Modeling and simulation are familiar tools to members of the IEEE Oceanic Engineering Society (OES). For the past forty years or so, since computers began to be available at government laboratories and academic institutions, the idea of using formal, abstract mathematical models solved numerically in digital computers on a domain of parameters to study aspects of the ocean has been widely accepted. Over this period the technical use of the word model has changed from designating a physically scaled object in kind, such as a model ship or a model harbor, hydrodynamically scaled for tow tests or sedimentation studies, to now mean a code for numerically solving the abstract model. Such codes may even be named, and treated as trade brands conveying utility or validity to potential users. This shift in focus has had a profound effect on research, education, and the practice of engineering in the ocean, and the evidence so far suggests that the tools and practice are far from maturity.
These changes have coincided with the revolutionary advances of processor and memory technology, and the collapse of the costs of a computing cycle, as contrasted with the escalating costs of going to sea. A few years ago it was common to hear apparently sober technologists, enthusiasts of modeling and simulation, argue that the expenses of sea-based expeditionary ocean studies could be successfully avoided by implementing complex, interdisciplinary numerical models. These would be essentially cascades of purely disciplinary models, that is, a meteorological model might produce wind stresses and rainfall that would drive a stratified ocean circulation model, perhaps perturbed by tidal forces and with a detailed bathymetry. Perhaps the hydrodynamic modeling would be faithful enough that internal waves would emerge in the interior and solitons be generated in packets on the modeled continental slopes in ways numerically indistinguishable from careful observations. These numerical solutions would be virtual oceans created inside inexpensive computers that could be exercised in every way the real ocean would, and more. Such synthetic environments would allow repeatable, controlled experiments that could not otherwise be accomplished in a real ocean. Other models, acoustic propagation models or ship navigation models, or fishery models for example, could be coupled to these with the benefit of rarely needing to go to sea.
Technologists were not the only ones to be tempted by the promise of modeling and simulation. Models, by there very nature, are dispassionate and quantitative. They do not “care” about their outcomes, do not make subtle adjustments as they run in order to achieve a desired result, they are repeatable and can be audited to assure their results. To an outsider they simply codify a set of knowledge, crafted by experts, and they provide reliable quantitative predictions for hypothetical circumstances. As such they are perfect tools for policy and decision-makers. In a complex, democratic society there are many conflicting interests competing for finite resources. Many social and governmental institutions are organized around the principle that these conflicts should be adjudicated according to objective measures not dominated by any interest. Quantitative methods have been developed extensively over the past two hundred years to meet these needs, often led by the engineering profession. Computerized models and simulations are at the apex of this sort of public analysis, and are often seen as being even more fair and more objective than other techniques by virtue of the fact that they can be run without intervention.
Such benefits, obviously worthwhile, remain a goal for modeling and simulation supporting engineering in the ocean, but the sea has so far resisted being so easily understood. Unfortunately, as these synthetic oceans have been created and tried there is a growing realization that many of the formal approximations that were useful to achieve simple, instructive, closed-form or manual numeric solutions, have shortcomings when applied to more complete simulations. The ocean is exceedingly complex. There are often seams between the models that hinder a simple cascade. There are problems of scale, bandwidth, and sampling that only submit to careful restrictions of the domain. There is coupling in the real ocean that is often ignored at our peril, but is not understood well enough to accurately model.
As we have attempted to achieve greater fidelity in numerical models we have continually been confronted by our mistaken ideas about the ocean. Non- linearities inserted into numerical codes were discovered to create extreme sensitivity to initial conditions, even in dynamical system with low dimensionality, an outcome anticipated by few in 1960 or even 1990. This sensitivity turns out to be a physical property long observed, but largely unrecognized until recent work in non-linear dynamics. Likewise, coupling between terms caused by variable and underdetermined boundary conditions, anisotropy, and lateral inhomogeneities have bedeviled many attempts to reconcile observed conditions in the real ocean to the models and simulations meant to represent them. The message of these failures is unmistakable, we have not yet achieved the sophistication needed to create seamless synthetic oceans, and will not except through continued heroic effort. Models, so far, are not substitutes for the real thing.
To make this thought more clear, and as a prototype for thinking about modeling and simulation, consider the example from another field, the orrery. In the eighteenth century an Englishman, probably George Graham, invented a clockwork mechanism that emulates the timing and phase of a Copernican solar system. Each of the major planets and each of their major satellites are represented by a ball, and are driven by a set of gears that moves each planet around the fixed sun. In order to run the model and simulate the motion of the planets one literally turns a crank to move the planets around the sun. The timing of this motion may be quite precise, but the finite size of the orrery is obviously not to scale in either the size of the planets or their relative distances. The model enjoys the distinction that although a considerable expertise is required to design and build the model, anyone can turn the crank and simulate the motion of the solar system. Likewise, innovations in the motion of the planets, such as reversing their direction, can be studied with the model.
This device was once called a planetarium, but that name is now applied to a different modeling and simulation environment where the positions of stars and planets are projected as points of light onto a darkened interior dome. The mechanical model of the solar system is named after Charles Boyle, the fourth Earl of Orrery, the project’s patron.
The model makes explicit the idea that the universe is a clockwork. There is no place in the expression of the abstract, and far more useful, formal Newtonian model of universal gravitation where mass and distance account for the observed motions rather than gear ratios. Although the orrery may be validated against an ephemeris, it cannot be verified against the Newtonian model. There is no way to add another planet or perturb an orbit to study sensitivities to small deviations. The model is rigid.
This paper reviews the interests of OES in modeling and simulation as indicated by publications since 1976 and conference presentations made since 1996. The paper proposes a framework for discussing modeling and simulation that separates the elements from the specific technical area supported by the modeling and simulation, and raises a concern about the universal need for verification and validation. Finally, there is a discussion of the importance of modeling and simulation to national policy debate and decisions, which is rarely the subject of OES technical journals.
Modeling and simulation are at the heart of modern engineering practice. Whether an electrical engineer uses Ohms Law or SPICE in designing and analyzing a circuit for an undersea current meter, or a mechanical engineer applies Newton’s Laws of Motion and the Navier-Stokes equation while devising a control system for an autonomous underwater vehicle, or a radar engineer assumes Bragg scattering while seeking to understand the performance of a synthetic aperture radar from above the sea, the construction and application of formal models, sometimes through simulation, is a critical professional activity for engineers. Regardless of division by academic discipline engineers face complex problems that require objective, quantitative analyses to facilitate design and policy decisions. Models, which are formal analogs of phenomena and systems, are the fundamental entities that allow these analyses.
Among highly interdisciplinary activities, such as those represented by the Oceanic Engineering Society, the objects of modeling may be diverse indeed. All of the physical properties and constraints of the ocean and its behavior apply, as do the phenomena of energy propagation and the dynamics of objects suspended in the ocean. Instrumentation and measurement of ocean properties, both directly and by remote observation, telemetry and communication within the ocean, control of ocean vehicles, organization of data archives for retrieval, analysis and display are only a few topics to which modeling and simulation apply within the scope of interests of OES.
When we refer to models we now most often mean a formal, mathematical structure that relates a mapping from a set of circumstances onto a set of consequences or outcomes. The idea is that there are casual relationships that can be represented by formal mathematical constructions such as maps or functions. By imposing other formal conditions, which often seem very well justified, such as measures and other properties of vector spaces, formal models can be manipulated by formal rules. These manipulations, while not saying anything explicitly truthful about the phenomena being modeled, have proven surprisingly robust in practice. This quality is the basis for mathematical physics, applied mathematics, and much of the formal, academic engineering represented in technical publications and societies such as OES.
Formal systems that are used in modeling are usually mathematical. Algebra, differential equations, integral equations, probability, the calculus of variations, dimensional analysis, are all traditional formal systems.
Over the past few decades they have been joined by game theory, linear programming, Monte Carlo techniques, fuzzy logic, neural networks, synthetic annealing, genetic algorithms and non-linear dynamics. In fact theoretical advances in non linear differential equations have produced formal results that explain solitons, intermittency, bifurcations of solutions, and a host of other results that mathematics simply could not accommodate fifty years ago.
These advances have the effect of moving constraints that existed before. Without changing the underlying requirement of the problem, now opportunities for a better understanding become available simply through advances in the underlying formal structure.
Formal modeling, the act of deciding on and imposing a mathematical form that is meant to represent a physical event or phenomenon, is a creative act of the imagination. Some models have enjoyed a long run a success when used to explain or predict events, Newton’s Second Law of motion comes to mind as the prototype of the effective formal model that is repeatedly reused to solve physical problems involving both static and dynamic loads and their consequences.
Because building models is an act of discretion intangible elements are part of the ultimate success of the model. Education is obviously critical. If the modeler is unaware of formal results that may be applied the problem he will be unable to exploit them. Experience is similarly critical. The nature of the phenomena being modeled, its domain and range and behavior are all clues to which models will be most effective. Beyond these there are matters of taste and talent. Because we like to think of models as objective conveyers of collective knowledge the role of intangibles such as these are seldom mentioned, yet the evidence of models suggests that it is not an activity done equally well by all practitioners. Despite the fact that some modelers may contribute more than others, modeling and simulation remains a field that is advanced by consensus. Models, no matter how brilliant, that fail to be adopted, simply fail. The formal model is the communication of an idea that has the property of being verifiable and testable. It predicts an outcome based on a circumstance.
Formal models may have an impact on knowledge by being run in two ways, the inverse and the forward problem. In the inverse problem a model is proposed with some form and some perhaps unknown parameters. Data is collected and the parameters are estimated to be most consistent with the data. The properties of the solution, as well as the most appropriate methods to find a solution, depend on the nature of the data themselves. This area of modeling and model application is well studied in some fields such as geophysical prospecting, medical and ocean tomography, for example. In its simplest form it is merely curve fitting, but it is also the basis for filling data bases and seeking automatic, adaptive systems and solutions.
Returning briefly to celestial mechanics for a clarifying example, the inverse problem is Kepler fitting ellipses under a Copernican, heliocentric solar system to Tyco Brahe’s observational data of planetary motion. Once in place the orbital parameters were solved allowing the model to be run forward to predict the future positions of the planets with respect to the fixed stars. One might say that inverse problems are fundamental to science, as we know it. The power of such models can be seen by noting the consequences of Newton’s question: “why ellipses?”
The forward problem is solved to find a value corresponding to an input. In this way the model is simply a function or a mapping of the domain onto a range. A single trajectory or sequence of points in the range that are a consequence of running the model iteratively on an independent variable, such as time, or according to contingent conditions, either stochastic or by the intervention of an outside agent, is what we mean by simulation.
Simulations are especially useful ways to explore the possibilities of models when the model is well known. This allows training simulators, hardware-in-the-loop stimulators, mission planning and rehearsal, and many other useful applications of models. Simulations may even exhibit emergent behaviors, that is show effects that were not explicitly programmed, for example in non-linear models, or cellular automata. In such cases it is insufficient to validate only the underlying models and databases of parameters, but the simulation’s dynamics must also be examined.
Modeling and simulation are well represented in the publications and activities of the OES. A survey of the INSPEC data base over the past thirty years or so, searching on the key phrase: (modeling or simulation) and ocean, reveals an exponential growth in publications from five in 1968 and two in 1969, to three-hundred-sixty-six in 1998, roughly doubling every two to five years. A similar search on IEEE Journal of Oceanic Engineering and (modeling or simulation) yielded 157 records the earliest of which were in 1976, J-OE volume 1, number 2. The growth in J-OE papers, while not as dramatic as in the literature as a whole, still reflects a steady increase: ten papers in the seventies; thirty-seven papers in the eighties; and one-hundred-four papers in the nineties. As is fitting for tools that are so integrated into practice, almost all of these papers focus on analysis or design topics related to a particular engineering problem or area of technical application, rather than issues common to modeling and simulation. An informal survey of the J-OE papers of the nineties selected in the INSPEC search suggests that signal processing, autonomous vehicle and ship control and propulsion, and ocean acoustics are the technical areas of interest that now receive the most attention in the J-OE.
An attempt to validate the impression created by the INSPEC search by actually inspecting the papers published in J-OE over the past three years was revealing, if not surprising. I failed to find a single paper that did not include an abstract, formal model of the matter under study. Formal modeling is central to the professional activities chronicled by this Society, and along with the unifying theme of the ocean, is the common element among the technical discourse of the Society.
In addition to the J-OE, OES sponsors an annual conference called OCEANS, often jointly with the Marine Technology Society. These conferences bring together many with common technical and policy issues around the ocean, and feature short technical presentations of papers on topics of current interest. Like J-OE, a survey of these conferences reinforces the theme that modeling and simulation are ubiquitous tools in practice. There are, however, special sessions regularly devoted to topics of interest to the M&S craft. Emerging issues such as data base organization, for example according to the principles of Geographical Information Systems, data visualization and display as ways to communicate numerical results to decision makers, the impact of new computing or data architectures, open standards, etc. In addition, there are papers on applications of new techniques, such as cellular automata, or genetic algorithms.
There is, however, little discussion of the impact of modeling and simulation as practiced by ocean science and engineering professionals on the quality of modeling and simulation practiced by public policy decision- makers. Although few are in a better position to examine and comment these uses of technical tools, OES has not engaged this aspect of M&S as a Society. Whether we should is an open question.
An Example of Modeling and Simulation in Public Policy: The U.S. Department of Defense Modeling and Simulation Master Plan.
On January 4, 1994 the Department of Defense issued a directive, number 5000.59 entitled “DoD Modeling and Simulation (M&S) Management.” This directive superseded the M&S management plan dated June 21, 1991, and a former directive establishing a Defense Modeling and Simulation Office (DMSO), while establishing a new Executive Council, and reestablishing DMSO. The directive established DoD policy that:
Investments shall promote the enhancements of DoD M&S technologies in support of operational needs and the acquisition process; develop common tools, methodologies, and databases; and establish standards and protocols promoting the internetting, data exchange, open system architecture, and software reusability of M&S applications. Those standards shall be consistent with current national, Federal, DoD- wide and, where practicable, international standards for open systems.
Furthermore, the Executive Council’s published vision for DoD M&S states that:
Defense modeling and simulation will provide readily available, operationally valid environments for use by DoD components:
- To train jointly, develop doctrine and tactics, formulate operational plans, and assess warfighting situations
- To support technology assessment, system upgrade, prototype and full-scale development, and force structuring.
Furthermore, common use of these environments will promote a closer interaction between the operations and acquisition communities -in carrying out their respective responsibilities.
The Navy, as a DoD component responsible for the representing the Ocean as well as naval interests, provides an additional vision:
In the 21st century, the United States Navy will use Modeling and Simulation to make better analytical decisions, improve warfighting skills, and develop superior systems to maintain the world’s most powerful maritime forces for the joint force commanders. Analysts will construct force structures; warfighters will train and prepare for war; and system designers and engineers will develop new systems and platforms, all through modeling and simulation in a synthetic battlespace credibly replicating the real world.
It all sounds so easy when they say it. Apparently there is little room left for hubris in the Pentagon.
These ambitions are built on the notion that this approach will save vast sums of money by avoiding the need to develop and test systems in the actual environment where they are to be used in situations that mean not only life or death for sailors, but perhaps life or death for a nation or an alliance. Does the state of the practice of modeling and simulation of the ocean environment, and systems that operate in the ocean justify this approach?
This question does not only apply to the U.S. Navy, but to all other policy makers with responsibilities that include environmental and climate health, public safety, fisheries management, and a host of other issues. Whether or not they have had the foresight to write down what they need from models of the ocean, the implications are clear for those who fund technological development and for technologists. We are going to need many verified, validated, correct models of the ocean to meet not only our own needs as we design and analyze systems for the ocean, but also high quality models and simulations to support serious policy questions. We must be cautious to not let important decisions be made on the basis of an Ocean Orrery, or if we do, to be alert to the consequencees.
Modeling and simulation are powerful tools for understanding and explaining the world around us. When used with discretion, modeling is a way of organizing knowledge accumulated through observation or deduced from underlying principles. Modeling is a way of enforcing constraints on otherwise wishful thinking. It is a powerful tool for explaining the physical, economic and practical world, and a synthetic measurement process that allows us to explore conditions that may otherwise be inaccessible. Modeling is a creative act of imagination, serving as a collective memory for organizing and storing what we know about physics, geometry and dynamics. Because modeling is a creative act it must be tempered by the discipline of verification for internal consistency, as well as validated against independent observation. Only when models replicate, in some way, outcomes in the real-world corresponding to similar entering conditions can we say the models are satisfactory.
When validated the model provides a synthetic, surrogate world that can be systematically explored without the inconvenience or risks normally associated with exploratory expeditions, although it is not clear that this approach actually saves money.
There is a core of work that is purely technical, and there is also an element of our work that is applied to decisions that have ramifications beyond the immediate technical concerns. These are essentially the decisions of when, whether and to what degree to commit public and private resources to a course of action. To participate in these decisions implies that we are able to coax special insights from our special understanding of - not only the pertinent physical phenomena but also their complex interaction and way they fit into a larger context. To do so is to make a bold claim about the nature of our knowledge - and to be modest regarding how much we really know.
1. Porter, T.M., Trust in numbers : the pursuit of objectivity in science and public life. 1995, Princeton, N.J.: Princeton University Press. xiv, 310.
2. Lorenz, E.N., The essence of chaos. The Jessie and John Danz lectures. 1993, Seattle: University of Washington Press. xii, 227.
3. Orrery, Brittanica Online, http://www.eb.com:180/cgi-bin/g?DocF=micro/442/4.html Access date: 22 June 1999.
4. Menke, W., Geophysical data analysis: discrete inverse theory. Rev. ed. International geophysics series ; v. 45. 1989, San Diego: Academic Press. xii, 289.
5. Johannes Kepler, Brittanica Online, http://www.eb.com:180/cgi-bin/g?DocF=macro/5003/52.html Access date: 22 June 1999.
6. INSPEC, 1997, SilverPlatter International N.V. www.lib.washington.edu/silverplatter/webspirs/bin/ Access Date: 22 June 1999.
7. IEEE Journal of Oceanic Engineering, 1996-1999. 21-24.
8. Marine Technology Society., American Society of Civil Engineers., and Institute of Electrical and Electronic Engineers., Oceans ‘97 MTS/IEEE: conference proceedings: 6-9 October 1997, World Trade and Convention Centre, Halifax, Nova Scotia, Canada. 1996, [Fort Lauderdale, Fla.]: Oceans ‘96 MTS/IEEE Conference Committee. 2 v.
9. Marine Technology Society., Oceanic Engineering Society (U.S.), and Institute of Electrical and Electronics Engineers., Oceans 96 MTS/IEEE: coastal ocean, prospects for the 21st century: conference proceedings, 23-26 September 1996, Broward County Convention Center, Fort Lauderdale, Florida. 1996, [Washington, DC?] Piscataway, NJ: Oceans ‘96 MTS/ IEEE Conference Committee; IEEE Service Center [distributor]. 4 v.
10. Oceanic Engineering Society (U.S.), Oceans ‘98: conference proceedings: 28 September-1 October, 1998, Nice, France, Acropoils Convention Center. 1998, Piscataway, NJ: Oceans ‘98 IEEE/OES Conference Organizing Committee. 3 v. (xxxi, 1853).
11. Defense, D.o., Department of Defense Directive 5000.59: DoD Modeling and Simulation (M&S) Management, 1994, Department of Defense. www.dmso.mil/docslib/mspolicy/directive.html. Access Date: 22 June 1999.
12. Office, N.M.a.S., NAVY Modeling and Simulation Master Plan, 1997. navmsmo.hq.navy.mil/policy/plans/navy/es.htm. Access Date: 22 June 1999.
E. C. Gough Jr. is a Principal Engineer at the Applied Physics Laboratory, The University of Washington where he leads research projects related to sonar and the ocean environment.
Mr. Gough has been at the University of Washington since 1989, when he joined as head of the Signals and Systems Department, and later served as head of the Environmental and Tactical Systems Department. There he led research projects in Arctic environmental acoustics, including studies of propagation and ambient noise. During this time Mr. Gough advised Submarine Development Squadron Twelve on applications of oceanography and environmental acoustics to anti submarine tactics. The department also conducted research and exploratory development in non-linear signal processing, time series analysis, environmental acoustics, and tactical oceanography.
From October 1996 through September 1998 Mr. Gough served as Technical Advisor to OPNAV N84, the ASW Requirements and Assessments Division of the Chief of Naval Operations Staff. He held this position as an IPA, on assignment from the University of Washington Applied Physics Laboratory.
From September 1993 through August 1995 Mr. Gough was Science Advisor to the Commander, U.S. Navy Sixth Fleet in the Mediterranean Sea, where he was responsible for developing and resolving the list of Command Technology Issues. While there his effort focused on C4I issues related to the Commander, Joint Task Force Afloat role of the flagship. The Navy Science Assistance Program of the Office of Naval Research sponsored this assignment.
Prior to joining the University of Washington, Mr. Gough was a Principal Scientist at Planning Systems Incorporated for twelve years. While there Mr. Gough led research projects, including field efforts, in Arctic environmental acoustics, low-frequency acoustics, signal processing and sonar engineering. From 1974 through 1976 Mr. Gough was an Engineer at Sperry Marine Systems in the Advanced Products Group.
Mr. Gough earned a Master of Applied Mathematics degree from the University of Virginia in 1977 and a B.Sc. Ocean Engineering degree, with Honors, from Florida Atlantic University in 1973. He was a cooperative education student at Woods Hole Oceanographic Institution in 1972 and 1973, and at the Smithsonian Institution’s Harbor Branch facility in 1971.
Mr. Gough’s honors include the Superior Civilian Service medal for his work at OPNAV N84, awarded by the Chief of Naval Operations in 1999. Mr. Gough is Chair of the IEEE Oceanic Engineering Society, and a member of the AdCom. He is also a member of the Society of Industrial and Applied Mathematicians.