Primitive Equation Modelling ofPlankton Ecosystems
421
accuracy of the routine used for calculating bio-optical feedback, and changing
from one to three size classes ofphytoplankton (Nogueira et al 2001).
18.17.4 Lessons learnt
What do we learn from this Ecological Turing Test of the WB model focusing on
the spring bloom observed by SeaWIFS? Firstly, the source of data was rich
enough to permit processing to extract products that were commensurate with the
simulation. Secondly, the ETT did reveal deficiencies in the original version of the
WB model, which was based on equations and parameters derived from the literature, and had never been "tuned" to improve comparison with observations.
Thirdly, the gap between observation and simulation was easily closed by improving the model in a rational way, namely by replacing deficient components (e.g.
optical biofeedback, cloud-radiation parametrization) with versions that were
known to be superior. Fourthly, there was no need to "tune" the model parameters
to close the gap between observation and simulation; such tuning runs counter to
the principles of primitive equation modelling, and should only be adopted as a
last, desperate measure when alI else has failed. (See Fasham and Evans 2000 for
the conventional view that tuning Eulerian models to observations is useful.)
FinalIy, the uncertainty in the observations is so large (mainly due to the need to
deal with patchiness) that after a few improvements the model was no longer vulnerable to further testing by the Sea WIFS observations (of that feature at that site).
That left the design of the model in limbo: it was clearly not perfect, but the
remaining deficiencies lay beyond detection by this particular SeaWIFS ETT. After
such correction there is no scope for assimilating Sea WIFS data in forecasts.
18.17.5 Imbalance
One final point about Ecological Turing testing of plankton ecosystem models.
We have seen that Lagrangian Ensemble modelling is remarkably stable, because it
integrates the biological equations along the trajectories of individual organisms.
There is a stable demographic attractor for balanced climate and a variety of nutrient loads (Fig. 18.4). To reveal the deficiency of an ecosystem model the Ecological Turing test must focus on that attractor. The test will be less useful if the
simulation is off the attractor. This can happen if the simulation has not yet settled
after erroneous initialisation, or because it describes a water column that has drifted
into the site of observation along an upstream gradient of ambient climate
(18.12.2). The WB model takes about three years to settle under stationary ambient
climate, so the ETT test described above was performed on the fifth year of the
simulation at a fixed location where the annual cycle of the environment is stationary. However, the Sea WIFS observations described a natural ecosystem, which is
inevitably in a state of transient response to changing ambient climate and weather.
It is very difficult to devise a procedure that will ensure that the natural ecosystem
was at its demographic attractor (or most probable demography, see 18.12.2) when
the observations were made. The best one can do is to choose a location and time of
year for which external stress is low, and has been low over the previous few years.
421
accuracy of the routine used for calculating bio-optical feedback, and changing
from one to three size classes ofphytoplankton (Nogueira et al 2001).
18.17.4 Lessons learnt
What do we learn from this Ecological Turing Test of the WB model focusing on
the spring bloom observed by SeaWIFS? Firstly, the source of data was rich
enough to permit processing to extract products that were commensurate with the
simulation. Secondly, the ETT did reveal deficiencies in the original version of the
WB model, which was based on equations and parameters derived from the literature, and had never been "tuned" to improve comparison with observations.
Thirdly, the gap between observation and simulation was easily closed by improving the model in a rational way, namely by replacing deficient components (e.g.
optical biofeedback, cloud-radiation parametrization) with versions that were
known to be superior. Fourthly, there was no need to "tune" the model parameters
to close the gap between observation and simulation; such tuning runs counter to
the principles of primitive equation modelling, and should only be adopted as a
last, desperate measure when alI else has failed. (See Fasham and Evans 2000 for
the conventional view that tuning Eulerian models to observations is useful.)
FinalIy, the uncertainty in the observations is so large (mainly due to the need to
deal with patchiness) that after a few improvements the model was no longer vulnerable to further testing by the Sea WIFS observations (of that feature at that site).
That left the design of the model in limbo: it was clearly not perfect, but the
remaining deficiencies lay beyond detection by this particular SeaWIFS ETT. After
such correction there is no scope for assimilating Sea WIFS data in forecasts.
18.17.5 Imbalance
One final point about Ecological Turing testing of plankton ecosystem models.
We have seen that Lagrangian Ensemble modelling is remarkably stable, because it
integrates the biological equations along the trajectories of individual organisms.
There is a stable demographic attractor for balanced climate and a variety of nutrient loads (Fig. 18.4). To reveal the deficiency of an ecosystem model the Ecological Turing test must focus on that attractor. The test will be less useful if the
simulation is off the attractor. This can happen if the simulation has not yet settled
after erroneous initialisation, or because it describes a water column that has drifted
into the site of observation along an upstream gradient of ambient climate
(18.12.2). The WB model takes about three years to settle under stationary ambient
climate, so the ETT test described above was performed on the fifth year of the
simulation at a fixed location where the annual cycle of the environment is stationary. However, the Sea WIFS observations described a natural ecosystem, which is
inevitably in a state of transient response to changing ambient climate and weather.
It is very difficult to devise a procedure that will ensure that the natural ecosystem
was at its demographic attractor (or most probable demography, see 18.12.2) when
the observations were made. The best one can do is to choose a location and time of
year for which external stress is low, and has been low over the previous few years.