Parameter variability can produce heavy tails in a model for the spatial distribution of settling organisms

Luis F. Gordillo · 2025

Read Parameter variability can produce heavy tails in a model for the spatial distribution of settling organisms by Luis F. Gordillo (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-16385.txt.

Introduction

How far young organisms drift before settling shapes whole ecosystems, and rare long trips matter most. Luis F. Gordillo and Priscilla E. Greenwood show that a simple drift model can yield those rare long trips if its settings vary from case to case. In their model, each organism wanders in two dimensions for a random time, with the wander and timing traits drawn fresh each time.

That modest variation alone produces heavy-tailed spread patterns, where far settlers are much more common than a fixed-setting model would give. The result offers a clean frame for grasping long-distance dispersal in movement ecology.

Catalog record

CREATORLuis F. Gordillo
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE23 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-16385

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Parameter variability can produce heavy tails in a model for the spatial distribution of settling organisms Luis F. Gordillo Abstract We show that a simple mechanistic model of spatial dispersal for settling organisms, subject to parameter variability, can generate heavy-tailed radial probability density functions. The movement of organisms in the model consists of a two-dimensional diffusion that ceases after a random time, where the parameters that characterize each of these stages have been randomized. Our findings show that these minimal assumptions can yield heavy-tailed dispersal patterns, providing a simplified framework that increases the understanding of long-distance dispersal events in movement ecology. Keywords: Dispersal, dispersal kernel, long-distance dispersal, heavy tails. 1 Introduction The mechanisms underlying long-distance dispersal in ecology remain partially understood despite the significant theoretical advances and extensive data collection over the past decades ( Bullock et al. 2017 ; Clark et al. 1999 ; Fandos et al. 2023 ). In part, this is due to the intricate interplay of various factors, including behavioral, individual and environmental complexities, which vary between scenarios ( Morales and Morán López 2022 ; Nathan et al. 2011 ; Schupp et al. 2019 ). As a result, dispersal kernels (or equivalently, density functions of locations) often oversimplify the underlying processes and sources of variability ( Bullock et al. 2017 ; Nathan et al. 2012 ). In the case of long-distance dispersal, kernels are frequently fitted with limited regard for the underlying biological mechanisms driving these patterns ( Bullock et al. 2017 ; Clark et al. 2005 ; Snell et al. 2019 ). Dispersal kernels with tails that decay more slowly than that of a Gaussian have been shown to arise when diffusivity is treated as a random variable drawn from a specified distribution ( Petrovskii et al. 2008 ; Petrovskii and Morozov 2009 ). However, such analyses overlook the fact that individual organisms, referred to here as individuals , eventually cease movement at random times, which significantly influences the shape of dispersal kernels ( Morales and Morán López 2022 ). When the stopping times are very long, individuals would be expected to diffuse further. In this paper, we show that the probability density of individuals’ final locations can have spatially heavy-tailed behavior, e.g. the tail decays as a power-law, emerging from a two-step model of movement and settlement when the distribution of settling times is heavy-tailed (in time). This is an alternative approach to long distance dispersal models, like Lévy walks, which describe movement patterns with step lengths that follow a heavy-tailed distribution ( Viswanathan et al. 2011 ) or advection-diffusion models accounting for turbulence ( Nathan et al. 2011 ). We start by recalling the classic paper by Broadbent and Kendall ( Broadbent and Kendall 1953 ), where a model for the dispersal of larvae of helminth parasites was proposed. Our paper is based on their approach. Broadbent and Kendall considered the spatial movement of each individual larva as a Brownian motion in the plane, with the origin as the initial position. Individual movement stops after an exponentially distributed time, which is the time the larva climbs onto a grass blade and is eaten by a grazer. The Broadbent-Kendall model is conceptually straightforward; its core idea has been well studied ( Lewis et al. 2016 ; Renshaw 1991 ) and adapted to contexts with similar dynamics ( Williams 1961 ; Yasuda 1975 ). The model yields the radial probability density function for the larva’s final position. This density depends on two parameters: a diffusion coefficient that governs movement and a parameter of the exponential distribution of the stopping time. In the Broadbent-Kendall model, the tail of the spatial density decays exponentially, so that it is light-tailed, which makes the model unsuitable for describing the outcome of long distance dispersal mechanisms. However, to introduce more flexibility to account for intrinsic or extrinsic variability, it has been found that such parameters can be represented by distributions ( Clark et al. 1999 ; Nathan et al. 2012 ; Petrovskii et al. 2008 ; Petrovskii and Morozov 2009 ). For instance, seed settling rates can vary with environmental conditions, while insect dispersal may differ due to physiological traits. This prompts the question: How does the stochasticity of the parameters affect the shape of the density function of Broadbent and Kendall? Here, we start with exponentially stopped diffusions and allow the parameters governing both diffusion and settling to follow familiar distributions. The resulting spatial tail of the dispersal distribution will have either light or heavy tail behavior depending on the distributions used for each parameter. We derive simplified analytical expressions for the spatial probability distributions obtained and compare their spatial tails with those derived using fixed parameters. This paper is organized as follows: In Section 2 we derive explicit analytical expressions for the radial probability density of individuals’ final locations, after randomizing each one of the two parameters and then both. In Section 3 the tail analysis is presented. Terminology. A light tail , or thin tail, distribution has a tail probability that decays at least as fast as an exponential function e^{-\theta x},\theta>0 . A heavy tail distribution has a probability tail that decays more slowly than an exponential function. 2 Randomizing the parameters 2.1 Random settling rate \lambda Our assumption is that each individual moves in the plane with Brownian motion and stops at a time \tau , the settling time . In ( Broadbent and Kendall 1953 ), \tau is exponentially distributed with parameter \lambda>0 , the settling rate . Suppose that \lambda is not the same for all individuals but has a probability distribution within the class of individuals, i.e. \lambda is a random variable with probability density f_{\lambda}(s) having support (0,\infty) . The distribution of the settling time is now H(t)=P(\tau\leq t)=\int_{0}^{\infty}P(\tau\leq t|\lambda=s)f_{\lambda}(s)ds=\int_{0}^{\infty}(1-e^{-st})f_{\lambda}(s)ds=1-\mathcal{L}(f_{\lambda}(s))(t), (1) where \mathcal{L} denotes the Laplace transform. Following Broadbent and Kendall 1953 , we assume that each individual undergoes diffusive movement in the plane, starting from the origin, and stopped at a time with distribution ( 1 ). We assume that the movement is isotropic and denote the diffusion coefficient by \sigma^{2}>0 . Let R(t) represent the distance of an individual from the origin at time t . In polar coordinates, the probability that R(t) lies within the interval [r,r+dr) is given by \phi(r,t)dr=P(r\leq R(t)<r+dr)=\frac{2r}{B^{2}t}e^{-r^{2}/B^{2}t}dr, (2) where B^{2}=2\sigma^{2} , to which we will refer as the diffusion parameter or diffusivity . By combining the expressions ( 1 ) and ( 2 ) we derive the radial probability density, g(r) , of the final location of an individual, g(r)=\int_{0}^{\infty}\phi(r,t)\cdot\frac{dH}{dt}dt=\int_{0}^{\infty}\phi(r,t)\cdot\mathcal{L}(sf_{\lambda}(s))(t)dt. (3) In the case when \lambda>0 is constant and H(t)=1-e^{-\lambda t} , Broadbent and Kendall identified the radial density g(r)=\frac{4\lambda r}{B^{2}}K_{0}\left(\frac{2r\sqrt{\lambda}}{B}\right), (4) where K_{0} is a modified Bessel function of the second kind. As will be verified in Section 3, this density is light-tailed. As applications of equation ( 3 ), let us derive the radial density of individuals, g(r) , for two cases: when \lambda is uniformly distributed and when it follows a Gamma distribution. In practice, when little prior information is available, \lambda might be assigned a uniform distribution over a plausible range, chosen based on physical or biological constraints. As noted in ( Clark et al. 1999 ), the Gamma

Read the complete text