Ecological Modelling 132 (2000) 115 – 124 www.elsevier.com/locate/ecolmodel Modelling animal movement as a persistent random walk in two dimensions: expected magnitude of net displacement Hsin-i Wu a, Bai-Lian Li b,*, Timothy A. Springer c, William H. Neill d a Department of Industrial Engineering, Center for Biosystems Modelling, Texas A&M Uni6ersity, College Station, TX 77843 -3131, USA b Department of Biology, Uni6ersity of New Mexico, 167 Castetter Hall, Albuquerque, NM 87131 -1091, USA c Wildlife International, 8598 Commerce Dri6e, Eastern, MD 21601, USA d Department of Wildlife and Fisheries Sciences, Texas A&M Uni6ersity, College Station, TX 77843 -2258, USA Abstract We present semi-empirical model of persistent random walk for studying animal movements in two-dimensions. The model incorporates an arbitrary distribution for the angles between successive steps in the tracks. Inclusion of a turning angle distribution enables explicit computation of the effect of persistence in the direction of travel on the expected magnitude of net displacement of the animal over time. We employed a form-analogous approach to obtain expressions for the expected net displacement and derived root mean square of the expected displacement of an animal at the end of a multi-step random walk in which turning angles were drawn from the Lemicon of Pascal, the elliptical, the von Mises, and the wrapped Cauchy distributions. The accuracy of these expressions for the expected magnitude of net displacement was tested by comparison with simulated results of persistent random walks where turning angles were drawn form the wrapped Cauchy distribution. Our results should be useful in predicting two-dimensional distribution of moving animals for which frequency distributions of the turning angles can be measured. © 2000 Elsevier Science B.V. All rights reserved. Keywords: Animal movement model; Expected net displacement; Random walk; Root mean square of displacement; Turning angle distribution 1. Introduction Movement (i.e. spatial displacement) has come to be recognized as a key element in the population biology of most organisms and the community dynamics of interacting species. Examples of * Corresponding author. Tel.: +1-505-2775140; fax: +1505-2770304. E-mail address: [email protected] (B.-L. Li). recent studies are in wind dispersal of plant seeds and pollen (Okubo and Levin, 1989), insect and animal movements (Kareiva and Shigesada, 1983; Li et al., 1987; Andow et al., 1990; Cain, 1991; Turchin, 1991), and microbial transport (Li et al., 1996). Quantifying movement patterns and relating them to the spatial distribution of organisms have been approached from two different perspectives: (1) theoretical analyses using diffusion and random walk models (Okubo, 1980; Levin, 1986; 0304-3800/00/$ - see front matter © 2000 Elsevier Science B.V. All rights reserved. PII: S 0 3 0 4 - 3 8 0 0 ( 0 0 ) 0 0 3 0 9 - 4 116 H.-i. Wu et al. / Ecological Modelling 132 (2000) 115–124 Turchin, 1991; Li et al., 1996), and (2) empirical studies employing computer simulations (Crist and Wiens, 1995). Both types of approaches are useful, but each also has its defects (Marsh and Jones, 1988). Because detailed studies of organism movement over long distances are technically and logistically very difficult, the need to develop quantitative methods for the description and prediction of such movements is of surpassing importance (Levin, 1987). Random walk as a simple model of diffusion processes has been adopted in study of many biological movements (Levin, 1986; Li et al., 1996). The classical random walk model deals with a particle moving in a series of steps, where the length of the step, time between steps, and direction of the step are independent of each other and those of preceding steps (Patlak, 1953). Marsh and Jones (1988) classified random walk models of animal movement into two classes, one in which step length and direction of movement are independent, the other in which they are not. Each class is further divided into two subclasses, ‘oriented models,’ in which the direction of each step is chosen relative to a fixed compass direction, and ‘unoriented models,’ in which the direction is chosen relative to the preceding step (Marsh and Jones, 1988). Most of the commonly used random walk models assume that the direction of travel during step i of a random walk is independent of the direction taken at step i− 1. However, the movements of many entities, including animals, exhibit persistence, i.e. the direction of travel during step i is dependent of the direction travelled in step i −1. Environmental variation can lead to changes in the persistence of movement and with a suitable model (e.g. Neill, 1979; Turchin, 1991), it is possible to relate the dynamics of animal distribution to environmentally induced changes in persistence of movement. The expected magnitude of the net displacement after n steps and the variance of displacement about the origin are both useful for quantifying movements (Cain, 1991), but analytical difficulties led early researchers to restrict study to the root mean squared displacement. A one-dimensional random walk model with corre- lation between step directions was developed by Taylor (1921), who provided an expression for the expected squared-displacement after n steps. Goldstein (1951) expanded this model and corrected an error in Taylor’s expression. Okubo (1980) also gave an expression for the expected squared displacement. But his result appears to be in error. Skellam (1973) derived a term for the expected squared displacement after n steps in two dimensions. As shown below, the correct expression for the two-dimensional case first derived by Skellam (1973) and followed by Kareiva and Shigesada (1983) is identical in form to the correct solution for the one-dimensional case (Goldstein 1951). Marsh and Jones (1988) also derived the correct expression for the expected squared displacement after n steps in two dimensions. Bovet and Benhamou (1988) developed an approximation for the expected magnitude of net displacement in 2D random walks, but their expression appears to be valid only if persistence is low. In this study, we propose a model of persistent 2D random walks that can be represented by drawing turning angles from a circular distribution where the probability of turning at an angle u between steps is proportional to the area swept out within the differential rotation of the radial axis about the origin. Four well-known circular distribution functions, which can be used to correlate the turning angle between steps, were used in this study. These distributions are the Lemicon of Pascal, the elliptical, the von Mises, and the wrapped Cauchy distribution. For each of these, we obtained expressions for directional correlation in terms of the geometric parameters of the distribution, and found how an expected net displacement and its variance can be calculated from the geometric parameters via a correlation. The wrapped Cauchy distribution is especially useful for simulation because its random deviates are easily obtainable. Our model could enable a new approach for the description and prediction of animal movements, especially those of organisms like pelagic fish that seem to maintain persistent courses of travel in a relatively featureless environment. & & H.-i. Wu et al. / Ecological Modelling 132 (2000) 115–124 =4 2. The mean squared displacement p f(u)cos udu 0 Let r i be the ith-step in a 2D random walk with fixed step size l and time interval t. The resultant displacement vector after n steps, Rb n, is Rb n = %r i (1) i The expected value for the square of the n-step displacement, E[R 2n], can be computed as follows: n n i=1 j=1 E[R 2n]= E % r i % r j n n−1 n n n =nl 2 + 2l 2 % % E[cos ui, j ], (2) i=1 j\i where uij is the angle between step r i and r j. In an ideal random walk, all turning angles between steps i and j within the range −p to +p are equally probable, so that E[cos ui,j ]= 0. In a persistent random walk, the turning angle is weighted and E[cos ui,j ]" 0. Under the simplifying assumption that angular correlation g exists only between successive steps, i.e. E[cos ui,j ]= g, & p −p f(u)cos udu =g (3) with f(u) being the angular weighting function, or the turning angle distribution. For animals moving in a relatively featureless environment, a symmetric weighting function is desired. A symmetric weighting function is an even function, thus Eq. (3) can be expressed 2 & p f(u)cos udu =g (4) 0 The expected cosine for the two-step angle of turning, E[cos ui,j + 2], can then be evaluated as E[cos ui, j + 2]= & p −p du% E[cos ui, j + s ]= g s (6) From Eq. (6), Eq. (2) can be simplified to n−1 n−1 f(u)du2 i=0 n & p −p f(u%)cos(u + u%) i=0 n 1+g 1− g n n −2g , for 05 g5 1 1−g (1−g)2 (7) The above equation takes the same form as the one-dimensional random walk with persistence (Goldstein, 1951; Skellam, 1973; Okubo, 1980). Eq. (7) is also the same equation in the section Model Ib of the Appendix A of Marsh and Jones (1988) (p. 126) when E[l]= l. (Note that Eq. 1+g 1− g n (5.20) E[R 2n]= l 2 n −2g 2 on p. 70 1−g (1−g)2 in Okubo (1980) is in error: the parameter g of the second term in the square brackets is squared.) 3. Evaluation of E[cos ui,j ] (5) since p− p f(u)sin udu =0 for f(u)sin u is an odd function. Likewise, the multi-step expected cosine for the angle of turning, E[cos ui,j + s ], can be evaluated by the same procedure to give the multi-step correlation parameter =l2 i=1 j\i n−1 f(u%)cos u%du%= g 2, 0 E[R 2n]= nl 2 + 2l 2 n % g i − n− % ig i = % E[r 2i ]+ 2 % % E[r i r j ] i=1 117 p n 4. Evaluation of E[Rn ] An expression for the expected net displacement after n-steps, E[Rn ], would be useful (for example, see Cain (1991)). In evaluating E[R 2n], one evaluates the average of R 2n over n turning angles. One might attempt to evaluate E[Rn ] by simply averaging the cosine of the turning angles that appear in a radical, but determining the average of n turning angles over functions inside a radical is impossible in general. Therefore, an alternative approach must be employed. Bovet and Benhamou (1988) derived an approximated expression (their Eq. (5)) for the expected magnitude of net displacement by decomposing tracks into x and y components and applying statistical distribution theory to the distributions. Their approximation, EBB[Rn ]: l ' 0.79n 1+b , 1−b (8) 118 H.-i. Wu et al. / Ecological Modelling 132 (2000) 115–124 has shortcomings that can be easily demonstrated. In Eq. (8), we have replaced g by b to distinguish the difference between the two parameters: b refers to the correlation parameter for the expected displacement while g denotes the correlation parameter for the expected square displacement. Suppose that an animal moves in a persistent (or ‘correlated,’ in terminology of Bovet and Benhamou (1988)) random walk with steps of unit length. After n =1 step, the animal will always be a step away from the origin. For n \ 1, E[Rn ] must always be less than n since E[Rn ]= n implies straight-line movement. If EBB[Rn ]/n is calculated for various parameter values of b, a pattern emerges as shown in Fig. 1. If b \ 0.11, then EBB[Rn ]\n for small n. As b increases, the value of n for which EBB[Rn ]\ n also increases. When b= 0.99, EBB[Rn ]\ n is for n up to 157 steps, and when b =0.999, EBB[Rn ]\ n is for n up to 1580 steps. Clearly, this approximation is of questionable value for highly persistent walks. McCulloch and Cain (1989) used Taylor series expansion of E[R 2n], which involves variance (Rn ), to obtain an approximated expression for E[Rn ]. Their approximation, except for the first few steps, is in very good agreement with a simulated result, as shown in their Fig. 2. They have successfully removed the deficiency encountered with the formulation of Bovet and Benhamou. However, their approximation requires evaluations of higher-orders of E[R m n ] with m] 4 because variance of at least R 2n is needed for the dominating expansion term. Our attempt in this study is to present an alternative approximation that requires only an evaluation of the successive step correlation parameter. We employ a form-analogous approach to approximate an expression for the expected displacement that is applicable to walks of any persistence (i.e. correlation) and works well for small n. As Rb n is the resultant displacement vector at the end of n steps, the magnitude of Rb n can be expressed as Fig. 1. The expected range of Bovet and Benhamou (1988), E[Rn ]/n, in relation to random walks of n steps. For persistent random walks where the correlation, b, between direction of movement in step i and step i +1 is high, the approximation proposed by Bovet and Benhamou clearly fails. H.-i. Wu et al. / Ecological Modelling 132 (2000) 115–124 119 suggest with the idea of testing the validity of Eq. (10). This is in parallel to an approach often used in the development of physical sciences: when detailed knowledge is lacking or when one encounters an intractable mathematical difficulty, simple expressions are adopted and tested against data. The Dirac delta function first introduced in 1930 (Marsden, 1974) and the Lennard–Jones potential for describing the interaction between two molecules (Reif, 1965; Barrow, 1988) are the two well-known examples. The expected value for the magnitude of the resultant after two steps, E[R2] can be evaluated E[R2]= l Fig. 2. With E[R2] being the expected magnitude of resultant displacement after two steps with equal length l, the angle u0 is the expected angle of deviation between successive steps. Rn + l ' n−1 n n+2 % % cos ui, j (9) i=1 j\i Notice the analogy between Eqs. 2 and 9. If cos ui,j is replaced with multi-step correlation parameter g j − i in Eq. (9), then the result would be the expected root mean square of Rb n, given that the expression of Eq. (9) is identical to Eq. (7). The successive-step angular correlation g for the expected square displacement is defined in Eq. (3), which led to the expression for E[R 2n]. If we could find a successive-step correlation parameter b, then by the analogy to the form of Eq. (7), the expected displacement after n-steps would become: E[Rn ]: l ' 1+b 1 −b n − 2b , for 0 5 b5 1 1−b (1 − b)2 (10) n This approximate expression for E[Rn ] was not derived through rigorous mathematical arguments, nor do we claim that the expression ' E l −1 n n+2ni = 1 j \ i cos ui, j proximated by l ' n is accurately ap- −1 n E[n]+ 2ni = 1 j \ i E[cos ui, j ]. Only because there is need for a simple applicable expression for the expected displacement, do we & p & −p = 4l p 0 where & u0 cos = 2 2 p 0 f(u) 2 + 2cos udu u u0 f(u)cos du =2lcos , 2 2 (11) u f(u)cos du 2 The angle u0 can be interpreted as the expected angle of deviation between successive steps as shown in Fig. 2. Let b be the cosine of the expected turning angle between successive steps, b= cos u0 = cos[2cos − 1 E[R2] E[R2]2 ]= − 1; 2l 2l 2 (12) this will be used as the successive-step angular correlation parameter for the expected displacement. Since it is impossible to evaluate the expected angle of deviation for a multi-step angle of turning, we follow the result given in Eq. (6) and use the form-analogous approach to propose that b s be interpreted as cos u0 j,j + s, or b s : cos u0 j, j + s (13) Once again, Eq. (13) was not obtained through rigorous mathematical derivation. The relationship between E[Rn ] and b for a wrapped Cauchy distribution is shown in Fig. 3. Note that E[Rn ] rendered from Eq. (10) is always less than n, for n\ 1. To test the validity of the approximated expression for E[Rn ], we compared results from an unbiased 2D random walk, E[Rn ]= np/2 (that 120 H.-i. Wu et al. / Ecological Modelling 132 (2000) 115–124 is same as Eq. (3) in Cain (1991)(p. 2138), and Eq. (10). For a uniform turn-angle distribution, f(u) = 1/2p, b= − 0.18943 is obtained from Eqs. 11 and 12. With this b value, Eq. (10) over-estimates the value obtained from the ideal 2D random walk by less than 7%. After the first couple of steps, Eq. (10) starts to under-estimate the ideal 2D random walk. The relative error increases as n increases and the under-estimation is capped by 6.85% as n approaches infinity. A computer simulation was also employed to generate persistent random walks from which values of E[Rn ] were calculated and averaged. Simulations were performed using the Pascal/VS programming language on an AMDAHL 470V8 computer and Mathematica on a SUN SPARC 1+ workstation. Random angles were generated by random deviates drawn from a wrapped Cauchy distribu- Fig. 4. The relative percent difference of the expected magnitude of the net displacement after n-steps, E[R2], calculated using Eq. (10), under-estimates the average simulated magnitude of the net displacement by at most 8%. Relative percent difference is defined as the difference between calculated and simulated results divided by the average magnitude of simulated displacement. Angles were drawn from a wrapped Cauchy distribution, with parameter o, which is equal to b. The number of steps covered by the walk is denoted by n. Symbols , , , and are for b=0.01, 0.7, 0.9, and 0.99, respectively. tion and the walks proceeded with constant unit step size. The simulated tracks were sampled to give 2500 resultants for each combination of n and b. The averaged magnitude of these resultants is compared with the calculated values of E[Rn ] in Fig. 4. The relative small errors and their lack of tendency to increase as n increases indicate that the proposed relationship of Eq. (10) is a reasonable approximation. Estimates of the variance of the magnitude of the net displacement V(Rn )= E[R 2n]− E[Rn ]2 Fig. 3. In a persistent random walk, the expected magnitude of the net displacement after n-steps, E[R2], increases as the angular correlation between steps, b, increases. There is good agreement between the calculated expected displacement from Eq. (10) (solid circles) and estimates of Bovet and Benhamou (1988) (solid triangles) for small b; agreement becomes increasingly poor as b increases. Expected net displacements by simulation (open circles) agree well with both approximations for small b but diverge from Bovet and Benhamou estimates as b increases. (14) are sensitive to errors in estimation of E[Rn ]. Fig. 5 compares estimates of one standard deviation using the approximated expression for E[Rn ] with estimates from a simulation. As Eq. (10) tends to under-estimate displacements given by the simulation that leads to larger estimates of variance than the simulation. Hence, statistical results using values from Eqs. 10 and 14 are conservative. 5. Symmetrical turning angle distributions Several useful symmetrical turning angle distributions (also known as circular distributions (Batschelet, 1981)) are reviewed. From a statistical H.-i. Wu et al. / Ecological Modelling 132 (2000) 115–124 121 point of view, the von Mises distribution is the circular distribution function of choice. However, only the wrapped Cauchy turning angle distribution provides a closed-form cumulative distribution function. For illustration, we present the case for the wrapped Cauchy distribution function here, which is modified from the elliptical function. Other distributions are included in the appendix. The geometric form of a wrapped Cauchy function, shown in Fig. 6, is expressed in polar coordinates as r=a ' 1− o 2 , 0 Bo B1 1+ o −2ocos u 2 (15) The total area swept by this radius is pa 2; thus, the probability density function for a wrapped Cauchy distribution is f(u) = 1− o 2 , 0B o B 1 2p(1+ o 2 − 2ocos u) (16) The correlation parameter gC, where the subscript C denotes Cauchy, equals o. The expression for E[R2] can be obtained by using Eqs. 11 and 16, E[R2]= 2l(1− o 2) p & p 0 cos(u/2) du 1 + o 2 −2ocos u Fig. 6. Equi-areal plots of the Cauchy distribution function with o =0.2 and 0.8. Although the total area enclosed by each figure is the same, the areas sustained by a given angle of deviation u are clearly different. = = 4l(1−o 2) p 2l(1+o) p o & 1 0 dx (1− o)2 + 4ox 2 tan − 1 2 o , 1− o (17) and thus the correlation parameter of angular deviation between successive steps for a wrapped Cauchy distribution is bC = Fig. 5. The variance of the expected magnitude of the net displacement, V[Rn ], predicted by Eq. (14) (solid circles) exceeds the variance obtained by simulation (open circles) for random walks with high persistence (upper pair of curves, b= 0.9) and low persistence (lower pair of curves, b =0.01). 2(1+ o)2 2 o tan − 1 op 2 1−o 2 −1 (18) The cumulative distribution FC(u) for this geometric function can be obtained by FC(u)= & u −p f(u%)du% 122 H.-i. Wu et al. / Ecological Modelling 132 (2000) 115–124 1+o u 1 1 tan + = tan − 1 1−o 2 p 2 (19) The random deviate n is then n= 2tan − 1 n 1− o 1 tan RN − p , 1+ o 2 (20) where 05RN 51 refers to random numbers drawn from a uniform distribution. From Eq. (20), the generated nth step coordinates xn and yn are, xn = xn − 1 + l cos n yn = yn − 1 +l sin n (21) 6. Conclusion and discussion The usefulness of the expected displacement in studying the spread of biological organisms is well recognized (McCulloch and Cain, 1989; Cain, 1991). Unfortunately, an exact expression for the expected displacement has not been derived successfully thus far. Many attempts to obtain an accurate approximation have been made in the past. We have succeeded in developing an approximation that is simple and easy to apply. In deriving E[R 2n], the expected squared displacement for an n-step 2D random walk, we assumed that angular correlation exists only between successive steps. This assumption results in a multi-step correlation that is simply the correlation parameter raised to sth power, where s is the number of intervening steps. This leads to an expression for E[R 2n], which has the same form as the E[R 2n] in the one-dimensional case. The derivation of E[Rn ] is not that mathematically fortunate, however. Yet an expression for E[Rn ] would be useful in many ecological studies. As we were unable to establish the relationship between the angular deviation for multiple steps and b, we followed the result obtained in Eq. (6) and propose that, using a form-analogous approach, the expected multi-step correlation for angular deviation can roughly be expressed as b s where s is the number of steps. The expected magnitude of Rb n, approximated by Eq. (10), and the calculated variance given in Eq. (14) are compared with simulation results for the wrapped Cauchy distri- bution using various values of the parameter b. The agreement between simulation and analytical results reasonably supports the validity of the relationship proposed in Eq. (10). Our expression for E[Rn ] extends the range of b over which estimates of the expected magnitude of the net displacement can be made, and is reasonably accurate even when n is small. McCulloch and Cain (1989) used Taylor series expansion of Eq. (10) to obtain their approximation while retaining the definition of g intact. Mathematically, their approach is very rigorous. However, evaluations of higher-orders of E[R m n] for m] 4 poses some difficulties for most of the statistical distribution functions. One alternative is to use a numerical approach as they suggested to generate required probability distributions. Our form-analogous approach has provided another alternative. The result is simple and only one parameter needs to be evaluated. In this study we have assumed fixed step lengths for simplification. Although step lengths for moving animals are highly variable, we believe this approximation is not sensitive to variable step lengths if n is sufficiently large. A variable length simulation is currently planned to study this effect. There are at present very few studies on the detailed movement patterns of animals over long distances. It is very difficult to be certain that the appropriate model has indeed been chosen in such a study (Marsh and Jones, 1988). Our formulation provides an alternative way for modeling animal movements with persistent random walks over long distances. Our results can also be very useful in estimating the expected net displacement of animal movements when the turn-angle distributions are the well-established functions. These results also are useful as a potential forecasting tool for the fishing industry and management agencies, to locate large aggregated fish schools in the relatively featureless ocean. Acknowledgements We thank W.M. Childress for his comments on an early version of this paper. Signed reviews by H.-i. Wu et al. / Ecological Modelling 132 (2000) 115–124 M.L. Cain and three anonymous reviewers were helpful. This work was partially supported by the National Science Foundation under the grants BSR-91-09240, DEB-93-06679 and DEB-9411976, and DOE/Sandia National Laboratories under contract BE-0229. This is Sevilleta LTER publication no. 138. Useful symmetrical turn-angle distributions for which cumulative distribution functions can be generated from random numbers (Cain, 1985) are presented below: The polar coordinate expression for the Lemicon of Pascal (Eq. 11.32, p. 44 of Spiegel, 1968) is given as (A1) The total area A enclosed by the boundary is & 1 A= 2 p −p r 2du = pa 2 o2 1+ 2 (A2) The turning angle distribution is obtained by taking the ratio of the area swept-out per unit angular displacement to the total area A, f(u) = r 2 (1+ocos u)2 = 2A p(2+o 2) gL = & −p f(u)cos udu = 2o 2+o (b) Elliptical distribution r= a(1− o 2) , 0BoB 1, 1− ocos u and the area pa 2 1 − o 2, the elliptical turning angle distribution function is f(u)= (A4) 4l p(2+o 2) = 8l p(2+o 2) & & u (1 + ocos u)2cos du 2 p 0 1 (1 + o −2ox 2)2dx, 0 where x =sin u/2, thus (1− o 2)3/2 , 0BoB 1 2p(1− ocos u)2 (A6) The turning angle correlation parameter for an ellipse, ge, is equal to the parameter of conic section, or eccentricity, o, ge = & p −p f(u)cos udu= o (A7) The E[R2] value can be evaluated as follows: u cos 2 2l(1−o 2)3/2 p E[R2]= du 2 p 0 (1−ocos u) & (A3) The correlation parameter of angular deviation between successive steps bL can be evaluated by mean of E[R2]. From Eq. (11), one can evaluate E[R2]= (A5) The expression for bL can then be obtained by substituting Eq. (A5) into Eq. (12). = The correlation parameter gL, here again the subscript refers to the distribution function employed, can then be expressed in terms of the parameter of the Lemicon of Pascal, o, p 8l 2 7 1+ o − o 2 . p(2+ o 2) 3 15 From the polar coordinate expression for an ellipse, Appendix (a) Lemicon of Pascal distribution r = a(1+o cos u) 0 B o B1 E[R2]= 123 4l(1−o 2)3/2 p ' & 1 0 dx (1− o+ 2ox 2)2 2l = 1 − o 2 p + (1+o 2)3 tan − 1 2o ' n 2o , 1− o (A8) and be can be obtained by using Eq. (12). (c) von Mises distribution The von Mises probability density function f(u) is the circular analog of the normal distribution and is frequently used in studying stochastic angular processes (Batschelet, 1981). It is given by f(u)= ekcos u , k =0, 2pI0(k) (A9) H.-i. Wu et al. / Ecological Modelling 132 (2000) 115–124 124 where k is the von Mises parameter and I0(k) is the modified Bessel function of order zero. The von Mises correlation parameter gM is given as gM = 1 pI0(k) & p ekcos ucos udu = 0 I1(k) , I0(k) (A10) since & p ekcos ucos udu =pI1(k) 0 (Eq. 9.6.19, p.376 of Abramowitz and Stegun, 1972), where I1(k) is the modified Bessel function of order one. The evaluation of E[R2] is given as E[R2]= 2l pI0(k) & 0 k = = p 4le pI0(k) ' ekcos ucos & p u du 2 2 e − 2kx dx 0 2 le k erf( 2k), pk I0(k) (A11) which leads to the expression for bM. References Andow, D.A., Kareiva, P.M., Levin, S.A., Okubo, A., 1990. Spread of invading organisms. Landsc. Ecol. 4, 177–188. Abramowitz, M., Stegun, I.A., 1972. Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables, National bureau of Standards, 10th ed. Wiley, New York, p. 1046. Batschelet, E., 1981. Circular Statistics in Biology. Academic, New York, p. 371. Barrow, G.M., 1988. Physical Chemistry. McGraw-Hill, New York, p. 859. Bovet, P., Benhamou, S., 1988. Spatial analysis of animal movements using a correlated random walk model. J. Theor. Biol. 131, 419–433. Cain, M.L., 1985. Random search by herbivorous insects: a simulation model. Ecology 66, 876–888. Cain, M.L., 1991. When do treatment differences in movement behaviors produce observable differences in long-term displacements? Ecology 76, 2137–2142. Crist, T.O., Wiens, J.A., 1995. Individual movements and estimation of population size in darkling beetles (Coleoptera: Tenebrionidae). J. Anim. Ecol. 64, 733–746. . Goldstein, S., 1951. On diffusion by discontinuous movements and on telegraph equation. Q. J. Mech. Appl. Math. 4, 129 – 156. Kareiva, P.M., Shigesada, N., 1983. Analyzing insect movements as a correlated random walk. Oecologia 56, 234 – 238. Levin, S.A., 1986. Random walk models of movement and their implication. In: Hallam, T., Levin, S.A. (Eds.), Mathematical Ecology: An Introduction. Springer – Verlag, Berlin, pp. 143 – 154. Levin, S.A., 1987. Ecological and evolutionary aspects of dispersal. In: Teramoto, E., Yamaguti, M. (Eds.), Mathematical topics in population biology, morphogenesis and neurosciences, Lecture Notes in Biomathematics, vol. 71. Springer – Verlag, Berlin, pp. 80 – 87. Li, B.L., Loehle, C., Malon, D., 1996. Microbial transport through heterogeneous porous media: random walk, fractal and percolation approaches. Ecol. Model. 85, 285 – 302. Li, Y., Li, B.L., Zhang, G., 1987. Insect Ecology. Hubei Science and Technology, Wuhan, p. 278. Marsden, J.E., 1974. Elementary Classical Analysis. Freeman, San Francisco, CA, p. 549. Marsh, L.M., Jones, R.E., 1988. The form and consequences of random walk movement models. J. Theor. Biol. 133, 113 – 131. McCulloch, C.E., Cain, M.L., 1989. Analyzing discrete movement data as a correlated random walk. Ecology 70, 383 – 388. Neill, W.H., 1979. Mechanics of fish distribution in heterothermal environments. Am. Zool. 19, 305 – 317. Okubo, A., 1980. Diffusion and Ecological Problems: Mathematical Models. Springer – Verlag, New York, p. 254. Okubo, A., Levin, S.A., 1989. A theoretical framework for data analysis of wind dispersal of seeds and pollen. Ecology 70, 329 – 338. Patlak, C.S., 1953. Random walk with persistence and external bias. Bull. Math. Biophys. 15, 311 – 338. Reif, F., 1965. Fundamentals of Statistical and Thermal Physics. McGraw-Hill, New York, p. 651. Skellam, J.G., 1973. The formulation and interpretation of mathematical and diffusionary models in population biology. In: Bartlett, M.S., Hiorns, R.W. (Eds.), The Mathematical Theory of the Dynamics of Biological Populations. Academic, London, pp. 63 – 85. Spiegel, M.R., 1968. Mathematical Handbook of Formulas and Tables (Schaum’s Outline Series). McGraw-Hill, New York, p. 271. Taylor, G.I., 1921. Diffusion by continuous movements. Proc. Lond. Math. Soc. 20, 196 – 212. Turchin, P., 1991. Translating foraging movements in heterogenous environments into spatial distribution of foragers. Ecology 72, 1253 – 1266.
© Copyright 2026 Paperzz