ARTICLE IN PRESS

0098-3004/$ - se

doi:10.1016/j.ca

$Code avail

CGEditor/inde�CorrespondE-mail addr

Computers & Geosciences 34 (2008) 247–258

www.elsevier.com/locate/cageo

An educational interactive numerical modelof the Chesapeake Bay$

Jessica R. Croucha,�, Yuzhong Shenb, Jay A. Austinc, Michael S. Dinnimand

aDepartment of Computer Science, Old Dominion University, Norfolk, VA 23529, USAbDepartment of Electrical and Computer Engineering, Old Dominion University, Norfolk, VA 23529, USA

cLarge Lakes Observatory, University of Minnesota, Duluth, MN 55812, USAdCenter for Coastal and Physical Oceanography, Old Dominion University, Norfolk, VA 23529, USA

Received 24 July 2006; received in revised form 23 March 2007; accepted 26 March 2007

Abstract

Scientists use sophisticated numerical models to study ocean circulation and other physical systems, but the complex

nature of such simulation software generally make them inaccessible to non-expert users. In principle, however, numerical

models represent an ideal teaching tool, allowing users to model the response of a complex system to changing conditions.

We have designed an interactive simulation program that allows a casual user to control the forcing conditions applied to a

numerical ocean circulation model using a graphical user interface, and to observe the results in real-time. This program is

implemented using the Regional Ocean Modeling System (ROMS) applied to the Chesapeake Bay. Portions of ROMS

were modified to facilitate user interaction, and the user interface and visualization capabilities represent new software

development. The result is an interactive simulation of the Chesapeake Bay environment that allows a user to control wind

speed and direction along with the rate of flow from the rivers that feed the bay. The simulation provides a variety of

visualizations of the response of the system, including water height, velocity, and salinity across horizontal and vertical

planes.

r 2007 Elsevier Ltd. All rights reserved.

Keywords: Modeling; Simulation; Visualization; Oceanography; Education; Interactive; Chesapeake Bay

1. Introduction

Geoscientists use numerical models to study avariety of complex physical systems such as oceancirculation, weather development, and landformevolution. Researchers investigate scenarios withthese models by specifying boundary conditions

e front matter r 2007 Elsevier Ltd. All rights reserved

geo.2007.03.017

able from server at http://www.iamg.org/

x.htm

ing author. Tel.: +1757 683 6001.

ess: jrcrouch@cs.odu.edu (J.R. Crouch).

that describe the external influences on a system andcompute predicted values for quantities of interestbased on the model’s equations. By carefullycontrolling the forcing, an investigator can isolatethe role of specific forcing parameters. Numericalsimulations therefore enable scientists to makeforecasts and perform virtual experiments that areimpossible to perform in reality.

The ability to investigate the role of specificforcing mechanisms should, in principle, makenumerical models ideal teaching tools, allowingeither an instructor to demonstrate a particular

.

ARTICLE IN PRESSJ.R. Crouch et al. / Computers & Geosciences 34 (2008) 247–258248

phenomenon, or students to engage in activelearning by using the model themselves. However,three major factors have kept these models out ofthe classroom. First, the complexity of thesemodels typically puts them out of the reachof all but a small handful of researchers whohave devoted large amounts of effort to under-standing the code and developing visualizationroutines for the output. Second, models aretypically run in an asynchronous fashion, withthe investigator first setting up the conditions forthe model, then running the model, and finallyvisualizing and interpreting the results. Thiscan make it difficult for students to generatean ‘‘intuition’’ about the behavior of a system.Finally, until recently, these models have requiredcomputing resources unavailable in a typicalclassroom.

While the expertise of professional scientists isneeded to design and implement numerical models,members of the larger community can improve theirunderstanding of the environment through interac-tion with simulation software. In communitiesaround the Chesapeake Bay, in particular, there isbroad interest in human impacts on the health ofthe bay’s complex ecosystem (Boesch et al., 2001;Officer et al., 1984). The environmental issuessurrounding the Chesapeake Bay make it anespecially important topic for public educationand discussion for ecological, economic, and poli-tical reasons.

The Chesapeake Bay simulation program pre-sented here is the product of a pilot project intendedto demonstrate the feasibility of making thecapabilities of the Regional Ocean Modeling System(ROMS), one of the ocean circulation researchcodes, accessible to non-expert members of thecommunity. The goal was to create a simulationprogram that would be both scientifically sophisti-cated and user-friendly. The design and implemen-tation of this program involved re-engineeringROMS to run interactively instead of in batchmode, creating a new user-friendly graphical userinterface, and integrating real-time three-dimen-sional data visualization routines. The resultingsimulation allows a user to control wind speed anddirection along with the rate of flow from the riversthat feed the bay. The program responds tochanging conditions, takes tidal effects into ac-count, and provides visualizations of water height,velocity, and salinity across the bay’s surface and incross-section.

The remainder of this paper is organized asfollows. Section 2 surveys previous work in educa-tion-oriented geoscience simulation, and Section 3describes the software architecture of the Chesa-peake Bay simulation program. Section 4 presentscase studies of Chesapeake Bay phenomena that canbe simulated, and Section 5 concludes this paperand discusses possibilities for extending the currentproof-of-concept program.

2. Previous work

Science education researchers report that compu-ter simulations with manipulable models canfoster inquiry-based learning (Windschitl, 2000).Such programs allow students to ask questions,form hypotheses, and perform experiments in asimulated environment. This activity facilitatesexploration and discovery, allowing students topractice the scientific method instead of merelyreading or hearing about the scientific work ofothers. Although not a replacement for laboratoryor field exercises, simulations allow supplementalexperiments to be performed quickly, requiringless equipment and student supervision thantraditional experiments. Topics like Newtonianmechanics lend themselves to simple equation-based simulation, and a wide variety of sucheducation-oriented simulation programs are avail-able (de Jong et al., 1999; Dede et al., 1996; Hartel,2000). Simulations for more complex phenomenacan be equally valuable but more difficult toimplement.

In the realm of geoscience education, a variety ofinstructional software is available, but reports of theeducational use of numerical simulation programsare relatively rare. Geoscience software that hasbeen effectively used for instruction includes role-playing games in which students play the part of ascientist investigating a virtual world (Saini-Eidukatet al., 2002; Renshaw and Taylor, 2000), and dataanalysis and visualization programs that allow usersto explore large sets of collected or simulated data.For example, Ramasundaram et al. (2005) pre-sented software that lets users interact with watertable data collected over a six year period in a 42-haflatwood forest. More recently, Winn et al. (2006)presented software that enabled students to visua-lize water movement and salinity data computed forPuget Sound using the Princeton Ocean Model(Blumberg and Mellor, 1987). This instructionalprogram allows exploration of the saved results of a

ARTICLE IN PRESSJ.R. Crouch et al. / Computers & Geosciences 34 (2008) 247–258 249

simulation, and lets users select one of two data filesthat were generated using different forcing condi-tions. The program does not allow a user to specifynew forcing conditions and compute new simulationresults. Similarly, the EdGCM project (Chandleret al., 2003) provides a graphical user interface forvisualizing data computed by a global climatemodel, but the program does not provide userswith an ability to change forcing conditions andcompute the impact of such changes.

The distinction between interactive exploration ofrecorded data and interactive numerical simulationis an important one. Both can be interesting andeducationally valuable, but in the first case thescenarios that a user can investigate are strictlylimited to the simulations that were recorded by thesoftware’s authors. In contrast, a simulation pro-gram that gives the user control over model forcingvariables and then computes the effects of thechosen settings provides access to a virtually limit-less number of scenarios. A user who is givencontrol over the forcing conditions can engage ingenuine scientific inquiry by asking new questions,forming hypotheses, and running simulationexperiments.

Few interactive numerical simulations exist thatare appropriate for non-expert users. Publishedreports of educational simulation software indicatethat the models employed tend to be based on lesscomplex governing equations than those in re-search-oriented simulations. An example is theclimate model presented by Bice (2001). Thisteaching model represents the earth’s climate usingtwo reservoirs of thermal energy: one for theatmosphere and one for the earth’s surface. Thismodel simulates gross climate change at a high levelof abstraction, but does not represent the spatialdistribution of temperature. Another example of aneducational simulation is WILSIM (Luo et al.,2004), a software program that models landformevolution due to the erosion, diffusion, and deposi-tion of soil over time. WILSIM uses a cellularautomata algorithm to represent the movement ofwater and soil across a virtual landscape. Thecellular automata model succeeds in simulatinglarge scale effects, but has some limitations com-pared to the more computationally intensiveapproach of solving the hydrodynamic equations.The authors report the simulation ‘‘is based on‘very simple approximations intended to capture thesynoptic effects of fluvial processes’ (Chase, 1992).’’The authors further warn that the simplifications

lead to incorrect model behavior such as ‘‘sedimentaccumulation at the bottom of the channels inrugged terrains, which is contrary to real worldsituations.’’ In general, model simplificationsease implementation and reduce computationalpower requirements, but also limit a simulation’sreliability.

An example of an equation-based, interactive,and education-oriented geoscience simulationprogram is the sea wave forecasting programby Whitford (2002a, b). This program allows auser to control model variables such as windspeed and duration and computes new simulationresults for user-specified settings. The programconsists of a suite of MATLAB functions, so somefacility using the MATLAB command line environ-ment is required in order to use this program.While this is a reasonable expectation for theuniversity science students who form the program’sintended user base, it does preclude use by morecasual users and anyone who lacks a MATLABinstallation.

In summary, most education-oriented geosciencesimulation programs use relatively simple modelspartially because implementation of complex re-search grade numerical models requires an unrea-sonable amount of development effort foreducational software. The alternative approachtaken by some authors has been to generate datafiles using a research-oriented numerical model andto provide students with a data visualizationprogram that allows them to interactively explorethe saved simulation results. The available educa-tional simulation programs therefore tend to workinteractively with a simplified model or be limited bystatic data files. The Chesapeake Bay simulationprogram presented in this paper was developedusing a different approach. Code for a graphicaluser interface and visualization functions wasintegrated with the code for a research grade oceancirculation model. This produced a user-friendlyprogram that allows users to set forcing conditionsand interactively compute new simulation results byengaging a sophisticated numerical model. One ofthe main objectives of this project was to find abalance between spatial/temporal resolution andexecution speed. While the system is not resolved atthe resolution that a research-quality implementa-tion of the Chesapeake might be, it still captures theessential character of estuarine circulation and runsquickly enough that it can retain the interest of acasual user.

ARTICLE IN PRESSJ.R. Crouch et al. / Computers & Geosciences 34 (2008) 247–258250

3. Simulation architecture and implementation

The three functional units that compose theChesapeake Bay simulation program are the gra-phical user interface, the numerical ocean circula-tion modeling code, and the visualization routines.The data flow between these is illustrated in Fig. 1.A description of the Chesapeake Bay is provided asinput data to the simulation. A description of eachof the simulation components follows.

3.1. Regional Ocean Modeling System (ROMS)

ROMS is an ocean circulation modeling programcreated and supported by members of the oceano-graphic research community (Shchepetkin andMcWilliams, 2005). The source code for ROMSversion 2.1 was the starting point for the develop-ment of the Chesapeake Bay simulation program.This research grade circulation model was selectedas the computational engine of the simulationbecause it supports a wide variety of circulationmodel properties. A subset of ROMS capabilities isaccessible in the first version of the interactiveChesapeake Bay simulation, but the expansion ofthis feature set will be straightforward since ROMSis already capable of modeling properties such asoxygen concentrations and pollutant distributions.

ROMS uses the hydrostatic primitive equationsto model ocean circulation. These equations are asimplification of the Reynolds-averaged Navier–

Fig. 1. Software architecture of interactive

Stokes equations and describe fluid flow on asphere, under the assumption that the depth of thefluid layer is much smaller than the sphere’s radius.ROMS uses a finite difference type of approach tosolve the hydrostatic primitive equations on a griddefined by terrain following coordinates calledsigma coordinates. This form of grid allows thethe topography of the ocean floor and coast to bedirectly incorporated in a model. As input, ROMSrequires parameters for the grid and a set of forcingconditions. Forcing conditions can include time andlocation dependent functions for variables such aswind speed and direction, atmospheric temperature,precipitation, river flows, tides, and pollutionsources. During the solution phase ROMS com-putes distributions for these and other variablesthroughout the three-dimensional volume of amodel at each time step.

ROMS is written in Fortran 90 and consists ofmore than 340,000 lines of source code. In order toproduce compact, efficient executables from a largecode base, ROMS makes extensive use of compilerdirectives to exclude portions of the code that areunnecessary for specific types of input models. Thiscode structure makes editing and recompilation ofthe code necessary for each model variation.Furthermore, ROMS is designed to run in batchmode, managing all model input and solutionoutput through files. Interactive adjustment tosimulation variables is not supported. Visualizationsof output data, if needed, are generally created by

Chesapeake Bay simulation system.

ARTICLE IN PRESSJ.R. Crouch et al. / Computers & Geosciences 34 (2008) 247–258 251

researchers during a post-processing step usingsecondary software such as MATLAB. Therefore,ROMS provides researchers with low-level controland great modeling flexibility but lacks the user-friendly facilities needed in software intended fornovice users. Indeed, graduate level oceanographersoften spend weeks to months learning to useROMS. The requisite level of user sophisticationprecludes ROMS use by most students and casualusers.

Modifications to ROMS were necessary in orderto make its powerful numerical modeling capabil-ities more accessible. The outermost loop in theROMS source code advances the simulation timeand calls routines to set up and solve equations forthe current time step. The primary modification toROMS was the removal of this time-stepping loopand the addition of new top level functions tomanage simulation time increments, update theforcing conditions at the beginning of each update,and pass the computed solution data to thevisualization routines at the end of each update.The time step size used by ROMS cannot becontrolled directly by a user because it impacts thenumerical stability of the solution routines. How-ever, ROMS can integrate the results from multipletime steps so that effects at different time scales canbe examined. User changes to the simulation updaterate are implemented as changes to the integrationperiod. Originally written as a standalone execu-table program, the modified ROMS code wascompiled into a linkable library called iROMS(interactive ROMS).

3.2. Chesapeake Bay model

The iROMS code used for the Chesapeake Baymodel was configured by enabling the algorithmsfor computing salinity levels, tracking floatingobjects, using a K-profile parameterization forvertical mixing (Large et al., 1994), controlling tidesat the open boundary, and managing point sourcesof momentum, temperature, and salinity. A com-putational grid was defined to represent the entireChesapeake Bay, stretching approximately200miles from Havre de Grace, Maryland, to CapeHenry, Virginia. Portions of several rivers feedingthe bay are represented in the grid, including theSusquehanna, Potomac, James, Rappahannock,and York Rivers.

The dimensions of a computational grid are animportant factor in the computational complexity

of the iROMS solution algorithm, so the desiredsimulation update rate influenced the design of agrid for the Chesapeake Bay model. The goal for theChesapeake Bay grid was to be small enough thatthe simulation would complete a solution step inless than a second while running on a PC and yet belarge enough to demonstrate interesting effects inthe bay. The chosen grid dimensions are 20� 50�10 (east–west � north–south � vertical layers).Latitude, longitude, and altitude coordinates arestored for each grid vertex. Research versions of theChesapeake Bay model typically have higher gridresolution and require offline computation that cantake hours to complete on a cluster computer. Withfuture advances in computational speed, the gridresolution for the Chesapeake Bay could easily beincreased. User-adjustable grid resolution is part ofplanned future work. Adjustable grid resolutionwould allow students to examine how the computedcirculation features vary with grid size and how therequired computation time is impacted by theresolution.

3.3. Graphical user interface

A user interface was implemented in Cþþ usingthe Fast Light Toolkit (fltk), an open source crossplatform library. The user interface code containsthe program’s main event loop that controls thesimulation’s execution and flow of data at thehighest level. Whenever a user changes the setting ofone of the on-screen controls, the event loopreceives notification of the change and updates theparameters passed to iROMS at the beginning ofthe next solution step. Calls to iROMS areimplemented by spawning new threads. The multi-threaded nature of the program enables the userinterface to provide smooth interactivity withoutblocking while waiting for iROMS to finishcomputing a solution step. Another benefit of themulti-threaded implementation is that the simula-tion can take advantage of a dual processor PC bycontinuously running iROMS calculations on oneprocessor while using the other processor forinterface and graphics functions.

The interface provides two categories of controls:those that affect the numerical model and those thataffect the visualization of computed results. Con-trols affecting the numerical model include aselection box for the update rate, sliders to set riverflow rates, and a slider and dial for setting the windspeed and direction. The interface displays the

ARTICLE IN PRESSJ.R. Crouch et al. / Computers & Geosciences 34 (2008) 247–258252

simulation time, updating it after each solution timestep. An update frequency in the range of 1–24 h canbe selected, influencing the degree of variability dueto tidal effects a user will notice. The rate of flowcan be set for the three largest rivers feeding the bay:the Susquehanna, Potomac, and James Rivers. Forthis simulation the wind is assumed to be constantacross the surface of the bay, and users may select asingle wind speed and direction.

Controls affecting the visualization of the com-puted results include check boxes to select aparticular variable for visualization and a slider toselect the depth of the grid layer to visualize. Eitherthe water’s salinity, velocity, height level, or theposition of floating markers can be selected forvisualization. Since the computed results are three-dimensional, plotting data on a map of the bayrequires a two-dimensional subset of the data beselected. The depth slider allows users to select anyof the grid’s 10 vertical layers for visualization. Theuser interface is shown in Fig. 2.

3.4. Visualization routines

Visualization routines were written using Open-SceneGraph, an open source cross platform gra-

Fig. 2. Interface for Chesapeake

phics library. Plan view and depth view windowsprovide orthogonal visualizations of the bay. Theplan view displays a map of the bay and surround-ing coastal areas, complete with rivers and nearbycities. The latitude/longitude coordinates associatedwith computational grid vertices correspond toðx; yÞ points on the map plot. Data values areassigned to map locations based on these coordi-nates and are linearly interpolated between vertices.The following types of visualizations can be selectedfor the plan view.

Water height is a scalar quantity represented byapplying color mapping to the bay. With a 1 hupdate rate, tidal variations in water height areevident using this visualization. The water heightvisualization is displayed in Fig. 3(a).

Salinity is another scalar quantity represented viacolor mapping. Variations in fresh water flowfrom the rivers feeding the bay cause pronouncedchanges in salinity. Salinity visualizations are shownin Fig. 6 for the example simulation described inSection 4.3.

Velocity vectors of unit length are drawn on thebay to show the direction of water flow. The vectorsare color mapped to indicate the magnitude of thevelocity. This alternative to using arrow length to

Bay simulation program.

ARTICLE IN PRESS

Fig. 3. Plan views. (a) Water height levels displayed via colormapping. (b) Velocity direction vectors with colormapped magnitudes.

J.R. Crouch et al. / Computers & Geosciences 34 (2008) 247–258 253

indicate vector magnitude eliminates the problem oflarge magnitude vectors obscuring neighboringvectors and cluttering the display. Changes in thewind tend to affect water velocity. This visualizationis shown in Fig. 3(b).

Particle tracking allows visualization of waterflow over time. Five floating virtual particles aredropped onto the surface of the bay every 20simulation days. The particle positions are trackedin iROMS and plotted at each time step. Thisvisualization shows how winds, rivers, and tides canaffect the path of floating objects over a multi-weektime scale. Particle markers can be seen in the planview portion of Fig. 2.

The depth view window provides a visualizationof a cross-section of the bay that follows a centerchannel running roughly north to south. The bay’sgeometry in this view accurately represents thetopography of the floor and the water’s time and

location dependent height variations. The salinitydistribution on the cross-section is displayed viacolor mapping, as shown in the lower left corner ofFig. 2.

3.5. Performance

The Chesapeake Bay simulation program runs atinteractive rates on currently available PC hard-ware. On a 3.2GHz dual Intel Xeon processor PC,25 simulation days pass per minute of user time. Ona laptop PC with a single 1.6GHz Intel processor,the simulation runs at a rate of 11 simulation daysper minute. This means that when the visualizationis set to update at the end of each hour of simulationtime, the results are refreshed every 0.10–0.23 s.

The program is currently available for down-load from the authors’ web site at http://www.d.umn.edu/�jaustin/CHIMP.html as a self-installing

ARTICLE IN PRESSJ.R. Crouch et al. / Computers & Geosciences 34 (2008) 247–258254

executable program that will run under the Micro-soft Windows operating system. Because the pro-gram contains no platform dependent source code,there is no obstacle to compiling the simulationprogram for other operating systems or other typesof computer hardware.

4. Case studies

The Chesapeake Bay simulation program is user-friendly and appropriate for a variety of usersincluding middle school, high school, and universitystudents, museum visitors, and other members ofthe community. It is capable of demonstrating thebasic features of the bay, including layered circula-tion, wind mixing, rotationally controlled flow,wind-driven flow, and tides. Three simulatedscenarios that illustrate interesting ChesapeakeBay phenomena are described in the followingsections. Specific user instructions for replicating

Fig. 4. Top: Initial salinity distribution in bay at beginning of simula

formed.

each scenario are included, along with a descriptionof the effects observed in the simulation.

4.1. Case study 1: normal estuarine salinity and

circulation

Estuaries such as the Chesapeake Bay arecontinuously fed by a source of fresh water at theirhead. The result is that normal estuarine circulationconsists of upper layers of relatively fresh waterflowing out of the bay and lower layers of heavier,saltier water flowing up into the bay. Due to mixingeffects, water closer to the head of the bay is lesssalty than water near its mouth.

User instructions: Set the update rate to 1 h, andset the Susquehanna River flow to a normal orhigher than normal level. Turn the wind off.Observe changes in the salinity over the next 90simulation days. Next, set the update rate to 12 h,

tion. Bottom: Salinity distribution in bay after stratification has

ARTICLE IN PRESSJ.R. Crouch et al. / Computers & Geosciences 34 (2008) 247–258 255

and examine the velocity vector visualization for thesurface and bottom layers of the bay.

Observed effects: Throughout the simulation theplan view shows low salinity at the head of theestuary and higher salinity at the mouth, whereoceanic water is being pulled into the estuary. Thecross-section initially shows homogeneous salinitythrough vertical water columns, with fresher waternear the head of the bay and saltier water near themouth. Without wind to mix the water, the denser,saltier water sinks to the bottom and salinitystratification becomes apparent in the depth view.The depth view also shows a fresh water/salt waterboundary near the head of the Susquehanna thatfluctuates with the tides. Fig. 4 shows snapshots ofthe salinity distribution that evolves in the depthview during this simulation. When the update rate ischanged to 12 h so that the simulation results areaveraged over a tidal period, the bay’s two layerflow becomes evident. The velocity vectors along thebottom of the bay show water flowing toward thebay’s head while the vectors in the surface layershow water flowing toward the bay’s mouth. Thissub-tidal two layer flow is an important feature ofstratified estuaries like the Chesapeake Bay.

4.2. Case study 2: wind effects

During the summer the Chesapeake Bay has fewstrong wind events, leading to stratification as seenin the first case study. The isolation of bottom layerof water from the surface combined with bacterialdecomposition of detritus results in oxygen depletion

Fig. 5. This cross-section of Chesapeake Bay during a strong north win

water out. Salinity stratification has been eliminated by wind mixing w

at the bottom of the bay. A severe storm will mix thewater column, improving oxygenation. Storm windsfrom the south can also affect water height nearMaryland by blowing water into or out of the northend of the bay.

User instructions: Set the update rate to 1 h. Turnthe wind off for 90 days, then turn on a strong northwind. Observe changes in the bay’s conditions usingthe salinity, height, velocity, and floating particlevisualizations.

Observed effects: The water’s surface height isvisibly depressed when the north wind is turned on,and the salinity stratification is lost as the windincreases the mixing between the upper and lowerlayers of water. The result is that vertical watercolumns take on more uniform salinity values, asshown in Fig. 5.

4.3. Case study 3: plumes

Because of the direction of the earth’s rotation,water leaving the Chesapeake Bay tends to flowalong the Virginia Beach coast and entering watertends to flow along the east side of the bay. Thisleads to a plume of high salinity along the southeastcoast of the bay. When a north wind blows, theplume tightens against the shore due to Ekmantransport in the surface layer. Ekman transportdescribes the tendency of surface water to travel tothe right of the direction of the wind in the northernhemisphere. In the case of a north wind, surfacewaters in the central bay move west, flattening theplume against the shore. Conversely, a south wind

d shows reduced water height at bay’s head due to wind pushing

ater.

ARTICLE IN PRESSJ.R. Crouch et al. / Computers & Geosciences 34 (2008) 247–258256

will cause the plume to spread out into the bay asthe surface water moves east.

User instructions: Start with the wind off andthe Susquehanna flow set low for 30 simula-tion days. Then set the Susquehanna flow to highfor 60 days. Finally, turn on a gentle wind fromthe north and observe changes in the salinitydistribution.

Observed effects: When the increased volume offresh water from the Susquehanna reaches themouth of the bay, it turns to the right and movessouth along the coast as a plume. When the northwind is turned on, the plume tightens up against theshore below the mouth of the Potomac, as shown inFig. 6. If the experiment is repeated using a southwind, the plume diffuses into the bay.

Fig. 6. (a) Surface salinity of Chesapeake Bay after weeks without wind

develops when scenario (a) is affected by a storm, simulated by increas

5. Conclusions and future work

As demonstrated through the three examplescenarios, the Chesapeake Bay simulation programaccounts for a variety of complex influences on thebay, including variations in tides, winds, and riverflows over time and the effects of the earth’srotation. This level of complexity is possible becausea research-grade numerical circulation modelingcode is employed as the computational engine ofthe simulation. In addition to providing instruc-tional software for the Chesapeake Bay, the moregeneral achievement of this simulation developmenthas been to demonstrate that it is possible to runsmall numerical models on PCs in an interactivemanner. Real-time interaction with an interesting

and rain. (b) A high salinity plume along southeast coast of bay

ed river flow and wind.

ARTICLE IN PRESSJ.R. Crouch et al. / Computers & Geosciences 34 (2008) 247–258 257

numerical model marks an important step beyondprevious work that provided interactive visualiza-tion of stored simulation data. This demonstrationof true interactive simulation should motivatefuture efforts to create educational software bypackaging scientific research code in an accessiblemanner.

The Chesapeake Bay simulation program hasreceived positive reviews from students, educators,and others after demonstrations that have includedan Earth Day event at Old Dominion University, anOceanography Day presentation at a primaryschool, and other formal and informal presentationswith students ranging from second grade to under-graduate university students. More formal evalua-tion of the program’s educational utility is plannedafter the development of curriculum materials toaccompany the program is complete. The possibilityof placing the program in a marine science museumexhibit is also being investigated.

Development efforts for the Chesapeake Baysimulation program are proceeding in two direc-tions. First, there is no reason that the simulationprogram needs to be limited to the Chesapeake Baylocation. ROMS research models have been devel-oped for a variety of bodies of water, and one goal isto make more of these models accessible through aninteractive simulation program. Instead of writingand compiling a new program for each location, theplan is to encapsulate all the pertinent informationabout a model’s geometry, grid characteristics, andforcing conditions (e.g., the number and names ofrivers feeding the bay) in a data file that can beloaded using a general purpose interface program.The loaded data will be used to configure an arrayof user interface controls and the visualizationoptions. A substantial amount of software designand engineering will be required to adapt theChesapeake Bay simulation program to accommo-date this level of generality, but the end result will bea much more flexible simulation tool. After themodel data file format is established and published,any researcher who develops a ROMS model will beable to adapt it for use in the interactive simulationenvironment. An online repository of downloadablemodels is envisioned for the future.

The second development direction is expansion ofthe set of ROMS features that are accessiblethrough the interactive environment. Much of thefunctionality available in ROMS remains untappedby the Chesapeake Bay simulation program. Visua-lization of pollution and oxygenation distributions

will be added, along with user controls to govern thenumber, location, and type of pollution sources andthe computational grid dimensions. A new type ofvisualization is planned that will allow a user tosimulate a real world experiment by dropping virtualbuoys into the water and plotting the data that ismeasured at those locations. These future develop-ments will make the interactive simulation environ-ment an even more powerful educational tool.

Acknowledgments

This project was sponsored by the VirginiaGovernor’s Research Initiative through the Officeof Research at Old Dominion University’s Office ofResearch. The authors thank John Klinck, LeeBelfore, and Elizabeth Smith for their assistance.

References

Bice, D.M., 2001. Using STELLA models to explore the

dynamics of earth systems: experimenting with earth’s climate

system using a simple computer model. Journal of Geoscience

Education 49 (2), 170–181.

Blumberg, A.F., Mellor, G.L., 1987. A description of a three-

dimensional coastal ocean circulation model. In: Heaps, N.S.

(Ed.), Three-Dimensional Coastal Ocean Models. American

Geophysical Union, Washington, DC, pp. 1–16.

Boesch, D.F., Brinsfield, R.B., Magnien, R.E., 2001. Chesapeake

Bay eutrophication: scientific understanding, ecosystem re-

storation, and challenges for agriculture. Journal of Environ-

mental Quality 30, 303–320.

Chandler, M., Shopsin, M., Richards, S., 2003. EdGCM: real-

time global climate modeling research for the classroom.

Geological Society of America 35 (6), 118.

Chase, C.G., 1992. Fluvial landsculpting and the fractal

dimension of topography. Geomorphology 5, 39–57 (special

issue on fractals).

de Jong, T., Martin, E., Zamarro, J.-M., Esquembre, F., Swaak,

J., van Joolingen, W.R., 1999. The integration of computer

simulation and learning support: an example from the physics

domain of collisions. Journal of Research in Science Teaching

36 (5), 597–615.

Dede, C.J., Salzman, M.C., Loftin, R.B., 1996. The development

of a virtual world for learning Newtonian mechanics. In:

MHVR ’94: Selected Papers from the First International

Conference on Hypermedia, Multimedia, and Virtual Reality:

Models, Systems, and Applications. Springer, London, UK,

pp. 87–106.

Hartel, H., 2000. xyZET: a simulation program for physics teaching.

Journal of Science Education and Technology 9 (3), 275–286.

Large, W.G., McWilliams, J.C., Doney, S.C., 1994. Oceanic

vertical mixing: a review and model with a nonlocal boundary

layer parameterization. Reviews of Geophysics 32, 363–403.

Luo, W., Duffin, K.L., Peronja, E., Stravers, J.A., Henry, G.M.,

2004. A web-based interactive landform simulation model

(WILSIM). Computers & Geosciences 30, 215–220.

ARTICLE IN PRESSJ.R. Crouch et al. / Computers & Geosciences 34 (2008) 247–258258

Officer, C.B., Biggs, R.B., Taft, J.L., Cronin, L.E., Tyler, M.A.,

Boynton, W.R., 1984. Chesapeake Bay anoxia: origin,

development, and significance. Science 223 (4631), 22–27.

Ramasundaram, V., Grunwald, S., Mangeot, A., Comerford,

N.B., Bliss, C.M., 2005. Development of an environmental

virtual field laboratory. Computers & Education 45 (1), 21–34.

Renshaw, C.E., Taylor, H.A., 2000. The educational effectiveness

of computer-based instruction. Computers & Geosciences 26

(6), 677–682.

Saini-Eidukat, B., Schwert, D.P., Slator, B.M., 2002. Geology

explorer: virtual geologic mapping and interpretation. Com-

puters & Geosciences 28 (10), 1167–1176.

Shchepetkin, A.F., McWilliams, J.C., 2005. The regional oceanic

modeling system (ROMS): a split-explicit, free-surface,

topography-following-coordinate oceanic model. Ocean

Modeling 9, 347–404.

Whitford, D.J., 2002a. Teaching ocean wave forecasting using

computer-generated visualization and animation—part 1: sea

forecasting. Computers & Geosciences 28 (4), 537–546.

Whitford, D.J., 2002b. Teaching ocean wave forecasting

using computer-generated visualization and animation—

part 2: swell forecasting. Computers & Geosciences 28 (4),

547–554.

Windschitl, M., 2000. Supporting the development of

science inquiry skills with special classes of software.

Educational Technology Research and Development 48 (2),

81–95.

Winn, W., Stahr, F., Sarason, C., Fruland, R., Oppenheimer, P.,

Lee, Y.-L., 2006. Learning oceanography from a computer

simulation compared with direct experience at sea. Journal of

Research in Science Teaching 43 (1), 25–42.