Persistence in a Two-Dimensional Moving-Habitat Model

Abstract

Environmental changes are forcing many species to track suitable conditions or face extinction. In this study, we use a two-dimensional integrodifference equation to analyze whether a population can track a habitat that is moving due to climate change. We model habitat as a simple rectangle. Our model quickly leads to an eigenvalue problem that determines whether the population persists or declines. After surveying techniques to solve the eigenvalue problem, we highlight three findings that impact conservation efforts such as reserve design and species risk assessment. First, while other models focus on habitat length (parallel to the direction of habitat movement), we show that ignoring habitat width (perpendicular to habitat movement) can lead to overestimates of persistence. Dispersal barriers and hostile landscapes that constrain habitat width greatly decrease the population’s ability to track its habitat. Second, for some long-distance dispersal kernels, increasing habitat length improves persistence without limit; for other kernels, increasing length is of limited help and has diminishing returns. Third, it is not always best to orient the long side of the habitat in the direction of climate change. Evidence suggests that the kurtosis of the dispersal kernel determines whether it is best to have a long, wide, or square habitat. In particular, populations with platykurtic dispersal benefit more from a wide habitat, while those with leptokurtic dispersal benefit more from a long habitat. We apply our model to the Rocky Mountain Apollo butterfly (Parnassius smintheus).

This is a preview of subscription content, log in to check access.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6

References

  1. Abramowitz M, Stegun IA (1970) Handbook of mathematical functions. Dover, New York

    Google Scholar 

  2. Adamski P, Witkowski ZJ (2007) Effectiveness of population recovery projects based on captive breeding. Biol Conserv 140:1–7

    Article  Google Scholar 

  3. Aitken SN, Yeaman S, Holliday JA, Wang T, Curtis-McLane S (2008) Adaptation, migration or extirpation: climate change outcomes for tree populations. Evol Appl 1:95–111

    Article  Google Scholar 

  4. Allee WC (1938) The social life of animals. Norton, New York

    Google Scholar 

  5. Alpin YA, Merikoski JK (2010) A simple proof for the inequality between the Perron root of a nonnegative matrix and that of its geometric symmetrization. Lobachevskii J Math 31:222–223

    MathSciNet  Article  MATH  Google Scholar 

  6. Austin M (2007) Species distribution models and ecological theory: a critical assessment and some possible new approaches. Ecol Modell 200:1–19

    Article  Google Scholar 

  7. Balakrishnan N, Lai C-D (2009) Continuous bivariate distributions. Springer, New York

    Google Scholar 

  8. Berestycki H, Diekmann O, Nagelkerke CJ, Zegeling PA (2009) Can a species keep pace with a shifting climate? Bull Math Biol 71:399–429

    MathSciNet  Article  MATH  Google Scholar 

  9. Best AS, Johst K, Münkemüller T, Travis JMJ (2007) Which species will successfully track climate change? The influence of intraspecific competition and density dependent dispersal on range shifting dynamics. Oikos 116:1531–1539

    Article  Google Scholar 

  10. Beverton RJH, Holt SJ (1957) On the dynamics of exploited fish populations. Her Majesty’s Stationery Office, London

    Google Scholar 

  11. Brooker RW, Travis JMJ, Clark EJ, Dytham C (2007) Modelling species’ range shifts in a changing climate: the impacts of biotic interactions, dispersal distance and the rate of climate change. J Theor Ecol 245:59–65

    MathSciNet  Google Scholar 

  12. Buckley LB, Tewksbury JJ, Deutsch CA (2013) Can terrestrial ectotherms escape the heat of climate change by moving? Proc R Soc B 280:1149

    Article  Google Scholar 

  13. Burrows MT, Schoeman DS, Buckley LB, Moore P, Poloczanska ES, Brander KM, Brown C, Bruno JF, Duarte CM, Halpern BS, Holding J, Kappel CV, Kiessling W, O’Connor MI, Pandolfi JM, Parmesan C, Schwing FB, Sydeman WJ, Richardson AJ (2011) The pace of shifting climate in marine and terrestrial ecosystems. Science 334:652–655

    Article  Google Scholar 

  14. Burrows MT, Schoeman DS, Richardson AJ, Molinos JG, Hoffman A, Buckley LB, Moore PJ, Brown CJ, Bruno JF, Duarte CM, Halpern BS, Hoegh-Guldberg O, Kappel CV, Kiessling W, O’Connor MI, Pandolfi JM, Parmesan C, Sydeman WJ, Ferrier S, Williams KJ, Poloczanska ES (2014) Geographical limits to species-range shifts are suggested by climate velocity. Nature 507:492–495

    Article  Google Scholar 

  15. Butchart SHM, Walpole M, Collen B, van Strien A, Scharlemann JPW, Almond REA, Baillie JEM, Bomhard B, Brown C, Bruno J, Carpenter KE, Carr GM, Chanson J, Chenery AM, Csirke J, Davidson NC, Dentener F, Foster M, Galli A, Galloway JN, Genovesi P, Gregory RD, Hockings M, Kapos V, Lamarque J-F, Leverington F, Loh J, McGeoch MA, McRae L, Minasyan A, Morcillo MH, Oldfield TEE, Pauly D, Quader S, Revenga C, Sauer JR, Skolnik B, Spear D, Stanwell-Smith D, Stuart SN, Symes A, Tierney M, Tyrrell TD, Vié J-C (2010) Global biodiversity: indicators of recent declines. Science 328:1164–1168

    Article  Google Scholar 

  16. Caswell H (2001) Matrix population models: construction, analysis, and interpretation. Sinauer, Sunderland

    Google Scholar 

  17. Caswell H, Neubert MG (2005) Reactivity and transient dynamics of discrete-time ecological systems. J Differ Equ Appl 11:295–310

    MathSciNet  Article  MATH  Google Scholar 

  18. Chen I-C, Hill JK, Ohlemüller R, Roy DB, Thomas CD (2011) Rapid range shifts of species associated with high levels of climate warming. Science 333:1024–1026

    Article  Google Scholar 

  19. Chevin L-M (2013) Genetic constraints on adaptation to a changing environment. Evolution 67:708–721

    Article  Google Scholar 

  20. Chevin L-M, Lande R (2010) When do adaptive plasticity and genetic evolution prevent extinction of a density-regulated population? Evolution 64:1143–1150

    Article  Google Scholar 

  21. Clark JS (1998) Why trees migrate so fast: confronting theory with dispersal biology and the paleorecord. Am Nat 152:204–224

    Article  Google Scholar 

  22. Collatz L (1942) Einschließungssatz für die charakteristischen zahlen von matrizen. Math Z 48:221–226

    MathSciNet  Article  Google Scholar 

  23. Corlett RT, Westcott DA (2013) Will plant movements keep up with climate change? Trends Ecol Evol 28:482–488

    Article  Google Scholar 

  24. Coster SS, Veysey Powell JS, Babbit KJ (2014) Characterizing the width of amphibian movements during postbreeding migration. Conserv Biol 28:756–762

    Article  Google Scholar 

  25. Cousens RD, Rawlinson AA (2001) When will plant morphology affect the shape of a seed dispersal “kernel”? J Theor Biol 211:229–238

    Article  Google Scholar 

  26. DeChaine EG, Martin AP (2004) Historic cycles of fragmentation and expansion in Parnassius smintheus (Papilionidae) inferred using mitochondrial DNA. Evolution 58:113–127

    Article  Google Scholar 

  27. Delattre T, Pichancourt J-B, Burel F, Kindlmann P (2010) Grassy field margins as potential corridors for butterflies in agricultural landscapes: a simulation study. Ecol Modell 221:370–377

    Article  Google Scholar 

  28. Delves LM, Walsh J (1974) Numerical solution of integral equations. Clarendon Press, Oxford

    Google Scholar 

  29. Dewhirst S, Lutscher F (2009) Dispersal in heterogeneous habitats: thresholds, spatial scales, and approximate rates of spread. Ecology 90:1338–1345

    Article  Google Scholar 

  30. Elith J, Leathwick JR (2009) Species distribution models: ecological explanation and prediction across space and time. Annu Rev Ecol Evol Syst 40:677

    Article  Google Scholar 

  31. Fagan WF, Lutscher F (2006) Average dispersal success: linking home range, dispersal, and metapopulation dynamics to reserve design. Ecol Appl 16:820–828

    Article  Google Scholar 

  32. Fahrig L (2003) Effects of habitat fragmentation on biodiversity. Annu Rev Ecol Evol Syst 34:487–515

    Article  Google Scholar 

  33. Fric Z, Konvicka M (2007) Dispersal kernels of butterflies: power-law functions are invariant to marking frequency. Basic Appl Ecol 8:377–386

    Article  Google Scholar 

  34. Gilbert-Norton L, Wilson R, Stevens JR, Beard KH (2010) A meta-analytic review of corridor effectiveness. Conserv Biol 24:660–668

    Article  Google Scholar 

  35. Guisan A, Thuiller W (2005) Predicting species distribution: offering more than simple habitat models. Ecol Lett 8:993–1009

    Article  Google Scholar 

  36. Haddad NM (2008) Finding the corridor more traveled. Proc Nat Acad Sci 105:19569–19570

    Article  Google Scholar 

  37. Hammersley JM, Handscomb DC (1964) Monte Carlo methods. Wiley, New York

    Google Scholar 

  38. Harsch MA, Zhou Y, HilleRisLambers J, Kot M (2014) Keeping pace with climate change: stage-structured moving-habitat models. Am Nat 184:25–37

    Article  Google Scholar 

  39. Hof C, Araújo MB, Jetz W, Rahbek C (2011) Additive threats from pathogens, climate and land-use change for global amphibian diversity. Nature 480:516–519

    Google Scholar 

  40. Hoffmann AA, Sgrò CM (2011) Climate change and evolutionary adaptation. Nature 470:479–485

    Article  Google Scholar 

  41. Horiguchi T, Fukui Y (1996) A variation of the Jentzsch theorem for a symmetric integral kernel and its application. Interdiscip Inf Sci 2:139–144

    MathSciNet  MATH  Google Scholar 

  42. Huey RB, Kearney MR, Krockenberger A, Holtum JAM, Jess M, Williams SE (2012) Predicting organismal vulnerability to climate warming: roles of behaviour, physiology and adaptation. Philos Trans R Soc B Biol Sci 367:1665–1679

    Article  Google Scholar 

  43. Hutson V, Pym J, Cloud M (2005) Applications of functional analysis and operator theory. Academic Press, London

    Google Scholar 

  44. Jentzsch R (1912) Über integralgleichungen mit positivem kern. J Reine Angew Math 141:235–244

    MathSciNet  MATH  Google Scholar 

  45. Jump AS, Peñuelas J (2005) Running to stand still: adaptation and the response of plants to rapid climate change. Ecol Lett 8:1010–1020

    Article  Google Scholar 

  46. Karlin S (1959) Positive operators. J Math Mech 8:907–937

    MathSciNet  MATH  Google Scholar 

  47. Karlin S (1964) The existence of eigenvalues for integral operators. Trans Am Math Soc 113:1–17

    MathSciNet  Article  MATH  Google Scholar 

  48. Kawasaki K, Shigesada N (2007) An integrodifference model for biological invasions in a periodically fragmented environment. Jpn J Ind Appl Math 24:3–15

    MathSciNet  Article  MATH  Google Scholar 

  49. Kierstead H, Slobodkin LB (1953) The size of water masses containing plankton blooms. J Mar Res 12:141–147

  50. Kolotilina LY (1993) Lower bounds for the Perron root of a nonnegative matrix. Linear Algebra Appl 180:133–151

    MathSciNet  Article  MATH  Google Scholar 

  51. Kot M, Phillips A (2015) Bounds for the critical speed of climate-driven moving-habitat models. Math Biosci 262:65–72

    MathSciNet  Article  Google Scholar 

  52. Kot M, Lewis MA, van den Driessche P (1996) Dispersal data and the spread of invading organisms. Ecology 77:2027–2042

  53. Kotz S, Podgorski K (2001) The Laplace distribution and generalizations: a revisit with applications to communications, economics, engineering, and finance. Birkhauser, Boston

    Google Scholar 

  54. Krein MG, Rutman MA (1948) Linear operators leaving invariant a cone in a Banach space. Uspekhi Mat Nauk 3:3–95

    MathSciNet  MATH  Google Scholar 

  55. Kubisch A, Holt RD, Poethke H-J, Fronhofer EA (2014) Where am I and why? Synthesizing range biology and the eco-evolutionary dynamics of dispersal. Oikos 123:5–22

    Article  Google Scholar 

  56. Latore J, Gould P, Mortimer AM (1998) Spatial dynamics and critical patch size of annual plant populations. J Theor Biol 190:277–285

    Article  Google Scholar 

  57. Lee T-T, Chang Y-F (1988) Solutions of convolution integral and integral equations via double general orthogonal polynomials. Int J Syst Sci 19:415–430

    MathSciNet  Article  MATH  Google Scholar 

  58. Leroux SJ, Larrivée M, Boucher-Lalonde V, Hurford A, Zuloaga J, Kerr JT, Lutscher F (2013) Mechanistic models for the spatial spread of species under climate change. Ecol Appl 23:815–828

    Article  Google Scholar 

  59. Lewis MA, Neubert MG, Caswell H, Clark JS, Shea K (2006) A guide to calculating discrete-time invasion rates from data. In: Cadotte MW, McMahon SM, Fukami T (eds) Conceptual ecology and invasion biology: reciprocal approaches to nature. Springer, Dordrecht, pp 169–192

    Google Scholar 

  60. Liira J, Paal T (2013) Do forest-dwelling plant species disperse along landscape corridors? Plant Ecol 214:455–470

    Article  Google Scholar 

  61. Loarie SR, Duffy PB, Hamilton H, Asner GP, Field CB, Ackerly DD (2009) The velocity of climate change. Nature 462:1052–1055

    Article  Google Scholar 

  62. Mardia KV (1970) Measures of multivariate skewness and kurtosis with applications. Biometrika 57:519–530

    MathSciNet  Article  MATH  Google Scholar 

  63. Marsden JE, Tromba A (2012) Vector calculus. WH Freeman, New York

    Google Scholar 

  64. Maruyama Y (2008) A measure of multivariate kurtosis with principal components. Commun Stat Theory Methods 37:2116–2123

    MathSciNet  Article  MATH  Google Scholar 

  65. Matter SF, Roland J (2002) An experimental examination of the effects of habitat quality on the dispersal and local abundance of the butterfly Parnassius smintheus. Ecol Entomol 27:308–316

    Article  Google Scholar 

  66. Matter SF, Roland J (2013) Mating failure of female Parnassius smintheus butterflies: a component but not a demographic Allee effect. Entomol Exp Appl 146:93–102

    Article  Google Scholar 

  67. Matter SF, Roland J, Moilanen A, Hanski I (2004) Migration and survival of Parnassius smintheus: detecting effects of habitat for individual butterflies. Ecol Appl 14:1526–1534

    Article  Google Scholar 

  68. Matter SF, Roslin T, Roland J (2005) Predicting immigration of two species in contrasting landscapes: effects of scale, patch size and isolation. Oikos 111:359–367

    Article  Google Scholar 

  69. Matter SF, Doyle A, Illerbrun K, Wheeler J, Roland J (2011) An assessment of direct and indirect effects of climate change for populations of the Rocky Mountain Apollo butterfly (Parnassius smintheus Doubleday). Insect Sci 18:385–392

    Article  Google Scholar 

  70. Miyagawa C, Seo T (2011) A new multivariate kurtosis and its asymptotic distribution. SUT J Math 47:55–71

    MathSciNet  MATH  Google Scholar 

  71. Neubert MG, Caswell H (1997) Alternatives to resilience for measuring the responses of ecological systems to perturbations. Ecology 78:653–665

    Article  Google Scholar 

  72. Parmesan C (2006) Ecological and evolutionary responses to recent climate change. Annu Rev Ecol Evol Syst 37:637–669

  73. Parmesan C, Yohe G (2003) A globally coherent fingerprint of climate change impacts across natural systems. Nature 421:37–42

    Article  Google Scholar 

  74. Pipkin AC (1991) A course on integral equations. Springer, New York

    Google Scholar 

  75. Porter D, Stirling DSG (1990) Integral equations: a practical treatment, from spectral theory to applications. Cambridge University Press, New York

    Google Scholar 

  76. Potapov AB, Lewis MA (2004) Climate and competition: the effect of moving range boundaries on habitat invasibility. Bull Math Biol 66:975–1008

    MathSciNet  Article  Google Scholar 

  77. Pouzols FM, Moilanen A (2014) A method for building corridors in spatial conservation prioritization. Landsc Ecol 29:789–801

    Article  Google Scholar 

  78. Press WH, Teukolsky SA, Vetterling WT, Flannery BP (2007) Numerical recipes: the art of scientific computing. Cambridge University Press, New York

    Google Scholar 

  79. Quintero I, Wiens JJ (2013) Rates of projected climate change dramatically exceed past rates of climatic niche evolution among vertebrate species. Ecol Lett 16:1095–1103

    Article  Google Scholar 

  80. Roland J, Matter SF (2007) Encroaching forests decouple alpine butterfly population dynamics. Proc Nat Acad Sci 104:13702–13704

    Article  Google Scholar 

  81. Roland J, Matter SF (2013) Variability in winter climate and winter extremes reduces population growth of an alpine butterfly. Ecology 94:190–199

    Article  Google Scholar 

  82. Roland J, Keyghobadi N, Fownes S (2000) Alpine Parnassius butterfly dispersal: effects of landscape and population size. Ecology 81:1642–1653

  83. Sansone G (2004) Orthogonal functions. Dover Publications Inc, Mineola, New York

    Google Scholar 

  84. Savage D, Barbetti MJ, MacLeod WJ, Salam MU, Renton M (2010) Timing of propagule release significantly alters the deposition area of resulting aerial dispersal. Divers Distrib 16:288–299

    Article  Google Scholar 

  85. Savage D, Barbetti MJ, MacLeod WJ, Salam MU, Renton M (2011) Can mechanistically parameterised, anisotropic dispersal kernels provide a reliable estimate of wind-assisted dispersal? Ecol Modell 222:1673–1682

    Article  Google Scholar 

  86. Schiffers K, Bourne EC, Lavergne S, Thuiller W, Travis JMJ (2013) Limited evolutionary rescue of locally adapted populations facing climate change. Philos Trans R Soc B Biol Sci 368:20120083

    Article  Google Scholar 

  87. Schloss CA, Nuñez TA, Lawler JJ (2012) Dispersal will limit ability of mammals to track climate change in the Western Hemisphere. Proc Nat Acad Sci 109:8606–8611

    Article  Google Scholar 

  88. Schoville SD, Roderick GK (2009) Alpine biogeography of Parnassian butterflies during Quaternary climate cycles in North America. Mol Ecol 18:3471–3485

    Article  Google Scholar 

  89. Schwenk AJ (1986) Tight bounds on the spectral radius of asymmetric nonnegative matrices. Linear Algebra Appl 75:257–265

    MathSciNet  Article  MATH  Google Scholar 

  90. Severini TA (2005) Elements of distribution theory. Cambridge University Press, Cambridge

    Google Scholar 

  91. Skarpaas O, Shea K (2007) Dispersal patterns, dispersal mechanisms, and invasion wave speeds for invasive thistles. Am Nat 170:421–430

    Article  Google Scholar 

  92. Srivastava MS (1984) A measure of skewness and kurtosis and a graphical method for assessing multivariate normality. Stat Probab Lett 2:263–267

    Article  Google Scholar 

  93. Travis JMJ, Dytham C (1999) Habitat persistence, habitat availability and the evolution of dispersal. Proc R Soc B Biol Sci 266:723–728

    Article  Google Scholar 

  94. Van Kirk RW, Lewis MA (1997) Integrodifference models for persistence in fragmented habitats. Bull Math Biol 59:107–137

    Article  MATH  Google Scholar 

  95. Van Kirk RW, Lewis MA (1999) Edge permeability and population persistence in isolated habitat patches. Nat Resour Model 12:37–64

    Article  MATH  Google Scholar 

  96. Visser ME (2008) Keeping up with a warming world: assessing the rate of adaptation to climate change. Proc R Soc B Biol Sci 275:649–659

    Article  Google Scholar 

  97. Wang J (2009) A family of kurtosis orderings for multivariate distributions. J Multivar Anal 100:509–517

    Article  MATH  Google Scholar 

  98. Wang M-L, Chang R-Y, Yang S-Y (1988) Double generalized orthogonal polynomial series for the solution of integral equations. Int J Syst Sci 19:459–470

    MathSciNet  Article  MATH  Google Scholar 

  99. Watson GN (1995) A treatise on the theory of Bessel functions. Cambridge University Press, Cambridge

    Google Scholar 

  100. Wood RJ, O’Neill MJ (2007) Finding the spectral radius of a large sparse non-negative matrix. ANZIAM J 48:C330–C345

  101. Wood RJ, O’Neill MJ (2004) An always convergent method for finding the spectral radius of an irreducible non-negative matrix. ANZIAM J 45:C474–C485

    MathSciNet  Google Scholar 

  102. Yalçınbaş S, Aynigül M, Akkaya T (2010) Legendre series solutions of Fredholm integral equations. Math Comput Appl 15:371–381

    MathSciNet  MATH  Google Scholar 

  103. Zhou Y, Kot M (2011) Discrete-time growth-dispersal models with shifting species ranges. Theor Ecol 4:13–25

    Article  Google Scholar 

  104. Zhou Y, Kot M (2013) Life on the move: modeling the effects of climate-driven range shifts with integrodifference equations. In: Lewis MA, Maini PK, Petrovskii SV (eds) Dispersal, individual movement and spatial ecology. Springer, Berlin, pp 263–292

    Google Scholar 

Download references

Acknowledgments

We thank Ying Zhou, Melanie Harsch, Rosie Leung, and Scott Rinnan for helpful comments. This material is based upon work supported by the National Science Foundation under Grant No. DMS-1308365 to MK. AP wishes to thank the Seattle Chapter of the ARCS Foundation for providing a fellowship supporting this work. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.

Author information

Affiliations

Authors

Corresponding author

Correspondence to Austin Phillips.

Appendix

Appendix

Nyström’s Method

Constructing Matrix \(\mathbf {A}\)

To discretize integral eigenvalue problem (7) using Nyström’s method, first define grid points

$$\begin{aligned} x_i&= -L/2 + i \cdot {\Delta } x,&i = 0, 1, \dots , n, \end{aligned}$$
(33)
$$\begin{aligned} y_j&= -W/2 + j \cdot {\Delta } y,&j = 0, 1, \dots , m, \end{aligned}$$
(34)

where \({\Delta } x = L/n\) and \({\Delta } y = W/m\). Create identical grid points \(x'_k\) and \(y'_{\ell }\) so that \({\Delta } x' = {\Delta } x\) and \({\Delta } y' = {\Delta } y\). Next, evaluate both sides of eigenvalue problem (7) only at grid points and approximate each integral on the right-hand side with a sum using, for example, the repeated trapezoidal rule. Because there are four sets of grid points, the resulting problem is, in a sense, four dimensional. We need to create additional indices to condense the system into a matrix eigenvalue problem with two dimensions. Define the indices

$$\begin{aligned} p(i,j) = i + 1 + (n+1)j, \quad q(k,\ell ) = k + 1 + (n+1)\ell . \end{aligned}$$
(35)

The mapping from (ij) pairs to p-values effectively reshapes a two-dimensional array into a one-dimensional vector (and similarly for the mapping from \((k,\ell )\) pairs to q-values), making the problem two-dimensional. Integral equation (7) becomes

$$\begin{aligned} \lambda u_p = R_0 \sum _{q=1}^{(n+1)(m+1)} A_{pq}u_q, \end{aligned}$$
(36)

where \(u_p\) is the p-th element of the reshaped vector u, and the element in row p, column q of matrix \(\mathbf {A}\) is

$$\begin{aligned} A_{pq}= \alpha _1 \, \alpha _2 \, k(x_i + c - x'_k,\, y_j - y'_\ell ) \,{\Delta } y_1 {\Delta } y_2, \end{aligned}$$
(37)

where

$$\begin{aligned} \alpha _1 = {\left\{ \begin{array}{ll} \dfrac{1}{2} \qquad &{}\text {if}\ \quad k = 0 \ \text {or} \ n,\\ 1 &{}\text {otherwise;} \end{array}\right. } \qquad \alpha _2 = {\left\{ \begin{array}{ll} \dfrac{1}{2} \qquad &{}\text {if}\ \quad \ell = 0 \ \text {or} \ m, \\ 1 &{}\text {otherwise.} \end{array}\right. } \end{aligned}$$
(38)

System (36) is now a standard matrix eigenvalue problem.

Avoiding Dispersal Kernel Singularities

If the dispersal kernel is singular at the origin, one way to avoid evaluating the singularity is to shift \(y'\) grid points onto the midpoints of y grid points,

$$\begin{aligned} y_\ell ' = -W/2 + {\Delta } y/2 + \ell \cdot {\Delta } y', \qquad \ell = 0, 1, \dots , m-1. \end{aligned}$$
(39)

The action of the integral operator K in eigenvalue problem (7) is now the composition of two operators, \({K}_1\) and \({K}_2\). The first calculates values at (xy) grid points, stored in some vector u. The second calculates values at \((x', y')\) grid points stored in another vector, v. Applying Nyström’s method to \({K}_1\) and \({K}_2\) as above yields a pair of matrices, \(\mathbf {A}_1\) and \(\mathbf {A}_2\), satisfying

$$\begin{aligned} u_{t+1} = \mathbf {A}_1 \,v_t, \qquad v_{t+1} = \mathbf {A}_2 \, u_t. \end{aligned}$$
(40)

Since \(u_{t+2} = \mathbf {A}_1\mathbf {A}_2 \, u_t\), the critical speed occurs when \(\rho (\mathbf {A}_1\mathbf {A}_2) = 1\).

Legendre-Expansion Method

Constructing Matrix \(\mathbf {B}\)

To construct matrix \(\mathbf {B}\) in eigenvalue problem (15), we must first recover the Legendre expansion coefficients \(b_{ijk\ell }\). Multiplying both sides of kernel expansion (14) by Legendre polynomials in each variable, integrating over the domain, and using orthogonality properties (11)–(12) gives

$$\begin{aligned} b_{ijk\ell }= & {} \left[ \dfrac{(2i+1)(2j+1)(2k+1)(2\ell +1)}{L^2W^2}\right] \nonumber \\&\times \int \limits ^{L/2}_{-L/2}\int \limits ^{L/2}_{-L/2}\int \limits ^{W/2}_{-W/2}\int \limits ^{W/2}_{-W/2} k(x+c - x', y - y')X_i(x)X_j(x')Y_k(y)Y_\ell (y')\nonumber \\&\times \, \mathrm{d}y' \,\mathrm{d}y \,\mathrm{d}x' \,\mathrm{d}x. \end{aligned}$$
(41)

Now, substitute kernel expansion (14) into integral eigenvalue problem (7).

$$\begin{aligned} \lambda u(x,y)&= R_0 \int ^{L/2}_{-L/2}\int ^{W/2}_{-W/2} \sum _{i,j,k,\ell = 0}^N b_{ijk\ell } \, X_i(x)X_j(x')Y_k(y)Y_\ell (y')\, u(x',y') \ \mathrm{d}y' \ \mathrm{d}x' \end{aligned}$$
(42)
$$\begin{aligned}&= R_0 \sum _{i,j,k,\ell = 0}^N b_{ijk\ell } \, X_i(x) Y_k(y) \left( \int ^{L/2}_{-L/2}\int ^{W/2}_{-W/2} X_j(x') Y_\ell (y') \ u(x',y') \, \mathrm{d}y' \, \mathrm{d}x'\right) \end{aligned}$$
(43)
$$\begin{aligned}&= R_0 \sum _{i,j,k,\ell = 0}^N b_{ijk\ell } \, X_i(x) Y_k(y) \, u_{j\ell }, \end{aligned}$$
(44)

where

$$\begin{aligned} u_{j\ell }&\equiv \int ^{L/2}_{-L/2}\int ^{W/2}_{-W/2} X_j(x') Y_\ell (y') \ u(x',y') \, \mathrm{d}y' \, \mathrm{d}x'. \end{aligned}$$
(45)

Multiplying both sides of Eq. (44) by Legendre polynomials \(X_r(x) Y_s(y)\) and integrating,

$$\begin{aligned}&\lambda \int ^{L/2}_{-L/2}\int ^{W/2}_{-W/2} u(x,y) X_r(x) Y_s(y) \, \mathrm{d}x \, \mathrm{d}y \end{aligned}$$
(46)
$$\begin{aligned}&\quad = R_0 \sum _{i,j,k,\ell = 0}^N b_{ijk\ell } \ u_{j\ell } \int ^{L/2}_{-L/2}\int ^{W/2}_{-W/2} X_i(x) X_r(x) Y_k(y) Y_s(y) \, \mathrm{d}y \, \mathrm{d}x. \end{aligned}$$
(47)

Using definition (45) and orthogonality,

$$\begin{aligned} \lambda u_{ik} = R_0LW \sum _{j=0}^N \sum _{\ell = 0}^N \dfrac{b_{ijk\ell } \ u_{ik}}{(2i+1)(2k+1)}. \end{aligned}$$
(48)

We have reverted to indices (ik) rather than (rs).

System (48) has four sets of indices and is, in a sense, four dimensional. We need to create additional indices to condense the system into a matrix eigenvalue problem with two dimensions. Let

$$\begin{aligned} p(i,k) = i + 1 + (N+1)k, \qquad q(j,\ell ) = j + 1 + (N+1)\ell , \end{aligned}$$
(49)

where \(i, j, k, \ell \in \{0, 1, \dots , N\}\). The mapping from (ik) pairs to p-values effectively reshapes a two-dimensional array into a one-dimensional vector (and similarly for the mapping from \((j,\ell )\) pairs to q-values), making the problem two-dimensional. System (48) becomes

$$\begin{aligned} \lambda u_p = R_0LW \sum _{q=1}^{(N+1)^2} \text {B}_\text {pq} u_q, \end{aligned}$$
(50)

where \(u_p\) is the p-th entry of the reshaped vector u. The element in row p, column q of matrix \(\mathbf {B}\) is

$$\begin{aligned} \text {B}_\text {pq} = \dfrac{b_{ijk\ell }}{(2i+1)(2k+1)}. \end{aligned}$$
(51)

Taylor Expansion Coefficients

The coefficients in approximation (20) found by expanding the kernel in a Taylor series are

$$\begin{aligned} \alpha&\equiv \int \limits ^{L/2}_{-L/2}\int \limits ^{W/2}_{-W/2} k(x,y) \,\mathrm{d}y \,\mathrm{d}x, \end{aligned}$$
(52)
$$\begin{aligned} \beta&\equiv \int \limits ^{W/2}_{-W/2} [ k(L/2, y) - k(-L/2,y)] \,\mathrm{d}y, \end{aligned}$$
(53)
$$\begin{aligned} \gamma&\equiv \int \limits ^{W/2}_{-W/2} \left[ \dfrac{\partial k}{\partial x}(L/2, y) - \dfrac{\partial k}{\partial x} (-L/2,y) \right] \,\mathrm{d}y, \end{aligned}$$
(54)
$$\begin{aligned} \delta&\equiv \int \limits ^{L/2}_{-L/2} \left[ \dfrac{\partial k}{\partial y}(x,W/2) - \dfrac{\partial k}{\partial y}(x,-W/2)\right] \,\mathrm{d}x . \end{aligned}$$
(55)

Optimal Aspect Ratio for the Symmetrized Gaussian Kernel

We will demonstrate an instance of the mesokurtic portion of Conjecture 2, that \(\theta _{\text {opt}} = 1\) for Gaussian dispersal kernels. We approximate the spectral radius \(\rho (\theta )\) using geometric symmetrization and an eigenfunction-one Rayleigh quotient (22). Symmetrization does not give an exact value for \(\rho (\theta )\), but it does give a tight lower bound and is analytically tractable. It may not be surprising that, should a maximal \(\rho (\theta )\) exist for a symmetric kernel, it might occur at \(\theta = 1\), when \(L = W\). What is less obvious (and what we will prove) is that a maximum does exist, and that it is unique.

Given a Gaussian kernel,

$$\begin{aligned} k(x+c - x', y - y') = \dfrac{1}{2\pi \sigma ^2}\exp \left[ \dfrac{-(x+c-x')^2}{2\sigma ^2}\right] \exp \left[ \dfrac{-(y-y')^2}{2\sigma ^2}\right] , \end{aligned}$$
(56)

the symmetrized kernel, using formula (21), is

$$\begin{aligned} k_G(x+c - x', y - y') = \dfrac{1}{2\pi \sigma ^2} \exp \left[ \dfrac{-c^2}{2\sigma ^2}\right] \exp \left[ \dfrac{-(x-x')^2}{2\sigma ^2}\right] \exp \left[ \dfrac{-(y-y')^2}{2\sigma ^2}\right] . \end{aligned}$$
(57)

The Rayleigh quotient approximation for the spectral radius with constant eigenfunction, as a function of L and W, is

$$\begin{aligned} \rho (L,W) = \dfrac{\int ^{L/2}_{-L/2} \int ^{L/2}_{-L/2} \int ^{W/2}_{-W/2} \int ^{W/2}_{-W/2} k_G(x+c - x', y - y') \, \mathrm{d}y' \, \mathrm{d}y \, \mathrm{d}x' \, \mathrm{d}x}{LW}. \end{aligned}$$
(58)

The numerator of approximation (58) is separable, and we may write

$$\begin{aligned} \rho (L,W) = C I(L) I(W), \end{aligned}$$
(59)

where

$$\begin{aligned} C = \dfrac{1}{2\pi \sigma ^2} \exp \left( \dfrac{-c^2}{2 \sigma ^2} \right) , \end{aligned}$$
(60)

and

$$\begin{aligned} I(L) = \sqrt{2 \pi } \sigma \, \text {erf} \left( \dfrac{L}{\sqrt{2}\sigma }\right) - \dfrac{2\sigma ^2}{L} + \dfrac{2\sigma ^2}{L} \exp \left( \dfrac{-L^2}{2\sigma ^2}\right) . \end{aligned}$$
(61)

I(W) is identical to I(L) with W replacing L. (We have not absorbed the constant C into I(L) and I(W) to make future calculations easier.)

We now have a constrained optimization problem: to maximize \(\rho (L,W)\) subject to the constraint \(LW = A\). Before using the method of Lagrange multipliers, we will describe a few properties of the function I(L) necessary for the proof. Then, we will construct a Lagrangian function and find the unique critical point of the associated system. Finally, we will use the associated bordered Hessian matrix to show the critical point gives a maximum.

The first two derivatives of I(L) are

$$\begin{aligned} I'(L) = \dfrac{2\sigma ^2 - 2\sigma ^2 \exp \left( \dfrac{-L^2}{2\sigma ^2}\right) }{L^2} \, > \, 0 \quad \text {for} \quad L > 0, \end{aligned}$$
(62)

and

$$\begin{aligned} I''(L)&= \dfrac{2L^2 \exp \left( \dfrac{-L^2}{2\sigma ^2}\right) - 4\sigma ^2 + 4\sigma ^2 \exp \left( \dfrac{-L^2}{2\sigma ^2}\right) }{L^3} \end{aligned}$$
(63)
$$\begin{aligned}&= \dfrac{2\exp \left( \dfrac{-L^2}{2\sigma ^2}\right) - 2I'(L)}{L}. \end{aligned}$$
(64)

Using the above derivatives, we can show the following useful lemma.

Lemma 1

$$\begin{aligned} I'(L) < \dfrac{I(L)}{L} < 1, \quad \mathrm{for} \quad L > 0. \end{aligned}$$
(65)

Proof

It is sufficient to show that I(L) is concave down and that \(\lim _{L\rightarrow 0^+} I'(L) = 1\). That is, if \(I'(L)\) is decreasing and bounded from above by one, I(L) must satisfy the lemma.

We first show that \(I''(L)\) is negative. From expression (63),

$$\begin{aligned} I''(L) = \dfrac{1}{L^3}\exp \left( \dfrac{-L^2}{2\sigma ^2}\right) \left[ 2L^2 - 4\sigma ^2\exp \left( \dfrac{L^2}{2\sigma ^2}\right) + 4\sigma ^2\right] . \end{aligned}$$
(66)

Since \(\exp (L) > L + 1\) when \(L \ne 0\),

$$\begin{aligned} I''(L) < \dfrac{1}{L^3}\exp \left( \dfrac{-L^2}{2\sigma ^2}\right) \left[ 2L^2 - 4\sigma ^2\left( \dfrac{L^2}{2\sigma ^2}+ 1\right) + 4\sigma ^2\right] = 0. \end{aligned}$$
(67)

Then, using L’Hôpital’s rule twice on expression (62), the desired limit quickly follows. \(\square \)

With the above lemma available, we now construct the Lagrangian function with objective function \(\rho (L,W)\) and constraint \(g(L,W) = LW - A = 0\).

$$\begin{aligned} {\varLambda }(L,W,\lambda ) = \rho (L,W) - \lambda g(L,W). \end{aligned}$$
(68)

The necessary first-order conditions for critical points are

$$\begin{aligned} {\varLambda }_L&= C I'(L) I(W) - \lambda W = 0, \end{aligned}$$
(69)
$$\begin{aligned} {\varLambda }_W&= C I(L) I'(W) - \lambda L = 0, \end{aligned}$$
(70)
$$\begin{aligned} {\varLambda }_\lambda&= LW - A = 0 \end{aligned}$$
(71)

(Marsden and Tromba 2012). Here, subscripts indicate partial differentiation. From (69) and (70),

$$\begin{aligned} \lambda = \dfrac{C I'(L) I(L)}{W} = \dfrac{C I(L) I'(W)}{L}. \end{aligned}$$
(72)

Rearranging the last equation,

$$\begin{aligned} \dfrac{ I'(L) L}{I(L)} = \dfrac{I'(W) W}{I(W)}. \end{aligned}$$
(73)

If we can show that the function

$$\begin{aligned} J(L) \equiv \dfrac{ I'(L) L}{I(L)} \end{aligned}$$
(74)

is one-to-one for \(L > 0\), then Eq. (73) implies \(L = W\), and with the constraint \(LW = A\), the only critical point is \((L,W) = (\sqrt{A},\sqrt{A})\). We will show J(L) is one-to-one by showing it is monotone decreasing.

From expressions (62) and (63) for \(I'(L)\) and \(I''(L)\),

$$\begin{aligned} J'(L) = \dfrac{1}{L}\left[ I I''+ \dfrac{I I'}{L} - (I')^2 \right] . \end{aligned}$$
(75)

(We assume differentiation is with respect to L on the right-hand side.) Substituting expression (64) for \(I''\),

$$\begin{aligned} J'(L)&= \dfrac{1}{L} \left[ I \left( \dfrac{2}{L}\exp \left( \dfrac{-L^2}{2\sigma ^2}\right) - \dfrac{2}{L}I'\right) + \dfrac{I I'}{L} - (I')^2\right] \end{aligned}$$
(76)
$$\begin{aligned}&= \dfrac{1}{L} \left[ \dfrac{2I}{L}\exp \left( \dfrac{-L^2}{2\sigma ^2}\right) - \dfrac{I I'}{L} - (I')^2\right] . \end{aligned}$$
(77)

The first and second inequalities in Lemma 1 now give, one after the other,

$$\begin{aligned} J'(L)&< \dfrac{1}{L}\left[ 2\exp \left( \dfrac{-L^2}{2\sigma ^2}\right) - \dfrac{I I'}{L} - (I')^2\right] \end{aligned}$$
(78)
$$\begin{aligned}&< \dfrac{1}{L}\left[ 2\exp \left( \dfrac{-L^2}{2\sigma ^2}\right) - 2(I')^2\right] \end{aligned}$$
(79)
$$\begin{aligned}&= \dfrac{2}{L}\left[ \exp \left( \dfrac{-L^2}{4\sigma ^2}\right) + I'\right] \left[ \exp \left( \dfrac{-L^2}{4\sigma ^2}\right) - I'\right] \end{aligned}$$
(80)
$$\begin{aligned}&= \dfrac{2}{L^3} \left[ \exp \left( \dfrac{-L^2}{4\sigma ^2}\right) + I'\right] \left[ L^2 \exp \left( \dfrac{-L^2}{4\sigma ^2}\right) + 2\sigma ^2 \exp \left( \dfrac{-L^2}{4\sigma ^2}\right) - 2\sigma ^2\right] . \end{aligned}$$
(81)

Since \(I'(L) > 0\), we only need to show that

$$\begin{aligned} M(L) \equiv L^2 \exp \left( \dfrac{-L^2}{4\sigma ^2}\right) + 2\sigma ^2 \exp \left( \dfrac{-L^2}{4\sigma ^2}\right) - 2\sigma ^2 < 0 \end{aligned}$$
(82)

when \(L > 0\) to prove J(L) is monotone decreasing. We have that \(M(0) = 0\), and

$$\begin{aligned} M'(L)&= -\dfrac{L^3}{2\sigma ^2} \exp \left( \dfrac{-L^2}{4\sigma ^2}\right) + 2L\exp \left( \dfrac{-L^2}{4\sigma ^2}\right) - 2L \exp \left( \dfrac{-L^2}{2\sigma ^2}\right) \end{aligned}$$
(83)
$$\begin{aligned}&= L\exp \left( \dfrac{-L^2}{4\sigma ^2}\right) \left[ -\dfrac{L^2}{2\sigma ^2} + 2 - 2 \exp \left( \dfrac{-L^2}{4\sigma ^2}\right) \right] \end{aligned}$$
(84)
$$\begin{aligned}&< L\exp \left( \dfrac{-L^2}{4\sigma ^2}\right) \left[ -\dfrac{L^2}{2\sigma ^2} + 2 - 2\left( -\dfrac{L^2}{4\sigma ^2} + 1\right) \right] \, = \, 0. \end{aligned}$$
(85)

Thus, \(M(L) < 0\), which makes J(L) monotone decreasing and one-to-one, which makes \(L = W\) the only solution of Eq. (73). Together with the area constraint, the unique critical point is \((L, W) = (\sqrt{A}, \sqrt{A})\).

To show our critical point maximizes the spectral radius, we will use the bordered Hessian matrix

$$\begin{aligned} H&= \begin{pmatrix} 0 &{}\quad -g_L &{}\quad -g_W \\ -g_L &{}\quad {\varLambda }_{LL} &{}\quad {\varLambda }_{LW}\\ -g_W &{}\quad {\varLambda }_{WL} &{}\quad {\varLambda }_{WW} \end{pmatrix} \end{aligned}$$
(86)
$$\begin{aligned}&= \begin{pmatrix} 0 &{}\quad -W &{}\quad -L\\ -W &{}\quad C I''(L)I(W) &{}\quad C I'(L)I'(W) - \lambda \\ -L &{}\quad C I'(L)I'(W) - \lambda &{}\quad C I(L) I''(W) \end{pmatrix}. \end{aligned}$$
(87)

(Recall that C is a constant.) For our \(3 \times 3\) matrix, a sufficient condition for the critical point to be a maximum is \(|H| > 0\), where the matrix is evaluated at the critical point (Marsden and Tromba 2012). \(| \cdot |\) denotes the determinant.

$$\begin{aligned} |H| = -CW^2I(L)I''(W) - CL^2I''(L)I(W) + 2CLWI'(L)I'(W) - 2C\lambda LW. \end{aligned}$$
(88)

At the critical point, \(L = W\), so we will use L to represent both quantities.

$$\begin{aligned} |H|&= -2CL^2I''(L) I(L) + 2CL^2 [I'(L)]^2 - 2CL^2\lambda \end{aligned}$$
(89)
$$\begin{aligned}&= 2CL^2\left( -I''(L) I(L) + [I'(L)]^2 - \dfrac{I'(L) I(L)}{L}\right) . \end{aligned}$$
(90)

However, we have already shown from steps (75)–(85) that

$$\begin{aligned} I''(L) I(L) - [I'(L)]^2 + \dfrac{I'(L) I(L)}{L} < 0 \end{aligned}$$
(91)

when \(L > 0\). The opposite quantity must be positive. Therefore, from expression (90), \(|H| > 0\) and the unique critical point \((L,W) = (\sqrt{A},\sqrt{A})\) maximizes the spectral radius. \(\theta _{\text {opt}} = 1\) for the symmetrized Gaussian kernel.

Effect of Switching Length and Width for the Symmetrized Generalized Gaussian Kernel

One way to illustrate Conjecture 1 is to plot the level curves of shifted kernels with different kurtosis values. If the level curves are stretched along the x-axis (the dimension of habitat length), a habitat with larger length than width is better because it captures more of the kernel’s probability mass (and, thus, more propagules). If level curves are stretched along the y-axis (the dimension of width), a habitat with larger width is preferable. We will use the geometric symmetrization approximation (21) on the generalized Gaussian kernel (26) and show kurtosis determines the direction the level curves are stretched. Our result provides evidence that, for the broad class of generalized Gaussian kernels, Conjecture 1 is true.

To simplify our approach, we will consider the square of the geometrically symmetrized kernel,

$$\begin{aligned} k_G^2(x+c-x',y-y') = = k(x + c - x', y - y') k(x' + c - x, y' - y). \end{aligned}$$
(92)

Evaluating (92) for the generalized Gaussian kernel (26) gives a distribution of the form

$$\begin{aligned}&k_G^2(x+c-x',y-y') \nonumber \\&\quad = \gamma \exp \left\{ -\dfrac{1}{2} \left[ (x+c-x')^2 + (y-y')^2\right] ^\beta \right\} \nonumber \\&\qquad \times \exp \left\{ -\dfrac{1}{2} \left[ (x'+c-x)^2 + (y'-y)^2\right] ^\beta \right\} , \end{aligned}$$
(93)

where \(\gamma \) is a normalizing constant. Since \(x'\) and \(y'\) simply shift the kernel, we will evaluate approximation (93) at \((x', y') = 0\).

$$\begin{aligned} k_G^2(x+c,y) = \gamma \exp \left\{ -\dfrac{1}{2} \left[ (x+c)^2 + y^2\right] ^\beta \right\} \exp \left\{ -\dfrac{1}{2} \left[ (x-c)^2 + y^2\right] ^\beta \right\} . \end{aligned}$$
(94)

As the speed of climate change c increases from zero, the level curves of kernel (94) stretch along one axis or the other when \(\beta \ne 1\) (Fig. 7). To measure the effect of the stretch, we will slice kernel (94) along each axis and compare the corresponding functions.

$$\begin{aligned} f_1(x)&\equiv k_G^2(x+c,0) = \gamma \exp \left\{ -\dfrac{1}{2} \left[ (x+c)^2\right] ^\beta -\dfrac{1}{2} \left[ (x-c)^2 \right] ^\beta \right\} , \end{aligned}$$
(95)
$$\begin{aligned} f_2(y)&\equiv k_G^2(0 + c,y) = \gamma \exp \left[ -(c^2+y^2)^\beta \right] . \end{aligned}$$
(96)
Fig. 7
figure7

Level curves of symmetrized generalized Gaussian kernels show the effect of increasing the speed of climate change, c. Plots (a), (c), and (e) used \(c = 0\). In plots (b), (d), and (f), we increased c to 1 km/year. We used \(\beta = 2\) to obtain a platykurtic kernel for (a, b), \(\beta = 1\) to obtain a mesokurtic kernel for (c, d), and \(\beta = 1/2\) to obtain a leptokurtic kernel for (e, f). The orientation of the level curves determines whether it is better to have a larger habitat length (for leptokurtic kernels) or larger width (for platykurtic kernels). For the Gaussian kernel, length and width have the same effect on persistence

Functions (95) and (96) are the cross sections of kernel (94) along the x- and y-axis, respectively. To compare the functions, let us, for example, set \(f_2\) as a function of x.

$$\begin{aligned} f_2(x) \equiv \gamma \exp \left[ -(c^2+x^2)^\beta \right] . \end{aligned}$$
(97)

We wish to find conditions under which, for example, \(f_1(x) > f_2(x)\), leading to a stretch along the x-axis.

$$\begin{aligned} \gamma \exp \left\{ -\dfrac{1}{2} \left[ (x+c)^2\right] ^\beta -\dfrac{1}{2} \left[ (x-c)^2 \right] ^\beta \right\} \, > \, \gamma \exp \left[ -(c^2+x^2)^\beta \right] , \end{aligned}$$
(98)

or

$$\begin{aligned} -\dfrac{1}{2}[(x+c)^2]^\beta - \dfrac{1}{2}[(x-c)^2]^\beta \, > \, -(c^2 + x^2)^\beta . \end{aligned}$$
(99)

Switching sides,

$$\begin{aligned} (c^2 + x^2)^\beta \, > \, \dfrac{1}{2}[(x+c)^2]^\beta + \dfrac{1}{2}[(x-c)^2]^\beta . \end{aligned}$$
(100)

The observation that

$$\begin{aligned} c^2 + x^2 = \dfrac{1}{2}(x+c)^2 + \dfrac{1}{2}(x-c)^2 \end{aligned}$$
(101)

suggests we can use a convexity argument with the function \(g(z) \equiv z^\beta \). In particular, Jensen’s inequality states that if (and only if) g(z) is concave down, then for any two values \(z_1\) and \(z_2\),

$$\begin{aligned} g\left( \dfrac{1}{2} z_1 + \dfrac{1}{2} z_2\right) > \dfrac{1}{2}g(z_1) + \dfrac{1}{2}g(z_2). \end{aligned}$$
(102)

g(z) is concave down only when \(\beta < 1\). Setting

$$\begin{aligned} z_1 = (x+c)^2, \quad z_2 = (x-c)^2, \end{aligned}$$
(103)

we see that inequality (100) is only true when \(\beta < 1\). The same argument applies for \(\beta > 1\), resulting in \(f_1(x) < f_2(x)\). In general,

$$\begin{aligned} f_1(x) {\left\{ \begin{array}{ll} > f_2(x) \quad \text { if} \quad \beta < 1,\\ = f_2(x) \quad \text { if} \quad \beta = 1,\\ < f_2(x) \quad \text { if} \quad \beta >1. \end{array}\right. } \end{aligned}$$
(104)

The parameter \(\beta \) controls the kurtosis of the generalized Gaussian distribution. \(\beta = 1\) gives a standard Gaussian. Increasing c has no effect on the Gaussian (Fig. 7c, d), so there is no preference for habitat length or width. \(\beta >1\) gives platykurtic kernels, for which increasing c stretches the kernel along the y-axis (Fig. 7a, b), favoring larger width. Finally, \(\beta < 1\) gives leptokurtic kernels, for which increasing c stretches the kernel along the x-axis (Fig. 7e, f), favoring larger length.

The use of convexity (102) above suggests a possible way to prove Conjecture 1. Recent definitions of kurtosis have centered around partial orderings of distributions that preserve certain ordering properties. One such ordering involves convexity (Wang 2009). If it can be shown that the partial ordering of two shifted kernels corresponds to an ordering of the importance of length and width (e.g., kernels with higher kurtosis favor habitat length over width), Conjecture 1 is proven. We will not attempt such a proof here, but convexity ordering provides a promising approach to our conjecture.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Phillips, A., Kot, M. Persistence in a Two-Dimensional Moving-Habitat Model. Bull Math Biol 77, 2125–2159 (2015). https://doi.org/10.1007/s11538-015-0119-z

Download citation

Keywords

  • Climate change
  • Two-dimensional habitat
  • Integrodifference equation
  • Persistence
  • Dispersal
  • Kurtosis