76
J. S. Godfrey
By comparing observed and modelled heat flux climatologies, Godfrey et al.
(submitted) and Yu et al. (submitted) have revealed two apparently very large problems: one with the ocean component, and the second with the atmospheric component,
of the net surface heat fluxes in the two types of model that are coupled in most of
today’s climate forecasting work.
In two papers, we (Godfrey et al., submitted) have compared observed heat
flux climatologies with the results published so far, from ocean models, on the annual
mean heat flux into the tropical Indian Ocean. The models’ fluxes are all markedly too
small—by a factor of about 2, if the WHOI climatology is correct. Two of these results
come from a model we have run; our model has very coarse resolution, but we get
similar net heat flux numbers to those from better models. In our own model we have
tracked the problem down to numerical “dispersion,” which causes the water column
to go (spuriously) convectively unstable when upwelling rates exceed a critical value.
The cause of this is well understood (e.g., Griffies et al., 2000), and it is likely to
occur in the other, higher resolution models. As a result, the absorption of heat into
the ocean—whose spatial and interannual variations are the major cause of natural
climate variability—are badly underestimated by the ocean component of coupled
models, at least in the tropical Indian Ocean.
Yu and colleagues’ results seem (on the face of it) to be still more drastic.
According to the WHOI climatology, the Indian Ocean absorbs about 50 W/m
2 on
annual and spatial average, north of 5
◦ S. Two “reanalysis” products, which assimilate
each day’s meteorological data into a recent-generation atmospheric model, give heat
fluxes of about 10 to 20 W/m
2 out of this region—i.e., these models seem to be
asking the Indian Ocean to be a heat source for the atmosphere, rather than a sink
as found in all the observation-based climatologies. I say “seem to be” because data
assimilation usually involves the introduction of artificial heat and moisture sources
into the model atmosphere, which will impact on its surface heat flux. A fairer test
would be to examine the heat flux climatology of a model run over observed SST, with
no data assimilation. Evidently, I am not qualified to comment on possible sources
of the problem with the atmospheric models (if there indeed is a problem): but if Yu
and colleagues’ result is found also in such model runs, they at least will offer a big
target for rectification efforts.
If the WHOI climatology lives up to its promise, I believe we first need to use
it to identify errors in net heat fluxes between the individual ocean and atmosphere
models and the observations, all over the world; and then (the hard part) to diagnose
and hopefully correct the sources of error in the models. A well-designed effort of this
kind should result in coupled models whose natural climatology is more realistic—and
whose natural variability is also more realistic, probably on all time scales.
The new climatology would also provide authoritative data on long-term trends
in heat fluxes into the ocean. For example, there is presently a massive trend of this
kind in the tropical Indian Ocean (Yu et al., personal communication), so big it is easily
detectable in the present WHOI climatology. Testing coupled models for their skill
in reproducing such long-term trends in heat fluxes will also provide a sophisticated
and powerful tool for assessing the quality of greenhouse models.
J. S. Godfrey
By comparing observed and modelled heat flux climatologies, Godfrey et al.
(submitted) and Yu et al. (submitted) have revealed two apparently very large problems: one with the ocean component, and the second with the atmospheric component,
of the net surface heat fluxes in the two types of model that are coupled in most of
today’s climate forecasting work.
In two papers, we (Godfrey et al., submitted) have compared observed heat
flux climatologies with the results published so far, from ocean models, on the annual
mean heat flux into the tropical Indian Ocean. The models’ fluxes are all markedly too
small—by a factor of about 2, if the WHOI climatology is correct. Two of these results
come from a model we have run; our model has very coarse resolution, but we get
similar net heat flux numbers to those from better models. In our own model we have
tracked the problem down to numerical “dispersion,” which causes the water column
to go (spuriously) convectively unstable when upwelling rates exceed a critical value.
The cause of this is well understood (e.g., Griffies et al., 2000), and it is likely to
occur in the other, higher resolution models. As a result, the absorption of heat into
the ocean—whose spatial and interannual variations are the major cause of natural
climate variability—are badly underestimated by the ocean component of coupled
models, at least in the tropical Indian Ocean.
Yu and colleagues’ results seem (on the face of it) to be still more drastic.
According to the WHOI climatology, the Indian Ocean absorbs about 50 W/m
2 on
annual and spatial average, north of 5
◦ S. Two “reanalysis” products, which assimilate
each day’s meteorological data into a recent-generation atmospheric model, give heat
fluxes of about 10 to 20 W/m
2 out of this region—i.e., these models seem to be
asking the Indian Ocean to be a heat source for the atmosphere, rather than a sink
as found in all the observation-based climatologies. I say “seem to be” because data
assimilation usually involves the introduction of artificial heat and moisture sources
into the model atmosphere, which will impact on its surface heat flux. A fairer test
would be to examine the heat flux climatology of a model run over observed SST, with
no data assimilation. Evidently, I am not qualified to comment on possible sources
of the problem with the atmospheric models (if there indeed is a problem): but if Yu
and colleagues’ result is found also in such model runs, they at least will offer a big
target for rectification efforts.
If the WHOI climatology lives up to its promise, I believe we first need to use
it to identify errors in net heat fluxes between the individual ocean and atmosphere
models and the observations, all over the world; and then (the hard part) to diagnose
and hopefully correct the sources of error in the models. A well-designed effort of this
kind should result in coupled models whose natural climatology is more realistic—and
whose natural variability is also more realistic, probably on all time scales.
The new climatology would also provide authoritative data on long-term trends
in heat fluxes into the ocean. For example, there is presently a massive trend of this
kind in the tropical Indian Ocean (Yu et al., personal communication), so big it is easily
detectable in the present WHOI climatology. Testing coupled models for their skill
in reproducing such long-term trends in heat fluxes will also provide a sophisticated
and powerful tool for assessing the quality of greenhouse models.