148
these issues, related to the observed depletion of ozone and enhancement
in sulphate aerosols.)
7 Unification and rationalisation of climate and weather prediction models
In this paper the notion of predictability has been reviewed, incompletely,
on timescales of days, seasons and decades. In doing so we have used
both simple and comprehensive models. Simple models are useful in order
to formulate hypotheses and to study basic mechanisms, but quantitative assessment of predictability can only be achieved with comprehensive
model studies. On timescales of days, the comprehensive models were
based on those used for numerical weather prediction; on timescales of
decades, global climate models were clearly most appropriate.
One of the successes of our subject in recent years has been the blurring of divisions between these two classes of model. Models that were
primarily developed as weather prediction models are now used for climate
prediction. Indeed some institutes develop unified models which are used
for both climate and weather forecasting in equal measure.
At ECMWF, I have long taken the view that if a weather prediction
model is to be successful in predicting blocking activity, for example, it
should be able to simulate blocking activity satisfactorily in terms of its
own climate. Similarly, it would seem reasonable to expect that a coupled
model used for forecasting EI Nino should be able to simulate EI Nino
variability in climate mode.
However, the converse is also true. As IPCC (1990) state, ' .. confidence
in a model used for climate simulation will be increased if the same model is
successful when used in a forecasting mode'. This conclusion is consistent
with the analysis of the climate predictability problem in section 7, where
it was suggested that in order that climate change be predictable, errors in
flux representation must be locally less than 4 W 1m 2 in regions such as the
warm pool. Tolerable errors elsewhere might exceed 4W 1m 2 significantly.
This moves the persective of climate change away from the global point of
view, towards the local point of view. In my opinion, the best way to reduce
these local flux errors is through comparison with detailed observations over
the relevant area, eg from the TOGA COARE experiment (Webster and
Lucas, 1992). However, such data are only made over limited periods of
these issues, related to the observed depletion of ozone and enhancement
in sulphate aerosols.)
7 Unification and rationalisation of climate and weather prediction models
In this paper the notion of predictability has been reviewed, incompletely,
on timescales of days, seasons and decades. In doing so we have used
both simple and comprehensive models. Simple models are useful in order
to formulate hypotheses and to study basic mechanisms, but quantitative assessment of predictability can only be achieved with comprehensive
model studies. On timescales of days, the comprehensive models were
based on those used for numerical weather prediction; on timescales of
decades, global climate models were clearly most appropriate.
One of the successes of our subject in recent years has been the blurring of divisions between these two classes of model. Models that were
primarily developed as weather prediction models are now used for climate
prediction. Indeed some institutes develop unified models which are used
for both climate and weather forecasting in equal measure.
At ECMWF, I have long taken the view that if a weather prediction
model is to be successful in predicting blocking activity, for example, it
should be able to simulate blocking activity satisfactorily in terms of its
own climate. Similarly, it would seem reasonable to expect that a coupled
model used for forecasting EI Nino should be able to simulate EI Nino
variability in climate mode.
However, the converse is also true. As IPCC (1990) state, ' .. confidence
in a model used for climate simulation will be increased if the same model is
successful when used in a forecasting mode'. This conclusion is consistent
with the analysis of the climate predictability problem in section 7, where
it was suggested that in order that climate change be predictable, errors in
flux representation must be locally less than 4 W 1m 2 in regions such as the
warm pool. Tolerable errors elsewhere might exceed 4W 1m 2 significantly.
This moves the persective of climate change away from the global point of
view, towards the local point of view. In my opinion, the best way to reduce
these local flux errors is through comparison with detailed observations over
the relevant area, eg from the TOGA COARE experiment (Webster and
Lucas, 1992). However, such data are only made over limited periods of