|
COUPLED
|
|
Coupled Ocean-Atmosphere Modelling
Since the last newsletter, the UGAMP Coupled Ocean-Atmosphere Modelling project has advanced considerably. The main model development phase has now been completed. We have in place a powerful and easy to use system for running coupled model integrations. This system is based around OASIS, which is a coupling interface originally developed at CERFACS in Toulouse, France. OASIS permits new models to be coupled straightforwardly with minimum modifications. This feature makes it easy to select component models of a complexity appropriate to the application at hand.The way in which a coupled model integration proceeds is illustrated in Figure 1.
The top-of-the-range configuration for the coupled model consists of the ECMWF IFS atmospheric model coupled to the MOMA ocean model. MOMA is based on a model originally developed by the NERC OCCAM project, which in turn was derived from the GFDL Cox-Bryan code. Within UGAMP, MOMA has been developed considerably by the addition of, for example, new physics parametrizations and new facilities for I/O.
Two main configurations of the ocean model have been set up. In the first version there are boundaries at 70N and 70S. In the second version, which is designed for tropical studies, the boundaries are at 29N and 29S. For the first experiments we are using the tropical version. Prior to carrying out fully coupled integrations, we are spinning up the ocean model by forcing it with climatological surface buoyancy fluxes and winds. Figure 2 shows the sea surface temperature and surface currents from the model midway through the fourth year of the spinup.
As soon as the ocean is fully spun up, we will carry out an integration in which it is coupled to the IFS model. A study of the fluxes produced by the IFS model when forced with observed SSTŐs suggests that we may expect some significant drift in the ocean model climate following coupling. The extent, nature, and insofar as possible causes, of any drift will have to be assessed. We will also investigate the variability produced in the model on a range of timescales.
The scope for specific studies investigating the role of ocean-atmosphereinteractions in climate and climate variability is very broad. To take just one example, Figure 3 shows variations in sea surface temperature in the Pacific ocean on the Equator. There are a number of striking features in the eastern part of the domain. Firstly, variations are dominated by an annual harmonic, whereas the variations in top-of-the-atmosphere solar forcing would lead one to expect a semiannual cycle on the Equator. Secondly, anomalies show a marked westward propagation. Thirdly, the amplitude of the anomalies is greatest near the east coast and diminishes westwards. Ocean-atmosphere interactions play a crucial role in explaining these features. With regard to the westward propagation, observations show that zonal wind stress anomalies also propagate westwards. A number of coupled mechanisms which may be involved have been suggested. These mechanisms are just some of the interactions that we will be investigating in the near future.
UGAMP Technical notes:1) A Flexible System for Coupled Ocean-At- mosphere Modelling, R. Sutton, J. Thuburn, I. Udall, January 1996.
2) The MOMA Ocean General Circulation Model, I. Udall, January 1996.
R. Sutton (Oxford), J. Thuburn (Reading, CGAM),
I. Udall
A coupled model of the middle atmosphere with interactive chemistry and dynamics
We have recently developed a new model of the middle atmosphere that incorporates interactive chemistry. In this model, ozone calculated from the chemistry scheme is fed back into the radiation scheme and influences the dynamics. The new model has been produced by coupling together the UGAMP Stratosphere-Mesosphere model (USMM) to the off-line chemistry and transport models, TOMCAT and SLIMCAT. The coupling interface is provided by OASIS, that has been developed at CERFACS in Toulouse, France for ocean atmosphere coupling.
The USMM is a mechanistic model, which permits the stratospheric evolution to be constrained to the observed tropospheric forcing by specifying the geopotential at the lower boundary of the model. The inclusion of a detailed ozone photochemistry in the USMM, including heterogeneous reactions on polar stratospheric clouds, will allow us to investigate the importance of feedbacks between perturbed chemistry and stratospheric dynamics.
The methodology of running the USMM and the Chemistry Transport Model (CTM) in tandem is illustrated in Figure 4. At the end of each coupling interval (six hours by default) the two models write out the relevant fields, i.e. winds and temperatures by the USMM and ozone by the CTM. The models then exchange signals with the coupler to say that they have finished their integration for that six hour period. Once the coupler has received signals from both models, it reads in the fields, interpolates then to the grid of the receiving model, and writes then out, before sending a signal to the models to tell them to resume integration for the next interval.
We believe that the approach that we have taken has the following advantages:
1) We can be flexible in our choice of spatial and temporal resolution for the two component models. For example we are able to run the transport chemistry model at lower spatial resolution than the dynamics model or with different time steps.
2) The use of the CTM allows us to use a more accurate and conservative advection scheme for the chemical tracers than the spectral scheme used by the USMM.
3) By taking two preexisting models in this way, we have been able to develop a new interactive model of the middle atmosphere relatively quickly. In addition, the flexibility of the approach means that it is easy to change component models. Initial testing has used TOMCAT coupled to the USMM with isobaric levels but future work will use SLIMCAT coupled to an isentropic version of the USMM.
4) Coupling between two different architectures is possible. For example, if SLIMCAT is adapted to run on a parallel computer, it would be pos- sible to run the CTM on a T3D while the USMM runs on a vector computer such as the YMP8 or the J90.
Figure 5 shows an early result from our initial testing of the model. It shows the difference in zonal mean temperatures between two runs of the USMM, one running in coupled mode using ozone from TOMCAT in its radiation scheme and one standing alone and using climatological ozone. In both runs, the USMM is forced with 100hPa geopotential heights from November 1992. Higher temperatures in the southern hemisphere stratosphere in the interactive run are caused by higher levels of ozone in this run than in the climatology. The upper stratosphere of the interactive run is also warmer in northern hemisphere high latitudes, but this difference can be ascribed to the inherent variability of the winter stratosphere in this region rather than any chemical changes.
We aim to use our model to investigate the importance of feedbacks between perturbed chem- istry and stratospheric dynamics during recent well observed northern hemisphere winter periods. Last winter had the lowest March stratospheric temperatures on record and it will be interesting to investigate the role of feedbacks between chemistry and dynamics in this and other winters. We will also be able to investigate future changes in ozone depletion as atmospheric concentrations of carbon dioxide increase and act to cool the stratosphere.
P. A. Stott and G. C. Watson (Edinburgh)
Data Assimilation Experiments with a Simple Coupled Ocean &
Statistical-Atmosphere Model
Some twin experiments have been carried out with a 2-layer ocean plus statistical atmosphere model built by Anderson and McCreary (1985), and extended by Balmeseda et al. (1994), to study El Nino Southern Oscillation, ENSO, phenomena. Such simple models are still the most successful at making hindcast predictions of ENSO events through the past 30 year period, Palmer and Anderson (1994).
The Model
The model consists of 2 active ocean layers representing the upper water column. The temperature in these layers and thickness can both vary. A deep third layer at rest lies below. The domain is from 30.75S to 30.75N and 122.25E to 68.25W with 1.5 degree resolution. The atmosphere is constructed statistically by blowing 30 years of Florida state university, FSU, winds over the ocean model and calculating the statistical relationship between the ocean temperatures produced and the surface winds. This relation can then be inverted for use when the model is run in "coupled mode" with the wind stress calculated from a given surface temperature distribution. The ocean model produces reasonable El Nino surface temperature signals over the 30 years FSU period in the central and western Pacific although the response in the east is less good, Balmeseda et al. (1994).
When the model is run in coupled mode and is initially excited with a westerly wind burst for 90 days in the west Pacific, the surface layer temperature anomaly in the central Pacific (Nino3 index) undergoes a periodic (4 year) oscillation seen in Figure 6a. Figure 6b shows the corresponding oscillations in the heat content (of each layer and total) over the Nino3 region where,
HC = HC1 + HC2 = h1(T1-T3)/H + h2(T2-T3)/H (1)
with T1, h1 being the temperature and thickness of the top layer 1; T2, h2 the same for layer 2, and T3 and H are constants. It can be seen that the interannual heat content oscillations are predominantly in the upper layer. This can be useful in the data assimilation experiment.
Initialization and Assimilation
When coupled models are initialised in order to perform a hindcast or forecast experiment it is usual to initialize them solely by specifying the winds over the previous few months in order to prepare the ocean component. Such a method can be successful if the tropical oceans are reasonably deterministic, i.e. are not governed by internal instability, Anderson et al. (1996). However with the advent of the TOGA array in the tropical Pacific, and with the success of recent satellite missions such a TOPEX, there is a wealth of direct ocean measurements which could be used in coupled ocean-atmosphere models. Altimeter data in particular is useful because the sea surface height can be regarded as a measure of total vertically integrated heat content in the oceans. Recent work on mid-latitude ocean models have shown that Altimeter data can be successfully assimilated in a largely conservative manner by seeking a redistribution of water masses (Cooper and Haines (1996)). The experiments below show that a similar idea can be useful in the tropics where a redistribution of heat content is the most important factor in future model evolution.
The experiments we have performed are in "twin" format where model data are reintroduced back into the same model from a differing phase of an ENSO lifecycle. In particular Figure 6c shows results of three such experiments on the evolution of the Nino3 index. In years 0-10 the evolution is identical to Figure 1. At year ten, when the model is entering a "La Nina" or cold phase (surface T in Nino3 area is low) some data from model year 8 is introduced in an attempt to reset the phase of the ENSO cycle.
Figure 6c shows the result of introducing T1 data alone from year 8. This might, for example, be measured by a satellite IR instrument such as the Along Track Scanning Radiometer of ERS1. The model atmospheric winds, which are sta- tistically related to T1, will also change immediately of course. The ENSO phase appears to be re-established though the amplitude of the model oscillation is weak and irregular but growing over the next 10 years. Certainly this would not help in predicting the oncoming ENSO immediately after the assimilation.
Figure 6c shows the result of introducing upper layer thickness h1 from year 8 without changing other variables. This thickness could be measured in practice by an array such as TOGA but it could not be measured remotely over wide areas. Again this assimilation fails because the El Nino phase is not produced. In this case the result is even more unsuccessful and the phase of ENSO is not reset.
Figure 6c assumes that surface temperature T1 and total heat content HC are known and inserted from the year 8 values. Both quantities can be measured remotely by satellite as the heat content is given by altimeter data. The problem is how to distribute the required changes in heat content in the vertical? Figure 6b shows that most interannual variations in total heat content are due to changes in the upper layer. Therefore in this assimilation experiment only the upper layer heat content, HC1, is changed. The new upper layer thickness, h1, is calculated by rearranging Eq. 1 assuming that HC and T1 are ob- served and T2 and h2 are left unchanged during the assimilation. This time the ENSO phase is successfully reproduced following the assimilation at year 10.
A more conventional technique for introducing altimeter information would be by vertically correlating the sea level data with thermocline water properties, Mellor and Ezer (1991), Pinardi et al. (1994), Fischer et al. (1995). Fischer et al. (1995) have already indicated that sea level data introduced into a multi-level coupled ocean model of the tropical Pacific might improve hindcast predictions of ENSO over the 1970's, 80's. The novelty of this new method for assimilating the data is in the conservation of the heat content imposed on layer 2. This should permit a more consistent approach to such assimilation experiments, for example permitting a series of assimilation steps with subsequent sets of satellite data over a period of time before the forecast is run onward to make a prediction. Other steps could be taken to recover and assimilate currents (although this is more difficult in the tropics where geostrophy does not hold, certainly on smaller scales).
We are currently working, in collaboration with David Anderson, on experiments using a new multi-level ocean model which retains the statistical atmosphere approach. We need to study how deep to apply the changes in surface heat content at the time of assimilation. Then hindcast experiments will be performed to test the assimilation strategy possibly using GEOSAT data for 1986-88 or TOPEX data for the 1990's. We believe such assimilation studies will be useful for working with very complex coupled GCMs which can only be run for short periods for practical reasons and therefore lend themselves to case studies which need to be prepared with good initial conditions.
Anderson and McCreary (1985) Slowly propagating disturbances in a coupled ocean-atmosphere model. J. Atmos. Sci. 42, 615-629.
Anderson, Sheinbaum, Haines (1996) Data assimilation in ocean models. Submitted Physics Rev. Manuscript available.
Balmeseda, Anderson and Davey (1994) ENSO prediction using a dynamical ocean model coupled to statistical atmospheres. Tellus 46A 497-511.
Cooper and Haines (1996) Data assimilation with water property conservation J. Geophys. Res. 101, C1, 1059-1078.
Fischer, Latif, Flugel, Ji (1995) On the benefit of sea level assimilation in the tropical pacific. Report 170, Max-Planck Inst. Hamburg.
Mellor and Ezer (1991) A Gulf Stream model and an altimetry assimilation scheme. J. Geophys. Res. 96 C5 8779-8795.
Palmer and Anderson (1994) The prospects of seasonal forecasting: a review paper. Quart. J. Roy. Met. Soc. 120, 755-793.
Pinardi, Miyakoda, Rosati, Gudgel (1994) A global ocean assimilation system for hydrographic data and satellite altimeter data, Proceedings Global Ocean Conference Vol1 Brighton 1994.
Keith Haines (Edinburgh)
Qin Zhang (Nanjing Institute of Meteorology, China Zhaoyong Guan, Nanjing Institute of Meteorology, Chin