Bulkin of Mathematical Biology, Vol. 59, No. 2, pp. 2.

55-262, 1997
Elsevier Science Inc.
0 1997 Society for Mathematical Biology
039x3240/97 $17.00 + 0.00

SOO92-8240(%)00071-7

MATRIX POPULATION MODEL WITH

DENSITY-DEPENDENT RECRUITMENT
FOR ASSESSMENT OF AGE-STRUCTURED
WILDLIFE POPULATIONS

?? A. L. JENSEN
School of Natural Resources & Environment,
University of Michigan,
Ann Arbor, MI 48109-1115, U.S.A.

A logistic density-dependent matrix model is developed in which the matrices contain only
parameters and recruitment is a function of adult population density. The model was
applied to simulate introductions of white-tailed deer into an area; the fitted model
predicted a carrying capacity of 21.5 deer, which was close to the observed carrying capacity
of 220 deer. The rate of population increase depends on the dominant eigenvalue of the
Leslie matrix, and the age structure of the simulated population approaches a stable age
distribution at the carrying capacity, which was similar to that generated by the Leslie
matrix. The logistic equation has been applied to study many phenomena, and the matrix
model can be applied to these same processes. For example, random variation can be added
to life history parameters, and population abundances generated with random effects on
fecundity show both the affect of annual variation in fecundity and a longer-term pattern
resulting from the age structure. 0 1997 Society for Mathematical Biology

1. Introduction. Matrix models have been applied to assess possible inter-

ventions for conservation of endangered species and to evaluate possible
outcomes of species introductions (e.g., Caswell, 1989; Crouse et al., 1987;
Heppell et al., 1994), but the matrix models applied in these studies are not
density-dependent models. Available density-dependent matrix models have
been complex models in which the transition matrix is a function of
abundance, and these models require many parameters in addition to those
of a life table and Leslie matrix (e.g., Leslie, 1959; Liu and Cohen, 1987;
Caswell, 1989). Jensen (1995, 1996) proposed simple density-dependent
matrix models in which the matrices consisted only of parameters, and no
additional parameters or functional forms beyond those of a life table and
Leslie matrix were necessary. The simple matrix models accurately de-
scribed population growth to a carrying capacity when the initial population
size was small, but small perturbations of a population’s age structure near
the carrying capacity can result in unrealistic age structures. In this study, a
density-dependent matrix model was obtained in which there was a unique
255
256 A. L.JENSEN

stable age distribution. The matrix model is still simple, and illustrates the
flexibility of constructing matrix models in which the transition matrices
contain only parameters and not population density. The model was applied
to a white-tailed deer population introduced into an area where white-tails
had been extirpated.

2. Model Development. The matrix model was developed in terms of

births and deaths just as Pielou (1969) developed the logistic equation. The
modelling approach was flexible, and the model could be developed with
death as well as birth density-dependent, but the model will be applied to a
white-tailed deer population in which only births are density-dependent, so
only births will be modelled as density-dependent. The vector of age
structure at time t + 1, N,, 1, is given in terms of births B,, 1 and survival
from time t, SN,, as

Nt+1=%+1 +m (1)

where S is a survival matrix with annual survivals on the subdiagonal and

zeros elsewhere. The vector for births can be written as

B t+l=RN,--GN,N,, (2)

where R is a fecundity matrix with terms m,s,_ 1, i = 1,2,. . . , k - 1, along

the first row, except for the last column, which is zero, and zeros every-
where else. mi is the number of young produced by an individual of age i,
si is survival from age i - 1 to i, and k is the oldest age attainable. The
matrix G, which measures the effect of density on fecundity, has terms gi,
i = 0,1,2,. . . , k - 1, along the first row, except for the last column, which is
zero, and zeros everywhere else. The matrix N1 is a diagonal matrix with
age structure at time t along the diagonal, and N, is the age-structure
vector at time t. Bringing the above equations together gives

N t+l =MN, - GN,N,, (3)

where it4 = R + S is the Leslie (1945) population projection matrix. For

species that reproduce annually, such as most wild populations, the Leslie
matrix can be written in terms of life table components as (e.g., Starfield
 MATRIX POPULATION MODEL 257

and Bleloch, 1986)

m1so WSl m3s2 *-- mkSk-I 0

SO 0 0 **- 0 0

M= 0 Sl
0 ... 0 0 (4)

0 0 0 *** sk-l ’

where ml, si, and k are as defined above. In equation (3), the matrices
contain only parameters and not population abundance.

3. Relation of New Model to Logistic Equation. Equation (3) is closely

related to the logistic equation. Adding and subtracting the age-structure
vector to the right side of equation (3) gives

N,+,=N,+(M-I)N,-GNN,, (5)

and at equilibrium, where N,, 1 = N,, equation (5) gives G =K-‘(M-I),

and, therefore,

N t+l = N, + (M-I)N, -K-‘(M-I)iV,N,. (6)

Equation (6) is analogous to the scalar discrete-time logistic equation (May,

1974), where (M - I) is analogous to the intrinsic rate of increase T. Both
the matrix G and the matrix K-’ are singular, but the matrix K-’ can be
calculated as G(M- I)-‘. In equation (3), the matrix K directly deter-
mines only the number of individuals of age 0, but the age structure at the
carrying capacity can be determined by projection from any initial age
structure.
The matrix M-I is analogous to the intrinsic rate of increase r of the
logistic equation, and the principal eigenvalue of M - I is 4 = A - 1, where
A is the principal eigenvalue of M and 4 = e’ - 1. If the population is
below the carrying capacity, the rate of population growth increases as A
increases, and A increases as the fecundity and survival terms of the matrix
M increase. The logistic matrix models exhibit stability characteristics
comparable to those of the discrete-time logistic equation explored by May
(1974); stability behavior of the matrix model was explored in detail by Liu
and Cohen (1987) and is discussed by Jensen (1995) and Caswell(l989).
For values of 4 that produce a constant population size, simulations
indicate there is a unique stable age structure at the carrying capacity for
258 A. L. JENSEN

all initial age structures, even when the initial population size is near the
carrying capacity. The age structure at the carrying capacity differs some-
what from the stable age structure of the Leslie (1945) matrix, but if the
population age structure is projected using the Leslie matrix with the
number at age 0 fixed at the number determined by equation (3), the stable
age structure generated with the Leslie matrix is the same as that gener-
ated with equation (3).

4. Application and Discussion. The matrix model with a density-depen-

dent recruitment function (equation (3)) was applied to describe the dynam-
ics of the George Reserve deer herd. The George Reserve is a 464-ha
reserve located in southeastern Michigan (McCullough, 1979). Deer had
been extirpated from the area before a 2.9-m high fence around the reserve
was completed in 1927. The following year, white-tailed deer from Grand
Island on Lake Superior were released on the property. In 1933, the first
annual deer drive was held and 160 deer were counted (McCullough, 1979).
Later studies indicated that the prehunt carrying capacity of the reserve
was about 220 deer (McCullough, 1979), and these studies also provided
age structure and harvest data (McCull6ugh, 1979).
It appears that for many animal populations, density dependence oper-
ates either on fecundity or early life stages (e.g., Charnov, 1993). McCul-
lough (1979) found that for the George Reserve deer herd, the carrying
capacity was determined by a linear relation between birth per individual
and population density, and that mortality was density-independent. McCul-
lough (1979) examined several recruitment relations with several different
variables and concluded that recruitment rate was best related to the total
number of adults as y = 0.986 - 0.005x, where y was the number of fawns
per individual (male and female) and n was the total number of animals,
adults and fawns, alive after the fall hunt; this indicates m, = 0.986 and
gi = 0.005 for a 11age groups. Combining these fecundity terms with survival
terms given by McCullough (1979) gives the projection matrix M as

‘0.7030 0.6359 0.6389 0.5522 0.5551 0.7661 0

0.713 0 0 0 0 0 0
0 0.645 0 0 0 0 0
0 0 0.648 0 0 0 0
0 0 0 0.560 0 0 0
0 0 0 0 0.563 0 0
, 0 0 0 0 0 0.777 0
 MATRIX POPULATION MODEL 259

and the elements of the matrix for density-dependent effects G are 0.005
along the first row, except for the last column, which is zero, and zero
everywhere else.
The fitted model (equation (3)) was applied to simulate the dynamics of
the George Reserve deer herd with an initial value of four age-l individuals
(Fig. 1). The d eer population had considerable capacity to increase, and
attained a carrying capacity of 215 deer after less than 30 years; this is close
to the carrying capacity of 220 deer estimated by McCullough (1979). With
equation (31, there was a unique stable age distribution (Table 1) and the
deer population approached this stable age distribution regardless of the
initial age structure. With a population size of 215 deer, the Leslie matrix
gives the stable age structure for ages 0 to 6 as 105, 56, 27, 13, 6, 2, and 1,
but projection with the Leslie (1945) matrix results in a continually chang-

250-

200--

s 150--

z
n
5
m
< lOO--

‘:I-,: YEARS AFTER

35
INTRODUCTION
Figure 1. Calculated abundance of white-tailed
initial value of four age-l individuals.
40 45 50

deer on George Reserve with

260 A. L. JENSEN

Table 1. Age structure of introduced white-tailed deer population

Years after introduction
Age 0 1 2 3 4 5 10 15 20 25 30

0 0 : 3 4 5 I 22 50 :;
: 04 2 : 21 31 41 13
6 31
17 29
3 0 8 1 0 0 0 3 9 17 21 22
z 0 0 0 0 0 01 42 49 11
6 12

6 0 8 8 0 0 0 0 1 3 4 ;

ing population size. If the population age structure is projected with the
Leslie matrix with the number at age 0 fixed at 76, the stable age structure
generated with the Leslie matrix is the same as that generated with
equation (3).
The carrying capacity in equation (6) has been modelled in several
different ways. Jensen (1995) assumed

Nr+l = N,+D(N,)N,, (71

and let D(N,) = (K - N,)/K, where K was the carrying capacity and Nt
was the sum of the elements in the age-structure vector at time t. Jensen
(1996) applied equation (6) with a carrying capacity for each age group
defined by letting K = diag{c,K}, where c, was the proportion of the
population of age x and K was the total population size at the carrying
capacity. With carrying capacities defined with D(N,) or diag{c,K}, unreal-
istic age structures could occur when a population returned to the carrying
capacity after a small perturbation.
The logistic equation has been applied to study many phenomena: time
lags (Hutchinson, 19481, extinction (Leigh, 1981), minimum viable popula-
tion sixes (Clark, 19731, harvesting (Schaefer, 1954), bioeconomics (Gordon,
1954), and the effects of random environmental variation (Beddington and
May, 1977). These same processes can be modelled and investigated with
the matrix model (e.g., Jensen, 1994, 1996). Jensen (1994) applied a
density-dependent matrix model with a time lag, and Jensen (1996) applied
a density-dependent matrix model to study optimal harvesting. The matrix
model gives a more realistic biological interpretation of many processes
that have been modelled with the scalar logistic equation. For example,
Beddington and May (1977) applied the logistic equation to examine the
effect of random environmental variation on population growth by adding a
normal distributed random variable to the intrinsic rate of increase in the
linear term. In the scalar logistic equation, the biological implication of the
random variable was uncertain, but it was assumed to represent random
environmental effects on fecundity (Beddington and May, 1977). In the
 MATRIX POPULATION MODEL 261

matrix model, random variation also can be added, but in the matrix model
the underlying biology is explicit. For example, to simulate the effect of
environmental variation on fecundity, a random variable can be added to
the fecundity terms as m, = mi + e, where e is a normal distributed random
variable with a mean of zero. A simulation was run for the George Reserve
deer herd with random variation in fecundity and a standard deviation of
0.25 (about 25% of mean fecundity). The coefficient of variation for
numbers of age 0 reported by McCullough (1979) was 0.24. The normal
random variables were calculated using the sum of 12 uniformly distributed
random variables. The population abundances generated with random
effects on fecundity show both the effect of annual variation in fecundity
and a longer-term pattern resulting from the time lag caused by the
population age structure (Fig. 21, and the fluctuations resemble the actual
deer abundances reported by McCullough (1979). The simple matrix form

350-

300--

E 250--
Y
n
& 200--

S
5 150--
S
z
Q lOO--

.1~;;:::::
50--

1
0 10 20 30 40 50 60 70 80 90 loo
YEARS AFTER INTRODUCTION
Figure 2. Calculated abundances of white-tailed deer on George Reserve with
initial value of four age-l individuals and a normal distributed random variable
added to fecundity.
262 A. L. JENSEN

of the logistic model adds biological realism with little additional

complexity.
Density-dependent matrix models can be a useful tool for modelling the
introduction of species and the effects on abundance of changes in mortal-
ity and survival parameters of endangered species. The model developed
here requires no additional parameters or mathematical functions, and it is
easy to apply to either laboratory or field populations when a combined life
and fecundity table and an estimate of the carrying capacity are available; it
provides a link between life table parameters, the logistic growth equation,
and Leslie’s original population projection matrix.

REFERENCES
Beddington, J. R. and R. M. May. 1977. Harvesting natural populations in a randomly
fluctuating environment. Science 197, 463-465.
Caswell, H. 1989. Matrix Population Models. Sunderland, MA: Sinauer Associated, Inc.
Charnov, E. L. 1993. Life History Inuariants. New York: Gxford University Press.
Clark, C. W. 1973. The economics of overexploitation. Science 181, 630-634.
Crouse, D. T., L. B. Crowder and H. Caswell. 1987. A stage-based population model for
loggerhead sea turtles and implications for conservation. Ecology 68, 1412-1423.
Gordon, H. S. 1954. The economic theory of a common property resource: the fishery.
J. Pol. Econ. 62, 124-142.
Heppell, S. S., J. R. Walters and L. B. Crowder. 1994. Evaluating management alternatives
for red-cockaded woodpeckers: a modeling approach. .I. Wildlife Management 58,479-487.
Hutchinson, G. E. 1948. Circular causal systems in ecology. Ann. N.Y. Acad. Sci. 50,
221-246.
Jensen, A. L. 1995. Simple density dependent matrix model for population projection. Ecol.
Model. 77, 43-48.
Jensen, A. L. 1996. Density dependent matrix yield equation for optimum harvest of age
structured populations. Ecol. Model. 88, 125-132.
Leigh, E. C. 1981. The average lifetime of a population in a varying environment. J. Theor.
Biol. 90, 213-239.
Leslie, P. H. 1945. On the use of matrices in certain population mathematics. Biometrika 33,
183-212.
Liu, L. and J. E. Cohen. 1987. Equilibrium and local stability in a logistic matrix model for
age-structured populations. J. Math. Biol. 25, 73-88.
May, R. M. 1974. Biological populations with non-overlapping generations: stable points,
stable cycles, and chaos. Science 186, 645-647.
McCullough, D. R. 1979. The George Reserve Deer Herd. Ann Arbor: University of Michigan
Press.
Pielou, E. C. 1969. An Introduction to Mathematical Ecology. New York: Wiley.
Schaefer, M. B. 1954. Some aspects of the dynamics of populations important to the
management of the commercial marine fisheries. Bull. Inter-Am. Trop. Tuna Comm. 1,
27-56.
Startield, A. M. and A. L. Bleloch. 1986. Building Models for Conservation and Wdlijie
Management. New York: Macmillan.

Received 4 December 1995

Revised version accepted 8 August 1996