 71
A.1 General distance equation ...... 71 A.1.1 Walking ...... 71 A.1.2 Swimming ...... 72 A.1.3 Flying ...... 72 A.2 Parameter estimation and model sensitivity ...... 74 A.2.1 Estimation of p0 ...... 74 A.2.2 Sensitivity analysis ...... 74
B DERIVATION OF DISTRIBUTIONS, A NOTE ON THE USE OF BAYES’ RULE, AND SUPPLEMENTARY SIMULATION RESULTS ...... 83
B.1 True Distance Distribution (TDD) and a Comment on the Use of Bayes’ Rule ...... 83 B.2 Robustness of Results to Search Conditions ...... 84 B.2.1 Target Density ...... 84 B.2.2 Signal Emission Rate ...... 84 B.2.3 Variation in Predator Scanning Times ...... 85 B.3 The Role of No-signal Events ...... 85
6 C MODEL OF SCENT PROPAGATION AND DEPENDENCE OF REGIME TRANSITIONS ON SIGNAL PROPAGATION LENGTH ...... 90
C.1 Scent Propagation ...... 90 C.2 Dependence of Regime Break on Signal Propagation Length ...... 91 C.3 Encounter Rate of a Predator with Perfect Sensing and Response, and Non-Zero Encounter Radius ...... 91 C.4 Encounter Probabilities in the Sparse Regime ...... 92
REFERENCES ...... 95
BIOGRAPHICAL SKETCH ...... 112
7 LIST OF TABLES Table page
A-1 Empirical values of the normalization constant ...... 75
A-2 Sensitivity of distance equations to variation in input parameters...... 76
A-3 Body mass and migration distance data ...... 76
8 LIST OF FIGURES Figure page
2-1 Schematic of migration process ...... 30
2-2 Migration distances ...... 31
2-3 Number of body lengths traveled ...... 31 2-4 Observed and predicted migration distances ...... 32
3-1 Schematic of predator search behavior ...... 46
3-2 Mean predator search times and variability about mean search time ...... 47
3-3 Typical search paths of simulated predators ...... 48 3-4 Information gain as a function of the ratio of visual to olfactory radius ...... 49
3-5 Area-restricted-search behavior of visual and visual-olfactory predators .... 49
4-1 Perfect sensing and response ...... 63
4-2 Scan points during search ...... 64
4-3 Encounter rates of purely random and signal-modulated predators ...... 64 4-4 Encounters rate of signal-modulated predators ...... 65
4-5 Empirical encounter probability as a function of target density ...... 66
B-1 Searchs time at low density ...... 87
B-2 Search times with reduced emission rate ...... 87 B-3 Search times and scanning phase length ...... 88
B-4 Likelihood funcions ...... 89
B-5 Search time with conditional response to olfactory signals ...... 89
C-1 Breakpoint between linear and sublinear regime ...... 94
9 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy NEW MODELS OF ANIMAL MOVEMENT
By
Andrew M. Hein
August 2013
Chair: James F. Gillooly Cochair: Scott A. McKinley Major: Zoology
Movement is an iconic feature of life; microorganisms swim up chemical gradients, motile predators search their environments for prey, and migratory animals make journeys that can take them across the planet. Advances in biomechanics and sensory biology have created opportunities to develop new mathematical models of animal movement that incorporate organismal biomechanics and sensory physiology. Such models are useful for understanding the ecological and evolutionary drivers of animal movement behavior, and also for predicting basic ecological rates and scales–for example, the rate of interactions among moving predators and their prey, or the spatial scale of movements made by seasonal migrants. This Dissertation is an attempt to develop such general models, and to use them to learn about both the origins and the implications of animal movement behavior. In Chapter 2, I began by investigating the physical constraints related to one of the most well studied movements that animals make: migration. I used a mathematical model to show how body mass influences the maximum distances that migrants travel through its effect on locomotion. I confirmed model predictions using a new global-scale dataset of animal migration distances. In Chapter 3, I sought to better understand how to model animal search behavior in the presence of noisy sensory signals, and how sensory information might affect the movement behavior of a searching animal. I developed a new mathematical framework for modeling the use of sensory data in movement decision-making. Results showed
10 that even a minimal capacity for sensing can give rise to movement behaviors that are commonly observed in nature, such as concentrated search effort near prey. Finally, in Chapter 4, I studied how movement behavior of searching animals changes as the density of their targets change. This work revealed that the ability of animals to gather and respond to sensory information can enable them to encounter prey at rates that differ fundamentally from those predicted by encounter rate models that ignore the use of sensory data.
11 CHAPTER 1 INTRODUCTION
The phenomenon of movement in general, and animal movement in particular, has
fascinated biologists for centuries (e.g. [1, 2]). Traditionally, animal movement has been studied either through detailed empirical work on particular species, or through highly abstracted mathematical models. Only recently, advances in fields such as sensory biology and biomechanics are beginning to facilitate the integration of organismal biology and mathematical theory of animal movement behavior.
Despite a rich history of investigation by theoreticians, many of the general mathematical models used to describe animal movement at the macro-scale rely on assumptions that are somewhat restrictive. For instance, some of the earliest models of animal movement were adopted from particle collision models in chemistry and used to predict encounter rates between predatory animals and their prey. These classical encounter rate models, developed by the pioneering theoretical biologist, Alfred Lotka and others, assume that predators and prey move randomly and independently of one another [3]. Lotka himself noted the inconsistency between this conception of
animal movement behavior, and the movement of animals in nature [3]. Of course,
generality often comes at the price of strong assumptions and the willingness of early ecologists to pay that price led to an enormous amount of development in the fields
of spatial ecology and coupled population dynamics (e.g., [4]). Still, ecologists like
Lotka and the visionary theoretician, John Skellam, imagined future work on animal
movement that would relax some of their own simplifying assumptions to allow for more
realistic depictions of organismal physiology and behavior [3, 5]. Accomplishing this goal requires an understanding of the elements of physiology and decision-making behavior
that are most relevant to animal movement. Since the development of early movement
models, researchers working in the areas of biomechanics and sensory biology have
made huge strides toward understanding the energetics of locomotion and the physics,
12 transmission, and processing of sensory signals. Developments in biomechanics theory, for example, have made it possible to write equations for the energetic costs of
locomotion as functions of speed and body size (e.g., [6, 7]). Empirical and theoretical
studies of sensory biology have gone a long way toward revealing how animals use
information to make movement decisions (e.g., [8–11]). These advances provide first-principles from which to derive new models of animal movement. In the chapters
that follow, I describe my attempt to contribute such models, and to use them to learn
about both the origins and the implications of animal movement behavior.
1.1 New Models of Animal Movement: Constraints of Physics, Constraints of Information
The way an animal moves around its environment must be determined, at least in part, by both the physical context of that movement and the background of information the animal has at its disposal. To better understand physical and informational constraints on animal movement, my collaborators and I have performed three theoretical studies to characterize these constraints in some generality. In Chapter
2, I describe our investigation of the biomechanical and energetic constraints related to one of the most well-studied movements that animals make: migration. We show how body mass–a fundamental characteristic of all animals–influences the maximum distances that migrating animals travel, through its effect on the physics of locomotion.
The data and models that we develop demonstrate that the dominant effect of body mass on migration distance emerges despite the differences among migratory species. One of the most interesting results of this study is the prediction that walking migrants of all sizes travel, on average, the same number of body lengths during migration (about
1.5 ×105 body lengths), as do swimming species of all sizes (1.7 × 106 body lengths).
Interestingly, this relationship does not hold for flying migrants, and the biomechanics of flight provide an explanation for this difference. A second problem is understanding how
to model animal search behavior in the presence of sensory signals [12]. Researchers
13 studying search and foraging movements have traditionally modeled movement using random walks. There has been much debate about what the most appropriate random walk models are. The assumption that underlies much of this work is that animals cannot get much useful information about the locations of their targets when target density is low. Thus, an animal must adopt some sort of statistical movement behavior that does not depend on the use of sensory cues [13]. In Chapter 3, we re-evaluate this assumption using a simulation model. In particular, we study the case of a searching predator that can measure only noisy olfactory cues from prey. We show that, so long as the range at which the predator gets noisy sensory data from prey is longer than the range at which it can capture prey, the predator can benefit tremendously from incorporating even minimal sensory data into movement behavior. We further show that a capacity for sensing and decision-making gives rise to commonly observed behaviors such as area-restricted search in regions that contain prey [14, 15]. A third and final question relates to how features of an animal’s environment influence its movement behavior. In Chapter 4, we study how movement behaviors of searching animals change as the density and spatial configuration of their targets change. Using simple mathematical models of sensing and decision-making along with simulations, we study the relationship between searcher-target encounter rate, and target density. The resulting relationships differ from classical mass-action models of species interactions, but are consistent with recent empirical data on prey encounter rates of predatory birds and fish. This study reveals the strong links between sensory data, movement behavior, and encounter rates of interacting species. Below I elaborate on the motivations for, and
findings of these investigations before describing them in full detail in Chapters 2-4.
1.2 Biomechanics, Energetics, and Animal Migration
Animal migration is one of the great wonders of nature, but the factors that determine how far migrants travel remain poorly understood. To address this issue, we develop a new quantitative model of animal migration and use it to describe the
14 maximum migration distance of walking, swimming and flying migrants. The model combines biomechanics and metabolic scaling to show how maximum migration distance is constrained by body size for each mode of travel. The model also indicates that the number of body lengths travelled by walking and swimming migrants should be approximately invariant of body size. Data from over 200 species of migratory birds, mammals, fish, and invertebrates support the central conclusion of the model that body size drives variation in maximum migration distance among species through its effects on metabolism and the cost of locomotion.
1.3 Sensory Information and Models of Animal Movement
Many organisms locate resources in environments in which sensory signals are rare, noisy, and lack directional information. Recent studies of search in such environments model search behavior using random walks (e.g., Levy walks) that match empirical movement distributions. We extend this modeling approach to include searcher responses to noisy sensory data. We explore the consequences of incorporating such sensory measurements into search behavior using simulations of a visual-olfactory predator in search of prey. Our results show that including even a simple response to noisy sensory data can dominate other features of random search, resulting in lower mean search times and decreased risk of long intervals between target encounters. In particular, we show that a lack of signal is not a lack of information.
Searchers that receive no signal can quickly abandon target-poor regions. On the other hand, receiving a strong signal leads a searcher to concentrate search effort near targets. These responses cause simulated searchers to concentrate search efforts near targets. This area-restricted search [15] behavior is a dominant feature of search movements of real predators such as oceanic birds [14, 16], which appear to use sensory signals to focus search efforts in productive areas and to avoid areas that lack prey. The model thus reveals that qualitatively realistic movement behavior can emerge even from very simple sensing and decision-making.
15 1.4 Linking Movement Behavior and Encounter Rates of Interacting Species
Most mobile animals search for resources, mates, and prey with the aid of sensory cues. The searching animal measures sensory data and presumably adjusts its search behavior based on those data. Yet, classical models of species encounter rates assume that searchers move independently of their targets. The assumption of independent movement leads to the familiar encounter rate kinetics used in modeling species interactions. Here, we use the example of predator-prey interactions to study how encounter rates change when predators use sensory information to find prey. We show that, even when predators pursue prey using only noisy, directionless odor signals, the resulting encounter rate equations differ qualitatively from those derived by classic theory of species interactions. Critically, predator sensory response lowers the sensitivity of encounter rate to prey density when prey density is low. This finding holds over a wide range of assumptions about predatory sensory capabilities, prey capture behavior, and the degree to which prey are clustered in the environment. Our results demonstrate how the exchange of information among interacting organisms can fundamentally alter the rates of physical interactions in biological systems.
16 CHAPTER 2 ENERGETIC AND BIOMECHANICAL CONSTRAINTS ON ANIMAL MIGRATION DISTANCE Each year, diverse species from around the planet set out on migrations ranging
from a few to thousands of kilometers in length [17–19]. Biologists have long hypothesized
that this variation in migration distance among species might be governed by differences
in basic species characteristics such as morphology and body size [1]. Although much
progress has been made in understanding how these characteristics are related to the mechanics of locomotion and to the migratory capabilities of individual species (e.g.
[20, 21]), success in understanding variation in migration distance among species has
been limited. This is because current models often require detailed information on the
morphology and behavior of migrants (e.g., [20, 22] ). This requirement has precluded a
quantitative analysis to determine the extent to which shared functional characteristics such as body size could be responsible for observed variation in migration distances
among species. As a result, the need for general theory and cross-species analyses of
migration has been strongly emphasized in recent years [23, 24].
Here, we present a model to describe constraints on animal migration distance. The model expands on past approaches [7, 25, 26] by incorporating (1) the body
mass-dependence of the cost of locomotion, (2) dynamic changes in the body masses
of migrants as they utilize stored fuel and (3) scaling of morphological characteristics
and maintenance metabolism among migrants of different body masses. In contrast
to past approaches, the model assumes that the number of re-fuelling stops made by migrants is unknown and may vary substantially among species. This facilitates
This chapter appeared as an article in the journal, Ecology Letters: Hein, A. M., C. Hou, and J. F. Gillooly. 2012. Energetic and biomechanical constraints on animal migration distance. Ecol. Lett. 15:104–110. Its reproduction here is authorized under the journal’s copyright policy.
17 prediction of statistical patterns of migration distance among species, even when the details of migratory behavior of individual species are unknown.
2.1 Model Development
We treat migration as a process in which a migrant travels a distance of YT (km) by breaking the journey into a series of N legs of length Yi , where i ∈ {1, 2, ..., N}
, Fig. 2-1A). Describing variation in migration distance among species thus requires
describing the processes that determine Yi , while accounting for among-species
variation in N. To accomplish this, we begin by making four simplifying assumptions (see Appendix A for detailed derivation and alternative assumptions). We assume (i)
that the total rate of energy use by a migrating animal, Ptot (W), is the sum of the rate
of energy use for general maintenance, Pmtn, and that required for locomotion, Ploc
(i.e. Ptot = Pmtn + Ploc = −dG/dt, where G = Joules of stored fuel energy), (ii) that migrants using a particular mode of locomotion are geometrically similar, such that linear
morphological characteristics (e.g. lengths of appendages) are proportional to M1/3 and
surface areas are proportional to M2/3 (where M is body mass (kg),[27] (iii) that migrant
metabolism provides the power required for locomotion, and (iv) that the number of
refueling stops made by individuals of each species is independent of body mass. During any given leg of a migration, the rate of change in migration distance per
unit change in body mass can be expressed as dYi /dM = (dYi /dt)(dtc/dG) =
−1 −vc/(Pmtn + Ploc ), where v is travel speed (m s ) and c is the energy density of
stored fuel (Joules kg−1). The distance traveled on a particular leg can be obtained by
integrating this expression from initial mass at the beginning of the leg, M0 (kg), to final
mass after all fuel energy has been used, M0(1 − f ), where f is the ratio of initial fuel M mass to 0, ∫ M0(1−f ) −v(M, )c Y = β dM. i P (M) + P (M, ) (2–1) M0 mtn loc β
Here, v, Pmtn, and Ploc have been rewritten to show their dependence on body mass
and on a small set of morphological traits, β (lengths and surface areas, e.g. wingspan,
18 body cross-sectional area), which determine the energetic cost of locomotion. This formulation allows for changes in speed and rate of energy use as the migrant loses
stored fuel mass.
Equation (2–1) can be used to predict how Yi varies among species by specifying
appropriate functions for v(M, β), Pmtn(M), and Ploc (M, β). We assume that Pmtn scales
3/4 with body mass as Pmtn = p0M , both within and among individuals, where p0 is a
normalization constant that varies by taxon [28, 29]. Biomechanics theory provides a
means of expressing Ploc and v as functions of M and β for migrants using a particular mode of locomotion (see below).
Generalizing to multi-leg migrations. Total distance traveled over the course of ∑N migration is given by the sum, i=1 Yi , where N is the number of migratory legs traveled by a given species (Fig. 2-1). N is unknown for the majority of migratory species.
To account for variation in N among species, we treat N as a random quantity with expected value, N . We treat Yi as fixed for a given species because we are interested in maximum migration distance. Following the law of iterated expectation, the expected distance traveled over N migratory legs is [ ] ∑N YT = E Yi = NYi , (2–2) i=1
where the operator, E, denotes the expected value [30]. Equation (2–2) shows that YT is
proportional to Yi , which is given by Equation (2–1).
2.1.1 Parameterizing Model for Walking, Swimming, and Flying Migrants
The model developed above is general and applies to migrants using any mode of
locomotion. Here, we parameterize the model for the three dominant modes of migratory
locomotion (walking, swimming, flight) by using standard models of locomotion to
describe the Ploc and v terms in Equation (2–1) (biomechanical models described in
19 detail in Appendix A). For walking migrants, Ploc can be described by
gM Pwalk = γ v, (2–3) Lc
−1 −1 where Lc is stride length (m), v is walking speed (m s ), γ is a cost coefficient (J N ),
and g is the acceleration due to gravity (m s−2,[31]) The only morphological variable in
Equation (2–3) is Lc , which is proportional to leg length [32]. We assume that walking migrants travel at speeds, v [33] and that they maintain these speeds over the course of
migration.
The power required for swimming can be described by the resistive model,
A v 2.8 P = b , swim δ 0.2 (2–4) Lb
2 where δ is a dimensionless cost coefficient, Ab is body cross-sectional area (m ),
−1 Lb is body length (m), and v is swimming speed (m s ,[6]). The set of relevant
morphological variables, β, is Ab and Lb. We assume that migrants swim at speeds that minimize the ratio, Ptot /v.
Power required for flight near minimum power speed can be described by the equation
2 −2 −1 3 P y = (1 + κ)[θM Lw v + ϕAbvf ], (2–5)
where κ is a dimensionless profile power coefficient, θ and ϕ are cost coefficients
2 (Appendix A), Ab is body cross sectional area (m ), Lw is wingspan (m), and κ is
2 proportional to Aw /Lw , where Aw is wing area [7]. The set of relevant morphological
variables, β, is therefore Ab, Lw , and Aw . We assume flying migrants travel at speeds
that minimize P y /vf [7].
Substituting Equations (2–3)-(2–5), corresponding migration speeds, and the
mass-dependence of maintenance metabolism into Equation (2–1) allows Yi to be
expressed as a function of initial mass M0, p0, and β for each mode of locomotion. In
each of the biomechanical models described above, the power required for locomotion
20 depends, in part, on a set of morphological lengths and areas, β, that do not change
as the migrant uses stored fuel to power migration. The dependence of Yi on β can
be eliminated by expressing morphological variables in terms of M0 based on the
assumption of geometric similarity (i.e. lengths, surface areas).
Substituting functions for Yi (Appendix A) into Equation (2–2) yields expressions for the expected maximum migration distances of walking
0.34 YT = y0M0 , (2–6)
swimming
−0.64 0.3 YT = y0p0 M0 , (2–7)
and flying [ ] p + k M0.42 Y = y log 0 1 0 T 0 0.42 (2–8) p0 + k2M0 migrants. Here y0 is a proportionality constant that varies by mode of locomotion, and k1 and k2 are empirical constants. Differences in the functional forms of Equations
(2–6) through (2–8) are caused by differences in the way Ploc depends on mass in
walking, swimming, and flying migrants. In the case of Equation (2–8), the predicted
relationship does not follow a simple power function in M0. This is because the cost of
flight increases more rapidly with increasing body mass than does the cost of walking or
swimming. The variable, p0, does not appear in the final form of the equation for walking
migrants because here we only consider the distance traveled by walking mammals,
for which p0 is roughly constant [34]. The exponents of the mass terms in Equations (2–6) through (2–8) describe how maximum migration distance changes as a function of M0 and reflect the mass-dependence of maintenance and locomotory metabolism.
The constant, y0, describes effects of mass-independent factors, such as the number of migratory legs, that affect the absolute distances traveled by migrants but do not affect the scaling of migration distance with body mass. The metabolic normalization constant, p0, and the morphological constants k1 and k2 can be estimated from empirical
21 measurements (see Materials and Methods). The framework described here uses body mass (Fig. 2-1B box a), morphology (Fig. 2-1B box b) and mode of locomotion (Fig.
2-1B box c) to determine migratory speed, and the metabolic costs of locomotory and
maintenance metabolism (Fig. 2-1B box d). Equation (2–1) ensures that changes in
speed and metabolism as the migrant uses stored fuel (Fig. 2-1B box e) are explicitly
incorporated into the prediction of Yi (Fig. 2-1B box f).
2.1.2 Model Predictions
Equations (2–6) through (2–8) make several quantitative predictions that can be
tested against data. First, each equation predicts that, after normalizing for p0, a single
curve can be used to describe expected maximum migration distance (in km) as a
function of M0 for species using each mode of locomotion. Second, each equation predicts how the number of body lengths traveled–a measure of relative distance [35]–varies with body mass. Migration distance and body length scale similarly with
1/3 mass in walking and swimming animals (i.e. YT roughly proportional to M0 , body
1/3 length ∝ M0 ) such that the number of body lengths traveled during migration, Ybl , is
1/3 1/3 0 described by Ybl = YT /(body length) ∝ M0 /M0 ∝ M0 . Thus, after normalizing for
differences in p0, the number of body lengths traveled by walking and swimming animals
should be approximately invariant with respect to M0. In flying animals, however, dividing
1/3 Equation (2–8) by M0 indicates that Ybl should decrease with increasing mass for all but the smallest flying migrants.
2.2 Materials and Methods
To evaluate the model, published measurements of maximum migration distances of
terrestrial mammals, fish, marine mammals, and flying insects and birds were collected.
Data from studies that met five criteria were included in the analysis: (1) reported movements could be considered to-and-fro migration or one-way migration [36], (2)
individuals were directly tracked by mark-recapture, telemetry or other means, groups
of individuals were tracked by repeated observation over the course of migration, or a
22 reliable estimate of distance traveled could otherwise be established, (3) maximum travel distances, maps, tracks or other information that allowed direct calculation of minimum estimates of the distances traveled by individual animals were reported, (4) there did not exist strong but indirect evidence from other studies (e.g. sightings of unmarked individuals, stable isotope data) suggesting that the maximum reported migration distance was substantially shorter than true maximum migration distance, and (5) in the case of flying species, studies reported migration distances of species that rely, at least partially, on flapping flight. The fifth criterion was imposed because the biomechanical model of flight used to derive our predictions applies most directly to flapping flight.
Migration distance and body mass data were included from a large dataset [37] for which all of the selection criteria could not be verified for all species. Including these data did not qualitatively affect our conclusions (see Results).
We estimated the constants k1 and k2 in Equation (2–8) using empirical studies of the morphology of flying insects and birds; however, the general form of Equation (2–8) and the resulting predictions are not strongly affected by variation in the empirical values used to estimate k1 and k2 (Appendix A). Empirical estimates of p0 were used in
Equations (2–7) throught (2–8) (Appendix A). Body mass data were used to estimate body lengths based on allometric equations (swimming mammals: [38]; others: [27]).
Body lengths were used to convert migration distance (km) into units of body lengths. To evaluate our first prediction, we fitted Equations (2–6) through (2–8) to migration distance data from walking (n = 33), swimming (n = 32), and flying migrants (n = 141), respectively. Equations (2–6) and (2–7) were fitted to log10-transformed distance and body mass data using ordinary least squares. Equation (2–8) was fitted to log10-transformed distance and body mass data using non-linear least squares (Gauss-Newton algorithm). Equations (2–6) through (2–8) have the general form:
d YT = y0h(M0 , p0), where h is a known function, y0 is a constant, and d is a scaling exponent. For each equation, two models were fitted: a model in which y0 was fitted
23 as a free parameter but d was set to the predicted value (i.e. d = 0.34, 0.3, 0.42, for
walking, swimming, and flying migrants, respectively), and a model in which both y0
and d were fitted. Model r 2 values reported below are based on the former method.
The latter method was used to generate 95% profile confidence intervals for the d
parameter. Prior to fitting, body mass values of swimming and flying animals were
0.3 −0.64 normalized to account for differences in p0 according to the equations Mnorm = M0 p0
0.42 −1 and Mnorm = M0 p0 , respectively. To test our second prediction–that the number of body lengths traveled was invariant of mass in walking and swimming migrants, but
decreased with mass in flying migrants–we fitted log10-transformed migration distance
(in body lengths) as a function of log10-transformed body mass (kg) using a quadratic
2 regression of the form, log10(Ybl ) = γ0 + γ1 log10(M0) + γ2 log10(M0) , where γi are regression coefficients [39]. Species were separated based on mode of locomotion
and by taxonomic groups differing in p0 (i.e. walking mammals, fish, marine mammals,
flying insects, and passerine and non-passerine birds were fitted separately). Statistical analyses were implemented using the nlme package [40] in R [41].
2.3 Results
Model predictions were evaluated using extensive data on maximum migration distances of animals from around the world (n = 206 species, Appendix A). Consistent
with our first prediction, maximum migration distance (km) varies systematically with
body mass for walking, swimming, and flying migrants (Fig. 2-2; r2 = 0.57, 0.65,
0.19, for walking, swimming, and flying species, respectively). The solid lines show predicted migration distance based on Equations (2–6) throught (2–8). There is a
tight correspondence between predicted relationships (solid lines) and fitted models
that treat both y0 and scaling exponents as free parameters (dashed lines and 95% confidence bands). In the case of walking and swimming animals, the data support model predictions of linear relationships in log-log space, with observed scaling exponents close to those predicted by Equations (2–6) and (2–7) (walking: predicted
24 = 0.34, observed = 0.36 95%CI [0.25,0.48]; swimming: predicted = 0.3, observed = 0.34 [0.28,0.41]). In the case of flying animals, data support the prediction that the
relationship is non-linear in log-log space reflecting the rapidly rising cost of flight
with increasing mass (Fig. 2-2C). Again, the observed mass exponent is close to
that predicted by Equation (2–8) (predicted = 0.42, observed = 0.43 [0.36,0.49]). Consistent with our second prediction, the number of body lengths traveled by swimming
and walking animals is independent of body mass (Fig. 2-3). On average, walking
mammals travel 1.5 × 105 body lengths (Fig. 2-3A). The slope and curvature terms
in the quadratic regression model does not differ from zero in walking mammals (n
= 33, p > 0.22) indicating that the number of body lengths traveled is uncorrelated with body mass in this group. Swimming animals travel an average of 1.7 × 106 body
lengths in a one-way migratory journey. The mean distance traveled by fish (triangles
in Fig. 3B) exceeds that traveled by swimming mammals (squares in Fig. 2-3B) by a
factor of 4 (fish: 2.1 × 106 body lengths; marine mammals: 5.3 × 105 body lengths, see Discussion), but the number of body lengths traveled is independent of mass in each
of these groups (slope and curvature does not differ from zero, fish: n = 20, p > 0.38;
swimming mammals: n = 12, p > 0.43). In flying migrants, the number of body lengths
migrated declines clearly with increasing body mass (Fig. 2-3C). In non-passerine birds
(n = 80), coefficients of linear and quadratic terms were both negative, and significantly
−5 different from zero (γ1 = -0.59, γ2 = -0.19, p < 2.2 × 10 ). In passerine birds (n =
45) and flying insects (n = 16) the γ1 term was negative and distinguishable from zero
−5 (passerines: γ1 = -0.63, p = 5.4 × 10 ; insects: γ1 = -0.16, p = 0.034). Results for flying migrants confirm our prediction that larger flying migrants generally travel fewer body lengths over the course of migration. The number of body lengths traveled decreases with increasing mass such that the smallest insects and birds travel around 1.4 × 108 body lengths whereas the largest birds travel around 5.2 × 106 body lengths. In other
25 words, the number of body lengths covered by moths, dragonflies, and hummingbirds is roughly 25-times that traveled by the largest ducks and geese.
A sensitivity analysis indicates that the agreement between model predictions and data are robust to deviations from geometric similarity and changes in the values of morphological and biomechanical parameters used to derive Equations (2–6)–(2–8) (Appendix A). In particular, the value of the exponent in metabolic scaling relationships has been a topic of much debate, with different authors reporting different exponents depending on the particular dataset and taxon studied and the method of analysis
(e.g. [34, 42]). However, sensitivity analysis shows that the shape of our predicted relationships, and the agreement between predictions and data are largely insensitive to changes in the value of the metabolic scaling exponent assumed (Appendix A).
Including data from [37] did not significantly change the estimate of the mass exponent
(0.36 95% CI [0.26,0.43] without data from [37], 0.43 [0.36,0.48] with data from [37]).
Including data from [37] decreased the model r2 from 0.37 to 0.19. 2.4 Discussion
When observed migration distances are plotted against predictions of Equations
(2–6) through (2–8), points from all three groups cluster around a 1:1 line (Fig. 2-4). The data shown in Figure 2-4 suggest that variation in maximum migration distances among species as distinct as Blue Whales (Balaenoptera musculus), Wildebeest
(Connochaetes taurinus), and Bar-tailed Godwits (Limosa lapponica) appears to be driven, in part, by the basic differences in metabolism, morphology, and biomechanics described by our model. The variation explained by the model reflects the influence of constraints on energetics and biomechanics imposed by body mass. There is a large body of work describing how morphology [6, 27], biomechanics [6, 21], and basic energetic properties such as maintenance metabolism [43, 44] are linked to body mass. Our model extends results of these studies by specifying how these quantities influence maximum migration distance of diverse species, thereby linking body mass
26 to migration distance. Our results show that constraints imposed by body mass are detectable in migration distance data, despite variation in migration distance among
species with similar body masses (i.e. variation about predicted relationships shown in
Figs. 2-2–2-4).
Migration distance data highlight the important role of basic differences in energetics in driving differences in migration distance among taxa. For example, the
number of body lengths traveled during migration is independent of body mass within
both swimming mammals and fish; however, fish travel an average of 4 times the
number of body lengths traveled by swimming mammals. Equation (2–7) shows that the
distances traveled by these groups depend on the metabolic normalization constant, p0, which describes mass-independent differences in the maintenance metabolic rates of
fish and marine mammals. In these groups, p0 differs by a factor of roughly 9.1 (p0 ≈
−3/4 −3/4 3.9 W kg in marine mammals, p0 ≈ 0.43 W kg in fish, see Appendix A), whereas
body length exhibits a similar relationship with mass in both groups (l ≈ 0.44M1/3) suggesting that the number of body lengths migrated by fish is greater by a factor of
(9.1)0.64 = 4.1, which is very close to the observed factor of 4. Thus, the difference
in the mean number of body lengths traveled by these groups may be driven by basic
differences in the cost of maintenance metabolism. Data also reveal patterns that do
not appear to be caused by the energetic and biomechanical factors considered here. For example, swimming is significantly less costly than flight in terms of the energy
required to travel a given distance [45], yet virtually all flying organisms travel distances
that are as great or greater than those traveled by most swimming species (Fig 2-4).
Whether this pattern is driven by differences in migratory behavior or other ecological or
evolutionary factors remains unknown and will likely be a fruitful area of future research. It is worth noting that other hypotheses may provide alternative explanations for
some of the qualitative patterns observed in migration distance data. For example, the
model predicts that migration distance (km) of larger flying species does not depend
27 strongly on mass. An increase in mass from 10−6 kg to 10−3 kg, increases expected migration distance by a factor of more than 8, whereas an increase in mass from 10−2
kg to 10 kg increases expected migration distance by a factor of less than 2. This
occurs because the energetic cost of flight increases rapidly with increasing mass
to the degree that the increasing fuel mass that can be carried by larger migrants provides a diminishing increase in migration distance. An alternative explanation for
this observation is that many subtropical and temperate habitats in the northern and
southern hemispheres are separated by 5 × 103 km –1 × 104 km and that many flying
migrants may not be under selection to migrate greater distances. In general, the
relationship between the distances traveled by migrants and the global distribution of suitable migratory habitats is poorly known but may ultimately influence the distances
traveled by many species.
While model predictions are supported by data, there is substantial unexplained
variation in Figures 2-2–2-4. Investigating why particular species deviate from predictions may be an effective way to identify ecological and evolutionary factors
that drive differences in migration distance but are not currently included in our model.
Our model ignores variation in fuel and morphology of species with similar masses and
does not consider the possibility that some migrants may seek to minimize the time
spent migrating. Two additional factors, in particular, are likely to contribute to observed residual variation. First, differences in the number migratory legs among otherwise
similar species will lead to variation in migration distance among species as indicated
by Equation (2–2). Second, species that interact strongly with abiotic currents during migration are likely to deviate from model predictions. The lack of information regarding the type and number of refueling stops made by migratory species, and the lack of information about the manner in which many flying and swimming migrants interact with abiotic currents represents an important gap in current knowledge. In the case of some well-studied species such as the arctic tern (Sterna paradisaea), it is clear that these
28 variables are important in facilitating extremely long-distance migrations. Individuals of this species stop at multiple highly productive foraging sites to refuel during migration
[18]. This species is also known to track global wind systems thereby taking advantage
of favorable air currents. In the case of species that migrate against abiotic currents,
migration distances might be expected to be shorter than our model predicts. Indeed, many of the swimming migrants that fall below the predicted line in Figure 2-2, are
anadromous fish such as shad (Alosa sapidissima), alewife (Alosa pseudoharengus), and river lamprey (Lampetra fluviatilis) that swim against water currents during upriver migrations. Increased understanding of the interactions between migrants and abiotic currents and the number of migratory stopovers will allow for extensions of the model that could further improve our understanding of the reasons for inter-specific differences in migration distance. In its current form, the model presented here provides a general expectation on maximum migration distance, which can be seen as a metric against which the distances traveled by particular species can be compared. The body sizes of migratory animals vary by over 11 orders of magnitude. The model presented here makes specific quantitative predictions about how this variation in size drives patterns of migration distance among species. It attributes differences in the distances traveled by migrants to systematic differences in metabolism and morphological traits that are tightly coupled to body size, and to differences in the underlying mechanics of walking, swimming, and flight. In doing so, it provides an analytically tractable framework for studying the influence of energetics and biomechanics on migration distance that is consistent with data on species ranging from the smallest migratory insects to the largest whales.
29 N A Yt = Yi i =1
Y1 Y2 Y3 YN ...
B a b Body Mass ( M0 ) Morphology ( )
d e P and P Mass loss as fuel mtn loc is used
c Mode of f Y locomotion i
Figure 2-1. (A) Total migration distance is the sum of the distances traveled on each of N migratory legs. (B) Migration distance on a single migratory leg. Body mass (a), morphology (b) and mode of locomotion (c) govern the rate at which a migrant uses stored fuel energy (d). This rate changes as migrant loses fuel mass (e), and determines the maximum distance covered during a single leg (f, Equation (2–1)). The relationship between a and b is governed by the mass-dependence of morphology. Total rate of energy use (d) is determined by the mass-dependence of maintenance metabolism and by the biomechanics of locomotion (Equations (2–3)-(2–3)).
30 iue2-3. Figure 2-2. Figure
Body lengths traveled Migration Distance (km) 10 10 10 10 nyadaeteeoentcretdfor corrected not therefore are and only in differences aesBadC oyms snraie codn oteequations the to according normalized is is mass mass M body body C, A, and panel B In panels parameters. with free (C) as regression fitted nonlinear intervals exponent or confidence B) 95% (A, and linear curves from fit best represent bands confidence ie eoema ubro oylntstaee yseisuigeach using species by traveled locomotion. lengths of body mode of number (diamonds). mean birds insects denote non-passerine flying Lines and C) (squares), and birds (squares), passerine mammals, mammals walking (triangles), and A) (triangles) by fish migration swimming during B) traveled lengths body of Number Equations of fits on based curves predicted flying are C) ( lines and Solid mammals A) insects. marine for and and mass birds fish body swimming normalized B) of mammals, function walking a as distance migration Maximum 10 10 10 10 10 10 2–6 0 1 2 3 4 5 3 5 7 9 norm 01 010 10 10 10 ● 2024 2 0 −2 )–( ● 10 ● ●● ● ● ● ● ● ● ● ● = ● −1 2–8 M 10 0 0.3 ● ● 0 odt with data to ) ● ● ● ● ● ● 10 p ● p ● ● − ● ● ● ● ● ● ● 1 ● ● ● ● ● ●●● 0 ● ● ● ● ● ● 0.64 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● mn rus aao akn nml r rmmammals from are animals walking on Data groups. among 10 ● ● ● ● ● ● ● ● ● ● ● 2 ● ● ● ● and ● ● A ● 10 A ● ● 3 M y 01 10 10 10 norm 0 10 te safe aaee.Dse ie and lines Dashed parameter. free a as fitted 202 0 −2 0 Normalised body mass body Normalised = Body mass Body (kg) 31 10 M 0.5 0 0.42 p 10 0 − 10 1 p 1 B 0 epciey ocretfor correct to respectively, , 4 . 10 10 1.5 B 6 10 10 −7 y −2.5 0 10 n h asscaling mass the and 10 −5 −2 10 10 −1.5 −3 10 M 10 −1 0 −1 10 k) In (kg). 10 −0.5 C C 5 10 4 10
3 ● 10 ●●● ●●● ● ●● ● ● ●●●
2 ● ●●●●● 10 ● ●● ● ● ●● Observeddistance (km)
1 ● ● 10 ●●● 0 10
10 0 10 1 10 2 10 3 10 4 10 5
Predicted distance (km)
Figure 2-4. Observed and predicted migration distances for the walking, swimming, and flying animals shown in Figure 2-2. Data from walking mammals (green circles), swimming fish (blue triangles) and marine mammals (blue squares), and flying insects (red triangles), passerine birds (red squares), and non-passerine birds (red diamonds) are shown. Black points and illustrations show the well-studied migrants Connochaetes taurinus (Wildebeest), Balaenoptera musculus (Blue Whale), and Limosa lapponica (Bar-tailed Godwit). Solid line indicates 1:1 line.
32 CHAPTER 3 SENSING AND DECISION-MAKING IN RANDOM SEARCH
Organisms routinely locate targets in complex environments. They can do this
by following gradients in the strength of sensory signals, provided such gradients are available and reliably lead toward targets [46]. But this is not always the case. In many
natural settings sensory signals are infrequent, noisy, and contain little directional
information [11]. For example, moths, sharks, and sea birds search environments that contain scent cues emitted by prey or mates, but these cues are often extremely sparse and subject to large fluctuations [9, 10, 47]. Under such sparse-signal conditions, it is not clear what behaviors allow organisms to efficiently and reliably locate resources.
Researchers have developed much of the theory of sparse-signal search by
studying mathematical models of searching organisms [12, 13, 48–51]. The dominant
paradigm for developing such models emerged from the random foraging hypothesis–the idea that searchers can encounter targets efficiently by adopting statistical movement
strategies that can be described as random walks ([12, 48], see [9, 11, 52] for alternative
approaches). This hypothesis, which has been applied to searching organisms ranging
from bees [12] to sea turtles [53], is often invoked when it is not possible or practical for
searchers to remember explicit spatial locations [48] and the typical distances between targets exceeds the searcher’s sensory range [54]. This framework has been used to
compare the performance of searchers moving according to different kinds of random
walk behavior. In particular, many studies have tried to determine whether searchers
moving according to Levy´ walks outperform searchers that move according to other types of random walk strategies (e.g. [13, 49–51]).
This chapter appeared as an article in the journal, Proceedings of the National Academy of Science: Hein, A. M. and S. A. McKinley. 2012. Sensing and decision-making in random search. Proc. Natl. Acad. Sci. USA. 109:12070–12074. Its reproduction here is authorized under the journal’s copyright policy.
33 If models are to yield insight into the behavior of searching organisms in nature, they must be simple enough to be studied, but should also capture the dominant features of search behavior. Implicit in the random foraging approach is the assumption that changes in a searchers’ movement behavior in response to sensory data are second-order effects, and that search behavior and performance are dominated by the features of the intrinsic (random) search strategy that the searcher employs. Here we explore an alternative hypothesis: that sensory processes can have a dominant effect on search performance, even when sensory signals are rare, noisy, and lack directional information.
Below we develop a general mathematical framework for modeling search decision-making. As in past models, the framework allows a searching organism to make movement decisions based on an intrinsic movement strategy (e.g. Levy´ walk), but allows such decisions to be modified based on noisy sensory data. It thus provides an explicit way to model changes in behavior in response to sensory measurements. We explore the effect of incorporating sensory data into search decisions using individual-based simulations of searching predators. We compare search times of simulated predators that make search decisions using random strategies alone (Levy´ walk and a novel diffusive strategy), to predators that modify their search behavior based on olfactory measurements. 3.1 Model Development
To study search decision-making, we consider an idealized model of a predator in search of prey. We wish to compare the behavior and performance of predators that search using a single intrinsic random strategy to predators that adaptively change their search behavior using the incomplete information gained from sensory measurements.
To evoke a strong intuition we consider two types of predator: a visual predator that makes movement decisions based on an intrinsic strategy and locates prey through a short-range, high acuity sense (vision), and a visual-olfactory predator that changes its
34 search behavior based on noisy olfactory data and detects prey at short range using vision. Predators wander through a large (periodic) two-dimensional habitat in which
the mean distance between prey is large. We assume prey emit a scent that can be
detected by nearby predators. Similar to previous approaches (e.g. [55] ), we assume
that search is divided into two phases: a local scanning phase and a movement phase (Figure 3-1A, [56]).
During the scanning phase, the predator locates any prey within its vision distance rv (Fig. 3-1A, solid inner circle) with probability one. This reflects the high local acuity
of vision. Visual-olfactory predators also scan for olfactory signals. The duration of
the scanning phase is denoted τv and τo for visual and visual-olfactory predators respectively. τv includes the time needed to visually search a region of radius rv and reorient before taking another step. τo includes the time taken to collect and process olfactory signals, visually search a region of radius rv , and reorient before taking another step. We define the olfactory radius ro (Fig. 3-1A, dashed outer circle) as the distance
where the predator registers an average of one scent signal per scanning period τo (see
below). We assume that each prey item emits scent at rate λ. During the movement
phase, the predator travels in a random uniform direction, a distance of l, at speed v.
Visual predators draw the step length l from a prescribed step length distribution θ(l),
examples of which are described in the next subsection. Visual-olfactory predators draw from a modified step length distribution defined below by Equation (3–1).
During the movement phase, we assume that the predator cannot locate prey or detect scent signals. Additionally, we assume that the predator only responds to the most recent scent signal information and does not store information about the locations it has visited. We study this limiting case where sensory signals are rare, lack directional information, and are not remembered by the predator because this is the scenario in which random search strategies are often invoked. We thus evaluate the scenario in which noisy sensory data are least likely to yield improvement over purely
35 random search. However, we point out that more sophisticated strategies are possible if predators remember past signal encounters or previously visited locations [11, 46, 57].
3.1.1 Searching Without Olfactory Data
To model predator movements, we begin with a model of decision-making in the
absence of any interaction with olfactory data. Researchers typically model the decision
process of random searchers by selecting two actions from prescribed probability
distributions: a step length l, and a turn angle ϕ. The details of these distributions
determine asymptotic properties of the search and strategies are often categorized by this asymptotic behavior: diffusive behavior, in which long-term mean-squared
displacement (MSD) scales linearly with time, and superdiffusive behavior in which MSD
increases superlinearly with time. An important feature of these strategies is that, unless
the searcher encounters a target, the distributions that define how searcher moves (i.e. the distributions of l and ϕ) are fixed. They are not altered in response to sensory
measurements.
We model the movements of visual predators using two types of strategies: a Levy´
strategy and a novel diffusive strategy. For both, we take the distribution of turn angles
between successive steps to be iid ϕ ∼ unif(0, 2π) [12]. The Levy´ strategy draws step
α−1 −α lengths from a Pareto distribution, θL(l) = (α − 1)lm l , with tail with parameter α and minimum step length lm (Fig. 3-1B solid curve, superdiffusive for 1 < α < 3 [12]). For
the second strategy, we introduce a new step-length distribution which we call the true
distance distribution (TDD) θT (l): a greedy strategy wherein the predator selects step lengths from the probability distribution of the distance to the nearest prey item (Fig.
3-1B dashed curve, see Supplementary Information (SI) Text for further discussion).
When prey are distributed according to a Poisson spatial process with intensity η in
−ηπl2 two dimensions, the TDD is given by the Rayleigh distribution θT (l) = 2ηπle . This strategy is quite distinct from the Levy´ strategy (compare curves in Fig. 3-1B) and later serves to illustrate the strong homogenizing effect of olfactory data on search behavior.
36 3.1.2 Incorporating Olfactory Data to Make Search Decisions
The key distinction between visual and olfactory senses in our model is that the
visual sense yields perfect information about the location of prey whereas the olfactory sense does not. Thus, including olfactory measurements allows us to model a predator’s
ability to gather and respond to partial information about target positions gleaned from
sensory measurements. Below we develop a model for incorporating olfactory signals
into search decision-making, but note that this framework could be modified to model
responses to other types of sensory cues. We hypothesize that predators utilize olfactory data through two steps. First,
a predator uses a signal observation to estimate the likely distance to the nearest
prey. Second, the predator modifies its intrinsic tendency to move in a particular way
(represented by θ(l)) based on this information. In keeping with recent models of
olfactory search, simulated predators collect olfactory data for τo units of time and
encounter H ∈ {0, 1, 2, ...} detectable units of scent [11, 57]. In order to act optimally,
a predator must make movement decisions based on two distinct uncertainties. First,
the predator’s distance to the nearest target is uncertain and is characterized by the
probability distribution ν. Second, for a particular ν, the optimal step length distribution θ is also uncertain. Identifying optimal predator behavior requires calculating a Bayesian
posterior for the distance distribution ν|H , and then determining the associated optimal step length distribution θ|H . This remains an unsolved and perhaps intractable problem.
Instead, we approximate this process. We wish to capture two elements of search decision-making: an intrinsic tendency to move in a particular way θ(l), and a likelihood function P(H = h|l) that translates an observed scent signal h into information about the distance to the nearest prey. A natural model for signal response that incorporates these features is a Bayesian update
37 of the step length distribution θ itself:
P(H = h|l) θ(l) θ(l|H = h) = ∫ ∞ . (3–1) P(H = h|l) (l) dl 0 θ
We refer to this as “signal-modulation” of the step length distribution θ(l). This approximation to the optimal strategy yields significant improvement in search performance (see Appendix B for further elaboration). 3.1.3 Interpreting Scent Signals
We assume the predator can estimate or intuit the probability of registering h units
of scent in τo units of time, as a function of its distance to the nearest prey. This amounts to being able to estimate the likelihood function P(H = h|l), which depends on the
process of scent propagation.
In the complex environments where many species search, turbulent fluctuations
in fluid velocity cause large local fluctuations in scent concentration [58]. When a prevailing wind or water current is present, predators can gain additional information
about the location of a scent source by measuring the velocity of the current [11, 47].
We consider the more difficult scenario in which there is no prevailing current. Under
these conditions, we model scent arrival as packets that appear at the prey position
x0 according to a Poisson arrival process and then move as a Brownian motion. From the predator’s perspective, this is equivalent to encountering a random number of units
of scent, H ∼ Pois(τo R(|x − xo|)), at its location x during a scanning phase of length
τo , where R is the rate of scent arrival defined by Equation (C–2) (see Materials and
Methods). Denoting l = |x − x0|, under these assumptions, the likelihood of h encounters
is
h −τo R(l) P(H = h|l) = [τo R(l)] e /h! (3–2)
Equation (3–2) depends on values of several physical parameters (e.g. the rate
at which detectable patches of scent decay) that may be difficult for a predator to infer
38 from measurements of its physical environment. We therefore take a qualitative view in prescribing the parameters of scent propagation. The most important qualitative feature
is the length scale ro , which corresponds to the distance at which a predator will register on average one unit of scent per scanning period τo . Heuristically, this is the distance at
which the predator is likely to detect a faint, yet non-trivial scent. A second qualitative
restriction is the expected number of encounters per unit τo at a distance of one body
length from the prey λa. Given these two measurements, the likelihood function can be
estimated.
The quantities ro and λa are much more readily measurable by a searching
organism than are the explicit parameters in Equation (C–2). It thus seems likely that these quantities may constitute part of an organism’s “olfactory search image” [59],
and may serve as the direct measurements useful for reinforcement learning.
3.2 Materials and Methods
3.2.1 Scent Propagation
To see how R(l) depends on the distance between predator and prey, let u(x) represent the mean concentration of scent at predator position x emitted by a prey
item located at position x0. An expression for the steady-state diffusion process without
advection is given by 0 = Du(x) − µu(x) + λδ(x0), where D represents the combined
molecular and turbulent diffusivity (m2s−1), µ represents the rate of dissolution of scent patches (s−1), and λ represents the rate of scent emission at the prey (s−1). In two dimensions, the mean rate of scent patch encounters by a predator of linear size a √ x R(l) = 2πD u(l) = µ located at is given by − ln(aψ) where ψ D [11]. This implies
λK0(ψl) R(l) = 2 , (3–3) −πψ ln(ψa)
where K0 represents a modified Bessel function of the second kind.
39 3.2.2 Simulation Details
The SI Text shows the robustness of results to changes in model parameters. For
each of the four search strategies (visual Levy,´ visual TDD, visual-olfactory Levy,´ and visual-olfactory TDD), we performed simulations in which predators explored a periodic
environment with 100 prey. Prey were positioned according to a Poisson point process
with the mean distance between prey chosen to achieve the desired density. In each
scanning phase, h was generated as a deviate from a Poisson distribution with mean
given by the product of τo and Equation (C–2) summed over all prey. In each simulation, the searcher was positioned at a random location and allowed to move through the
environment until it came within a distance of rv of a prey item during its scanning phase. For each strategy, we performed 1000 simulations and recorded the time until
first prey encounter in each simulation. Predators were assumed to travel at a constant speed of one body length per unit time. Environments were constructed so that prey density had a mean of 1 prey per 106 squared body lengths, a realistic low density for prey, but qualitative results hold for lower prey densities (see Appendix B). In the case of
the Levy´ strategies, we repeated simulations across a range of α values from α = 1.2
to α = 3. Note that the optimal value of α for the Levy´ predator was α = 3 for which the long-term behavior is expected to be Gaussian [12]. In all figures, Levy´ strategies with
the optimal value of α are shown unless otherwise noted.
3.3 Results
3.3.1 Visual-Olfactory Predators Find Targets Faster and More Reliably Than Visual Predators
Figure 3-2A shows mean search times of simulated visual and visual-olfactory predators (search time = time until first target encounter). Visual predators that use the Levy´ strategy (Fig. 3-2A, solid line, see also Materials and Methods) have lower mean search times than predators that use the TDD strategy (Fig. 3-3A, dashed line).
However, when conditions are such that the olfactory radius ro is greater than the vision
40 radius rv , visual-olfactory predators find prey faster than their visual counterparts (Fig. 3-2A; circles represent results from visual-olfactory Levy´ with optimal α, where optimal α
was in the range 2.6-3.0 for all ro /rv ; diamonds represent visual-olfactory TDD strategy).
Mean search time of visual-olfactory predators continues to decrease as the distance
over which prey scents can be detected increases. Visual-olfactory predators have lower mean search times than visual predators
primarily because they rarely search for long periods of time without finding prey.
Figure 3-2B shows that the tails of the search time distributions for the visual-olfactory
predators (Fig. 3-2B, circles) decay roughly exponentially at a rate that is much faster
than the decay rate of the visual predators (Fig. 3-2B, squares). At least two factors contribute to the difference in performance between the two
predator types. First, visual-olfactory predators learn from “no-signal” events. They
respond to these events by leaving regions that do not contain targets. Second, as has
been observed in many species in nature [14, 47], visual-olfactory predators perform area-restricted search [15] and concentrate search effort in regions that contain prey.
Below we discuss how both of these behaviors emerge naturally through responses to
sensory signals.
To characterize changes in predator behavior in response to sensory data in the
following sections, we use a metric of information gain: the Kullback-Leibler divergence (KL, [60]). The magnitude of the change in behavior of a visual-olfactory predator
when it receives a signal of strength h relative to its intrinsic behavior θ(l), is given by ∫ KL = θ(l|h) log(θ(l|h)/θ(l)) dl. A literal interpretation of the quantity KL is the following: suppose an observer must decide, based on empirical data, whether a searcher is using olfactory data or not. The KL gives a mean rate of gain of information obtained by observing a visual-olfactory searcher moving in response to a signal of magnitude h. In
regimes where the signal contains little useful information (for example when ro /rv ≈
1 and h =0), the behavior is not modified greatly from θ(l). The resulting KL value is
41 small. However, when information is substantial (say when h = 5, for small ro /rv ) the KL is larger.
3.3.2 Visual-Olfactory Predators Learn From No-Signal Events
Figure 3-3 shows typical search paths of the four strategies through a target field in the regime where ro > rv . When searching such an environment, a predator will frequently be too far from prey to receive scent signals. For example, the inset panels in Figure 3-3C and 3-3D show that the number of signals received in scanning phases
is typically zero, with signals of greater than zero only occurring when the predator is close to prey. Intuitively, it may seem that a predator gains little information from these
no-signal events. Yet, by not receiving a scent signal, the predator gains a vital piece of
information: prey are not likely to be nearby.
Figure 3-4A shows step length distributions of visual-olfactory predators after receiving no signal. Both strategies exhibit a low probability of making small steps. The
Levy´ strategy in particular, is strongly affected; Figure 3-1B shows that this strategy
has a high probability of taking small steps between re-orientations. Yet, when the
visual-olfactory Levy´ predator receives no signal, it is unlikely to make a small step (Fig.
3-4A, Figure B-4). Figure 3-4B shows that when h = 0, KL increases as the olfaction
radius becomes larger. In fact, as ro /rv becomes large, both strategies change more in response to no-signal events than when h = 5 (Fig. 3-4B, circles (h = 0) cross above
squares (h = 5) for both strategies). For ro /rv sufficiently large, the change in behavior
in response to no-signal events allow visual-olfactory predators to avoid performing area restricted search (ARS) when they are far from prey (Figure 3-5). The visual Levy´
predator, on the other hand, spends 24% of its steps in ARS but only 2.4% in ARS near
targets. Avoiding these wasted steps strongly affects search time. Even by responding
only to no-signal events and ignoring cases in which h > 0, a visual-olfactory Levy´
predator can find prey much more rapidly than a visual Levy´ predator (Fig. B-5).
42 The observation that no-signal events contain valuable information is qualitatively similar to an observation from optimal foraging theory regarding a forager searching a discrete patch for hidden resources. In that scenario, the more time the forager spends in the patch without encountering resources, the more certain it becomes that the patch does not contain resources [61]. Our model extends this idea to searchers moving through continuous spatial environments using two sensory modalities and reveals that the change in a searchers behavior in response to no-signal events depends critically on the length scales of these sensory modalities.
3.3.3 Visual-Olfactory Predators Concentrate Search Effort Near Targets
From Figure 3-3A and 3-3B, it is clear that visual predators behave similarly in regions that are near and far from prey. Visual-olfactory predators, on the other hand, make more short exploratory steps in the vicinity of prey (Fig. 3-3C,3-3D). The strong change in strategy that occurs when a visual-olfactory predator receives a nonzero scent signal is reflected in the large value of KL for all values of ro /rv (Fig. 3-3B). Both visual-olfactory strategies increase their probability of making a short step when they encounter a nonzero scent signal (Fig 3-3A). Because of this, visual-olfactory predators perform ARS near targets and are more likely to encounter nearby prey than are visual predators (Figure 3-5).
3.4 Discussion
The framework presented here allows one to include responses to partial information gained from noisy sensory measurements when modeling random search. Our results reveal that analysis of the length scales of sensory modalities, in this case ro and rv , is crucial to determining whether such a sensory response will dominate search performance. The distinction between different types of intrinsic strategies (e.g. Levy´ vs TDD [49, 50]) is important when it is genuinely not possible to learn about resources from a distance (ro /rv ≤ 1). However, when ro /rv >1, searchers that dynamically modify their behavior in response to sensory data experience a qualitative improvement
43 in search performance. This holds over a wide range of the parameters of the scent model and other features of predator behavior (Figures B-2). This finding suggests a
connection between sensing, decision-making, and search performance, even under
sparse-signal conditions.
Moreover, behaviors such as area-restricted search near prey [14] emerge naturally from responses to sensory information. Visual-olfactory predators preform this behavior
in our simulations by turning more frequently when they receive scent cues. Historically,
ARS has been explained as a consequence of a predator concentrating search effort
in areas where it has previously found prey. This is beneficial if prey are clustered in
space [15]. Yet, we show that this behavior can also emerge when prey are not spatially clustered, if predators change their movement behavior in response to noisy sensory
data. Recent evidence suggests that some species may initiate area-restricted search
in this way. For example, wandering albatrosses appear to alter turning patterns after
encountering prey scent, effectively concentrating their search effort in local regions [47]. Greater frigatebirds forage primarily in highly productive mesoscale eddies [16]. They
appear to track these eddies, at least in part, using scent cues.
In our simulations Levy´ predators intersperse periods of local search with
large-scale relocation movements. Movements of many species including foraging
marine fish and reptiles [53], and ants in search of colony-mates [62] exhibit this qualitative pattern [48, 53, 62]. This is often cited as a feature of Levy´ walks that makes
them effective strategies for encountering targets. Yet, our results show that Levy´
predators spend much of their time searching locally in regions that do not contain
prey (Fig. 3-5). On the other hand, visual-olfactory predators appropriately match their
behavior to their proximity to targets, leading to shorter search times. In light of our results, a natural hypothesis is that searching organisms utilize different movement
behaviors depending on their perceived distance to targets. It has been shown that
strategies that mix movements with different length scales can outperform strategies that
44 draw movements from a single distribution, but that such mixed movement behavior can be difficult to distinguish from a Levy´ strategy [51]. Indeed, recent analyses have begun to find evidence of mixed behaviors in movement data (e.g. [63]). Our framework provides a means of studying how such mixed behaviors can emerge through interactions with sensory information.
45 A B ro 3 rv 10 -3 3 ) l 10 θ( -9 ) l 2 θ( l1 10 1 10 2 10 4 l
l2 0 1 0500 1000 1500 body lengths ( l)
Figure 3-1. Schematic of predator search. A) During the scanning phase of the search, a prey encounter occurs if the predator is within a radius of rv (solid inner circle) of a prey item. The predator also detects scent signals emitted by prey within a radius of ro (dashed outer circle) at an average rate of ≥ 1 per τo units of time. The predator then turns a random uniform angle between 0 and 2π. During the movement phase, the predator moves a distance of l units determined by its step length distribution. B) Step length distributions corresponding to visual Levy´ (solid curve, α = 3, lm = 1 body length) and TDD (η = 1/(1000)2 body lengths, dashed curve) strategies. Inset shows distributions on log-log scale.
46 4 A 5 ×10 5 4
● ●●
3 ×10 3 ● ● ●
4 ● ● ● meansearch time ● ● 10 0 2 4 6 8 10 olfaction radius/vision radius
●
1000 B ●
● ●
●
10 100 ● frequency1 + ● ● ● ● ● ●● ● ● ● ● ● ● 1 0 1×10 5 2×10 5 3×10 5 search time
Figure 3-2. Predator search times. A) Mean search time as a function of the ratio of the olfactory radius (ro ) to vision radius (rv ). Solid orange line (visual Levy),´ dashed blue line (visual TDD), orange circles (visual-olfactory Levy),´ and blue diamonds (visual-olfactory TDD) each represent mean search time of 1000 replicate simulations. Confidence bands represent ±2 SEM. The following parameters values were used: a = 1, rv = lm = 50a, τv = 1 s, τo = 30 s, mean inter-target distance was 1000a, and λa = 100 units of scent per τo (see text for description of parameters, also SI Text). B) Empirical distribution of search times of visual Levy´ (orange solid line, squares), visual TDD (blue dashed line, squares), visual-olfactory Levy´ (orange solid line, circles), and visual-olfactory TDD (blue dashed line, circles) strategies. In the case of the visual-olfactory strategies, frequencies are shown for ro /rv = 4. Note the large number of searches resulting in long search times for visual predators.
47 A B
C D ) ) 20 h 20 h 15 15 5 10 scent signals ( signals scent scent signals ( signals scent 0 0 5 10
Figure 3-3. Typical search paths through a scent field with log10(1 + mean number of scent encounters per unit τo ) indicated by grayscale (darker grey denotes more encounters). In white regions, mean number of encounters is effectively zero. Paths for A) visual Levy,´ B) visual TDD, C) visual-olfactory Levy,´ and D) visual-olfactory TDD are shown. Color scale of path changes from blue to red with increasing time. Inset panels in C and D show the number of hits received during each scanning period with colors corresponding to colors in search paths. ro /rv = 4 in all panels; all other parameters as in Fig. 2A.
48 AB 15
10 ● ● ● ● ● ● ● ● ● h = ● ● ) 5 ● ● ● ●
h ● ● ●
-1 ● ● ● ●
= ●
10 ●
10 ● H | ● l l -3
( ● 5 θ 10 h = 0 -5 ● 0 Informationgain (KL) 10 0 1000 2000 0 5 10 15 body lengths ( l) olfaction radius/ vision radius
Figure 3-4. Effect of olfactory data on step lengths. A) Step length distributions after signal modulation when h = 0 and when h = 5. B) Information gain as measured by the Kullback-Leibler divergence between the visual strategy and the corresponding visual-olfactory strategy when h = 0 (squares) and h = 5 (circles) as a function of ro /rv . Dashed curves represent visual-olfactory TDD strategy. Solid curves represent visual-olfactory Levy´ strategy. Note the increasing information gain when h = 0. In both panels α, lm, and ν as in Fig. 1B, all other parameters as in Fig. 2A.
V Lévy V-O Lévy V TDD V-O TDD
0 10 20 30 0 20 40 60 80 % steps spent in % proximity area-restricted search events resulting in prey capture
Figure 3-5. Effect of olfactory data on area-restricted search (ARS). Left bars show % steps spent performing ARS. Shaded regions show the % of ARS searches that occur within 4rv of prey. For the visual Levy´ predator (top bar), contrast the large % of steps spent in ARS, with the small fraction of these steps spent near prey (top bar, shaded region). ARS defined as any period in which predator makes ≥ 5 consecutive steps within a region of radius 4rv . Right bars show % of proximity events in which predator locates prey. Bars for visual-olfactory (V-O) predators show that they successfully locate nearby prey more frequently than do visual (V) predators. Proximity events defined to by any period of ≥ 1 consecutive steps within 2rv of a target.
49 CHAPTER 4 SENSORY INFORMATION AND ENCOUNTER RATES OF INTERACTING SPECIES
Classical models of species interactions assume that encounters are governed by a process akin to mass-action; individuals move along random linear trajectories and encounter one another when they come within a critical distance [3, 64]. Under these assumptions, an individual searcher encounters targets at a rate proportional to the density of targets ρ [3, 65]. Recent work has extended the study of encounter rates to consider searchers that follow movement paths that are not linear trajectories, encounter targets probabilistically, destroy targets after encounters, and search intermittently [66–68]. Under a variety of circumstances, these models too predict that a searcher will encounter its targets at a rate proportional to target density (for a list of conditions, see [67]). A vital assumption both of older and newer models is that the searching organism moves independently of the locations of targets. In the context of predator-prey interactions, this implies for instance, that predators do not alter their movement behavior in response to sensory cues emitted by their prey.
Of course, the assumption that searchers move independently of targets is made for mathematical convenience. The question is whether models that rely on this assumption capture the salient features of encounter rate kinetics in nature. Empirical studies have shown that shutting down particular sensory modalities such as chemosensing or flow sensing can dramatically decrease search performance (e.g., [10]), and that sensory cues appear to influence both small-scale [69] and large-scale [16, 47] search behavior. While such studies more rigorously confirm the intuition that the use of sensory data should improve search performance, little is known about how sensing can influence the qualitative relationship between encounter rate and target density. Here, we argue that sensory response can have a dominant effect on the rate of encounters between searchers and their targets, not only by increasing encounter rate, but also by qualitatively changing the dependence of encounter rate on target density.
50 Below we adopt the language and intuition associated with predators searching for prey. We assume that a predator samples the environment for sensory cues passively
emitted by prey, and adjusts its movement behavior according to explicit mathematical
models presented below. This approach builds on a recently developed framework for
modeling search decision-making [70] to model the flow of sensory information from prey to predators. We consider three scenarios: (1) perfect sensing and response: the
predator can ascertain the precise locations of prey from the sensory data it receives
and responds optimally, (2) imperfect sensing and response: the predator detects
noisy scent signals emitted by prey and alters its movement behavior in response,
and (3) purely random search: the predator does not use sensory information to guide its movement decisions. Models (1) and (2) represent upper and lower bounds,
respectively, on the acquisition and use of information about prey positions. Our central
finding is that there is a systematic shift away from a linear encounter rate function at
both of these bounds, suggesting that the collection and use of any form of sensory data may fundamentally alter encounter rate kinetics. We discuss the role of information in
governing predator-prey encounter rates, but note that our general methodology could
be applied to rates of encounters in other types of ecological interactions (e.g., between
mates, competitors, mutualists).
4.1 Materials and Methods
4.1.1 Encounter Rate and Search Behavior: Some Definitions
Studies of biological search strategies typically describe how the type of movement
behavior used by a searching organism affects the time needed to encounter its first
target τ, or the rate of target encounters . For consistency with past work, we define