An adjoint model is a transposition of a computer model's code that creates a different set of output related to the original output. For instance, instead of analyzing pressure levels, the adjoint MM5 analyzes pressure gradients, so that zones with strong gradients (and thus high winds) are highlighted.
Only one adjoint solution is possible for any given mathematical model. The complexity of weather and climate models makes the creation of their adjoint versions a difficult undertaking. However, the results provide a unique way to get at phenomena hidden within the standard data fields. "It helps you understand the original model in a computationally efficient way," says Zou. Adjoint versions are especially helpful when new sources of data are being assimilated into a model, such as special observations taken near a point of interest. Quantities that are being inferred--such as three-dimensional wind fields from Doppler radar--can be represented more directly in an adjoint model.
Zou's four-year NCAR appointment to develop the adjoint MM5 was supported primarily by the Global Positioning System meteorology program, or GPS/MET. The UCAR-based project used a space-borne receiver to intercept GPS signals, with the signal delay then analyzed to infer meteorological qualities of the atmosphere up to 60 kilometers (36 miles).
Several major research centers are at work developing adjoint versions of their models. According to Zou, the European Centre for Medium-Range Weather Forecasts hopes to make an adjoint version of its flagship global model available operationally by the end of 1997. However, NCAR's is the first adjoint version of a mesoscale model made available to the research community at large. It was officially unveiled on 24-25 July at a tutorial in Boulder that was "very successful," says Zou.
Two publications helped the participants get acquainted with the new model: a user's guide to the standard version of MM5 and an introduction to adjoint modeling. "It takes time for people to understand and use the system," Zou notes. The two publications are available as NCAR Technical Notes [ordering info here].
According to Bill Kuo, head of M3's Mesoscale Prediction Group, even while Zou is in Florida, the adjoint modeling effort "will continue within our group. We now have François Vandenberghe leading that effort. Of course, we'll maintain a close cooperation with Xiaolei Zou at Florida State University. We still have many joint projects, and we expect the cooperation to continue for many years."
Developed in the late 1980s, RegCM was the first community model to unite the high resolution of a mesoscale model with the long time scales of a global climate model. It was created by an NCAR group headed by Robert Dickinson (now at the University of Arizona). They nested version 4 of the NCAR/Pennsylvania State University mesoscale model (MM4) within version 2 of the NCAR community climate model (CCM2). The MM4 had to be tweaked slightly to make it suitable for research over periods longer than a few days; for instance, a radiation scheme was added to provide seasonal variations in solar input. With the resulting blended model, scientists could feed the data from long-term global simulations into a regional context and obtain more detail on what might occur at that scale.
RegCM has proven itself a flexible tool, used by more than 35 university researchers for studies of past, present, and future climates over a variety of regions, including the continental United States, Europe, Africa, eastern Asia, Australia, and the Aral Sea. NCAR scientist Linda Mearns (CGD) has assembled a new group within the Environmental and Societal Impacts Group that is using the RegCM for agricultural and other impacts studies.
Some computer time for RegCM was made possible by the Model Evaluation Consortium on Climate Assessment (MECCA). This international collaborative effort, managed by the Electric Power Research Institute, installed a Cray Y-MP at NCAR in the early 1990s expressly for long-term global and regional climate simulations. "The computer time from MECCA provided critical resources for testing and developing RegCM," says Filippo. (A new summary report on MECCA, Assessing Climate Change, has just been released by Gordon and Breach Science Publishers.
Linda and Filippo are now partners with William Chameides (Georgia Institute of Technology), joining with other researchers in the United States and China in a NASA-organized project to examine China's Yangtze Delta as a "metro-agro-plex," a place where rapid urbanization rubs shoulders with high-yield agriculture. "The best analogy in the U.S. is New Jersey," says Linda. She, Filippo, and colleagues will evaluate changes in land use while including industrial emissions in the newly upgraded RegCM. According to Filippo, "Our hope is not to get an answer on the exact future climate but to get a scenario of how rapid urbanization, industrialization, and agricultural practices might interact with the regional climate in a global change context. For example, northern China has been warming at a much faster pace than southern China, and they're not sure why."
The Journal of Geophysical Research-Atmospheres plans to devote a special section of its November 1998 issue to papers documenting the upgraded RegCM and related research. "We have many more users now, and many of them have never used the RegCM before," notes Giorgi. "The nice thing is that this will be the first time we've had all these studies assembled in one place."