Encyclopedia of Physical Science and Technology (R. A. Meyers, Ed.), Academic Press, Third Edition, Vol. 5., pp 565-581.
Modern paradigms for modeling and visualization of environmental systems
António M. Baptista
Oregon Graduate Institute of Science and Technology
- INTRODUCTION TO ENVIRONMENTAL OBSERVATION AND FORECASTING SYSTEMS
- A PILOT ESTUARINE ENVIRONMENTAL OBSERVATION AND FORECASTING SYSTEM
- OBSERVATION NETWORK
- MODELING INFRASTRUCTURE
- INFORMATION AND VISUALIZATION INFRASTRUCTURE
- APPLICATIONS
- PARADIGMS IN TRAINING AND EDUCATION
- OUTLOOK
GLOSSARY
Environmental Information Technology (EIT): ensemble of concepts and technologies
designed to deliver quantifiably reliable
environmental information at the right time and in the right form to the right users.
Environmental Observation and Forecasting Systems (EOFS): EIT tools that
combine real-time sensor measurements
with advanced numerical models
to describe with progressively increasing reliability and detail,
critical aspects of the dynamics of complex environmental systems.
Numerical modeling: the process of integration of numerical codes, system topology, forcings and quality controls towards the description of a physical or environmental process.
Circulation forecasts: predictive description of circulation processes into the future, through numerical modeling.
Circulation hindcasts: retrospective description of circulation processes, through numerical modeling
Visualization: the process leading to the development of a mental or graphical description of static or dynamic processes.
PARADIGMS IN MODELING AND VISUALIZATION OF ENVIRONMENTALSYSTEMS are shifting dramatically. Modeling is no longer perceived only as the rigorous but narrow process by which a set of differential equations and associated boundary conditions are transformed into systems of algebraic equations and solved by a computer. Visualization is no longer oriented primarily to specialists or restricted to a particular medium, and aesthetic sophistication is increasingly less important than contents and timeliness.
Modern modeling and visualization paradigms are being driven by the need to integrate quantifiably reliable scientific information in decision-making, in operational procedures and in cross-disciplinary research, as well as by the opportunities resulting from fast-evolving computer, sensor, and information technologies. Environmental Observation and Forecasting Systems (EOFS) provide an excellent illustration. In these systems, models produce critically important information contents, and visualization provides essential interfaces between information and multiple end users. But both modeling and visualization are integrated in an Environmental Information Technology (EIT) infrastructure, which ultimate goal is delivering the right information to the right user, in the right format, at the right time and with bounded uncertainty. This article discusses resulting benefits, requirements and unresolved challenges.
I.INTRODUCTION TO ENVIRONMENTAL OBSERVATION AND FORECASTING SYSTEMS
The concept of systematic, human-computed weather forecasting dates back to the start of the twentieth century, when British meteorologist Lewis Richardson [1] proved to be ahead of the technology of his time: it took him three months to predict weather for the next 24 hours. But the seed was planted, and with the post-World War II advent of electronic computers, systematic weather forecasting became a practical tool.
Arguably the original form of EOFS, weather forecasting is a pervasive commodity of modern life, influencing decisions as simple as an individual’s choice of the day’s clothing or as complex as the setting of regional energy strategies. In weather forecasting, time-sensitive data gathered from sensors or generated through computer models are integrated and interpreted towards a description of present and future atmospheric conditions. Established visual or audio interfaces routinely translate such description to both the general public and customized audiences such as farmers or mariners, leaving ample opportunity for the additional development of on-demand interfaces. Data and products are generated or flow through a worldwide network of information systems, managed by a disparate array of public and private entities.
Only in the later part of the twentieth century was the vision of systematic forecasting extended to other environmental systems. However, both society needs and advances in computer power and connectivity have in recent years fueled decisive advances in the development of real-time EOFS, in particular for ocean, coastal ocean, estuarine, and river systems ([2], [3], [4], [5], [6]). As they mature and self-organize, these EOFS are revolutionizing the way scientists share information about the environment and represent an unprecedented opportunity to break traditional information barriers between scientists and society at large.
EOFS have enormous potential for social pay-off, and, as with weather forecasting, success will lead to increased user expectations. As an example, the Marine Prediction Center of NOAA regularly generates forecast products for mariners, extending over large regions of the Pacific and Atlantic oceans, and describing combined atmospheric and sea conditions (e.g., Table 1). These products fulfill U.S. responsibilities with the World Meteorological Organization and Safety of Life at Sea Convention, and have had direct daily impact on safety at sea, protection of life and property, and enhancement of economic opportunity. Yet, such products are often no longer considered sufficient. Coordinators of marine search and rescue operations or of oil spill response teams, for instance, seek to have detailed, on-demand information on marine currents and winds wherever and whenever an accident might occur, aided as needed by real-time computer simulations to narrow the search region or to track the spill. The pay-off will be in additional lives saved, more effective ecosystem protection and in reduced operational costs. The necessary capabilities, including information quality control and uncertainty analysis, are within the technical reach of the next generation of EOFS.
Next-generation EOFS will also change the timing and subjectivity of regional environmental management decisions. For instance, decision-makers contemplating changes in hydropower management strategies in a regulated system like the Columbia River (e.g., Fig. 1), will expect at least some of the impacts of alternative scenarios to be compared quickly and on demand, in preparation for and during their deliberations. These impacts are currently investigated through typically lengthy and often controversial studies. If EOFS are implemented with a long-term perspective and appropriate regional consensus, the pay-off will be both in the timeliness of critical information, and in focus on what is indeed ambiguous or open to social choice. This focus is urgently needed: controversy is too often extended to aspects of the supporting science that are fundamentally unambiguous, preventing emphasis on the objective analysis of the relative social merits of possible alternatives.
To meet the evolving vision and functionality introduced above, next-generation EOFS will be large-scale, extendable, multi-purpose systems shared among many diverse applications, and will have unprecedented ability to respond quickly on demand, and to progressively accumulate knowledge from application to application. Many of these systems will be integrated in national or worldwide networks (e.g., Fig. 2). Continued advances in computational and visualization methods and in computer power and connectivity, complemented by new paradigms for research, training and information access, will be necessary to fulfill this vision.
II.A PILOT ESTUARINE ENVIRONMENTAL OBSERVATION AND FORECASTING SYSTEM
Estuaries and adjacent coasts are poster cases for the need for new paradigms and technologies for interdisciplinary research, education and information access. Indeed, these are interface ecosystems of great physical and ecological complexity, boasting strong variability at multiple scales in space-time. Estuaries are often the focus of intense and conflicting human activity: navigation, resource harvesting, effluent disposal, recreation, energy production, etc. They are also natural integrators of characteristics and impacts of adjoining watersheds, airsheds, and continental shelves. “Pristine” estuaries may be unexplored indicators of global climate change trends and their biological implications. Coastal plumes extend the influence of freshwater inputs beyond the mouth of the estuary. In the case of river-dominated systems, such as the Columbia River, this influence extends for hundreds of miles along the coast (Fig. 1).
Effective management of the nation’s estuaries will rely progressively more on real-time and prognostic information on physical characteristics, environmental stressors, and ecological indicators. EOFS are important enabling systems for this vision. Pilot implementations ([2], [3], [5]) have demonstrated promising ability to combine real-time continuous data monitoring and computer modeling into detailed spatial and temporal descriptions of past, present, and future estuarine conditions. However, pilot EOFS are often limited by factors such as over-specialization, design rigidity, insufficient assessment of information quality and uncertainty, sub-optimal monitoring, sub-optimal trade-offs in accuracy/cost of forecasts, insufficient storage capacity and high maintenance costs.
Intrigued by EOFS challenges and opportunities, the Center for Coastal and Land-Margin Research of the Oregon Graduate Institute initiated in 1996 the development of CORIE, a pilot system for the Columbia River estuary and plume. CORIE ([3], [7], [8]), was designed explicitly as multi-purpose, multi-user regional infrastructure for science and management, thus differing from many other estuarine EOFS, which have historically been driven primarily by navigation concerns ([2]).
At the heart of CORIE are three integrated components (Fig. 3): a real-time observation system (Section III), a modeling system (Section IV) and an information management and visualization system (Section V). These three systems combine to automatically generate an array of primary products, including time-sensitive displays of sensor data and of computer forecasts, which are often openly distributed through the World Wide Web (henceforth, web). Data from sensors and models are also processed in customized fashion, either on- or off-line, for particular scientific, management or operational applications (Section VI).
Modeling and visualization are integral aspects of CORIE, but special paradigms and requirements do apply. Modeling, observation and information management and visualization are not disjoint components, conducted in isolation by separate teams. Rather, all components are closely integrated, requiring tight interaction among team members with different primary expertise: modelers are often in the field, and field personnel participate in the analysis of modeling results.
Visualization is prevalent across the information domain, often as a mechanism for fusion of multiple data sources, and typically as the culmination of a time-sensitive, end-to-end flow of data from information producers (e.g., models, sensors) to a diverse range of human users. Visual sophistication is increasingly less important than contents, robustness, and timeliness. Quantification of uncertainty is highly desirable, but difficult to automate, thus being often unsatisfactory in its detail or lagging in time.
Modeling is more than the rigorous but narrow process by which a set of differential equations and associated boundary conditions are transformed into systems of algebraic equations and solved by a computer. Indeed, the modeling process often requires automated acquisition of external forcings, automated comparison against or fusion with field data, and code execution within pre-defined schedules. Models are typically required to run at least 5-10 times faster than real-time. While model results can always be examined in customized manners, an array of visual products is generated automatically as a part of default forecasting procedures.
III.OBSERVATION NETWORK
The CORIE observation network consists of multiple estuarine stations and one offshore station, each with a variable number of sensors. Most stations measure water levels, temperature and salinity, several stations measure velocity profiles, and a few measure wind. Sampling intervals are in the 1-15 minute range, with a sample consisting of the average over a short period (e.g., 30 s) of measurements made at several Hertz. Most stations have telemetry, typically based on spread-spectrum radio. Selected stations have field computers, enabling local storage and facilitating two-way communication.
By design, the configuration and instrumentation of the observational network has and will continue to evolve over time. While a backbone of anchor stations are considered permanent, other stations or sub-networks appear and disappear as dictated by scientific, engineering or management needs and by available funding. For instance, we are currently implementing two inter-related extensions of the network (Fig. 4), which became time-sensitive due to an unusually dry hydrological year and the potential for record low river discharges during the 2001 Spring freshet season. The first extension will enable the characterization of the vertical density structure in the two deep channels of the lower estuary, and is motivated primarily by the need for detailed validation of the CORIE circulation models. The second extension, concentrated on a lateral bay near the upstream end of salinity propagation, will provide anchor measurements of salinity, temperature, and pressure for third-party habitat surveys, in the context of investigations of the influence of the estuary on salmon survival.
Timely visualization of the data is an integral concept of CORIE, with deep implications for cost-effective system maintenance and for innovative data usage. For instance, the physical integrity of most stations, their power reserves, and the contents and quality of their data streams (e.g., Fig. 5) can be examined at least daily, enabling early detection of problems and their eventual resolution. Anyone, anywhere in the globe, can have free access to pre-determined representations of most CORIE data within seconds of their collection, providing that they have web access. Reasons to want such access will vary: a researcher in an oceanographic vessel may use velocity, backscatter and salinity data to track in real-time the progression of estuary fronts or other estuarine features; a bar pilot may use water levels or velocities to plan a docking operation; a fisherman may look at salinity and temperature data to choose the day’s fishing spot. Alternatively, users may choose to look at statistics and other forms of analysis of accumulated data. Timeliness is in this case measured not by time since collection, but by the time and/or effort required for obtaining the desired information in the desired form. An increasing number of CORIE products attempts to anticipate needs for access to processed data (Fig. 6).
Estuarine and offshore stations pose different logistical challenges, thus also eliciting different definitions of access timeliness. For instance, we are particularly concerned with the physical integrity of the offshore station, OGI01, which is often exposed to severe sea conditions during winter and is routinely exposed to ship traffic and seasonal fishing activities. Our telemetry strategy for this station consists of routine transmission, via satellite and at multiple times a day, of the location of the surface buoy where most instrumentation is mounted. Actual oceanographic datacan be transmitted in the same fashion, but because of cost and bandwidth limitations, telemetry of these data is typically limited to periods where quasi real-time access is operationally valuable – e.g., to assist scientific cruises. For estuarine stations, inexpensive telemetry solutions based on spread-spectrum radio enable routine real-time transmission and web-display of all the data being collected.
Associated with the notion of timely access is the challenge of timely quality control. A classic scientific paradigm is that of carefully extracting the most possible information from a limited data set, with modest consideration to the time that the analysis might take. CORIE and similar observation systems are adding a new paradigm: that of automated quality assessment of continuously flowing data streams from large arrays of sensors. We are finding that this challenge has to be addressed very differently for different sensors. Water levels, water temperature and velocity profiles tend to be resilient data streams, for which failure modes are often abrupt and susceptible of early detection by automated quality assessment methods. By contrast, conductivity sensors are commonly subject to biofouling, often leading to slow, progressive signal degradation that mimics aspects of low-frequency natural variability (Fig. 7). Without novel methods for early detection of biofouling, automated quality assessment of conductivity (and, by association, salinity) data will remain unsolved. Also unsolved is the problem of eventual correction of biofouled records. Future methods to detect and correct biofouled data may involve identification of changes in temporal patterns of modeling error at sensor locations.
IV.MODELING INFRASTRUCTURE
Our approach to CORIE modeling reflects the philosophy that numerical modeling must, in the context of EOFS, be an end-to-end task. Our main objective is, at this stage of the development of CORIE, to produce (a) daily forecasts of circulation in the Columbia River estuary and plume, and (b) archival simulations of circulation, in a slightly larger domain, for modern, pre-development, and future-scenario conditions.
Consistently with the above philosophy and objectives, we designed a computational infrastructure that includes five key components (Fig. 8).
A.Numerical codes
Numerical codes solve for the governing equations describing an environmental process, given an appropriate set of topological inputs, initial conditions and boundary conditions. All numerical codes used in the CORIE system are based on the solution of the 2D or 3D shallow water flow and/or transport equations, by either finite elements or finite volumes, on unstructured grids. Our design of the modeling system incorporates code redundancy as a mechanism of quality control, and similar purpose codes are conceptually interchangeable.
Table 2 identifies some of the main CORIE circulation codes, and highlights their differences. Because the modeling products include daily (and, in the future, real-time) forecasts, and multi-year hindcasts, computational performance is very important: simulations that run about 10 times faster than clock time are a desirable target. Table 3 illustrates the computational efficiency of the codes that are currently used for generation of 2D and 3D circulation forecasts, respectively, and shows a clear improvement of computational performance: the most recent 3D code runs substantially faster than an older 2D code, using the same computer hardware.