1. Introduction
The 2008 Summer Olympic and Paralympic Games will be held in Beijing, P.R. China. The success of both indoor and outdoor events and the well being of spectators during the Games will be affected by weather conditions and ambient air pollution levels. The Chinese Meteorological Agency (CMA) plans to provide comprehensive environmental forecasts in support of the Games. The Institute of Urban Meteorology (IUM) and the Beijing Meteorological Bureau (BMB) are two organizations that have been tasked with the development of such capabilities under the direction of CMA. At the suggestion of Dr. Ying-Hwa (Bill) Kuo of the National Center for Atmospheric Research (NCAR) in Boulder, CO, representatives from IUM and BMB are interacting with scientists at many U.S. institutions to transfer current techniques and new ideas into action for the 2008 Olympic Games weather support.
This report contains a description of scientific meetings held in Beijing, P.R. China for portions of five business days from September 6-10, 2004. Three University of Washington (UW) atmospheric scientists traveled from Seattle, Washington to Beijing for one week to explain the real-time, regional environmental prediction capabilities in place at the UW Department of Atmospheric Sciences under the direction of Dr. Clifford Mass. Unfortunately, due to a medical emergency, Dr. Mass was unable to make the journey to Beijing as had been expected. Research Scientist, Richard Steed, and Ph.D. candidates, Justin Sharp and Eric Grimit, arrived in Beijing without their leader and were faced with the daunting task of presenting Dr. Mass' material in addition to their own. Despite the unexpected challenges, the scientific exchanges were extremely successful and future collaboration between UW and IUM scientists was informally agreed upon. The UW scientists enjoyed an enriching cultural experience in addition to the scientific exchange.
2. Scientific Experience
In total, eight talks of more than one hour each were given by the three UW representatives during the first four days of the exchange. The presentations dealt with the components of the UW regional environmental prediction system, and various research projects in support of the effort. Issues related to the setup, funding, and organization of regional environmental prediction centers were discussed. The topic of resource allocation (high-resolution deterministic forecasts versus lower-resolution forecast ensembles) was addressed using the research studies and experiences of the UW group. Strategies for post-processing both deterministic and probabilistic mesoscale forecasts of near-surface weather variables were highlighted. Technical aspects of model error sources were discussed in detail, eliciting a great deal of interest from the IUM scientists. On the final day, IUM scientists presented findings from their research in China. A discussion of future possibilities for collaboration and recommendations of strategies for Olympic Games weather support ensued. The following sections contain summaries of each presentation and the major points of discussion.
a. Monday, September 6: BMB, Academic Hall
IUM director Yingchun Wang provided welcoming remarks in both English and Chinese, expressing gratitude on behalf of IUM and BMB. Justin Sharp introduced the visiting scientists from UW to the group and extended heartfelt thanks for the invitation to Beijing and the willingness on the part of IUM to organize the sessions and the sightseeing. Mr. Sharp then proceeded with the first part of an introductory talk titled, "Overview of the Pacific Northwest Environmental Prediction System", originally to be given by Professor Cliff Mass.
The first part of this overview talk included a brief introduction to weather in the U.S. Pacific Northwest. Pacific Northwest weather is primarily controlled by the combination of the upstream influence of the Pacific Ocean and the substantial topography. Pacific Northwest mesoscale weather phenomena are modulated strongly by the interaction of the large-scale, synoptic flow with the small-scale terrain features. Mesoscale forecast models perform quite well under this type of regime, if the synoptic-scale flow is accurately predicted. Mr. Sharp showed examples of skillful forecasts of Puget Sound Convergence Zone precipitation with MM5. However, an example case showing extreme synoptic-scale forecast error at lead times of only 48 hours was also presented, reminding the audience that mesoscale forecasts under such situations are only useful if the large-scale flow has been accurately predicted by the global forecast model used to initialize the mesoscale model.
Eric Grimit outlined the major components of the Pacific Northwest Environmental Prediction System. The system consists of the MM5 atmospheric model, a collection of synoptic-scale atmospheric model forecasts from operational weather centers worldwide, the DHSVM hydrology model, the CALGRID air quality model, the Oregon State land-surface model, an extensive collection of regional observations from local networks, and guidance based on research from major field programs (e.g., COAST, IMPROVE) and the work of other groups at UW. Mr. Grimit provided some technical details about the atmospheric portion of the regional environmental prediction system. The UW MM5 real-time deterministic runs are configured as three nested domains with one-way interaction down to 4-km grid spacing and 38 vertical levels. Initial conditions and lateral boundary conditions are derived from the NCEP GFS model. Analysis-nudging is used solely on the 36-km domain to keep the larger-scale MM5 forecast close to the global model solution. The 12-km and 4-km domains are allowed to freely evolve, subject to the boundary conditions coming from the 36-km forecast. The physics options used include the Kain-Fritsch cumulus parameterization, MRF planetary boundary layer scheme, Reisner II mixed-phase microphysics, and the CCM2 radiation parameterization with a simple modification to the effective ice radius. The UW MM5 runs in real-time at both 0000 and 1200 UTC cycles and a large number of products are posted to a website (http://www.atmos.washington.edu/mm5rt).
The setup of the regional, short-range ensemble forecast system (down to 12-km grid spacing) was also discussed briefly. Using initial conditions and lateral boundary conditions from synoptic-scale analyses and forecasts of global operational weather centers, an ensemble of 48-hour MM5 forecasts are also generated twice per day. A more detailed talk on the ensemble system was given later in the visit.
The computational resources required to run these systems was also touched upon. Three Linux clusters of 20, 32, and 40 processors are the main workhorses, but other systems also play roles. Approximately 20 Terabytes of disk storage is also in use.
Richard Steed proceeded with descriptions of the integrated regional prediction system including hydrology, air quality, land-surface, pavement temperature, and smoke models. The Distributed Hydrology Soil Vegetation Model (DHSVM) is run at 150 m resolution using MM5 temperature, wind, and precipitation forecasts for individual watersheds. DHSVM forecasts provide streamflow predictions for most major Pacific Northwest rivers. In cooperation with Washington State University, the AIRPACT regional air quality modeling system uses MM5 forecast output and CALGRID to predict the dispersion of O3, NOX, VOC, and particulate matter around the Puget Sound region.
The road weather information system was introduced. This effort, which was discussed in more detail another talk during the visit, is a partnership between the UW and the Washington State Department of Transportation (WSDOT), aimed at providing the public and the WSDOT with a combined source of weather data, road conditions, model forecasts, and other information. Lastly, Mr. Steed discussed the smoke and fire management system, which is a joint project between UW and the U.S. Forest Service. MM5 forecast grids are used as inputs for running Eulerian and Lagrangian smoke plume and dispersion models. Often, the U.S. Forest Service uses MM5 forecast guidance in the fire-lighting operations to combat wildfires.
The discussion that followed the overview talk was oriented toward common forecast problems that face both IUM and UW. Certainly, air quality is a major concern in Beijing, and pollutant forecasts will be critical during the 2008 Olympic Games. Efforts are already underway to meet this need. Dust storms can reduce visibility, especially during the spring when strong winds blow from Mongolia into northern China. Other main forecast problems for Beijing during the summer include high maximum temperatures and occasional heavy rainfall. Although it was agreed that IUM and BMB face different forecast challenges than the UW group due to the general difference in climate between the two regions, it was also recognized that many of the forecast methods are transferable. One issue facing both groups is the sparseness of 3D observational data upstream. Ensemble methods are sure to play a large role in the future forecasting techniques of both groups to quantify initial condition uncertainty.
The second talk of the day dealt with regional weather prediction efforts in the U.S. and was given by Mr. Grimit. The talk outlined the organization of weather prediction efforts in the U.S., which were centralized at a few major operational centers before 1990. The ingredients for regional numerical weather prediction became available with the advent of UNIX/VMS workstations and the development of mesoscale models (e.g., MM5, RAMS, and ARPS) in the 1990s. The recent rapid increase in local computational power has helped to dramatically increase the number of groups in the university, government, and private sectors running local mesoscale models. The attraction to this regional approach was that local models could be tailored for the region in question. Substantial regional variations in climate, weather, and surface characteristics preclude a "one size fits all" strategy. In the U.S., a hybrid approach to numerical weather prediction seems most effective where regional centers are managed by a national entity. However, this type of infrastructure does not currently exist. Therefore, the long-term viability of existing regional prediction centers is in doubt.
The discussion after this talk centered on regional prediction efforts in China. The exact organizational structure was not completely lucid after the discussion, but it seemed that China already had the appropriate components in place. Under the CMA, each Chinese province or autonomous region has a meteorological bureau that handles local environmental forecasts within that region. An example is the BMB, which handles such forecasts for Beijing only. In general, it seemed that the Chinese structure was more amenable to the presence of regional weather prediction centers that cooperate with the national organization. Here, it seems that the U.S. could learn from the Chinese.
b. Tuesday, September 7: IUM, Meeting Room
Director Wang began the day's meetings with a 15-minute talk about IUM. The institute was established in 2001 and is the only research institution in China dedicated to urban meteorology. IUM is one of eight research institutes established under the CMA within China's Ministry of Science and Technology. The goals of IUM are to conduct applied research associated with urban meteorology using innovative observational, data processing and modeling techniques and to promote urban meteorological research within China. There are four major branches of IUM including: the observation and data processing (ODP) branch, the forecast technique (FT) branch, the urban ecological meteorology (UEM) branch, and the urban meteorological disaster (UMD) branch.
The objectives of the ODP branch of the IUM are to design observing networks to meet the requirements for meteorology and climate data in the urban environment, to develop new observing techniques, and to collect, quality control, distribute, and archive massive amounts of observational data. The FT branch is tasked with the development of advanced modeling techniques at fine spatial-scales for the urban area, data assimilation at such fine spatial-scales, post-processing and verification of model forecast output, and severe weather nowcasting. The UEM branch deals with the relationship between urban development and climate change, interactions between the weather and the urban ecological systems, impacts of weather on human activities and health, air quality forecasting, urban biological and medical meteorology, and meteorological factors in city planning. The focus of the UMD branch is on mitigation of meteorological disasters (e.g., heavy rain events, sustained periods of very high temperatures, and sand storms), emergency planning, response, and recovery in the urban area, risk evaluation and consultation on policy-making and city-planning.
As a result of vigorous hard work and willingness to interact widely with scientists from the U.S., Germany, Australia, Denmark, Korea, and Taiwan, IUM has accomplished a great deal in only three short years of existence. IUM has developed a mesoscale NWP system using MM5 for operational weather forecasts in the Beijing area, an air quality forecasting system, a dust-weather and sand storm prediction system for operational use at BMB, and a highway visibility monitoring system and forecasting technique. IUM has also conducted research on the relationship between weather, air pollution, and city planning. IUM scientists collaborated with NCAR scientists during the Sino-U.S. Workshop on Convective Storm Nowcasting and are continuing that exchange to develop a very short-range prediction system for convective weather systems in Beijing. IUM is also working with NCAR scientists to develop techniques for ground-based GPS retrieval of water vapor information.
Two projects are underway at IUM to develop a weather support system for the 2008 Summer Olympics. The first project is a study on a comprehensive database storage system for 4D meteorological observations. The second project is a study on precision forecasting techniques in the urban area, including wind forecasts around buildings.
Mr. Sharp presented his extensive observational and modeling research on gap winds in mountainous terrain. Gap flow often plays a profound role in defining the weather and climate within and downstream of mesoscale channels. Although Mr. Sharp's research focuses on one particular gap, the Columbia River Gorge, his results can be generalized to many gaps around the world including those near Beijing. The presentation began by highlighting the specific effects of Columbia Gorge gap flow on the city of Portland. In contrast to the southwesterly wind direction that is typical of the Pacific Northwest in the winter, east-southeast is the prevalent wind direction in Portland because of the frequent occurrence of easterly gap flow. This gap flow can also bring freezing temperatures and support conditions that result in snow, sleet and freezing rain. A case study of a freezing rain event was described. Mr. Sharp showed that the MM5 mesoscale numerical weather prediction model, run at sufficiently high resolution could realistically recreate Gorge gap flow. A resolution of at least 1.33 km grid-spacing was required. Mr. Sharp also noted the importance of selecting model parameterizations that were appropriate for the weather regime being simulated. In particular, he showed that in the case of Columbia Gorge gap flow it was vital that the upstream atmospheric profile is correctly simulated. The upstream source of air is the Columbia Basin, which in winter is often enveloped by supercooled fog or low stratus cloud. Microphysical parameterizations that cannot represent supercooled water (e.g. the Dudhia Simple Ice Scheme) will rapidly precipitate the cloud water as snow leading to gross inaccuracies in the surface radiation budget and thus the temperature and stability of the low-level air.
The rest of Mr .Sharp's presentation focused on results from a comprehensive investigation of the parameter space affecting Columbia Gorge gap flow events. Simulations were run using the MM5 with three nested domains, where a large inner domain with a grid spacing of 1 km covers the entire Columbia Gorge and surrounding terrain. By using thermal wind balance, the entire model domain can be initialized using vertical soundings at one or two specific grid points. This method is validated by showing that the gap flow of the December 2000 case can be approximately reproduced using idealized initial conditions based upon a single point sounding on each side of the Cascades. Numerous other sounding profiles were then used to systematically analyze the gap flow that develops within the Gorge for different upstream and downstream conditions. Key results from this "phase space" investigation were presented. Based upon these results it was proposed that the dynamical and structural characteristics of the flow through the Gorge appear to be somewhat analogous to the hydraulic responses predicted by shallow water theory. However, inconsistencies with this theory were also found and research is ongoing to determine more precisely the mechanisms governing the flow.
A talk on post-processing of mesoscale model output was given by Mr. Steed and Mr. Grimit. Mr. Steed mainly discussed a series of grid-based bias removal strategies being considered by the UW group for eventual implementation on the real-time deterministic MM5 forecast system. A relatively simple approach using a running-mean estimate of forecast bias at observation locations from the previous 14 days appears to work quite well for bias correction of 2-m temperature and relative humidity forecasts, although research is ongoing. Each grid point uses the trailing 14-day forecast bias estimates from the five closest observation locations that share the same land use category (as designated by the MM5) and elevation (to within a certain tolerance).
Mr. Grimit discussed additional post-processing techniques under development for the UW mesoscale ensemble system. The bias correction approach (and variants) discussed by Mr. Steed improves probabilistic forecast quality, but does not fix the problem of low dispersion. Truth still falls outside the range of the ensemble forecasts more often than it should. Mr. Grimit briefly discussed an ensemble model output statistics (EMOS) approach, developed by Dr. Tilmann Gneiting at the UW statistics department, that can be interpreted as a multiple linear regression approach that accounts for forecast bias in mean and variance and the spread-skill relationship. A new technique gaining broad acceptance in the ensemble post-processing community worldwide is called Bayesian Model Averaging (BMA). BMA was also developed at the UW statistics department under the direction of Dr. Adrian Raftery. BMA fixes the low dispersion problem and allows for multi-modal probability forecasts (unlike EMOS). Plans are currently underway at UW to implement BMA for real-time post-processing of UW mesoscale ensemble gridded forecasts.
The discussion after Tuesday's presentations was sparse. One comment was that it was surprising that the UW group was planning on using such a simple method of bias correction (the 14-day running mean approach). It was stated by someone from the IUM group that a similar approach had also been tried, but that the proportion of forecast degradations was too high in tests and so the method was abandoned. The UW group responded that this issue was being actively scrutinized and that the proportion and magnitude of forecast degradations would be minimized in any operational implementation.
c. Wednesday, September 8: IUM, Meeting Room
Mr. Sharp began the day with a talk regarding the benefits and major issues of high-resolution prediction. The degree to which higher resolution improves weather forecasts was discussed using examples taken from the UW realtime modeling system that produces twice daily regional forecasts at 36, 12 and 4 km. Examples of ultra-high resolution simulations from research simulations of the Columbia Gorge, Stampede Gap and the Strait of Juan de Fuca were also shown. The improvement in structure with increasing resolution was clear, with sharper fronts and more realistic interaction of weather systems with the terrain. An example of prediction of a Puget Sound convergence zone was compelling evidence that higher resolution allowed prediction of mesoscale weather features that were not resolved by national operational models.
The difficulty of objectively verifying high-resolution simulations was illustrated. In particular, it was shown how the tighter structures produced by high-resolution runs result in greater penalty for timing and position errors than the same errors in a lower resolution forecast. Alternative approaches to verification were discussed including possible development of a feature based verification system.
The drawbacks of high-resolution forecasts were also discussed. Given the uncertainty of model initial conditions, the limitations of physical parameterizations at high resolution and imperfect model numerics, high-resolution may amplify errors, producing a poor forecast that appears to be very accurate. Apart from deceiving uneducated users, a few forecast failures can lead forecasters to reject high-resolution forecasts of extreme events that subsequently verify. It was suggested that medium resolution ensemble forecasts could be used to complement high-resolution deterministic forecasts to provide probabilistic and uncertainty information. A balance in the use of computational resources between high-resolution and ensemble forecasts probability is needed. The exact formulation of that balance is the subject of continued research at UW and some of this work was the subject of the next presentation by Mr. Grimit.
Mr. Grimit provided a comprehensive description of the UW mesoscale ensemble (UWME) system. The development of UWME was born out of several years of experience with high-resolution (4-km) MM5 forecasts. Although structures from higher-resolution forecasts appeared sharper and more accurate by eye, due to small spatial or temporal shifts compared to the verifying features, traditional statistical verifications showed only marginal improvement over lower resolution forecasts. It was also observed that large differences in the initializations of major operational forecast systems produce dramatic differences in the respective MM5 forecasts based on them. Thus, an ensemble system was created to quantify the mesoscale forecast uncertainty due solely to the differences in large-scale analyses and forecasts from major operational weather centers. The initial version of UWME contained only five members. Currently, UWME is composed of eight members run at both 0000 and 1200 UTC cycles over nested domains of 36- and 12-km grid spacing. The forecast duration is 48 hours.
Another eight-member ensemble system, called UWME+, attempts to account for forecast errors coming from the model as well as the initial uncertainty. Physics diversity is incorporated into the system by using multiple parameterizations for sub-grid physics, as they are available in MM5. Both land and water surface conditions are perturbed. Fixed parameters describing surface roughness, moisture availability, and albedo are varied as well. Ensemble verification shows that the UWME+ system is superior to the UWME, but both exhibit low dispersion and thus require statistical post-processing to create calibrated probabilistic guidance.
Mr. Steed discussed specific problems with MM5 physics. By running an operational mesoscale forecast model each day during the past several years, scientists at UW have gained significant insight into the limitations and advantages of many of the physical parameterizations in mesoscale models. In observing model behavior using different combinations of physics options in the MM5, several important issues have come to light. Precipitation microphysics in the model has been a source of significant model error in the NW US. By switching from the simple-ice scheme to the mixed-phase (Reisner II) scheme, cloud and precipitation forecasts appear to be much more realistic. Major field projects investigating microphysics have been spawned from these insights, including IMPROVE I and II, which included intensive observing periods to collect the information necessary to improve the microphysics in mesoscale models. Without the experience of examining many simulations, these important projects may not have happened.
PBL schemes in MM5 (and other models) are also very problematic. Forecasts of low-level parameters such as 2m temperature and 10m wind have significant problems, a large part of which are a result of poor PBL parameterizations. Another issue which has become apparent is the interaction of systematic errors in one physical scheme with errors from another. For example, the simple radiation scheme in the MM5 produces a large cold bias when run in combination with Reisner II microphysics. When switched to the CCM2 radiation scheme, the cold bias is dramatically improved. This bias in the simple radiation scheme was not apparent when run in combination with simple-ice microphysics. These sorts of insights can only be found through running many model simulations, and an operational forecast system run each day provides just this opportunity.
There was extensive discussion and interest on these topics. The UW use of convective parameterizations on higher resolution nests were of particular interest.
d. Thursday, September 9: IUM, Meeting Room
Mr. Steed provided an overview of the road weather prediction effort. The University of Washington and Washington State Department of Transportation have worked together for several years to improve weather information as it applies to surface transportation. Weather has an enormous impact on transportation in Washington. Many lives are lost in weather related accidents, and millions of dollars are spent to maintain the highways, especially during snow events in wintertime. To combat these problems, the UW has developed several web sites which provide high resolution weather observations and forecasts to the traveling public. Another area of work at the UW is developing improved forecasts for government highway officials. This application includes predictions of pavement surface temperature and weather from a mesoscale ensemble forecast system. The user interface developed for highway officials produces automatic warnings if threatening weather conditions are observed or predicted. Many lessons have been learned at UW through interactions with the Department of Transportation in the area of weather application development. The single greatest challenge ahead is developing methods to provide probabilistic forecasts for transportation officials. Using a mesoscale ensemble forecast system, this challenge should be met, but only with significant work developing probabilistic forecast techniques, improved ensemble techniques, and education of the end users.
There then followed a discussion of road weather prediction possibilities at IUM.
e. Friday, September 10: IUM, Meeting Room
On the final day of meetings it was IUM's turn to present some of their work to the UW scientists. Dr. Chao-Lin Zhang discussed the current configuration of the mesoscale forecast system at IUM and the future plans for NWP during the 2008 Olympic Games. The MM5 model is run at 45- and 15-km with one-way nesting over a 103x103 grid point area around Beijing. The forecasts are generated twice per day and run out to 36 hours. The physics options chosen include the Grell cumulus parameterization, simple ice microphysics, the Blackadar planetary boundary layer scheme, the 5-layer soil model, and Dudhia radiation accounting for interactions with cloud. Initial conditions and lateral boundary conditions are taken from the global T213 forecasts from the National Meteorological Center (NMC) at CMA.
The proposed system for Olympic Games weather support will be implemented by IUM and may use the Weather Research and Forecasting (WRF) model rather than MM5. IUM expects to implement triply-nested version with two-way interaction on 27-, 9-, and 3-km grids. There will be 37 vertical levels. The forecast duration will be anywhere from 12-24 hours, and will be updated with new runs every 3 hours. Physics parameterizations had not yet been chosen and IUM scientists seemed to be interested in the UW findings regarding specific schemes (as discussed above).
Next, Dr. Xiaoling Zhang discussed the air pollution prediction effort at IUM. Dr. Zhang described a simple multiple linear regression prediction method for various pollutants based on a past history of MM5 forecasts and air pollutant observations. Forecast variables included in the MLR include PBL depth, 850-hPa temperature, and relative humidity, surface wind, surface weather, and inversion presence.
Zaiwen Wang presented a post-processing methodology currently being researched at IUM. It uses a form of artificial intelligence called the Support Vector Method (SVM; developed in China). Mr. Wang presented evidence that SVM post-processing dramatically reduces forecast bias, but offered no comparisons with simpler competing approaches.
Dr. Min Chen presented her work on digital filter initialization of mesoscale model forecasts with MM5. The objective of Dr. Chen's work was to reduce or remove imbalances in the wind and mass fields to prevent false gravity and sound wave generation during the spin-up period. Dr. Chen incorporated digital filtering with a Dolph-Chebyshev filter as a weak constraint in 4DVAR aimed at minimizing the mean absolute surface pressure tendency during a specified time window in spin-up. The MM5 spin-up problems were drastically reduced using this methodology.
3. Cultural Experience
Possibly the most important part of the visit from our personal perspective was the cultural enrichment. Chinese culture is very different to that of the US or Europe and it was fruitful to be exposed to that culture as guests of an organization such as BMB. The Chinese are proud of their heritage and of their achievements and our hosts went to great lengths to expose us to as much of China as was possible in our short stay. We were taken to many of Beijing's major tourist attractions including the Summer Palace, the Temple of Heaven, the Forbidden City and the Great Wall.
We were treated to banquets and evening entertainment. The objective was clearly to express their friendship and impress us with the generosity being bestowed. As a result, we not only developed a good working relationship with our hosts but also individual friendships. The language barrier was frustrating at times, but since our hosts had a far greater grasp of English than we had of Chinese it was our obligation to be patient. Our attempts at basic Chinese were both appreciated and a source of great amusement!
We developed an insight into the Chinese way of doing things. For example, the importance that is placed on punctuality, the value of structural hierarchy and the negotiation that goes into nearly every purchase. Familiarity with the differences between our cultures helps foster understanding.
4. Concluding Remarks
Our visit to China was rewarding and exciting. It was also productive and built the foundation for future collaboration between IUM and the University of Washington. We informally agreed that future collaboration would be fruitful. A first step may be for the University of Washington to host a visiting scientist from IUM. This would allow for a continued exchange of ideas while allowing the Chinese scientist the opportunity to absorb US culture. Since IUM was particularly interested in assistance from the University of Washington in developing a road weather prediction system this might be a good project for an exchange scientist to work on.
We are very grateful to our hosts at IUM and BMB for making the trip not only a forum for the exchange of scientific ideas, but also an immersion into Chinese culture. We would also like to thank the National Science Foundation for funding our travel and acknowledge Bill Kuo for his important role in organizing the visit and his help as a cultural interface.
|
