DATA ASSIMILATION IN BIOGEOCHEMICAL MODELLING
533
the loss side. The net impact of marine biology on biogeochemical cycles
is, however, controlled by the balance of production and loss processes.
Because marine phytoplankton seems to invest relatively more into
fence structures (mineral cell walls, spines, chains and colonies) than do
land plants, which seem to compete more for fastest growth, one might
even argue that marine ecosystems are more sensitive to loss processes
than are terrestrial ecosystems [Smetacek, 2001].
4.
Motivation for data assimilation
In a situation in which our understanding of marine ecosystem dynamics is still relatively poor and in which observations and data types are
distributed unevenly, data assimilation may be seen as promising tool
to interpolate in time and space as well as among different data types.
Dynamical, albeit hypothetical, rules, e.g. in form of model equations,
help to go beyond purely statistical interpolation and to link the observational information according to these rules. As such models have various
poorly known parameters and functional relationships, data assimilation
can at the same time provide a vehicle to estimate these parameters and
parameterisations. This is conceptually different from state estimation
that attempts to find a model state that agrees best with the observations and possibly a previous model forecast.
State estimation is used frequently in the field of meteorology to initialise new forecast simulations. For marine biogeochemistry, this aspect
is generally less relevant, although it has already been applied for operational planning of research cruises [Popova et al., 2002]. Forecast times
are typically limited to a few weeks. The dissipative character of the
dynamics that we believe to hold for marine ecosystems and that we use
in our models [Popova et al., 1997] and the strong seasonal and intraseasonal forcing in form of light, temperature, and mixing regimes lead to
a quick memory loss of the initial conditions in typically much less than
a year.
With respect to improving longer term forecasts, e.g. for climate prediction purposes, it seems to be more promising to rely on parameter
estimation (and “parameterisation estimation”) to improve our quantitative understanding of marine ecosystem dynamics. Data assimilation
then provides a tool to test various hypothetical model dynamics against
the available observations in an organised and quantitative way. The following sections attempt to give some overview over recent activities in
this area.
de
,
,
533
the loss side. The net impact of marine biology on biogeochemical cycles
is, however, controlled by the balance of production and loss processes.
Because marine phytoplankton seems to invest relatively more into
fence structures (mineral cell walls, spines, chains and colonies) than do
land plants, which seem to compete more for fastest growth, one might
even argue that marine ecosystems are more sensitive to loss processes
than are terrestrial ecosystems [Smetacek, 2001].
4.
Motivation for data assimilation
In a situation in which our understanding of marine ecosystem dynamics is still relatively poor and in which observations and data types are
distributed unevenly, data assimilation may be seen as promising tool
to interpolate in time and space as well as among different data types.
Dynamical, albeit hypothetical, rules, e.g. in form of model equations,
help to go beyond purely statistical interpolation and to link the observational information according to these rules. As such models have various
poorly known parameters and functional relationships, data assimilation
can at the same time provide a vehicle to estimate these parameters and
parameterisations. This is conceptually different from state estimation
that attempts to find a model state that agrees best with the observations and possibly a previous model forecast.
State estimation is used frequently in the field of meteorology to initialise new forecast simulations. For marine biogeochemistry, this aspect
is generally less relevant, although it has already been applied for operational planning of research cruises [Popova et al., 2002]. Forecast times
are typically limited to a few weeks. The dissipative character of the
dynamics that we believe to hold for marine ecosystems and that we use
in our models [Popova et al., 1997] and the strong seasonal and intraseasonal forcing in form of light, temperature, and mixing regimes lead to
a quick memory loss of the initial conditions in typically much less than
a year.
With respect to improving longer term forecasts, e.g. for climate prediction purposes, it seems to be more promising to rely on parameter
estimation (and “parameterisation estimation”) to improve our quantitative understanding of marine ecosystem dynamics. Data assimilation
then provides a tool to test various hypothetical model dynamics against
the available observations in an organised and quantitative way. The following sections attempt to give some overview over recent activities in
this area.
de
,
,