400 John Woods
18.9.2 Competitive exclusion
In the first experiment (Al Battran et al. 1998), we changed the values of only
one parameter. This is called the "infant mortality" parameter. It determines the
amount of energy (e microJoules) stored in the energy pool of a newly born phytoplankter. Reproduction tends to occur in the afternoon, followed by the hours of
darkness when energy loss by respiration is not compensated by energy gain from
photosynthesis. So having a sufficient energy reserve at birth, e, ensures that the
newly born phytoplankter does not die by energy starvation during the first night
(hence infant mortality). But there is a demographic cost: more energy must be
stored between successive reproductions. So varying e trades birth rate against
death rate. In the default version of the WB model, e = O, so primary production is
not optimised. A first guess at the optimal value of e suggests it should equal the
nocturnal respiration, which is 12 microJoules. The numerical experiment was
designed to discover the optimal value as an emergent property of the Azores VPE.
The experiment was initialised with 16 populations of phytoplankton, differing
only in their value of e. That was the basis for the inter-population variation in
competitive advantage. Each population started with the same number of plankters,
but after five years, the relative abundance had changed, with the greatest number
in those populations that had the value of e that was best fitted to the ambient c1imate. Fig. 18.5a shows the synoptic abundances on the 1 March of year 5 for an
ensemble offive VPEs (created with different seed values, see 18.10). The best fitted value of e was 16 Jll, but the effect of demographic noise is to reduce the demographic consequences of competition between the populations with e in the range
14 to 18 1lJ. The abundances of phytoplankton populations in this range had
increased at the expense ofthose with either lower or higher values. The less wellfitted plankton were driven to extinction with twenty years. This first numerical
experiment illustrated the capability of the LE method to simulate natural selection
in a Virtual Plankton Ecosystem.
The Azores VPE was recomputed using a revised version of the WB model in
which e has its best-fitted value of 16Jll. With primary production optimised in this
way, the spring bloom peaked five days earlier that in the e=O version. That contributed 10% towards eliminating the error identified by comparison with SeaWIFS observation (18.17).
18.9.3 Evolution by mutation
The second experiment (Woods & Barkmann 1995) was performed with a single
population of phytoplankton all initialised with e = O. This experiment allowed
mutation to occur in the value of e. Mutation events occurred randomly with a
mean frequency of once every 100 reproductions. (This is m4ch faster than occurs
in nature, the acceleration being necessary for computational reasons.) We see in
Fig. 18.5b that the distribution of e va1ues diffused through e-space over twenty
years, with a peak at the optimal value of 14 Jll. Note the simi1arity ofthe distributions in Fig. 18.5a and the last pane1 of Fig. 18.5b. Natural se1ection yields the
same distribution in parameter space, whether starting from a prescribed set of
18.9.2 Competitive exclusion
In the first experiment (Al Battran et al. 1998), we changed the values of only
one parameter. This is called the "infant mortality" parameter. It determines the
amount of energy (e microJoules) stored in the energy pool of a newly born phytoplankter. Reproduction tends to occur in the afternoon, followed by the hours of
darkness when energy loss by respiration is not compensated by energy gain from
photosynthesis. So having a sufficient energy reserve at birth, e, ensures that the
newly born phytoplankter does not die by energy starvation during the first night
(hence infant mortality). But there is a demographic cost: more energy must be
stored between successive reproductions. So varying e trades birth rate against
death rate. In the default version of the WB model, e = O, so primary production is
not optimised. A first guess at the optimal value of e suggests it should equal the
nocturnal respiration, which is 12 microJoules. The numerical experiment was
designed to discover the optimal value as an emergent property of the Azores VPE.
The experiment was initialised with 16 populations of phytoplankton, differing
only in their value of e. That was the basis for the inter-population variation in
competitive advantage. Each population started with the same number of plankters,
but after five years, the relative abundance had changed, with the greatest number
in those populations that had the value of e that was best fitted to the ambient c1imate. Fig. 18.5a shows the synoptic abundances on the 1 March of year 5 for an
ensemble offive VPEs (created with different seed values, see 18.10). The best fitted value of e was 16 Jll, but the effect of demographic noise is to reduce the demographic consequences of competition between the populations with e in the range
14 to 18 1lJ. The abundances of phytoplankton populations in this range had
increased at the expense ofthose with either lower or higher values. The less wellfitted plankton were driven to extinction with twenty years. This first numerical
experiment illustrated the capability of the LE method to simulate natural selection
in a Virtual Plankton Ecosystem.
The Azores VPE was recomputed using a revised version of the WB model in
which e has its best-fitted value of 16Jll. With primary production optimised in this
way, the spring bloom peaked five days earlier that in the e=O version. That contributed 10% towards eliminating the error identified by comparison with SeaWIFS observation (18.17).
18.9.3 Evolution by mutation
The second experiment (Woods & Barkmann 1995) was performed with a single
population of phytoplankton all initialised with e = O. This experiment allowed
mutation to occur in the value of e. Mutation events occurred randomly with a
mean frequency of once every 100 reproductions. (This is m4ch faster than occurs
in nature, the acceleration being necessary for computational reasons.) We see in
Fig. 18.5b that the distribution of e va1ues diffused through e-space over twenty
years, with a peak at the optimal value of 14 Jll. Note the simi1arity ofthe distributions in Fig. 18.5a and the last pane1 of Fig. 18.5b. Natural se1ection yields the
same distribution in parameter space, whether starting from a prescribed set of