706
Herterich et al.
geostrophy improves the quality of the reconstruction considerably.
Experiments with smaller data sets (e.g. T and
S given at the sea surface only) show that the model's need for additional hints is stronger in this situation. Up to now it is necessary to start from an
ocean state which is not too far from the desired
result (e.g. from a modem state) when using the
amount of data the CLIMAP atlas provides. We
hope to overcome this problem by feeding more
data into the model, firstly more by number, secondly more by kind, e.g. the velocity at the ocean
floor (estimated from sediment properties) could
easily be assimilated in this kind of inverse model;
however, the estimates themselves are quite inaccurate.
2RAIOM currently uses the data from the
CLIMAP atlas for reconstructing the LGM; it
would profit directly from the enlarged data set
CLIMAP 2000 would provide. On the other hand,
2RAIOMcouid be used to check CLIMAP 2000's
data for validity: If a certain data point cannot be
reconstructed inversely within its error boundaries,
the quality of this point has to be checked carefully.
One goal of CLIMAP 2000 is to provide complete maps instead of just sets of data points. Instead of just inter- and extrapolating the available
data, 2RAIOM can be used in the framework of
CLIMAP 2000 to extend the data to regions where
measurements are too costly or impossible.
Reconstruction of vegetation zones
The idea of the inverse approach described here
is to use pollen counts from ocean drilling cores
combined with wind fields resulting from atmospheric modelling to reconstruct the vegetation distribution on the nearby land for various time slices.
In six cores taken off the coast of northwest Africa, pollen counts for various species as a function of depth have been obtained (Dupont 1993;
Dupont and Weinelt 1996; Jahns et al. 1998). The
ECHAM atmospheric general circulation model
has been run with paleoclimatic boundary conditions corresponding to different time slices (Lorenz
et al. 1996). From the wind fields of ten model
years, trajectories are computed (Fig. 15) which
have a length of four days and end at the core sites.
For a given vegetation distribution, a model can
be constructed which calculates the pollen concentration in the various cores by following the trajectories. The transport is too complex to be modelled
in detail; various processes are parameterized and
the appropriate parameter values are calibrated by
using the modem vegetation and comparing calculated and observed pollen counts in sediment surface samples. However, we are interested in calculating the vegetation distribution using the core
pollen counts. This is a very ill-posed problem, since
the number of unknowns (for a given species and
time slice) is the number of grid points on land in
the area covered by the trajectories and is much
larger than the number of data points (the number
of cores). There are thus (infinitely) many vegetation distributions which can reproduce the pollen
counts (and even more because only a reproduction within the errors is required).
Therefore we make use of an inverse approach
(Turchin et al. 1971), combining the data with a
priori information in order to choose the most believable solution. This can be done by demanding
that, of all possible solutions, the "correct" one is
that which is smoothest or most uniform or contains only the minimum required vegetation density.
Additional constraints can be obtained by disallowing different species to occur in the same area
which are known a priori not to do so, e.g. rain forest
and desert species. Preliminary results indicate that
even using just three cores, one constraint (minimizing the required vegetation density) and with no
calibration, the method can deliver "believable"
results.
Conventional vegetation models determine the
vegetation type at a certain location depending on
the climatic conditions (temperature, precipitation,
wind etc.) taken from an atmospheric model. The
results of atmospheric modelling are in tum influenced by the vegetation present (Claussen 1994).
This can lead to multi-equilibrium solutions, i.e.
putting rain forest in a specific region results in climatic conditions which preserve the forest, while
desert in the same region yields a climate preventing vegetation. The inverse approach described
herein offers a method which is more independent
ofthe results of atmospheric modelling (not completely, as wind fields are needed) and takes into
Herterich et al.
geostrophy improves the quality of the reconstruction considerably.
Experiments with smaller data sets (e.g. T and
S given at the sea surface only) show that the model's need for additional hints is stronger in this situation. Up to now it is necessary to start from an
ocean state which is not too far from the desired
result (e.g. from a modem state) when using the
amount of data the CLIMAP atlas provides. We
hope to overcome this problem by feeding more
data into the model, firstly more by number, secondly more by kind, e.g. the velocity at the ocean
floor (estimated from sediment properties) could
easily be assimilated in this kind of inverse model;
however, the estimates themselves are quite inaccurate.
2RAIOM currently uses the data from the
CLIMAP atlas for reconstructing the LGM; it
would profit directly from the enlarged data set
CLIMAP 2000 would provide. On the other hand,
2RAIOMcouid be used to check CLIMAP 2000's
data for validity: If a certain data point cannot be
reconstructed inversely within its error boundaries,
the quality of this point has to be checked carefully.
One goal of CLIMAP 2000 is to provide complete maps instead of just sets of data points. Instead of just inter- and extrapolating the available
data, 2RAIOM can be used in the framework of
CLIMAP 2000 to extend the data to regions where
measurements are too costly or impossible.
Reconstruction of vegetation zones
The idea of the inverse approach described here
is to use pollen counts from ocean drilling cores
combined with wind fields resulting from atmospheric modelling to reconstruct the vegetation distribution on the nearby land for various time slices.
In six cores taken off the coast of northwest Africa, pollen counts for various species as a function of depth have been obtained (Dupont 1993;
Dupont and Weinelt 1996; Jahns et al. 1998). The
ECHAM atmospheric general circulation model
has been run with paleoclimatic boundary conditions corresponding to different time slices (Lorenz
et al. 1996). From the wind fields of ten model
years, trajectories are computed (Fig. 15) which
have a length of four days and end at the core sites.
For a given vegetation distribution, a model can
be constructed which calculates the pollen concentration in the various cores by following the trajectories. The transport is too complex to be modelled
in detail; various processes are parameterized and
the appropriate parameter values are calibrated by
using the modem vegetation and comparing calculated and observed pollen counts in sediment surface samples. However, we are interested in calculating the vegetation distribution using the core
pollen counts. This is a very ill-posed problem, since
the number of unknowns (for a given species and
time slice) is the number of grid points on land in
the area covered by the trajectories and is much
larger than the number of data points (the number
of cores). There are thus (infinitely) many vegetation distributions which can reproduce the pollen
counts (and even more because only a reproduction within the errors is required).
Therefore we make use of an inverse approach
(Turchin et al. 1971), combining the data with a
priori information in order to choose the most believable solution. This can be done by demanding
that, of all possible solutions, the "correct" one is
that which is smoothest or most uniform or contains only the minimum required vegetation density.
Additional constraints can be obtained by disallowing different species to occur in the same area
which are known a priori not to do so, e.g. rain forest
and desert species. Preliminary results indicate that
even using just three cores, one constraint (minimizing the required vegetation density) and with no
calibration, the method can deliver "believable"
results.
Conventional vegetation models determine the
vegetation type at a certain location depending on
the climatic conditions (temperature, precipitation,
wind etc.) taken from an atmospheric model. The
results of atmospheric modelling are in tum influenced by the vegetation present (Claussen 1994).
This can lead to multi-equilibrium solutions, i.e.
putting rain forest in a specific region results in climatic conditions which preserve the forest, while
desert in the same region yields a climate preventing vegetation. The inverse approach described
herein offers a method which is more independent
ofthe results of atmospheric modelling (not completely, as wind fields are needed) and takes into