A Scientific Strategy for U.S. Participation in the GOALS (Global Ocean-Atmosphere-Land System) Component of the CLIVAR (Climate Variability and Predictability) Programme (1998)

Modeling is an important means of understanding the physical structure of the coupled climate system and assessing its predictability. Modeling serves as a means of synthesizing the long-term observations, process studies, and empirical and diagnostic studies of GOALS and focusing them on the program's central objective—the development of skillful prediction of seasonal-to-interannual variations of the coupled ocean—atmosphere—land system. Therefore, a comprehensive modeling program is an essential component of GOALS.

To realize the full potential of models as an aid in achieving the scientific objectives of GOALS, a complete hierarchy of models needs to be developed and applied to various problems. The panel feels that global General Circulation Models (GCMs) and intermediate-scale models are the most appropriate models for quantitative climate simulations and prediction experiments. Intermediate and mechanistic models are recommended to isolate fundamental processes. Empirical models constructed on the basis of relationships found in the historical record are thought to be best used in the explorations of predictability and to act as a benchmark with which the performance of other models can be compared.

The panel suggests that the modeling efforts under TOGA, which dealt with time and space scales similar to those addressed by GOALS, can serve as examples of the range of activities envisioned for GOALS and as the starting point for modeling in the new program. It is recalled that skillful forecasts by empirical models identified some of the predictable components of the climate system. Later, highly simplified models disclosed the fundamental processes that influence the components of the coupled system and provided insight into its predictability. These processes were studied more deeply with sophisticated stand-alone

component models and in simplified coupled ocean—atmosphere models. The simplified coupled models were used with moderate success in prediction. They indicated the robustness of the predictability of the coupled system and demonstrated the utility of deterministic models as climate prediction tools. However, the simplified coupled models used were usually of limited geographic extent and depended on mimicking or approximating the physical processes involved in ENSO. The only product of many simplified models was a forecast of SST in the tropical Pacific Ocean region. They did not take into account other large-scale circulation features (e.g., the monsoons), nor did they provide attendant forecasts of precipitation, insolation, or circulation either local to the Pacific basin or remote from it.

To address the scientific objectives of GOALS and seek forecasts for the global domain, more complete modeling is thought to be necessary. It is especially important to develop models that have the ability to simulate the annual cycle and interannual variability and to predict seasonal-to-interannual variations of circulation and rainfall, two significant benchmarks for GOALS modeling.

In addition to continuing the effort to understand better the coupled system and develop more comprehensive models of it, modeling efforts in GOALS should also be applied to tasks such as determination of the predictability of the coupled system, experimental prediction, data assimilation, and observing system design.

The GOALS modeling process and its relationships to the other elements of GOALS are described in Figure 4-1.

Five principal types of activities are recommended for the GOALS modeling components. These are:

1. model development;
2. predictability and sensitivity studies;
3. experimental prediction;
4. development of data assimilation systems; and
5. observing system simulation.

In general terms, the panel believes that improvements are needed in all of the component models of the climate system in order to increase prediction skill in the seasonal-to-interannual time frame. In particular, the interface fluxes of various quantities such as heat, moisture, and momentum need to be represented more accurately in the models.

Atmospheric General Circulation Models need to be improved to a stage where, when driven with prescribed observed SST, they simulate realistically the observed annual cycle and interannual variability of the surface wind stress and heat flux, as well as the global atmospheric circulation and rainfall. Better

atmosphere and ocean models are expected to evolve from a systematic program of diagnosing and evaluating models with improved parameterizations. Similarly, Oceanic General Circulation Models need to be improved to a stage where, with prescribed surface stresses and heat fluxes, they simulate realistically the observed annual cycle and interannual variability of SST, upwelling, upper-ocean heat content, convection, and subduction. The improvement of Land Surface Process Models (LSPMs) to a stage where they represent adequately the interaction between the land surface and vegetation and the atmosphere is necessary. Serious consideration must be given to the incorporation of a dynamic land-surface component to existing models such that the characteristics of the land surface and the state of the biosphere evolve with changes in atmospheric circulation, clouds, precipitation, radiation, and so forth. On the large scale, LSPMs should also be constructed to account for the significant time lags between winter snow or ice and their melt in spring, leading to ground water recharge in spring. The coupling of land surface processes with the oceans through river outflow should also be examined. On shorter time scales, especially in the tropics, LSPMs should be able to characterize moisture recycling processes. Evapotranspiration through the vegetation canopy and albedo changes and feedbacks are important processes that have to be modeled better.

In addition to improving the individual component models of the climate system, the panel recommends an equally strong effort in coupling the component models together so that they better approximate the natural system. In this connection, the composite models should be tested by assessing their ability to simulate and predict key parameters and variables. Significant improvements are considered necessary in coupled ocean—atmosphere—land models so that they can realistically simulate the observed annual climate cycle and its higher frequency components, as well as the statistics of the observed interannual variability of tropical SST, land surface temperature, the global atmospheric circulation, and rainfall. These models will need to take into account land-surface hydrology processes, a requirement best met by close cooperation with GEWEX modeling groups working on the parameterizations of these processes.

To minimize the complications that arise when linking component models together, the panel recommends that coupled models should be developed with modular "plug compatibility." This would enable a broader participation of the scientific community in the development of models and also enable the exchange of various parameterization schemes between models and between scientists. If possible, protocols for "plug-compatibility" standards should be developed.

Though somewhat simpler than GCMs, models of intermediate complexity should also be constructed. These models can be compared quantitatively to observations for specific applications. Their uses include isolating fundamental processes, sensitivity studies, exploratory predictions, and ensemble predictions where computational costs are important. For GOALS applications, greater attention to land processes, physical parameterizations, and consistent energy bud-

gets and atmospheric eddy fluxes is highly recommended. Simpler models should be developed for the study of specific processes thought to be important in nature. By making it possible to study processes in a controlled setting, such models can lead to the identification and characterization of those mechanisms that affect seasonal-to-interannual prediction.

With regard to applications and human dimensions, the panel recommends that auxiliary models be designed to predict societally important quantities not routinely produced by seasonal-to-interannual climate forecast models. These models can be based on both empirical and physical techniques and could involve predictions of quantities including regional rainfall and storm track activity that are not predicted by intermediate models or, for example, the likelihood of tropical storms or extreme events that are not predicted by coupled GCMs. Auxiliary models include those used to make projections of agricultural yield, water availability, fish productivity, energy demand, economic impact, and so on.

Improved strategies should be developed for nesting high-resolution regional models and global climate models to infer detailed structures of regional climate anomalies forced by global boundary conditions predicted by coupled ocean—land—atmosphere models. Furthermore, the systematic and periodic intercomparison of models should be continued in order that the physical sciences and the user communities have available an ongoing assessment of the status of models.

The determination of the limits of predictability on the seasonal-to-interannual time scale includes not only traditional numerical experiments designed to examine global assessments of the limits of predictability, but also investigations of the geographical, seasonal, regime, and field dependence of model predictions. Individual component and coupled models of a broad range of complexities will be required to contribute to this task. Whether ENSO predictability is sensitive to high-frequency atmospheric forcing needs to be determined as well as the degree to which ENSO irregularity and limits to predictability might result from chaotic behavior in the slow components of the coupled system.

The panel recommends that the dynamic and physical mechanisms that contribute to internal and forced variability on seasonal-to-interannual time scales should be investigated. The aim of the investigation is to enhance our understanding of the causes of internal and forced variability so that results from comprehensive models and empirical studies can be interpreted. For example, what is the role of extratropical SST anomalies in climate variability? What is the predictability of the annual statistics of tropical storms? In particular, quantitative estimates need to be made of the range of time for which initial conditions of the atmosphere are important. This determination is crucial for the predictability of midlatitude variations. Another example is the determination of how the probability distributions of planetary waves respond to tropical heating anoma-

lies, interactions with the annual cycle, ENSO cycles, monsoon variability, and climate change.

Other activities recommended by the panel include experiments to determine the contributions of the upper-ocean heat content (tropical and extratropical) and land surface moisture and heat content to the sources of memory for seasonal-to-interannual variability. Also recommended are assessments of the relative sensitivity of climate system response (and its time evolution) to uncertainty in initial conditions and boundary conditions.

Central to GOALS is the development and implementation of experimental prediction projects directed at establishing the seasonal-to-interannual prediction capabilities of models, and to enhance the information content of their forecasts in regard to modeling applications. For example, it is important to determine the skill of state-of-the-art models in hindcasting past variations of the climate system. This should include an evaluation of component models with specified boundary conditions in addition to the evaluation of the performance of coupled models. Techniques such as ensemble forecasting could be utilized effectively for estimating the probability and distribution of future climate system states. The panel also feels that improved statistical and dynamic techniques need to be developed for assessing and predicting the skill of model forecasts. To this end, the panel recommends the establishment of standardized procedures and protocols for the ongoing evaluation of experimental forecasts and the comparison of these forecasts with empirical/statistical prediction methods.

Of particular importance is the identification of regions, circulation features, or phenomena that have above-average predictability on seasonal time scales. Moreover, real-time predictions of climate variations using different models and techniques should be carried out in addition to retrospective and research mode model experiments.

Data assimilation, which is also discussed in Chapter 6 under "Data Analysis and Assimilation; Data Reanalysis," is an activity that cuts across more than one GOALS element. The improvement of data assimilation methods, especially for the ocean and land components (and the coupled system), and demonstration of the usefulness of these techniques for experimental prediction are important aims of GOALS research. The panel emphasizes that four-dimensional data assimilation is essential for defining mutually consistent states of the atmosphere, the ocean, and the land surface as required by global coupled models. This should also enhance the predictability of the coupled system. The large cost of GCM

data assimilation implies that this activity needs to be carefully coordinated with the operational centers, including NCEP, the IRI, and NASA, among others.

Models are useful for estimating the utility of particular measurements. To the extent that models accurately simulate the statistics of the full spectrum of energetic space and time scales that occur in nature, they can be used to set resolution requirements that will avoid undersampling and aliasing. Ocean models are just approaching the point where they can contribute such assessments of observations for climate research purposes, but the empirical approach, where deliberate attempts to oversample are made in order to check for aliasing, is still needed. Atmospheric models, having been tested more thoroughly over the years, are currently more reliable for this purpose. Modeling experiments can also be used to determine the minimum observational system needed to induce data assimilation systems to generate the observed circulation fields to within acceptable tolerances. Scenarios for observations during the Global Weather Experiment (GWE; also known as FGGE) were evaluated with atmospheric general circulation models, for example. A unique requirement for GOALS is to assess elements of the observing system in regard to their utility for supporting predictions on seasonal-to-interannual time scales. The most efficient mix of observing techniques that will deliver the required accuracy and space—time resolution will be sought. The corresponding need for observing system simulation experiments with both component models and coupled models is a particular challenge to GOALS.

The panel strongly recommends the establishment of a GOALS Modeling Working Group (GMWG), in order to involve a broader cross-section of the scientific community and to encourage and coordinate a program for model intercomparison, climate system simulations, and experimental predictions. The GMWG should encourage a multi-model approach and, in particular, facilitate the interchange of fully documented and fully accessible models and data sets. The GMWG should also develop protocols for plug compatibility between models and standardized procedures for the ongoing evaluation of model simulations and experimental predictions. Other activities should include plans and estimates for computing and communication needs and the coordination of specialized numerical experiments. Under CLIVAR, the international Numerical Experimentation Group (NEG) is devoted to modeling issues on seasonal-to-interannual time scales. The GMWG, in consolidating the efforts of a broad group of U.S. modelers, should coordinate closely with CLIVAR/NEG to optimize international modeling efforts and cooperation.