Global Dynamics
The objective of the Global Dynamics Section (GDS) is to further develop the scientific understanding of the dynamical and physical mechanisms and theoretical predictability of large-scale atmospheric variability on timescales of days to decades. This process will allow construction of the scientific basis for predicting transient, global circulations in the atmosphere beyond present practical limits. GDS scientists take three approaches to their research: (1) numerical and theoretical experimentation with a hierarchy of physical models ranging from the barotropic vorticity equation to coupled atmosphere-ocean models; (2) diagnostic analyses of the cause of atmospheric climatic variability and its theoretical and practical predictability in simulation and forecast experiments using the Community Climate Model (CCM), Regional Community Climate Model (RegCCM), and the Climate System Model (CSM); and (3) sensitivity analyses of numerical prediction models to atmospheric initial and boundary conditions using variational techniques that will aid in the design of improved methods of data assimilation, for both conventional and non-conventional meteorological data, e.g., precipitation, soil moisture, and sea surface temperature (SST).
Predictability and Prediction Studies of Weather and Climate Variations
The studies described below are highlights of the research in GDS devoted to the prediction and predictability of climate variations and extreme events. These studies are integral to our section goals of extending and defining the spatio-temporal domain over which scientifically and societally useful forecasts can be made. GDS scientists have continued their interest in the inherent predictability of atmospheric phenomena and have utilized their expertise gained in ensemble prediction techniques to address the prediction of extreme events.
Akira Kasahara continues to work with Jun-ichi Tsutsui and Hiromaru Hirakuchi (both of Central Research Institute of Electric Power Industry (CRIEPI, Japan) to investigate the genesis of tropical cyclones (TCs) using T63 and T170 versions of the Community Climate Model version 2 (CCM2) with an Arakawa-Schubert cumulus parameterization. One unique aspect of this study is the use of precipitation rates estimated from satellite data in the diabatic and cumulus initialization for CCM. This technique allows us to test how well the prediction model can reproduce the genesis of TCs. In fact, it has been demonstrated with input data from European Centre for Medium-Range Weather Forecasts (ECMWF) analyses that the genesis of Typhoon Flo in September, 1990, over the western Pacific could not be reproduced without the use of this initialization technique. The predictability of the TC genesis is being investigated by using different input datasets of ECMWF, the National Centers for Environmental Prediction (NCEP), and the Japan Meteorological Agency (JMA) analyses, as well as the two different model resolutions. One interesting finding is that the ECMWF analysis that does not use TC bogusing produced much better 6-day forecasts of Typhoon Flo than NCEP and JMA analyses, both of which adopt TC bogusing. It is speculated that conventional TC bogusing methods without consistent diabatic balance between the dynamic and thermodynamic fields may be detrimental to medium-range forecasts of TCs, even though the appearance of analyses may improve by use of bogusing. Such a potential dynamic imbalance may prevent the containment of diabatic heating in the warm core of an incipient vortex. Thus, the development of TCs is eventually terminated due to the separation of diabatic heating from the cyclone core.
An important area of applied predictability research is the quantification of the uncertainty in medium-range weather forecasts through ensemble techniques. In an effort to further our understanding of the nature and limitations of current ensemble forecasting methods, Dave Baumhefner has addressed the question of ensemble sensitivity to different types of initial perturbations. Samples of the NCEP medium-range forecasting (MRF) operational ensemble forecasts (11 members at T62 resolution) and ECMWF operational ensemble forecast system (33 members at T63 resolution) made daily for the winter of 1995-96 were collected and analyzed. Cases were selected from this set and rerun with the CCM3 model ensemble system. The NCEP system uses a "bred mode" perturbation, the NCAR system uses an analysis difference simulator, and ECMWF uses a singular vector decomposition. The forecast skill of the three systems was analyzed. At longer timescales the dispersion of all ensembles were very similar, indicating the method of perturbation was not an important factor.
The early motivation for ensemble prediction studies in GDS was the problem of extended-range prediction, both as a means to dynamically ascertain probabilistic information in a system with limited deterministic predictability and to sharpen signals forced by anomalous boundary conditions. Baumhefner, with Joseph Tribbia, has continued research in seasonal forecast skill, concentrating on a forecast comparison project set up by the Center for Ocean-Land-Atmosphere Studies (COLA) in which several models will be tested for skill in the three-month time frame. Sixteen winter cases have been run with 10-member ensembles using CCM3. These forecasts were all forced by observed SSTs. The seasonal skill of these runs was evaluated and various methods of systematic error removal were tested. The forecast midlatitude patterns of flow were, on average, not very skillful; however, in 6 of the 16 cases, the skill was quite good. The probability distributions of the ensemble as defined by the individual member forecast values of the PNA index always included the observed value.
For studies of the predictability and prediction of interannual variability, the dominant climate variations in this time range are associated with the El Niño Southern Oscillation (ENSO). It is possible that even for seasonal predictions a prognostic SST field is necessary, and thus the predictability of the coupled upper-ocean/atmosphere must be examined. Tribbia, Peter Gent (Oceanography Section, OS), and Jeff Lee have completed the study of the predictive skill of a model system suitable for studying ENSO, i.e., CCM3 coupled to the CSM tropical Pacific model. This effort will diagnose the skill of this system in reproducing the major warm and cold events occurring over the last 15 years. To date successful forecasts have been made of the evolution of the 1982-83 warm event and the 1984 cold event with nine months lead time.
Diagnostic and Theoretical Studies of Variability and Validation
Within GDS the purpose of diagnostic analyses is twofold, diagnosis is used to test theoretical ideas concerning the mechanisms responsible for climate variations and their relative import and also test (i.e., validate) the behavior of comprehensive climate models like the CCM against that of the observed climate system. Naturally, the aforementioned prediction studies can also be viewed in this latter context. Additionally, several particularly insightful examples of past GDS studies exemplifying these two types of diagnoses are detailed below.
The climate system shows variability on seasonal to interannual timescales both in the tropics and in the extratropics. The tropical variability is dominated by the ENSO signal, which has received a lot of attention in recent years. The extratropical variability also has a remotely-forced component that is related to the tropical ENSO signal, especially in the Pacific-North American region. However, a large fraction of climate variability in the extratropics is not associated with ENSO, and is generated by middle/high latitude processes alone. Quantifying this intrinsic extratropical variability, analyzing its predictability, and understanding the mechanisms behind it is one of the research goals of GDS.
Ramalingam Saravanan has carried out a diagnostic study using the CSM that sheds some light on the mechanisms of midlatitude climate variability. A hierarchy of general circulation model (GCM) integrations were analyzed in the study, corresponding to different degrees of coupling between the ocean and the atmosphere--the 300-year coupled integration using the CSM being at one end of the hierarchy and the uncoupled CCM3 integrations forced by the climatological annual cycle of SST being at the other end. The former represents full coupling between the ocean and atmosphere, and the latter represents intrinsically atmospheric variability. At an intermediate level of coupling, CCM3 integrations forced by the observed monthly-mean SST were also considered. At each level of the hierarchy, the simulated atmospheric low-frequency variability was compared to the observed low-frequency variability, as represented in the NCEP/NCAR reanalysis data. The quality of the simulations improved with the increasing degree of coupling, with the coupled CSM integration providing the most faithful simulation. However, even the uncoupled CCM3 integration captured the spatial patterns of the variability rather well, although the amplitudes were somewhat off the mark. This means that the spatial patterns of atmospheric low-frequency variability in the coupled climate system are essentially the same as those in the uncoupled atmosphere. Coupling to an interactive ocean simply alters the amplitudes of the different modes of atmospheric variability.
The patterns of surface heat flux associated with the dominant modes of atmospheric low-frequency variability, the Pacific-North American pattern, and the North Atlantic Oscillation were also analyzed in this hierarchy of CSM and CCM3 integrations. The surface heat flux patterns show a close correspondence to observed spatial patterns of SST variability in the midlatitudes, indicating that stochastic low-frequency variability in the atmosphere may be the primary mechanism behind observed midlatitude climate variability.
(This figure (20K) shows the empirical
orthogonal functions (EOFs) of monthly-mean 500 millibars geopotential
height corresponding to the observed PNA-like pattern in the North
Pacific region during
The non-dimensional normalized EOFs have been multiplied by the square-root of the corresponding EOF variance to show the dimensional standard-deviation of Z500 at each grid-point associated with the EOF. The EOF number and the fractional variance associated with it are shown in the figure titles. Contour interval = 20m.)
For studying the low-frequency evolution of the atmosphere, linear models have proven to be a very useful tool. The wide range of diagnostic methods that are applicable to linear operators have made it possible to investigate comprehensively many aspects of atmospheric dynamics on intraseasonal and longer timescales. Of course, the more closely the dynamics of a linear model match the actual dynamics of the atmosphere, the more worthwhile the model is as a tool. For the past few years, Grant Branstator has been developing a linear model of the atmosphere that is designed to represent all of the linear dynamical processes affecting atmospheric evolution and that should be the most complete linear model that can be constructed. This is accomplished by using a empirical approach in which the dynamical equations are formulated to reproduce the atmosphere’s dynamics as observed in a long record of past behavior rather than formulating them from physical principles. Given his interests in external sources of low-frequency variability (e.g., the effect of SST anomalies), the fact that such a model approximates the response of a fully nonlinear GCM to equatorial heating much more accurately than a conventional linear model does, indicates that it is a promising means of studying various problems in low-frequency dynamics. Recent diagnosis of this model has helped to determine why it is more accurate than conventional models that are formulated by simply linearizing the equations of motion. Branstator has been able to determine that even though the state variable in his model is the barotropic streamfunction, implicit in its dynamics are the effects of divergence anomalies and of feedbacks from momentum fluxes associated with high-frequency transients. The latter, in particular, is a process that until now has been difficult to include in linear models, even though its importance in low-frequency dynamics is widely recognized.
Experiments by Branstator with this empirically-based model have highlighted the sensitivity of intraseasonal and longer timescale dynamics to the vertical structure of perturbations. Since no one vertical structure dominates in nature, this sensitivity has prompted him to extend the barotropic model that he initially used to test the feasibility of the empirical approach for constructing linear models to a multilevel model with vorticity, divergence, temperature, and surface pressure as its state variables. Initial experiments indicate that this model is significantly more skillful at approximating the equatorially-forced solutions of a GCM than was its barotropic counterpart. One drawback arises from the fact that the more degrees of freedom in such a model, the longer the dataset must be that is used in its construction. One approach that Branstator has found to be promising for reducing data requirements is to assume that the same dynamics are valid at all longitudes. The model that results from such an assumption, which is analogous to a model linearized about a zonally symmetric state, needs about an order of magnitude shorter record of atmospheric states but represents the dynamics that largely govern many kinds of situations.
For the most part, investigations of variability on longer than synoptic timescales have concentrated on winter, probably because that is when these fluctuations have their largest amplitude. Recently, Branstator has begun a project to document the seasonal cycle of such variability and to study dynamical processes that may produce this cycle. Initial findings indicate that basic properties like seasonal changes in the geographical distribution and amplitude of variability are reproduced rather well in CCM3, making it possible to use this model as a surrogate for nature in this study. The hypothesis that the seasonal cycle in variability can be largely explained by dynamical effects caused by the seasonal cycle in time-averaged conditions is now being tested. The results of this work are serving as motivation for a related joint study between Branstator and Jorgen Frederiksen (Commonwealth Scientific and Industrial Research Organization, CSIRO). Some of the dominant wintertime low-frequency structures in nature are thought to be related to dominant eigenmodes of the barotropic vorticity equation linearized about the mean wintertime circulation. To take into account seasonal changes in the system's dynamics, they are instead calculating the modes of the same equation but linearized about a state that slowly evolves in time. With this modification to the calculation, the leading modes are modulated by the seasonal cycle in a way that qualitatively matches the seasonal cycle of variability found in nature.
In recent years, there has been increasing attention devoted to climatic variability on decadal to centennial timescales. Understanding the mechanisms behind this variability is crucial to explaining regional climate shifts and also for distinguishing between natural and anthropogenic climate change. An important related question is the change in high-frequency variability concomitant with either climate change or decadal variability. Several research questions in this area are studied in GDS and discussed below.
During the first quarter of 1997, CRIEPI conducted, in collaboration with CGD scientists, a 125-year simulation using the CSM on a 32-processor SX-4 at the NEC Laboratory in Japan with assistance from NEC’s U.S. subsidiary, HNSX Supercomputers. (CRIEPI provided the financial support for computer time.) In this simulation, carbon dioxide levels in the CSM were specified to increase 1%/year, roughly in line with current global trends. Thus, a doubling of present-day levels of CO2 is reached after 70 years and a tripling after about 110 years. The global average temperature at that point was roughly 2°C above present values.
One objective of the above simulation is to investigate the changes in the climatological statistics on the behaviors of TCs due to the increase of CO2 in the atmosphere. Kasahara, working earlier with Tsutsui, conducted the survey of tropical cyclone-like vortices in a long-term simulation run with the CCM2 in which the observed SST distributions are used. It was found that the seasonal variations of tropical cyclogenesis are well reproduced by the standard T42 CCM2. However, after examining the daily outputs from the 125-year increased CO2 run with the standard (T42) CSM, it was found that tropical cyclone-like vortices in the atmospheric model (CCM3) of CSM are too weak to study meaningfully the statistics of simulated TCs.
One reason CCM2 was superseded by CCM3 is CCM3's improved climatology. This, however, does not necessarily mean that all aspects of atmospheric circulations are simulated better by the CCM3 than by the CCM2. In fact, the resolution of T42 may be too coarse for transient medium-scale phenomena, such as TCs. In this connection, we plan to examine the outputs from the Atmospheric Model Intercomparison Project II (AMIP II) simulations with a semi-Lagrangian linear-grid version of CCM3 at the resolutions of T63, T127, T191, and T235, which are being conducted under the direction of Robert Dickinson (University of Arizona). This will help us to determine what is the minimum resolution of CCM3 that is required to study the behavior of tropical cyclone-like vortices.
Because of the need for higher spatial resolution, the Regional Climate Model (RegCM) has been used in some studies of variability change. Filippo Giorgi, in collaboration with Linda Mearns (Environmental and Societal Impacts Group, ESIG), Christine Shields, and Maria Rosaria Marinucci, has completed an analysis of five-year long, present day, and 2xCO2 regional model experiments over two regions, one encompassing the Central Plains of the U.S. and one covering the Southeast U.S. The boundary conditions for the model runs are provided by the CSIRO GCM. These runs, which exhibit greatly improved performance compared to previous experiments with the RegCM, will provide climate change scenarios for use in impact assessment studies as part of two separate projects, one sponsored by the Department of Energy (DOE) and one by the Environmental Protection Agency (EPA).
Another application of the RegCM has been to study the climate and hydrologic budget of the Aral Sea Basin. The objective of this work is to study the effects of different climatic conditions and lake sizes, on the water budget of the Aral Sea. Different multi-year simulations with the RegCM coupled to an improved version of the lake model within RegCM, driven by ECMWF analyses of observations, and with different lake sizes, have been completed and are being analyzed. Giorgi is performing this study in collaboration with Lisa Sloan and Eric Small (both of the University of California, Santa Cruz).
RegCM has also been used in seasonal and interannual climate variability studies for Eastern Africa. The RegCM has been adapted to Eastern Africa and has been used to simulate the fall-winter short rain season over the region and its interannual variability. The model was driven by ECMWF analyses of observations for 12 October-December seasons and showed good skill in reproducing spatial surface climate patterns, as well as the basic features of interannual variability. Studies of deforestation and lake effects (Lake Victoria) have also been completed as part of this work. The work is in collaboration with Fred Semazzi and Liqiang Sun (both of North Carolina State University).
Sensitivity, Development, and Assimilation Studies
GDS has been at the forefront of scientific applications of variational techniques in large-scale dynamic meteorology and forecasting. The main applications have been in the study of the smooth assimilation of nonstandard initial data into forecast models and to quantitative sensitivity analysis. Adjoint models are the only computationally reasonable tool for determining the general sensitivity of most measures of aspects of model output with respect to small perturbations of model input. Knowledge of such sensitivity is critical for a wide range of applications, including data assimilation, targeting observations, predictability investigation, and general guidance in determining "what is important." The primary adjoint model used in GDS is version 1C of the Mesoscale Adjoint Modeling System (MAMS) developed by Ronald Errico and Kevin Raeder.
Errico and Raeder made several changes to MAMS, producing a new version, MAMS2. While the first version was principally based on the mesoscale model version 4 (MM4) formulation, with re-design to improve efficiency, the new one more closely follows the CCM to have an improved physical formulation. In particular, MAMS2 uses a more properly designed vertical difference scheme to have appropriate energy conservation properties and weaker, artificial diffusion. Their limited evaluation indicates that precipitation forecasts made with MAMS2 are qualitatively similar to both observations and forecasts produced by the mesoscale model version 5 (MM5) at comparable resolution at much less the computational expense. MAMS2 is being used at the Naval Research Laboratory (NRL) to develop observation targeting strategies, e.g., as implemented during the Fronts and Atlantic Storm Track Experiment (FASTEX).
Errico and Raeder also documented how well the tangent linear version of MAMS2 could reproduce the behaviors of perturbations in the nonlinear version when discontinuously-formulated, moist physical processes were important. When the most unstable singular vectors were perturbed with initial magnitudes the size of analysis uncertainty, quantitative agreement between their 24-hour evolutions were poor when convection dominated the physics but very good otherwise. Even for the cases of quantitatively poor results, there was good qualitative agreement. This was a very strict test to pass, because the magnitudes of the perturbations grew by factors of ten or more. Tests using the adjoint version of MAMS to estimate perturbed precipitation accumulations from significantly large, initial condition perturbations revealed high accuracy in regions of strong cyclogenesis where precipitation was strong.
One dynamical application of adjoint methods is in the construction of optimally growing structures, or singular vectors, for evolving flow fields. Errico and Raeder with Martin Ehrendorfer (University of Vienna) compared singular vectors produced by moist and dry versions of MAMS2 using several norms to measure perturbation growth. They found that: (1) extremely fast-growing instabilities appear where convection is dominant; (2) the growth of some strong instabilities in the dry model are significantly increased in the moist model, although the structures of the instabilities are qualitatively similar; (3) the explicit linearization of moist processes was necessary to capture these effects; and (4) results can be very sensitive to specification of weights in the norm.
Also, Errico, with Rolf Langland and Ron Gelaro (NRL), examined the theoretical relationships among singular vectors, Lyapunov vectors, bred modes (as produced at NCEP), and initial condition errors. These pertain to several operational and proposed applications, including ensemble forecasting, data assimilation, and observation targeting. Their analysis indicates that there is minimal theoretical support to suggest either Lyapunov vectors or bred modes are presently useful for estimating analysis errors or their statistics: the former do not appropriately consider the constraining influence of observations or effects of nonlinearity and the latter inadequately incorporate the effects of either the dynamics or the observations. Also, they note that singular vectors produced using properly formulated models and norms produce geostrophically-balanced perturbations, unless moist convection is explicitly considered and dominant, unlike has been claimed by others. In a related study, Errico and Carolyn Reynolds (NRL) determined that convergence of singular vectors to backward (also termed "local") Lyapunov vectors takes at least 5 days in a 3-level, T21 quasi-geostrophic model, longer than the 2 days estimated from much simpler or lower resolution models.
Another application of adjoint methods is in data assimilation. Errico, Doug Nychka (Geophysical Statistics Project, GSP), Zhan-Qian Lu (GSP), and Luc Fillion (Atmospheric Environment Service, Montreal) examined the statistical aspects associated with the variational assimilation of precipitation data. They used the common nonconvective precipitation model that has a first-order discontinuity at 100% relative humidity. For an assimilation system they considered a general, Bayesian formulation. They showed that precipitation assimilation systems described in the literature mistakenly ignore model error statistics and the constraint that precipitation is non-negative in defining an appropriate measure of fit of observations to predicted values. Also, minimum variance estimates and maximum likelihood estimates differed significantly due to the multi-modal nature of the posterior probability distributions. Further, they showed that the common model is biased. A simple modification to reduce the bias had the added benefit of rendering the model more continuous.
Assimilation of nonstandard fields is also straightforward using adjoint methods for solving the variational formulation of the problem. Tomislava Vukicevic is working with Peter Hess and Sasha Madronich (both of the Atmospheric Chemistry Division, ACD) to study the impact of surface emissions, boundary conditions, and transport (particularly convective transport) on tropospheric distribution of chemical species over the North Pacific. In this study they use a four-dimensional variational (4DVAR) data assimilation technique and an adjoint sensitivity method in conjunction with a regional Chemical Transport Model (CTM). This model uses a MM5 forecast to drive the transport.
In this study they are particularly investigating the use of clouds and other data to improve simulations of cloud distribution in model simulations. This will lead to improved simulations of cloud transport, particularly convective transport, which is of great importance for accurate representation of chemical species distribution in remote locations such as the North Pacific. The atmospheric model (MM5) and the CTM simulations involve timescales of 5 to 10 days and spatial scales of the order of 10000 km. Consequently, this study of the use of the model and data to improve the understanding of tropospheric processes is also relevant to more than the general problem of medium-range predictability.
More traditional sensitivity studies are also being carried out in GDS, as exemplified by the following examples.
As part of a hierarchy of coupled ocean-atmosphere models to study low-frequency variability in the climate system, Saravanan, Gokhan Danabasoglu (OS), and James McWilliams (OS and UCLA) have developed a coupled atmosphere-ocean sector model with an idealized land surface and ocean basin geometry. The atmospheric component is a global two-level spectral model, using the moist primitive equations and simplified physical parameterizations. The oceanic component uses the NCAR CSM Ocean Model configured in a 60 degree wide Atlantic-like sector geometry. Extended integrations of the coupled model show preferred interdecadal timescales in many of the oceanic variables. This interdecadal signal in the ocean is associated with, partly excited by, and possibly even feeds back non-locally on an North Atlantic Oscillation-like mode of variability in the atmosphere. The regions of most active variability are found to coincide with regions of deep convective activity in the ocean, suggesting that the variability may be sensitive to the parameterization of convection in the ocean model. Indeed, subsequent sensitivity studies have shown that the characteristics of the oceanic variability are significantly influenced by the particular choice of the convection parameterization used in the ocean model, where the less plausible choice induces a spurious centennial modulation of the interdecadal variability.
Tsutsui and Hirakuchi (both of CRIEPI) working with Kasahara completed a study on the sensitivity of two cumulus parameterizations on the prediction of TCs using the RegCM2 developed by Giorgi and his colleagues, based on the Pennsylvania State University-NCAR mesoscale model. The two cumulus parameterizations are the relaxed Arakawa-Schubert (RAS) and the modified Kuo schemes. The diabatic and cumulus initialization schemes, with the use of precipitation rates estimated from satellite data, are used to create the input data using the ECMWF analysis for Typhoons Flo and Ed in September, 1990. Prediction experiments are conducted for the two typhoons at various stages of their life cycles. One finding is that the impact of the diabatic and cumulus initialization is more significant than the choice in the cumulus parameterizations. At least during the first 24 hours, the predicted precipitation patterns with the two cumulus schemes are very close due to physical initialization. While the two cumulus schemes produce predicted precipitation patterns that agree well with observations, the Kuo scheme has a tendency to produce intense and spotty patterns, whereas the RAS scheme produces relatively weak and widespread precipitation patterns.
Also in RegCM, Giorgi, in collaboration with Anji Seth (University of Arizona), has been studying the effects of domain size and lateral boundary conditions on the simulation of summer climate over the continental U.S. This work has shown that care needs to be taken in the selection of the domain, especially when performing sensitivity experiments to internal model physics (e.g., soil water content). Giorgi has also examined the effect of the inclusion of CCM3 physics parameterizations (radiative transfer, convection, and land surface processes) within the RegCM. The convection and radiation schemes have been implemented and tested for seasonal to multi-year simulations over the continental U.S. and Asia. The implementation of the Land Surface Model (LSM) is still under way. This work, which will constitute the basis for the next RegCM version that will be released to the user community, has been done in collaboration with Shields and Keiichi Nishizawa (CRIEPI). Also underway is the development of a variable resolution model configuration by Joshua Qian (North Carolina State University).
As can be seen from the above example, sensitivity experiments are often the result of model development.
Giorgi is continuing the development of an on-line coupling of a tracer transport/removal scheme within the RegCM framework. This coupled model, which is presently being tested, will be used to study regional chemistry/aerosol-climate interactions over East Asia as part of the NASA-sponsored China Metro-Agro-Plex (MAP) project. The regional model will be coupled to a hierarchy of chemistry/aerosol models of increasing complexity to assess the effects of sulfate aerosols on regional climate. The model will also be used for regional air quality studies. This work is in collaboration with William Chameides (Georgia Institute of Technology), Congbin Fu (Chinese Academy of Sciences), Prasad Kasibhatla (Duke University), and a number of students and postdocs.