422 John Woods
That will reduce the off-attractor uncertainty in the data set used to represent the
natural ecosystem.
18.17.6 The Azores
For that reason, I prefer to perform Ecological Turing Tests with observations
made at the Azores, where the mean circulation is weak, which reduces the annual
change in ambient climate, and where the Azores Anticyclone reduces weather
stress. The Azores also lies on the line of annual balance in the surf ace heat flux,
which makes it possible to ignore the advective heat flux divergence in simulating
the ecosystem at the demographic attractor. None of these advantages is found at
Bermuda, where data are collected for testing plankton ecosystem models
(Michaels and Knap 1996; McGillicudy et al. 1998). Nor are they found at the
Ocean Weather Stations in the North Atlantic (Fasham 1993).
18.17.7 Conclusion
In conclusion, Ecological Turing Testing provides a formal way to discover
when a model is deficient. The choice of feature for testing may suggest which
aspects of the model need to be improved. However, experience with Sea WIFS
observations suggests that the uncertainties in the feature as described by analysis
of observations is so large that only gross inadequacies of the model will be
revealed by existing observations. Urgent attention should be given to designing
new experiments that will yield observations that can be used for more stringent
testing of ecosystem models. That is the most important task for biological oceanographers.
18.18 Source errors in Primitive Equation modelling
18.18.1 Introduction
Priorities for R&D aimed at improving ecosystem models depend on accurate
analysis and ranking of the sources of errors. In the previous section I showed how
we quantify the output errors, by comparing selected features in the simulation
with observations. That process is limited by uncertainties in the observations. In
practice there are not many features that can be tested with currently available
observations, because they provide a rather uncertain description of any chosen
feature, and many features cannot be observed with present techniques. Furthermore, the difference between simulated and observed versions of a feature seldom
teU us what caused the error. Nevertheless, it is essential to identify the causes of
errors in simulating the ecosystem. If we cannot learn enough from verification of
external features, we must start at the other end, by assessing the source errors in
the modelling procedure. That is the subject of this section.
That will reduce the off-attractor uncertainty in the data set used to represent the
natural ecosystem.
18.17.6 The Azores
For that reason, I prefer to perform Ecological Turing Tests with observations
made at the Azores, where the mean circulation is weak, which reduces the annual
change in ambient climate, and where the Azores Anticyclone reduces weather
stress. The Azores also lies on the line of annual balance in the surf ace heat flux,
which makes it possible to ignore the advective heat flux divergence in simulating
the ecosystem at the demographic attractor. None of these advantages is found at
Bermuda, where data are collected for testing plankton ecosystem models
(Michaels and Knap 1996; McGillicudy et al. 1998). Nor are they found at the
Ocean Weather Stations in the North Atlantic (Fasham 1993).
18.17.7 Conclusion
In conclusion, Ecological Turing Testing provides a formal way to discover
when a model is deficient. The choice of feature for testing may suggest which
aspects of the model need to be improved. However, experience with Sea WIFS
observations suggests that the uncertainties in the feature as described by analysis
of observations is so large that only gross inadequacies of the model will be
revealed by existing observations. Urgent attention should be given to designing
new experiments that will yield observations that can be used for more stringent
testing of ecosystem models. That is the most important task for biological oceanographers.
18.18 Source errors in Primitive Equation modelling
18.18.1 Introduction
Priorities for R&D aimed at improving ecosystem models depend on accurate
analysis and ranking of the sources of errors. In the previous section I showed how
we quantify the output errors, by comparing selected features in the simulation
with observations. That process is limited by uncertainties in the observations. In
practice there are not many features that can be tested with currently available
observations, because they provide a rather uncertain description of any chosen
feature, and many features cannot be observed with present techniques. Furthermore, the difference between simulated and observed versions of a feature seldom
teU us what caused the error. Nevertheless, it is essential to identify the causes of
errors in simulating the ecosystem. If we cannot learn enough from verification of
external features, we must start at the other end, by assessing the source errors in
the modelling procedure. That is the subject of this section.