You are on page 1of 29

J. Math.

Biology (1981) 12:265-293

,Journal of

Mathematical 131ology
by Springer-Verlag1981

Optimal Fishery Policy for Size-Specific, Density-Dependent Population Models


Louis W. Botsford
University of California, Bodega Marine Laboratory, P. O. Box 247, Bodega Bay, CA 94923, USA

Abstract. Optimal harvest policy is derived for a size-specific population model based on the continuity equation. In this model both growth and recruitment rates can depend on either the number of individuals of specific sizes in the population or the level of food available. Necessary conditions for values of fishing pressure and size limit that maximize the present value of the resource are obtained and are interpreted in economic terms. The general solution obtained here reduces to solutions obtained previously for some special cases: the single age-class model, and the linear age-dependent model. Solutions involving constant fishery policy axe sought for several different, specific versions of the general model. In each of these versions a constant policy solution is not optimal. This implies that for a general, realistic model the policy that maximizes present value is a time-varying or "pulse fishing" policy. The theoretical and practical implications of the results are discussed in the light of existing results. Key words: Fishery policy - Pulse fishing size specific models - Age-specific models - Fishery economics

1. Introduction

In spite of the fact that individual growth, reproductive and mortality rates profoundly affect the dynamic behavior of exploited fish populations, they are seldom explicitly included in models for analysis of optimal harvest strategies. For example, the logistic model, applied to fisheries.by Graham (1935), Schaefer (1954) and others, reflects a compensatory decrease in population growth rate as population size increases, but does not explicitly include the density-dependent changes in growth, reproductive or mortality rates that are responsible for the decrease. A second example is the widely used model of Beverton and Holt (1957). Although they examined the effects of various density-dependent mechanisms on

Present address: Departmmat of Wildlife and Fisheries Biology, University of California, Davis, CA

95616, USA

0303 - 6812/81/0012/0265/$05.80

266

L.W. Botsford

optimal policy obtained from their dynamic pool model, analytical solutions of this problem using their model have not included density-dependence (cf., Goh, 1973; Clark et al., 1973). A third example, the stock-recruitment model developed by Ricker (1954), reflects the density-dependent effect of stock on recruitment level, but does not explicitly include the responsible biological mechanism. Analytical solutions to the optimal harvest problem using this model do not include multiple years classes nor growth and survival beyond recruitment. Several workers (of., Waiters, 1969; Silliman, 1969; Beverton and Holt, 1957) have combined a stockrecruitment model with a model of individual growth and mortality, and examined responses to different harvest policies through simulation. However, there have been few attempts to analytically determine optimal harvest policy using models of similar complexity (a recent exception is Getz, 1979). The tools necessary to formulate a more realistic analogue of a fish population are available in current constructs of theoretical biology. Age-size specific models developed by Sinko and Streifer (1967) reflect specifically the growth, mortality and reproductive rates of individuals in a population. Their model, a form of the continuity equation, reflects the way that growth, reproduction and mortality rates influence the number of animals at each size and age as a function of time. Models of density-dependent effects in growth, mortality and reproductive rates (assuming they are food-related) can be based on concepts of individual bioenergetics (of., Warren and Davis, 1967; Warren, 1971; Paloheimo and Dickie, 1970). The work described herein is an analysis of optimal harvest policy for a more realistic, general population model based on these two kinds of models. After obtaining necessary conditions for optimal harvest in the general model specific cases are examined in detail. Models as complex as those used here are generally not used in fishery analysis due to the lack of adequate data to completely specify them. Indeed, the amount of data needed to completely specify a complex model of a specific population-is seldom available for a real fishery. However, rather than attempting to define optimal policy for a specific population, the intent here is to expose economic aspects of population behavior that are occluded by the simplifications inherent in other models. These may then suggest altered strategies based on current data, or new data that are vital to proper management~ Some mathematically similar analyses have been pursued in terms of Leslie matrix models (1945); however, results obtained are of limited value due to the linearity of the model (see Mendelssohn, 1976, for a review). Because the model is linear, only a population with dominant eigenvalue (1) greater than one (i.e., a nondecreasing population) could yield a sustained harvest. This model not only yields unreasonably optimal policy but is biologically unrealistic in that it reflects a population that if left alone would increase exponentially. The optimal harvest problem has been made mathematically more tractable by including constraints (e.g., not reducing i below one or other more general linear constraints) (cfo, Rorres and Fair, 1975; Rorres, 1976). The crucial limitations of this model axe now commonly recognized (cf., Mendelssohn, 1976; Beddington and Taylor, 1973; Reed, 1980)o The approach taken here is based on the belief that more useful results are obtained from models that include essential, biologically realistic nonlinearities, than are obtained from linear models with arbitrary, auxiliary constraints.

Optimal Size-Specific Fishery Policy

267

2. The Model The model used was chosen to realistically reflect the dynamic biological behavior and important economic characteristics of an exploited population. Since growth, reproductive and mortality rates of individuals in aquatic populations usually depend more on size than on age, and individual size is the variable of economic importance, a continuity equation in a size variable was chosen as the basic model. A review of relevant features of fish and crustacean populations (Botsford, 1978) indicated that the most important density-dependent effects in commercially important populations are felt through mortality in the younger stages and individual growth rate (usually of juveniles). These rates are therefore modeled as density-dependent. Density~epcndent effects on recruitment and growth rates are modeled here in two ways: (a) food-dependent and (b) population-dependent. In the former-instance, recruitment and growth rates depend on the amount of food available to the population, which is assumed to depend only on the rate at which food is produced or made available to the population and the rate at which the population consumes food~ The production rate is assumed to be constant and the total population consumption rate is the sum of consumption rates o f each individual in the population. In the simpler, population-dependent case individual recruitment and growth rates depend on a weighted sum of individuals of each size present in the population. Animals of different sizes in the population are weighted according to their relative effect on each density-dependent rate. For example, if densitydependent recruitment were due to cannibalism, the effect of different size individuals on the recruitment would be modeled by a weighting function that reflected the propensity of an individual of a certain size to cannibalize the young (cf., Botsford and Wickham, 1978)o Equal in importance to a realistic biological model is a realistic description of the effect of fishing on a population and the economic benefits derived therefrom. The relationship between fractional mortality rate due to fishing and the cost of fishing is classically assumed to be a direct proportion, independent of population size. Fractional mortality rate is assumed to be proportional to fishing effort which is in turn assumed proportional to cost. In practice each of these assumptions may not hold due to such effects as schooling behavior or inadequate definition of effort (cf., Rothschild, 1977; MacCall, 1976; Clark and Mangel, 1979). These assumptions also ignore the capital investment necessary to maintain fishing capability even when not actively fishing (see Clark et al., 1979, for a discussion and analysis of this problem). Although these potential drawbacks are acknowledged, they are not dealt with here. Meaningful, practical results can be obtained using the linear assumption and, if this assumption then appears to be of crucial importance, cases for which it does not hold can be then treated. In addition to the preceding considerations which arise in most fishery 9optimization problems, the models used here require description of the dependence of fishing effort on individual size. Size selectivity in fishing depends on mesh size in nets, escape ports in traps and onboard sorting in some other cases. Mesh size in nets and escape ports in traps select for animals above a certain size whereas

268

L.W. Botsford

onboard sorting is more flexible. The cost of fishing each size is usually equal and the cost is usually not increased by increasing the range of sizes fished. Therefore, while size specific models allow the flexibility of a different fishing effort at each size, the problems encountered in modeling tl~e associated cost are complex. In the work described here, a compromise between effort being completely size-dependent and completely size-independent is used. Briefly, constant effort is applied over optimal size ranges, with the cost of effort independent of size range. Fishing effort, E(t), is defined to be the fractional mortality per unit of time at time t. This effort is applied only over a specified size range. The lower and upper bounds on the Size range at time t are denoted by m~(t) and mu(t), respectively. (In the following control analysis the notation reflects only this single size range being fished; the possibility of two or more unconnected ranges is discussed later). The exvessel value of a fish of size m at time t is denoted p(m, t). The rate at which revenues are generated due to fishing is then
~mu(t)

A(t) - E ( t ) [

ml(t)

n(m, t)p(m, t)dm,

(I)

where n(m, t) dm is the number of individuals between sizes m and m + dm at time t. Note that fractional mortality (fishing effort) is constant over the size range (rn~(t), m~(t)). In most practical situations the cost of a unit of effort is independent of effort. The cost of fishing is therefore assumed to be proportional to effort. The proportionality constant K represents the product of the cost per unit nominal effort and the amount of real effort per unit nominal effort (e.g., dollars per boatday times fractional mortality rate per boat-day). The rate at which revenues are spent through fishing is then

is(t) -----KE(t).

(2)

The complete model can now be expressed in terms of the optimal harvest problem: to choose the functions E(t), m~(t) and mu(t) so that the present value of the resource,

J'=-f2e-tt[J;(t)-js(t)]dt

f:e-n'E(t)L

Jm,,,) p(m, t)n(m, t) dm

_+,

(3)

is maximized subject to the constraint

0 <~E(t) < e _ x ,
and constraints that define population behavior. In this expression e = the base of the natural logarithm '6 --- the discount rate and Em~ = a constant, the upper limit of fishing effort.

(4)

Optimal Size-Specific Fishery Policy

269

The constraint that defines the population is a size-specific, continuity equation population model,
On(m, Q . . . Ot . 0 Om [g(m,t)n(m,t)] - [d(m) + E(t)]n(m,t); mt < m < m.,

(5a) (5b)

O - - [9(m, t)n(m, t)] - d(m)n(m, t); Om

m ~< mt,

m >t m~,

which requires boundary conditions


g(mo, t)n(mo, t) ----Ri[n(', t), t], n(m, O) = no(m),

(6) (7)

where 0(m, t) = the growth rate of an individual of size m at time t


d(m) = the fractional mortality rate of an individual of size m

mo = the size at which young enter the model space (i.e., are recruited to the population) and
R~[n(., t), t] = the recruitment rate at size density function n(-, t) and time t (for the ith version of the model).

The Fast term in the population model describes the effect of growth rate on size density function n and the second term describes the effect of mortality. Fractional mortality rate includes fishing mortality only for sizes between mr(t) and m,(t). The boundary conditions (6) and (7) describe behavior at the size boundary (m = me) for all time, and the time boundary (t = 0) for all size. Although not explicitly reflected in the notation used, both growth rate g and recruitment rate R( can depend on either food level F(t) (food-dependent) or the other members in the population (population-dependent). In the latter case these rates depend on effective population size, which is defined by
C(t) --

f.
o

c(m)n(m, t) dm

(8)

where c(m) is a weighting function reflecting the relative effect of an animal of size m on growth or recruitment rate. Several versions of R~ are considered in the following analysis. For cases in which recruitment is assumed independent of adult stock, recruitment is assumed to be specified
R t = R(t).

(9a)

For linear reproduction


R2 =

b(m)n(m, t) din.

(9b)

270 For food-dependent recruitment


R3 = R2f[F(t)].

L.W. Botsford

(9c)

For population-dependent recruitment


R4 = R2fc[C(t)].

(9d)

Finally the constraint that defines behavior of food level F(t) is


dF/dt -- ~ + / d t 0

w(m, Nt))n(m, t) dm

(10)

which requires a boundary condition r(O) = Fo where w(m, F(t)) -. the consumption rate of an individual of size m at food level F(t),
dF+/dt = the rate at which food is made available to the population

(11)

(assumed constant). This expression reflects the rate of increase of food available to the population, which is assumed to be the sum of a positive component due to production and a negative component reflecting total food consumption by the population~ Thus, the control problem is basically to maximize the present value of the resource subject to one or two constraints: one being a size-specific continuity equation model with various reproductive boundary conditions, and the other being the food equation when applicable. In general, growth and reproductive rates can depend on food level or other members of the population. To avoid potential confusion due to additional notation, in the following derivations both growth and recruitment rates in the model are assumed to depend on food level F(t) [i.e., i -- 3 in Eq. (6) (see Eq. 9c)] and g is denoted by g(m, F). However, final results are given for several different cases of interest, and the required changes in the derivation are obvious.

3. Necessary Conditions for Optimal Harvest


The approach to deriving necessary conditions is based on a standard approach to applied optimal control of lumped parameter systems (cf., Bryson and Ho, 1969, or Citron, 1969)o The control problem can be discretized in the size variable and treated as an actual lumped parameter system with analogous results (cf., Botsford, 1976). The lumped parameter approach is basically to adjoin constraints to the cost functional, choose values for multipliers (adjoint variables) that cause variations due to variations in state variables to be zero, and choose the control variables so that the variation due to variations in the control variable are zero. The derivation of necessary conditions for optimal control of this system is described in Appendix A. The resulting adjoint equations are:

Optimal Size-Specific Fishery Policy


0-----~-- - [a(m) + E(t)]2,(m, t) - arm, ~ )

271

O2,(m, t)

~ 02,(m, t) "~m

e-rtE(t)p(m) mr(t) <~m <~m.(t), (12a)

- 2.(mo, t) b(m)f[F(t)] + 2F(t)w(m, F); O2,(m, t) t) 0-----~ = d(m)2,(rn, t) - g(m, F) a2.~m,

2,(mo, t)b(m)f[F(t)]

+ 2v(t)w(m,F); d2F = _ dt n(m, t) O0 o

m > m,(t),

m < mr(t),

(12b)

F) 02.(m, t) . . . Ow(m, F) Om - zr(t) .~

- 2.(too, t) b(m)f'[F(t)]J dra. Necessary conditions for control variables rnt and mu are

(13)

e-rtp(m~, t) = 2.(rn~, t), e-~tp(mu, t) = 2.(rnu, t).

(14a) (14b)

A similar necessary condition would result for each assumed fishing size limit. These could result in several unconnected size ranges that are fished. The coefficient of the variation in control variable E(t) is the switching function

S(t). S(t)~e-6' f'"')n(m,t)p(m,t)dm


dml(t)
-

K e - " - f=u,o 2.(m, t)n(m, t)dm.


dml(t)

(15)

(In cases involving several fishing size ranges, the integrals in S(t) would be over all sizes being fished.) The control rule for E(t) is

E(t) = Emx E(t) = 0


The boundary conditions are

for for

S(t) > 0, S(t) < 0,

(16a) (16b) (16c)

E(t) = singular control for S(t) = 0.


2.(m, T) = 0 for all m along the boundary t -- T and 2r(T) = 0.

(17a)

(17b)

From this point on it is assumed that market value does not vary with time (i.e.,

p(m, t) = p(m)).
Control is singular when S(t) = 0 for a non-zero period of time. In order for this to occur, the conditions dS/dt = 0 and d2S/dt z = 0 must hold. Although the former condition does not explicitly involve the control variable, E(t), the latter condition will for first-order singular arcs (Bell and Jacobson, 1975). For most cases of the problem solved here, only first-order singular arcs are discussed.

272

L. W. Botsford

The derivation of the expression

6[fsii:;P(m)n(m,t)dm-K ]
=

,Iml(O

-- e6'2v(t)w(m,F) + e~'2.(mo,t)b(m)f[F]} din,


from dS/dt = 0 is described in Appendix B. Derivation of E(t)=--d-~IG(m.,t)n(m.,t)([-~--g(mu, F))

(18)

dm~

-G(m,,t)n(m,,t)(-~-o(m,:_b")) 1
--~1 f ' ' ~ n(m, t) I g(m, F)OG(m't)OG(m't) 96Ka.,,~o ~ + 0------~-where G(m, t) is defined to be

d(m)G(m,t) 1 din,

(19)

G(m, t) =-o(m, F) d~ _ p(m) d(m) - ea'2F(t )w(m, F) + e6'2.(mo,t)b(m)f[F] - 6p(m),


(20) from dZS/dt ~ -- 0 is also described in Appendix B. SinceE(t) would not appear in this expression for the case 6 = 0, the problem is not of first order singularity in that case. This special case is not pursued further here. The necessary conditions for optimality derived thus far are for food-dependent recruitment and growth rate. The derivation is only slightly different if either of these is modeled as population dependent. If growth rate is population-dependent rather than food-dependent the term

c(m)

~,.
o

n(m',t) ~

.01.(m', t) 00( m' , t)

O ~ din'

will appear in addition to 2r(t)w(m, F) wherever the latter appears in the adjoint equations (12a, b) and the conditions for a singular arc (18, 19). If recruitment is population-dependentrather than food-dependent,the appropriate expressions for Ri (i.e., R4 in Eq. (9d)) will lead to appearance of the term

2.(mo, t)[b(m)f,[C(t)]+ c(m)f'c[C(t)]f.~b(m')n(m',t)dm" ]


instead of 2.(too, t )b(m)f[ F(t )] wherever the latter appears in the adjoint equations and conditions for a singular

Optimal Size-Specific Fishery Policy

273

arc. If neither recruitment nor growth rate are modeled as food dependent, the dependence on F in the model and the adjoint equation will not be present. In cases for which the growth rate does not depend on F, the partial differential equation (12) for the adjoint variable 2,(m, t) can be solved using the method of characteristics. Also, although the case of food dependent survival of young is used in the derivation, for the alternate case in which survival of the young depends instead on the number of older animals, the reproductive term in the partial differential equation for the adjoint variable would be replaced by a term that depends on the size density n(m, t). Therefore, although the expressions derived here hold for the latter case, the fact that the actual solution for the adjoint variable depends on the dependent variable n(m, t) must be kept in mind. The derivation of the solution is described in Appendix C. 2~(m, t) (for a growth rate that depends on size only) can be expressed in terms of age s rather than size m as

X(s, to) =

~(0,.-~ + to)[~(z)f[F(z + to)] o'(s---'ff


-- 2r(~ +/o)~(z,F(~ + to)) a--~-s)d*, s > su(t) (21a)

X(s, to) =

f?

[p(~)E(~ + to)e -~'176 + ~(0,~ + to)f[F(~ + to)]~(~)

- ,~F(~ + tol~(z, F(T +

to)l] ~d~ + J~(s~,to)~(su) ' .~(s)


(21b)
,t~

st(t) <~s <~s,(t)


X(s, to) = [Jr(o, ~ + to)f[F(~ + to)]~;(~) - ,t~(~ + to)~(~, F(~ + / o ) ) ]
+ X(s,, to) a(a(~t ; ~

s < st(t);

(21 c)

where
$(s, to) - ~.,(m(s), s + to)

re(s) = the size of an individual at age s,


$ =

age,

to = time at which size equals mo,

p('c) -~ p(m(z)),
9 (~, ~ - w(m(r), ~3, R~) - b(m(~)),

s~(t) and s~(t) are the ages that correspond to size mz(t) and mu(t), respectively, and
cr(s) is defined by
a(So)

9 'If

=- exp -

~o

[a(x) + E(x + to)] dx

19
'

s~(t) <~ to + s <~su(t)

274

L. W, Botsford

--- exp SO

[~(x)] dx ;

s < st(t),

s > s~(t)

(22)

where if(s) ~ d(m(s)). The ordinary differential equation for the adjoint variable 2F caia be solved in a way similar to the method of solution for equation (21). The solution (which is also in Appendix C) is

; 2,(O= f ,(m,s)[O. __f(s)) s)] oCmo, O2.(m,


-(2(mo, s ) b ( m ) f ' [ F ( s ) ] ) d m e x p [ - ~ f , , o n ( m , x )
<.
Ow(m, F(x))

-~--f

amaxjas.
(2.3:)

__]_

Parallel to parameter optimization problems, setting the first variation of tl~e objective functional to zero can yield either a minimum or a maximum value. Examination of the second Variation is therefore necessary in order to differentiate between a maximum and a minimum. A special condition must be used for singular control. For a first order singularity, if E(t) were the only control variable:the necessary condition for a maximum would be (Bryson and Ho, 1969)

1 0 :d':
o~ \al ,/
From Appendix B

0 dZS dE dt z
Hence

dS ~ + 6Ke -6'. dt

(25)

0 dZS OE dt 2 <~0

(26)

when dS/dt = 0. The condition for a maximum is satisfied for the optimal values of control variables m, and m, If those were specified and E(t) were the only control variable, the condition for a maximum would be satisfied. Based on this fact and the observations in the next section indicating that the choice of mz(t) and m,(t) produces a maximum rather than a minimum, the necessary conditions obtained herein yield a maximum, as desired.

Interpretation
An intuitive interpretation of the results thus far in terms of modern capital theory aids greatly in extracting the practical consequences of the necessary conditions on management policy. Economists have referred to the entity represented by the adjoint variable in several ways: as the marginal value of the capital stock at time t, as the marginal user cost for the loss in value when a capital asset is reduced by one marginal unit, and as the shadow price of capital (Clark, 1976). These in-

Optimal Size-Specific Fishery Policy

275

terpretations refer to the fact that the adjoint variable represents the change in total future revenues for a unit change in its corresponding state v~iriable. Equations (21 a - c ) can be interpreted similarly for the adjoint variable 2,. This adjoint variable is the total future value of an individual of age s, including the value due to fishing revenues, the value due to reproduction and the value due to food not eaten. The adjoint variable 2, thus represents the change in future value of the resource incurred by removing an individual of age s. Note that 2F, the marginal value of a unit of food level, appears in the term reflecting the future value of food not consumed. Similarly ~'(0, t), the marginal value of an individual of age zero, appears in the term reflecting the future reproductive value. The adjoint variable 2~, can be interpreted similarly. The two terms in the inner integral [Eq. (23)] are each a product of the amount that a change in F affects the density functions (growth and survival of young) and the amount that these affect future value (through 2,). These terms, being functions of size, are integrated over all size. The integral of the result over all future time is multiplied by a decreasing exponential. This factor reflects the fact that there is a fractional change in 2r with time that is proportional to the sensitivity of consumption rate to F. The differential equations describing the rate of change of the adjoint variables with time can be interpreted as minus the rate of depreciation of capital (Dorfman, 1969; Clark, 1976). Along the optimal path this must equal the sum of the contribution to the rate at which profits are realized and the rate at which the value of the stock is enhanced (Dorfman, I969). In the first differential equation for 2,, (12a), the third term on the fight-hand side is the contribution to profits while the remainder of the terms and all terms in (12b) are contributions to value (through future profits). In the differential equation for 2~, (13), there are no contributions to profits since food is not harvested or provided. The rate of change of 2~ therefore contributes to value only. Interpretation of the condition for singular control (dS/dt = 0) can be best described through definition of net biovalue V
V =(m)n(m, t) d m - K.

(27)

This definition is similar to the definition of net biovalue for the single age-class, Bcvcrton-Holt model in Clark et al. (1973). Net biovalue is the net revenues gained per unit fishing effort per unit time. The left-hand side of the equation derived from dS/dt = 0 [Eq. (18)] is5V. The right-hand side of thisequation is dV/dt due to nonfishing causes (i.e.,removal by fishing is not included as a decrease in V). The first term includes the increase in value due to growth, the second the decrease in value due to mortality, the third the decrease in value due to food consumed, and the last the increase in value due to reproduction. Singular control thus maintains the natural rate of change of net biovalue with time equal to the rate of growth of the monetary equivalent of net biovalue at interestrate 6 (i,e., dV/dt = 6 V). This makes sense intuitively.If dV/dt were greater than 6 V, biovalue would be increasing at a fasterrate than an equivalent investment, hence itwould be best to postpone fishing and let it grow. If dV/dt were less than 5 V, the resources should be fished at the m a x i m u m rate and thus be converted to capital which can increase in value faster.

276

L.W. Botsford

Assuming Emax is large enough, the strategy described as fishing at E,~az when dV/dt > 6 V is equivalent to staying on the line dV/dt = 6 V. The optimal values of control variables m~ and m~ are such that the current market value of an individual at either boundary size equals the total future value to be derived from that individual. Note that the expression for dV/dt [the right-hand side of Eq. (18)] does not include any terms describing the flow of value due to growth of individuals across the fishing boundaries. This is due to the fact that the conditions for optimal values of the boundaries prescribe that the value of an individual at the size boundary is zero (i.e., the sum of its current and future value is zero). A second consequence of the size limit conditions is that the current market value of all individuals within the range of sizes to be fished is greater than their future value~ The decision concerning how much to fish each size of individuals can thus be viewed as a two-step process~ The current market value of the fish at present must be greater than its future value if left unfished, in order for it to be within the fishable range. Once it is within the fishable range, the decision concerning how much fishing pressure should be applied is then based on the total number within the fishable range. The latter decision is based on considerations discussed earlier regarding E(t). These decisions are, of course, not made independently since the control law for the size limits depends on the future value of fishing effort, the control law for fishing effort depends on the current value of size limits, and the size limit controls also affect the conditions for singular control.

4. Solutions to Several Cases

The: necessary conditions for optimal harvest can be solved for several different cases of both theoretical and practical interest. Analytical solutions can be obtained for the first two cases, whereas numerical techniques are required for the remaining cases. The first case, the dynamic pool model (Beverton and Holt, 1957), is solved to demonstrate that the general solution obtained here yields the same results as those obtained previously by Clark et al. (1973) and others for a simple model. The second case is solved to reaffirm the lack of realism and practical value obtained from solutions to a linear age*specific population model. This solution amplifies the comments of Mendelssohn (1976) and Reed (1979) by presenting this problem in a new light. The remaining cases are new solutions applicable to fishery and aquaculture problems.

Single A#e Class


The ftrst case involves optimal harvest of a single age class of identical size with no density-dependent effects (similar to the model of Beverton and Holt, 1957). Since there is only a single age class we can set to = 0 in both the adjoint and state equations. Since specifying both size and time is redundant in this case, the number of animals at time t is defined to be N(t) (with N(0) = No). Similarly both growth and mortality rates can be expressed as functions of time only. Based on a solution to the size specific continuity equation obtained in Botsford (1978), N(t) can be expressed as

Optimal Size-SpecificFishery Policy

277

N(t)=~Noexpl-f'o(~(s)+ E(s))ds]. m(t,)=f ,le-'('-'"g(z)m(z)expl-f ,i(ff(s)+E(s))ds]dz,


where t~ is the fishing age limit and p(t) = km(t). From (15), S(t) can be expressed as
S(t) = e-~*km(t)N(t) - Ke -~' - N(t)A(t).

(28)

The expression for the adjoint variable contains only the fishing term; from the ~djoint equation ( 2 1 a - c ) and fishing size limit conditions (14a, b), assuming that there is only one solution to the latter, this condition can be expressed as (29)

(30) (31)

From (18), dS/dt = 0 implies


; 6 k m ( t ) N ( t ) ~ 6K = N(t)ko(t) - N ( t ) k m ( t ) d ( t )

which can be solved for N(t).


N*(t) =

6K k[g(t) - re(t)d(t) - 6m(t)]

(32)

This is a slightly more general version of Clark's expression for the same problem [Eq. (8.37) in Clark, 1976]. The optimal control solution can be determined from these expressions. At t = O, S(t) is assumed to be negative in all cases of interest, and the control E(t) is zero. The minimum size limit is reached at t = t~; however, fishing does not necessarily begin. The fishing pressure E(t) is still zero because at t = fi, fishing size limit control indicates that the first and third terms of S(t) are equal, hence S(t) is still negative (except in the case K = 0). When N(t) reaches N*(t) singular control begins. Fishing pressure is adjusted so that N(t) = N*(t), assuming the required value is not larger than Em~,. Since 2(0 in the third term of S(t) is zero at the time fishing is discontinued, it is discontinued when the first two terms are equal. Thus, when N(t) has decreased to K/km(t), fishing is discontinued. Several different control histories can arise for various parameter values. These are discussed in Clark (1976), Clark et al. (1973), and Goli (1973), and therefore are not repeated here (see Clark, 1976, for graphical representations of the various cases).
Linear Model

For the second case, the state equation for the population is the continuity equation with the linear reproductive boundary condition and without density-dependent growth rate. Since growth rateis not density dependent, the size-specific continuity equation is equivalent to the age-specific continuity equation. The latter, when combined with a linear reproductive boundary condition, is equivalent to a renewal equation which is the continuous-time, continuous-age analogue of the Leslie matrix (Leslie, 1945). For ~b defined as
q~b=f~b(a)a(a)da,

(33)

278

L.W. Botsford

where b(a) = reproductive rate at age a and or(a) = number surviving from age zero toage a, if ~b > 1 solutions are unstable about zero (i.e., increasing exponentially), if ~b < 1 solutions are stable about zero (i.e., decreasing exponentially), and if ~b = 1 the system is at a neutral equilibrium level. Since the cases for which the unfished population is either at aneutral equilibrium or stable about zero could not support a sustained fishery, they are not of interest here. If any fishing were warranted in these cases, it would begin at t = 0, be discontinued after it was no longer profitable to fish, and would not be warranted for any greater time. The case in which the unfished state variable is unstable about zero is of more potential practical interest. Both the adjoint and singular control (dS/dt = 0) equations contain a reproductive term, but no food consumption term. The control will depend on behavior of the adjoint variable, which contains a term involving X(0, t). From equation (2 l a - c) if there is no fishing ~ ~,(0, t) = ~(O, t + ~)~(z)cr(~)d~. -(34)

If ~'(0, t) is viewed as a function of the second variable, this is a renewal equation with time reversed. Stability conditions are the same as for the linear state equation except they are reversed (i.e., ~ > 1 implies a decreasing exponential solution, 9 ~ < 1 implies an increasing exponential solution and ~b -- 1 implies equilibrium). This knowledge of the behavior of the adjoint variable can be used to qualitatively characterize the optimal control history (Fig. 1). The expression for S(t) [Eq. (15)] contains 2.(m, t) in the third term, The adjoint variable in turn contains a term involving 2,(too, T) [or X(0, 7')]. From the stability characteristics
discussed earlier,since w e are concerned only with the unstable population, if~(0, t) ispositivenear t = T, itwillincrease exponentially in value as t ~ 0. (This stateme~ is based on behavior over a relativelylong time period, i.e.,based on the dominant eigenvalue.) The negative third term of S(t) will be large near t = 0 (assuming no fishingexcept near t -- T), and willdecrease as t ~ T. The implied control history is E(t) = 0 at t = 0, because S(t) < 0. This willcontinue until somewhere near t -- T (say to) at which point S will equal zero and singular or m a x i m u m control will be

,,n <r

>=

n.*

0 S(t)<O no fishinq TIME

to S(t)=O fishing

Fig. 1. Typical control history for linear reproduction. The population is fished only between to and T. Since
~(0, t) is increasing exponentially as time decreases from to, S(t) < 0[Eq. (15)'1 hence fishing for t < to is not optimal

Optimal Size-SpecificFishery Policy

279

optimal until t = T. For the sake of simplicity, this discussion has thus far not involved the discount rate fi: If6 is large, the "initial condition" on ,~(i.e., its value at t = to) is small, hence S equals zero sooner. Thus 6 acts in the same way but in opposition to the dominant root of the renewal equation for ~(0, t). The behavior is therefore determined by the difference between them. Again, these qualitative conclusions are based on the dominant behavior of the adjoint variable, and some exceptions could possibly be found. However, the general result has intuitive appeal. If one has a stock that can increase exponentially at a rate greater than the discount rate, the best harvest strategy is to let it increase until just before the end of the planning period, then harvest. This behavior is similar to that observed for other age-specific, linear models (cf., Mendelssohn, 1976, Rorres, 1976, and Reed, 1979).

More Realistic Cases

Analytical solutions to the remaining cases do not appear possible. Since numerical solutions involving ordinary differential equations for size-discrete versions of the adjoint variable (2) and state variable (n) would lead to a two-point boundary value problem of high dimension, they are not attempted here. Instead a simpler form of solution is sought for each case. In some similar control problems the solution approaches a singular arc, remains on the singular arc for a non-zero period of time, then leaves the singular arc before the end of the time horizon (T) (Bryson and Ho, 1969). While on the singular arc, necessary conditions for optimality are that: (1) the switching function be zero, (2) the time derivative of the switching function be zero, (3) the adjoint equations be satisfied, and (4) the state equations be satisfied. For the remaining cases, the possibility of a singular solution on (tl, t2) (0 ~< tl < t2 ~< T) will be examined. To further simplify the problem, the solution sought will be one in which the state variables are at equilibrium and controls are constant with time. The optimal harvest solution for the logistic model (Clark, 1976) involves a singular arc preceded and followed by non-singular periods. The population level is constant with time throughout the singular arc. A constant fishery policy (i.e., constant size limits and fishing pressure) would be easier to implement than a time-varying policy. Because of the existence of a constant-policy solution in the logistic case and its desirability from a practical standpoint, singular solutions in which control variables are constant and the population is at equilibrium are sought here. The approach taken for each solution is to express the necessary conditions in age-specific form, then present the attempted singular-equilibrium solutions in terms of a commonly used numerical procedure. This procedure consists basically of determining values of controls that satisfy the adjoint equation, the state equations and the condition that the time derivative of the switching function be zero, then seeking a boundary condition (at t2) on the adjoint variables so that the switching function is zero on the singular arc. (This procedure can be followed, for example, in determining the optimal control in the logistic harvest problem.) This procedure is followed as a mnemonic device to demonstrate heuristically that

280

L . W . Botsford

singular equilibrium controls cannot satisfy necessary conditions for optimality for the remaining cases. For the simplest case, constant specified recruitment, the size limit condition (14) can be expressed as

e-~'km(s3 = X(sl, t)
where from (21 a - c)

(35)

~(sl, t ) = - -6+Ek f+2 -'o m(.c)a(.c)e-6(,-,,)d~ ea(s,) , ,

e-aZk

fr-+o

E(to + ~)m(z)a(~)e- +('- !+2-,o))dz

(36)

where time t = to + sl, the time of birth plus current age andp(s) = kin(s). The first term applies to the singular arc and the second to the interval (t2, T). The condition dS/dt = 0 I-Eq. (18)] can be expressed as

R~ := [;i ~km(z)a(r) d~ - K If+~ R~a(~)k[~(z) - m(~) ~(~)] dT


I I

(37)

where we have used the fact that at equilibrium n(a, t) = R+a(a), where R, is the specified constant recruitment value. From Eq. (15) the condition S(t) = 0 can be written

e - 6'R~

f okm(-c)a(z) dz ~- Ke-6' = R+ ;+oa(z)~(z, t) dz

(38)

in which ,~(z, t) can be expressed by Eq. (36) with st replaced by ~. Assuming that the values of E and st could be chosen to satisfy the size limit condition (35) and the condition dS/dt = 0 (37), the values of ~ along the end boundary (t = t2) that yield S(t) = 0 on the singular arc would then be sought. For the case considered here the last condition cannot be satisfied for the following reason. The adjoint variable (future value) contains two terms: (1) the total value obtained by fishing of theage class for age.4 greater than s+ at the singular fishing pressure up to time t2 and (2) the total value obtained from fishing that same age

I
O mOX

--- --//'S - - - I - - characteristic AGE

~-tl "' r/~'t2-Omox


TIME

t2 T

Fig. 2. Age versus time over the planning period from time zero to time 7". For singular equilibrium control to be possible o n (tt, t2) the values of all individuals (adjoint variable) must depend on how their age class is fished for t > tv Since individuals born at any t < t2 - ~ will not be alive at t2 (e.g., an individual born at t3 singular equilibrium control is not possible on (tt, t2)

Optimal Size-SpecificFishery Policy

281

class from time tz to Tat fishing pressure Er~x or zero. Assuming that individuals do not live beyond a maximum age, ar.a, for all time t < tz " o-max, no individuals currently alive will survive to be fished at t > t2 (Fig. 2). Hence the value of X(a, t) is independent of J[(a, tz), and consequently depends on!y on values of E and sl. Thus, even though values o f E a n d s~may be found that satisfy dS/dt = 0 and the size limit condition, the condition S ( t ) = 0 cannot be satisfied (except fortuitously) for constant values of E and s~. Because of this it appears that singular-equilibrium control with constant values of control variables is not possible. Furthermore, from this behavior of the adjoint variable, one would suspect that some sort of pulse fishing in which the period of the pulse was less than twice the maximum age of an individual would be optimal. This supports the conjecture of Clark (1976) concerning a similar problem (recruitment at discrete times). For the second case, population-dependent recruitment, the expression for the adjoint variable is the same as (35) except for an additional term [from the discussion following Eq. (20)].
X(sl, t) = e - ~'E k ~t2 -,o m( z )a( z )e - ~" -~').dz
a(s,) 3,,

fT-to --" E(to + "c)m('r)a(z)e-e~'-l'2-'~ dz + a ( ~ - to) ~,,-,o

e-ttk

r-,o

a(O

*/$!

~(0, t + to)-7-~-~[b(t)fi[C.] - c(t)f~[C.]R,~b] dt u~..Jo

(39)

where f',[C,] is the derivative of f~ with respect to its argument evaluated at equilibrium, ~b is defined by Eq. (33) and R, is the equilibrium recruitment level. The term containing ~(:) reflects the positive influence of fecundity on recruitment while the term containing ?(:) reflects the negative influence of density-dependent effects on recruitment. The condition dS/dt = 0 will also have an additional term
5 R,

[;;

km('c)a(z) dz - K
I

3;
=

R,a('c)k[(z)
1

m(z)3(z)] dz

+
I

,t(0,~)e%(~)E~(~)f~[C,]
(40)

- ~(z)f;[C,]R, qr~b]dz.

This additional term appears similar but differs from the additional term in Eq. (39) in that the integral in (40) is over age at current time while the integral in (39) is over age at future time (i.e., along a characteristic). The approach to a singular-equilibrium solution in this case is to choose values of E, sl, ~[(0,t) and X(~, t2) (Vz), that satisfy dS/dt = O, S ( t ) = 0, the size limit condition and the adjoint equation. We assume that for any value of ~'(0, t) we can choose values of E and s~ that satisfy dS/dt = 0 [Eq. (40)] and the size limit condition [Eq. (35) with ~(s~, z) expressed by Eq. (39)]. From the condition S(t) = 0

282

L.W. Botsford

[Eq. (38) with X(r, t) expressed by Eq. (39) with "c substituted for st], the time dependence of ~'(0, t) must be of the form Ce-~' with C a constant. However, from the expression for X(0, t) derived from the adjoint conditions ~.(0, t) = e
-6, E k

a(sl) J,,

~u-,o

m(r)a(z)e -~*-s'~ dz

f,_,k [,-,o
+ a(t2 - to) .lu-,o E(to + T)m(r)a(r)e -~*-~'2-'~ dr +

; 7 0I(0,r + to)a(r)[G(z)f~[C.] .... I a(sl) d~,

C--(r)f;EC.]R.e~] dr.

(41)

A.solution of the form J[(0, t) -- Ce -6t, for t < tz - a ~ would be

X(O, t) = e -6' I
-

f?o

m(r)a(r)e -~*-~~ dx (42)


-

e-%(r)E~;(r)f~EC,]

e(r)f'EC,]R,~b]

Since a(r) is zero for r > a~x this expression determines the value of the constant C. Solution of the condition S(t) = 0 for the constant C required to set the switching function to zero on the singular arc shows that it is-not the same as the constant C required by the adjoint equation. Therefore it appears that these could not both be satisfied on (tl, tz). Thus for this problem also a singular-equilibrium is not possible, basically because values of the adjoint variable on the singular arc are not related to values at the end of the arc due to the limited lifetime of individuals in the population. Because of this, it again appears that a form of pulse fishing, with periods less than twice the maximum lifetime of fish will be the optimal policy. For the third case, food dependent recruitment, the adjoint condition for 2F and the state equation for food variable Fare additional necessary conditions. The size limit condition is expressed by Eq. (35) and the condition S(t) = 0 is expressed by Eq. (38) each with the appropriate expression for X taken directly from Eqs. ( 2 1 a - c ) (i.e., this is the case used in the derivation of necessary conditions). Because they are quite similar to those for the previous case, necessary conditions for this case are not repeated here. Although these expressions involve 2F(t), their similarity to the previous case is apparent after factoring X(0, t) = Ce -6' from 2~,(t) to obtain 2k ~ 2F(t)/Ce -~t. The same argument as was followed in the preceding case can now be followed in this case with one exception. Rather than the end result being two different expressions for the constant C, we could attempt to adjust the value of 2~, (which appears in each) so that they would be equal. However, this solution requires ;t~,to be constant with time. Examination of Eq. (23) shows that it is not constant. Hence for this case also a singular solution does not appear possible. For each of these cases only a single size limit (a lower limit) has been considered. There may be instances in which two or more size limit conditions can

Optimal Size-SpecificFishery Policy

283

be satisfied. However, this possibility does not seem to remove the basic impediment to obtaining a singular-equilibrium solution. A change in the limits of integration in S(t) = 0 and an additional term in ~[(0, t) do not allow a common solution to both. Although-they have not been investigated in detail, it appears that the same difficulties would be encountered in attempting a singular equilibrium solution for the remaining cases (i.e., those involving density-dependent growth rate). At equilibrium, growth rate would be constant with time and the necessary conditions would be similar to those for the cases described here.

5. Discussion

For the assumed forms of the fishing cost function and harvesting controls, necessary conditions for ~0ptimal harvest have been developed for a general population niodel that reflects a wide range of population dynamics and behavior. These conditions indicate that when multiple age or size-classes and densitydependent growth or recruitment rate are explicitly included in population models optimal policy is quite different than for simple, less realistic models. From a practical point of view the most intriguing aspects of the results obtained here are the implications of the lack of constant fishery policy solutions. This result differs substantially from standard solutions to the optimal fishery policy problem obtained with the logistic model. The latter solutions include a singular arc of constant policy at an equilibrium level (cf., Clark, 1976; Cliff and Vincent, 1973; Goh, 1969/1970). The logistic model is widely regarded as an adequate population model from which general, robust conclusions can be drawn. It reflects the self-inhibiting behavior that limits population level, but does not explicitly include the specific density-dependent mechanisms responsible for this behavior of their age-dependence. The results obtained here demonstrate that when these mechanisms are explicitly included in a population model, thereby making it more realistic, a constant equilibrium policy is no longer optimal. The difference in results stems from the fact that in the simple, single state variable case the condition dS/dt = 0 and the solution to the adjoint equation are identical at equilibrium, whereas in the cases discussed herein this is not true. Because of this, as discussed in the previous section, S(t) = 0 and dS/dt = 0 cannot both be satisfied on the singular arc. Although these reasons for the different results are not particularly satisfying intuitively, it is not difficult to accept the fact that one could do better by harvesting a population to a low level, then letting it grow rapidly to a high level with a minimum of density-dependent interference, than by maintaining a constant level and removing biovalue at a constant rate (especially when one considers that costs are incurred only when actively fishing). The practical implications of the results obtained are quite sensitive to some of the assumptions on which the analysis was based. For example, the cost functional accounts only for variable costs. Inclusion of fixed costs by adding the cost of maintaining capital equipment through periods of no fishing could change results. On the other hand, the fact that boats that fish several species could switch from one to the other at different times would decrease this effect. Other factors that could

284

L.W. Botsford

increase this premium on steady operation of a fishery are effects due to processing capability and market demand (of., Anderson, 1977). For multiple age class models, a version of the specified, constant recruitment model with recruitment at yearly intervals has been analyzed by Clark et al. (1973) and Clark (1976). They have shown that for fishing costs equal to zero, both age selective and non-selective policies lead to a policy of pulse fishing. For the selective case the same result can be obtained from Eqs. (14) and (18) (Botsford, 1978) for the continuous recruitment case. The result states that catching all individuals when they reach the age at which
6mt = g(mO - m~ d(mi)

(43)

is satisfied is optimal. Clark (1976) did not analyze the case of non-zcro cost of fishing but made the conjecture that optimal policy would also be a pulse fishing policy for this case (and non-selective gear). Hannesson (1975), through simulation, examined the case in which fishing costs are not zero, and concluded that puls~ fishing was slightly better than the stationary fishing policy tried. Waiters (1969) numerically obtained the maximum yield of models with individual growth and stock-recruitment relationships for two cases: one with and one without fishing selectivity. In the former case a constant policy was optimal while in the latter a pulse fishing policy was optimal. Pope (1973), using a similar model with no gear selectivity, also numerically determined optimal yield to be a policy involving pulse fishing. Getz (1979), also using a model that includes a stock-recruitment relationship and individual growth~ recently obtained a numerical solution for maximum yield with a description of harvesting similar to that used here; constant fishing mortality for all sizes greater than a minimum size to be determined. Fishing mortality and maximum size were assumed constant with time. Necessary conditions for the optimal level of effort can be viewed as providing the best "balance" between biological and economic rates. This is most simply explained using the logistic model because it reflects the general self-limiting, compensatory behavior of population growth yet has only a single dependent variable. Clark (1973a, b~ 1976) and Clark and Munro (1975), using the logistic model, developed a useful interpretation of the optimal harvest problem as an adjustment of the biological rate of increase of the population to a level that depends on how much consideration is given to future harvests (i.e., the value of ~). For ~ > 09 the assumed sole owner of the fishery obtains less than the potential maximum economic rent, because future profits are worth less than current profits. The amount by which optimal level is less than the maximum economic rent depends on both the interest rate and the productive rate of the fished populations. For the logistic model this dependence was simply described by Clark (1976) as the bionomic growth ratio ~, the ratio of discount rate to intrinsic growth rate of the stock (the constant r which is the limit of the specific growth rate as the population level goes to zero, and is also the maximum possible specific growth rate). As this ratio increases, the optimal level decreases and vice versa. This analysis also leads to the rationale behind extinction being the optimal policy. When the cost of fishing is independent of population size, if~ is less than unity, the optimal stock level is zero.

Optimal Size-Specific Fishery Policy

285

That is, when the productive capability of the population is less than the discount rate, the optimal policy for the sole owner is to convert the resource to capital (see also Watt, 1968, p. 125). Necessary conditions obtained here are similar, but they explicitly divide general biological growth into its actual age- or size-specific components. Optimal levels as determined from the logistic model depend biologically on the relationship between the intrinsic rate of increase of the population (r) and the maximum population level (K). This relationship for the logistic equation is a specific one that, while it possesses certain gross similarities to behavior of actual populations, is not based on specific growth, reproductive or mortality rates in the population. The improvement provided by the work presented here is that results are based on explicit consideration of how biological population growth occurs. There is, however, a practical difference between the results obtained here and those obtained using the logistic model. The "biovalue" considered in setting exploitation rate includes only those individuals between the fishing size limits, whereas in the logistic model it includes all individuals in the populations. Thus, optimal policy derived using the logistic model would differ significantly from optimal policy for the same population using the models analyzed here. The way in which optimal values of control variables achieve the optimal balance between current and future profits, and how growth, reproductive, mortality, and discount rates affect this balance can be most easily seen in the necessary condition for singular control, dS/dt = 0 [Eq. (18)], but is also affected by the size limit condition [Eq. (14)]. For any fixed values of mr and m~, E is adjusted so that t h e "population level" (actually the present market value of animals between mt and m~) equals the ratio of growth rate (RHS of Eq. (18)) to the discount rate (6). The optimal size limits assure that only individuals whose current market value (LHS of Eq. (14)) is greater than the total value derived from them in the future (RHS of Eq. (14)) are fished. The optimal selection of age or size of capture as determined by the size limit condition is also related to existing theory. Size limits are determined by current market value relative to total future value to be derived from an animal alive at that size. Future value involves consideration of direct value through fishing, reproduction, density-dependent effects and food consumed. Although they are not often explicitly included in formulation of fishery policy, most of these components of value of an individual have at least been qualitatively considered in management. For example, the strategy of fishing as heavily as possible without destroying the reproductive ability of the stock in a sense considers reproductive value of individuals. Also, consideration of the value of food not eaten by fish that are removed by harvest was emphasized by Paloheimo and Dickie (1970). A remarkable example of prior consideration of the ideas expressed here is a short, concise analysis by R. H. MacArthur that leads to a consideration of reproductive value that is similar to that expressed here (MacArthur, 1960). Starting from Fisher's definition of reproductive value of an individual of a certain age (Fisher, 1958) (identical to the reproductive component of the adjoint variable at the same age), he reached the conclusion that an age selective predator (harvester) should take animals with the largest ratio (value to predator/ reproductive value).

286

L.W. Botsford

Comparison of current market value with direct future value due to fishing in setting an age limit also resulted from the analysis of Clark et al. (1973) for a single age class. Again relative value of future catches was described through the discount rate. When the discount rate is zero, maximum effort is applied at the age of maximum biomass. This is the classical case considered by Beverton and Holt (1957) and involves equal consideration of current and future values. As the discount rate increases from zero, the age of first capture decreases. As discount rate increases, there is less and less consideration of future value. Similar variations in discount rate will produce similar changes in the lower size limit in the results obtained here. However, the results are conceptually more similar to those obtained by Goh (1973), who treated the case for which fishing costs were zero and effort was at a limiting maximum value. The greater similarity is due to the facts that in the optimal policy derived here, the size limit condition does not involve cost of fishing, and the value of fishing pressure is in a sense specified by another necessary condition (as~at = 0). A possible theoretical interpretation of the results obtained here is a combination of existing theories of optimal adjustment of self-compensating biological rates with existing theories of optimal selection of fishing age or size limits. Thus, these results combine the perspective of the logistic or production model with the perspective of the single age class models. While these comparisons aid conception of the new results, there are important differences. The results differ from those obtained from the logistic equation in that only individuals within the size limits would be considered in setting fishing pressure. They differ from the single age class results in that the size limits are effectively set at the values corresponding to the case of zero fishing cost and specified fishing level. In summary, the results obtained have both theoretical and practical implications for fishery research. On the theoretical side, interpretation of the necessary conditions for optimality extends and unifies previous results. On the practical side, the increased biological realism of the models used apparently leads to time-varying fishery policy. This result may change as the fishing cost and market price models are made more realistic. The exact forms of solutions and the influence of these economic assumptions on them are currently being pursued.
Acknowledgements. I am indebted to A. L. Suer for her review of this manuscript. This work is an
extension of part of my Ph.D. dissertation at the University of California, Davis. I am grateful to W. A. Gardner, A. J. Krenner and Ho E. Ranch for thV, advice and assistance in that endeavor. Finally, I ir would like to thank Colin Clark for his comments on this work. This work is a result of research sponsored by NOAA Office of Sea Grant, Department of Commerce, under Grant #NOAA-M01-184 R/F52. The U.S. Government is authorized to produce and distribute reprints for governmental purposes notwithstanding any copyright notation that may appear hereon.

Appendix A
Necessary Conditions

Adjoining constraints to J yields the augmented performance functional

Optimal Size-Specific Fishery Policy

287

e - ~'E(t)
0 mI

pn dm

+ f

O0 2.(m, t) I - n am m!
"~

On g Om

dn - En -

Onldm Otl

L am-~
+ 2.(m, t) n~m -- g~m -- dn - atJ

dm

+ 2,(t)m-~T-

r,,.

L
=

wndm ---~

"]}
c~t

dr.

(A-l)

Integration of

~,.
mo

= :..gn I~ -

(A-2)
(A-3)
"

2.--dt Ot r dF

~.,nl r

--

(A-4)

by parts, the relationship

n~ ()..O) n2.am Og
=

02. norm

and a result using the boundary condition (6)


2.9n I,.~ = 0 - 2(too, t)Ri

field

'=

{e-6'E[f..,P"dm+

+J-,L am

;[.o
"'

-02, ngom

dn2. +

.al J d m + .].~k Om. r-rnoO at


[2~(T)F(T) - ,td0)F(0)].

dn2. +

n ~ J

dm

+ 2F---~--,- 2r
o

wndm + F--~- + 2.(m o, t)Ri dt o

r2.(m, T)n(m, 7")


(A-5)

2.(m, O)n(m,

0)] d m

The variation in J with respect to a variation in E is


AJ =

rE
.10

AE e -6~
L

or\~l ml

pndm-

.)

288

L.W. Botsford

+Jo LJ.o~, am
+

t"r An F { ' , / g ~ -- d2. 02.

+~-//

a,~.]d m

J.A am
am

Ti-J
d2. - E2, + - ~ + e-6'Ep dm ('r 2r f' o'a~ ] a, + jo ''[- ('~ Fn -e~"-~m LJ,..L og-

+J=,\

+ 2.(mo, t)V.(R)-

+ 2.(too, tl~-ff + - ~ - j - An(m, T)

rio

2.(m, T)dm _---AF(T)2r(T)

+ Ainuf ~ [E(t)p(mu, t)e-"n(m~,t) - E(t)2.(m~,t)n(mu, t)] act + Amz ~oo[E(t)p(mz, t)n(ml, t)e -6' - E(t)2n(mt, t)n(mt, t)] dt
where for Rt given by (9a)

(A-6)

V.(RO 0,
=

(A-7a)

for R2 given by (9b)

V,,(R2)=
for R3 given by (9c)

b(m) din,

(A-7b)

V,,(R2) = f' b(ra)dmf[F( t)]


and for R, given by (9d)

(A-Tc)

V.'R,'=f~{b(m)f.[:ic(mg"(m'.t)am ']
+f c(m')n(m',t)dm c(m)
o

b(m')n(m',t)dm' din. (A-7d)

Setting the coefficients of the variation in state variable F(t) and n(m, t) equal to zero yields the adjoint equations (12a, b, 13). Setting the coefficients of the variation in control variables mzand mM equal to zero yields the control equations (14a and b). The coefficient of the variation in control variable E(t) is the switching function, S(t) [Eq. (I5)]. Boundary conditions are obtained by setting variations in boundary values to zero (17a, b).

OptimalSize-SpecificFisheryPolicy
Appendix B

289

Singular Control To determine dS/dt, differentiating (15) yields


--

dS = -- 6e_,, fm"" p(m)n(m, t)dm + e -6' [p (m.)n(m., t)--~dm. dt j,.,,) - p(mOn(mt, t)--d- + J..,~o [p(m)n(m, t)] dm.~ + 6Ke -6' idin. - 2n(m., t)n(m., t)--d- -- 2.(ml, t)n(ml, t) dm~ fIdt

O From (Sa) and (5b) we obtain


J,,,,~o

+ /

r ""~ ~ [2.(m, t)n(m, t)] dm] 0


~

(B-l)

~[p(m)n(m,t)] =p(m){-O-~[o(m,F)n(m,t)]-(d(m)+ E(t))n(m,t)} (B-2) O[2.(m,t)n(m,t)] at


~.(m, t) { O-~[9(m,F)n(m,t)]
-

(d(m) + E(t))n(m, t)} + n(m, t) tO2.(m,at ) .___

(B-3)

From (12a), (12b), (B-I), (B-2) and (B-3), we obtain


e_6t 0

[p(m)n(m, t)] - ~ In(m, t)2.(m, t)]

=e-6'p(m)[-O-~[g(m,F)n(m,t)]-d(m)n(m,t)] -$.(m,t){-O--~[g(m,F)n(m,t)] } +n(m,t)g(m,F)


-

a2,,(m, t)

O---'---m(B-4)

n(m, t)[2F(t)w(m, F) - 2.(mo, t)b(m)f[F(t)]].


02. 2.O--~(9n) + gnom = 0-~(2.9 n)

Using the product rule twice we obtain


(B-5)

and

p(m) [_ O_~[g(m, F)n(m, t )] _ d(m)n(m, t )] = ( _ O_~(pgn) + ng-ff-'mdP p dn) . (B-6)

290

L. W. Botsford

Equation (B-4) and size limit control conditions (14a) and (14b) yield

dt

~ml(O

e-~'n(m, t) g(m, F)dm - p(m)d(m)

-- e-6t2v(t)w(m, F) + e6'A,,(mo,t)b(m)f[F]l dm

-- 6

f mu(t) dm~{O

e-~tp(m)n(m, t)dm + 6Ke -6t.

(B-7)

Hence the condition, dS/dt = O, implies Eq. (18). With the definition

@ G(m, t) =- g(m, 1:)~-~ - p(m)d(m) - e-6t2e(t)w(m, F) + e6tg(mo,t)b(m)f[F~ - 6p(m)


differentiation of (B-7) yields

d2S am~ -~f = e_6tIn(m~, t)G(m~, t)--~- - n(m,, t)G(m,, t ) ~ ] "-- e-6t |

['m~(O

Jm~(t)

n(m, t)G(m, t)dm - 62Ke -6t - e -6~

fm~(O
dmz(t)

x [ { - o~ ~q(m, F)n(m, t)] - (d(m) + E(t))n(m. t)} G(m, t) +n(m,t) OG?'t). 1 den.
From (18) the second term in (B-9) is 62Ke-~; thus using (B-9)

m~{O

Om

r " n OG dm

(B-10)

(B-9) becomes

d 2S /Ore, dr-- = e -6' I G(m., On(re., t)~-~- - g(m..F) ) T .{On, - G(m,, t)n(m,, t)~-~-~-- g(m,, F) )1 + e-" ~..o, n(m. t) ,~o I -aG(m't) OG(m't) x g(m, ~) ~m ~ Ot
Hence (18) and d2S/df -- 0 field (19).

] '(d(m) + E(t))G(m, t) din.

Optimal Size-SpecificFishery Policy

291

Appendix C
Adjoint Variables
With the definition

D(m, t) =- e-6'E(t)p(m) + 2,(mo, t)b(m)f[F(t)]- 2~(t)w(m, lZ); ml(t) <~m <~m~(t), D(m, t) ~- 2.(mo, t )b(m) f [ F( t ) ] - 2F(t )w(m, F); mz(t) > m, m~(t) < m.
(C-2) Equation (12a) can be expressed as
-

(C-l)

02,(m, t) O2.(m, t) + g(m, F ) - = [d(m) + E(t)];.,(m, t) - D(m, t). Ot Om

(C-3)

This can be expressed in terms of an age variable s defined as s = t - to for the cases in which 9 does not depend on time [i.e. ms(s) is the solution to dmJds = 9(m~(s))]. Under this condition (C-3) becomes

dX(s,to) = ds
where

Ea(s) +

E(s + to)]X(s, to) - / 3 ( s , s + ~(s)=-d(m(s)),

to)

(C-4)

X(s, to) - 2.(m(s), t),

ff)(s, s + to) = D(m(s), t).


This equation can be solved for each to. Multiplication of both sides by exp[-fi[~(x) yields + E(x + to)]dx 1

d [2-(S,to)exp[- f i [~(x)+ E(X + to)]dx]l


=-~(S,
which implies

to + s)exp[- fi [~(x)+ E(X+ to)]dx1

(c-5)

X(S,to) = I- fl D(~,v+ to)expl- fi [a(x)+ e(X + to)]dxl + C] x exp[- fl E~(x)+ E(x + to)]dx] 9
From the boundary condition (17a), ~[(T - to, to) = 0; thus (C-6)

c= f~-'~

f~Ca(x)+ E(x + to)]dx]D(z,Z + to)dz

(C-7)

292

L.W. Botsford

and
~[(S, to) = B(Z, "Cq- t o ) e x p -f,

[-3(x) + E(x + to)] dr. d~.

From the definition o f ~ and D (C-l, C-2, C-4) this can be expressed for each of the three age ranges that correspond to m < ml(t), m > rn~(t) and mr(t) <~ m <<.rn.(t) (i.e. s < st(t), s > s,(t) and st(t) <~ s <<.s~(t), respectively). The resulting expressions are (21a), (21b) and (21c). The solution of the ordinary differential equation il 3) for the adjoint variable ;tr is similar. With the definitions
fl(t) -~

f.

n(m,t) gw
o

7;

F) dm ~ - 2~(mo, t)b(m)f'[F(t)] dm

(C-8) (C-9)

A ( O --- -

oin(m, 0

(13") can be expressed as


d;t__.s ~_ 2r(t)fl(t) +A(t). dt

(C-IO)

The solution to (C-IO) is

2r(t)=If'ofdS)exp[flft(x)dxldsClexp[-f'oA(X)dx]. +
From the boundary condition (17b) ~F(t) = -

(C-11)

;"
t

A(s)exp

if'
9

A(x)dx
t

(C-12)

Equation (23) then follows from the definitions (C-8) and (C-9). References
Anderson~ L~ Tho economics of fisheries management, p. 214. Baltimore: Johns Hopkins University Press 1977 Beddington, J~ R., Taylor, D. B.: Optimum age spedfic harvesting of a population. Biometrics 29, 801-809 (I973) Bell, D. J., Jacobson, D. H.: Singular optimal control problems, p. 190. London: Academic Press 1975 Beverton, R. J. H., Holt, S. J.: On the dynamics of exploited fish populations. Fish. Invert. London, Set. 2, 19, 533 (1957) Botsford, L. W.: Modeling and optimization of aquatic productive systems. Ph.D. Thesis Proposal, unpublished (1976) Botsford, L~ W.: Modeling, stability, and optimization of aquatic productive systems~ Ph.Do Thesis, University of California, Davis (1978) Botsford, L~W., Wickham, D. E.: Behavior of age-specific, density-dependent models and the northern California Dungeness crab fishery~ J. Fish. Res. Board Can~ 35, 833-843 0978) Bryson, A. E., Ho, Y.: Applied optimal control, p. 481. Waltham: Ginn and Company 1969 Citron, J. J.: Elements of optimal control, p. 266. New York: Holt, Rinehart and Winston, 1969 Clark, C~ W. : The economics of overexploitation. Soence 181, 630-634 (1973a) Clark, C. W.: Profit maximization and the extinction of animal species. J. Polit. Econ. 81, 950-961 (1973b) Clark, C, W., Munro, G. Ro: The economies of fishing and modern capital theory: A simplified approach. J. Environ. Econ. Manage. 2, 92 - 106 (1975)

Optimal Size-Specific Fishery Policy

293

Clark, C. W . : Mathematical bioeconomics: The optimal management of renewable resour~s, p. 352. New York: John Wiley and Sons, Inc. 1976 Clark, C. W., Mangel, M.: Aggregation and fishery dynamics: A theoretical study of schooling and the purse seine tuna fisheries. U.S. Fish. Bull. 77, No. 2 (1979) Clark, C. W., Clarke, F. H., Munrg, G. R.: The optimal exploitation of renewable resource stocks: Problems of irreversible investment. Econometrics 47, No. 1 (1979) Clark, C., Edwards, G., Friedlander, M.: Bcverton-Holt model of a commercial fishery: Optimal dynamics. J. Fish. Res. Board Can. 30, 1629-1640 (1973) Cliff, E. M., Vincent, T. L. : An optimal policy for a fish harvest. J. Opt. Theory and Application 12, 493 - 496 (1973) Dorfman, R.: An economic interpretation of optimal control theory. Amer. Econ. Rev. 59, 8 1 7 - 831 (1969) Fisher, R. A.: The genetical theory of natural selection. New York: Dover t958 Getz, W. M.: Optimal harvesting of structured populations. Math. Biosci. 44, 2 6 9 - 291 (1979) Gob, B. S . : Optimal control of a fish resource. Malayan Scientist 5, 65 - 7 0 (1969/1970) Goh, B. S.: Optimal control of renewable resources and pest populations. Prec. 6th Hawaii Internat. Conf. on System Sciences, January 1973 Graham, M.: Modern theory of exPloiting a fisher~ and application to North Sea trawling. J. Cons. Perm. Int. Explor. Mer. 10, 264-274 (1935) Hannesson, R.: Fishery dynamics: A North Atlantic cod fishery. Can. J. Econ. 8, 151 - 1 7 3 (1975) Leslie, P. H.: On the use of matrices in certain population mathematics. Biometrika 33, 183-212 (1945) MaeArthur, Re H.: On the relation between reproductive value and optimal predation~ Prec. Nat~ Acad. Sci. USA 46) 143-145 (1960) MacCall, A. D.: Density-dependence of catehability coefficient in the California Pacific sardine (Sardinopssa#ax caerulea),purse seine fishery. Cal-Co-op Oceanic Fish. Invest. Rep. 18, 136-148 (1976) Mendelssohn, R.: Optimization problems associated with a leslie matrix. Amer. Natur. 110, 339 - 349 (1976) Paloheimo, J~E., Dickie, L. M.: Production and food supply. In: Marine food chains (J. He Steele, ed.). Berkeley-Los Angeles: University of California Press 1970 Pope, J. G.: An investigation into the effects of variable rates of the exploitation of fishery resources. In: The mathematical theory of the dynamics of biological populations (M. S. Bartlett, R. W. Hiorns, eds.) ppo 347. New York: Academic Press 1973 Reed, W. J.: Optimum age-specific harvesting in a non-linear population model. Biometrics 36, 579-593 (1980) Ricker, W. E.: Stock and recruitment. J. Fish. Res. Board Can. 11, 559-623 (1954) Rorres, C. : Optimal sustainable yields of a renewable resource. Biometrics 32, 945-948 (1976) Rorres, C., Fair, W. : Optimal harvesting policy for an age-spe~r population. Math. Biosci. 24, 31 - 47 (1975) Rothschild, B. J.: Fishing effort. In: Fish population dynamics (J. Gulland, ed.), p. 372. New York: John Wiley and Sons 1977 Schaefer, M. B.: Some aspects of the dynamics of populations important to the management of the commercial marine fisheries. Bull. Inter-Amer. Trop. Tuua Comm. 1, 2 6 - 5 6 (1954) SiUiman, R. P.: Analog computer simulation and catch forecasting in commercially fished populations. Trans. Amer. Fish. See. 98, 560-569 (1969) Sinko, J. W., Streifer, W. : A new model for age size structure of a population. Ecology 48, 918 - 9 1 9 (1967) Waiters, C. J.: A generalized computer simulation model for fish population studies. Trans.Amer. Fish. SOeo 98, 505--512 (1969) Warren~ C. Eo : Biology and water pollution control, p. 434. Philadelphia: W. B. Sounders and Co., 1971 Warren, C. E., Davis, G. E.: Laboratory studies of the feeding, bioenergetics and growth offish~ In: The biological basis of freshwater fish production (S. B. Gerking, ed.). New York: John Wiley and Sons 1967 Watt, Ko E. F.: Ecology and resource management, p. 450. New York: MacGraw-Hitl 1968 Received February 1/Revised August 25, 1980

You might also like