A Three-Tier System Architecture Design and Development for Hurricane Occurrence Simulation Shu-Ching Chen , Sneh Gulati , Shahid Hamid , Xin Huang , Lin Luo , Nirva Morisseau-Leroy , Mark D. Powell , Chengjun Zhan and Chengcui Zhang School of Computer Science, Florida International University, Miami, FL 33199, USA Email: chens, xhuan001, lluo0001, czhan002, czhang02 @cs.fiu.edu Department of Statistics, Florida International University, Miami, FL 33199, USA Email: [email protected]Department of Finance, Florida International University, Miami, FL 33199, USA Email: [email protected]Cooperative Institute for Marine and Atmospheric Science, University of Miami, Coral Gables, FL 33124, USA Email: [email protected]Hurricane Research Division, NOAA, Miami, FL 33149, USA Email: [email protected]Abstract— As an environmental phenomenon, hurricanes cause significant property damage and loss of life in coastal areas almost every year. Research concerning hurricanes and their aftermath is gaining more and more attentions nowadays. The potential changeability of hurricane data and hurricane models requires robust, maintainable and easily extensible software system for hurricane simulation. With focuses on the design and the implementation of the components at each layer, this paper describes a hurricane simulation system built on the three- tier architecture to achieve good flexibility, extensibility and resistance to potential changes. Index Terms— Three-tier architecture, Web-based system, Database, Hurricane I. I NTRODUCTION An important step in hurricane analysis and prediction is building computer models of a hurricane. Usually, statistical models are built from the historical hurricane data and then the analysis and the prediction can be performed based on these models. Unfortunately , the number of documented hurricanes is limited. For example, in HURDAT database [11], which is maintained by the National Hurricane Center in Miami, Florida and the National Climatic Data Center in Asheville, North Carolina, there are only 1274 hurricanes. One way to supplement for the number of hurricanes is to run simulations based on the statistical models built from historical data. The projection data from the simulation can be integrated with the real hurricane data for further use such as hurricane track modeling and loss and damage estimation. Therefore, the computer system for the purpose of hurricane modeling and projection is of great significance in aiding the hurricane analysis and hurricane hazard prediction. This paper presents our work in designing and building a web-based distributed software system that can be used for the modeling and projection of hurricane occurrences. Our system can let the user build statistical models of the hurricane occurrence from the historical data stored in the database, which is a part of the system. It also provides simulation and projection functionality so that the user can run simulations and projections based on the statistical models. Compared with other relevant systems, which are discussed in section II, the proposed system has the following features: 1) It is a large-scale system which can handle the huge amount of hurricane simulation data and the intensive computation required for analysis and projection; 2) It aims to support both professional and general-purpose users in a very convenient way; 3) Our system is built upon an object-relational database management system, Oracle9i, which is one of the core system components to store and manage the historical hurricane data, the hurricane data model and the projection results. 4) Since the hurricane data are constantly being updated and the mathematical models for the hurricane data are also potentially changeable, a three-tier architecture is adopted as our system’s fundamental architecture to provide the transparency among the data layer, business logic layer and the user interface layer, thus making our system more flexible, maintainable, robust and resistant to the potential changes in the lifetime of the system. The rest of this paper is organized as follows. First, the related works are discussed briefly. Next, the system is intro- duced from the architectural point of view in Section III. Then the design and implementation of the major system compo- nents, namely the user interface layer, application logic layer and database components, are described in a comprehensive way in Section IV, Section V and Section VI respectively. Finally, Section VII gives the conclusion.
Transcript
A Three-Tier SystemArchitectureDesignandDevelopmentfor HurricaneOccurrenceSimulation
Abstract— As an envir onmentalphenomenon,hurricanes causesignificant property damage and loss of life in coastal areasalmost every year. Research concerning hurricanes and theiraftermath is gaining more and more attentions nowadays. Thepotential changeability of hurricane data and hurricane modelsrequires robust, maintainable and easily extensible softwaresystem for hurricane simulation. With focuseson the designand the implementation of the components at each layer, thispaper describesa hurricane simulation systembuilt on the thr ee-tier architecture to achieve good flexibility , extensibility andresistanceto potential changes.
Index Terms— Thr ee-tier architecture, Web-based system,Database,Hurricane
I . INTRODUCTION
An important step in hurricaneanalysisand prediction isbuilding computermodelsof a hurricane.Usually, statisticalmodelsarebuilt from thehistoricalhurricanedataandthentheanalysisand the predictioncan be performedbasedon thesemodels.Unfortunately, thenumberof documentedhurricanesis limited. For example, in HURDAT database[11], whichis maintainedby the National Hurricane Center in Miami,Florida and the National Climatic Data Centerin Asheville,North Carolina, thereare only 1274 hurricanes.One way tosupplementfor the numberof hurricanesis to run simulationsbasedon the statistical models built from historical data.The projection data from the simulation can be integratedwith the real hurricanedatafor further usesuchashurricanetrack modeling and loss and damageestimation.Therefore,the computersystemfor the purposeof hurricanemodelingandprojectionis of greatsignificancein aiding the hurricaneanalysisandhurricanehazardprediction.
This paperpresentsour work in designingand building aweb-baseddistributed software systemthat can be usedforthe modeling and projection of hurricaneoccurrences.Our
systemcanlet theuserbuild statisticalmodelsof thehurricaneoccurrencefrom the historical data stored in the database,which is a part of the system.It alsoprovidessimulationandprojection functionality so that the user can run simulationsandprojectionsbasedon the statisticalmodels.
Comparedwith otherrelevantsystems,which arediscussedin sectionII, theproposedsystemhasthefollowing features:1)It is a large-scalesystemwhich can handlethe hugeamountof hurricanesimulation data and the intensive computationrequiredfor analysisandprojection;2) It aimsto supportbothprofessionaland general-purposeusersin a very convenientway; 3) Our systemis built uponanobject-relationaldatabasemanagementsystem,Oracle9i,which is oneof thecoresystemcomponentsto store and managethe historical hurricanedata, the hurricanedatamodel and the projection results.4)Sincethe hurricanedataareconstantlybeingupdatedandthemathematicalmodelsfor thehurricanedataarealsopotentiallychangeable,a three-tierarchitectureis adoptedasour system’sfundamentalarchitectureto provide the transparency amongthedatalayer, businesslogic layerandtheuserinterfacelayer,thus making our systemmore flexible, maintainable,robustand resistantto the potential changesin the lifetime of thesystem.
The rest of this paper is organizedas follows. First, therelatedworks arediscussedbriefly. Next, the systemis intro-ducedfrom thearchitecturalpoint of view in SectionIII. Thenthe designand implementationof the major systemcompo-nents,namelythe userinterfacelayer, applicationlogic layerand databasecomponents,are describedin a comprehensiveway in Section IV, SectionV and Section VI respectively.Finally, SectionVII givesthe conclusion.
I I . RELATED WORK
There has been a lot of researchon modeling naturalatmospherichazards.Ever since 1948 when Charney et al.made the first successfuldynamical-numericalforecast,nu-merous researchershave contributed to weather modelingandpredictionandestablishedNumericalWeatherPrediction(NWP) modelsand systems,such as ARPS [1] and RAMS[14]. The Advanced Regional Prediction System (ARPS),developed at the University of Oklahoma, is a completeatmosphericnumericalmodeling/predictionsystemdesignedto explicitly representconvective and cold-seasonstorms.RAMS [14], or the Regional AtmosphericModeling System,is a highly versatilenumericalcodeproducedto simulateandforecast meteorologicalphenomena.RAMS was developedby researchersat ColoradoStateUniversity and the ASTERdivision of Mission ResearchCorporation.Like most of themeteorologicalphenomenamodeling and projection system,ARPSandRAMS have the following components:1) A dataanalysispackageto preprocessthe datafrom observation forfuturecomputation;2) An numericalatmosphericmodelwhichperforms the actual simulation and prediction; 3) A post-processingpackagethathandlesthevisualizationandanalysisof results.
Althoughimprovementin thestudyof NWP systemsfavorsmore and more accurateweatherforecasting,most of thosesystemshave the following limitations:
1) There are very few databasemanagementand ware-house techniquesused in those systems.Due to thetremendousamountof data and time-consumingmod-eling processwhich are two inherentproblemsin nat-ural hazardmodeling and prediction, it could be veryhelpful to apply databasemanagementor data ware-housetechniquesto store,retrieve andmanagethe dataand modelsefficiently. Currently, most of the so-calledGIS “Databases”,such as Global Ecosystems(GEP)Database[3] and State Soil Geographic(STATSGO)Database[15], aremerelycollectionsof datasetsinsteadof being stored in and managedby a real DatabaseManagementSystem(DBMS).
2) Most of thesesystemsare stand-alonesystems.Eachapplicationis running on one or several machines,andthey are totally independentfrom each other. Thus itis difficult for different users to shareand exchangeinformation.
On theotherhand,somesoftwareproductsfor thehurricanedamageand loss assessmentalreadyexist. One of the mostprominent tools is HAZUS [2] [10]. HAZUS stands forHazardsU.S. and was developedby the FederalEmergencyManagementAgency (FEMA) as a standardized,nationalmethodology for natural hazardslossesassessment.UsingGeographicInformation Systems(GIS) technology, HAZUScan supportestimatesof damageand lossesthat result fromvariousnaturaldisasterssuchaswind andflood. Someusefuldatabases,such as a national-level basic exposuredatabase,arebuilt into the HAZUS systemthat allows the userto run a
OC4J Container
Oracle Database
JNI
Web BrowserHTTP/SSL
User Interface
Web Server
JavaBeans
JDBC
DatabaseApplication Logic
IMSL LibraryMath/Statistical
Modulein C/C++
Fig. 1. Detailedarchitectureof the system
preliminaryanalysiswithout having to collect additionallocaldata.It alsoprovidesthefunctionalityto allow theuserto plugtheir own datainto the databases.
Although HAZUS is powerful anduseful, it is not suitableto be usedby general-purposeuserswho mayknow a little ornothingabouttheprofoundmechanisms.And asa stand-aloneGIS system,the necessarysoftware such as the commercialGIS packagelike ArcView, andhardwareneedto be installedin every machineon which the HAZUS systemruns, whichin turn increasesboth the expensesandmanuallabor.
I I I . SYSTEM ARCHITECTURE
To achieve the systemrobustness,flexibility andresistanceto potentialchange,the popular three-tierarchitectureis de-ployed in our system.The architectureis composedof threelayers:the userinterfacelayer, the applicationlogic layer andthe databaselayer. The three-tierarchitectureaims to solve anumberof recurringdesignanddevelopmentproblems,henceto make the applicationdevelopmentwork more easily andefficiently. The interface layer in the three-tier architectureofferstheusera friendly andconveniententryto communicatewith the systemwhile the application logic layer performsthe controlling functionalitiesand manipulatingthe underly-ing logic connectionof information flows; finally, the datamodeling job is conductedby the databaselayer, which canstore, index, manageand model information neededfor thisapplication.
User Interface Tier The first tier is the user interface tier.This tier managestheinput/outputdataandtheir display. Withthe intention of offering greaterconvenienceto the user, thesystemis prototypedon the Internet.The usersareallowed toaccessthesystemby usingany existing webbrowsersoftware.The user interface tier containsHTML componentsneededto collect incoming information and to display informationreceived from the application logic tiers. The web visitorscommunicatewith the web server via applicationprotocols,suchasHTTPandSSL,sendingrequestsandreceiving replies.In our system,the major web-scriptinglanguageexploited in
designingthepresentationlayer is theJava Server Pages(JSP)technique� [7] . Thedetaileddesignandimplementationof thistier will be discussedin detail in SectionIV.
Application Logic Tier The application logic tier is themiddle tier, which bridgesthe gapbetweenthe userinterfaceandthe underlyingdatabase,hiding technicaldetailsfrom theusers.An Oracle9iApplication Server is deployed. Its OC4Jcontainerembedsa webserver, which respondsto events,suchas data receiving, translating,dispatchingand feed-backingjobs[12] [13]. Componentsin this tier receiverequestscomingfrom the interfacetier and interpretthe requestsinto aproposactions controlled by the defined work flow in accordancewith certain pre-definedrules. Java Beansperform the ap-propriate communicationand calculation activities, such asgetting/pushinginformation from the databaseand carryingout the necessarycomputing work with respect to properstatisticaland mathematicalmodels.JDBC [5] is utilized forJava Beansto accessthe physicaldatabase.In the interestofthequicksystemresponse,C/C++languageis usedto programthe computingmodulesthat are integratedinto the Java codevia JNI [6]. The detailson this tier are in SectionV.
DatabaseTier The databasetier is responsiblefor modelingand storing information neededfor the systemand for opti-mizing the dataaccess.Data neededby the applicationlogiclayer are retrieved from the database,then the computationresultsproducedby theapplicationlogic layerarestoredbackin the database.Since data are one of the most complexaspectsof many existing information systems,it is essentialin structuringthe system.Both the facts and rules capturedduring datamodelingandprocessingare importantto ensurethe data integrity. An Oracle9i databaseis deployed in oursystem,andtheObjectRelationalModel is appliedto facilitatedatareuseandstandardadherence.SectionVI will give moredetailsaboutit.
IV. USER INTERFACE
The intendedsystemis prototypedinto Internet; therefore,the designand implementationof the systemuser interfacemainly becomesa job of designingand implementingwebpages.The userscan gain accessto the systemthroughanycommonlyusedcommercialbrowsersuchasInternetExplorer,Netscape,etc.
Due to its “unlimited” expressive power andnaturalcoher-encewith theJ2EEarchitecture,JSPweb-scriptingtechnologyis adoptedto implementthe web pages[7] [8]. JSP, sitting ontop of a Java servletsmodel,caneasilyandflexibly generatethe dynamic contentof a web page.The basic idea of JSPis to allow Java code to be mixed with static HTML orXML templates.The Java logic handlesthe dynamiccontentgenerationwhile themarkuplanguagecontrolsthestructuringand the presentationof data.
Figure2 shows thebasiccourseby which the userinteractswith the system.One individual JSPpageis implementedfor
System Boundary
System Interface
display simulation result
submit simulation request
submit modelling request
display statistical models
select dataset
Log onto the system
The User
Fig. 2. Interactionflow betweenthe userand the system
Fig. 3. Datasetselectionweb page
each individual user-systeminteraction. First, a log-in JSPpageis usedto take careof the userlog-in process.The userneedsto provide the usernameand password for the systemto authenticatethe user.
One meteorologicalfact is that the statistical propertiesof hurricanesvary with different year ranges.For example,the statisticalpropertiesof stormsin El-Nino yearsare quitedifferentfrom thosein non-El-Ninoyears.Therefore,differentstatisticalmodelsare necessaryfor different year ranges.Inour system,all the historicalhurricanerecordsin the databaseare categorized into five datasetsaccordingto meteorologiccriteria, which are: 1) 1851-2000,2) 1900-2000,3) 1944-2000, 4) ENSO and 5) Multi-Decadal. Dif ferent statisticalmodelsare built for individual datasets.Therefore,after theuserlogs onto the system,anotherJSPpageshown asFigure3, allows the user to selectthe dataset. After a userselectsthe dataset, anotherJSPpagelets the usersubmit a requestto the applicationlogic layer to build the statisticalmodel ofthe hurricanefor the selecteddata set. Then the applicationlayer builds the model, storesthe model into databaseandsendsthe modelto the userinterfacelayer which displaysthe
web server
OC4J container
application logic layer
Math/Statistical Module
JavaBeandatabase
IMSL library
JNIJNI
JavaBeansimulation
JavaBeanmodelling
Fig. 4. Basicstructureof applicationlogic layer
resultantstatisticalmodel to the user by using anotherJSPpage.After that, the usercan requestthe applicationlayer torun simulation(projection)basedon thestatisticalmodel.Theapplicationlayer doesa simulationasthe response,storesthemodel into the databaseand feedbacksthe simulationresultsto theuserinterfacelayer, which displaysthemvia a JSPpage.
V. APPLICATION LOGIC LAYER
The applicationlogic tier is the middle tier which bridgesthegapbetweentheuserinterfaceandtheunderlyingdatabase,hiding technicaldetails from the user. It communicateswiththe userinterface,performsthe statisticalmodelingandsimu-lation, andinteractswith the databaselayer suchasretrievinghurricanedata from the databaseand storing the statisticalmodelandsimulationresultsinto the database.
A. ApplicationLogic Layer Overview
Figure 4 shows the basic componentsin the applicationlogic layer and the relationshipsamongthosecomponents.
An Oracle9iApplication Server is deployed to supply thefundamentalservicesthatallow componentsto concentrateonbusinesslogic without concernfor low-level implementationdetails. It handlesnetworking, authentication,authorization,persistence,and remote object access.Its OC4J containerembedsa web server, which respondsto eventssuchas datareceiving, translating,dispatchingand feed-backingjob [12][13]. The Java Beansperformall the actualwork in businesslogic. The “databaseJavaBean” utilizes JDBC [5] to accessthephysicaloracledatabaseto retrieve andstorethehurricanedata,statisticalmodelsandsimulationresults.The “modelingJavaBean” is responsiblefor building the statistical modelsfor the hurricanedatafrom the user’s specifieddataset.The“simulation JavaBean” runs simulationsusing the statisticalmodels.For the sake of performance,time-consumingcom-putation tasks,namely the statistical model calculation andsimulation, are actually achieved by using C/C++ codesin“Math/StatisticalModule”, which runson linux platform.TheC/C++ codeis seamlesslyintegratedinto correspondingJavacodesin the Java Beans(the “modeling JavaBean” and the
“simulation JavaBean”) via the JNI (Java Native Interface)mechanism[6]. The commercialsoftware IMSL [4] providesthe high-performanceC routinesfor mathematicalandstatis-tical calculation.
B. AnnualHurricane Occurrencemodeling
Annual HurricaneOccurrencemodelingaims to model thenumber of hurricanesoccurring per year (AHO) and thehurricanegenesistime (AHO). Therefore,for eachdata setthere are actually two statistical models: one for the SGTand the other for the SGT. After the modelingprocedure,thestatisticalmodelsarestoredinto thedatabasevia the“databaseJavaBean”. The modeling algorithms are implemented in“Math/StatisticalModule” which is called by the “modelingJavaBean”and the “simulation JavaBean”
1) AHO Modeling: Since different statistical models arebuilt for different dataset,the user first needsto selectonedatasetfrom the five categories through the user interfaceas aforementioned.Let � datasamplesin the user-specifieddatasetretrieved from the databasebe denoted by � ��������� ������������ �����! , where � is the numberof yearsinthedatasetand
���denotesthe numberof hurricanesoccurring
in the�#"%$
year in the dataset.The statisticalmodel of AHOis built basedon the � datasamples.According to domainknowledgein meteorology, the beststatisticaldistribution ofthe number of hurricanesoccurring per year is either thePoissondistribution or the Negative Binomial distribution.First, the parametersof both the Poissondistribution andNegative Binomial distribution are estimatedfrom the datasamples . Thenthechi-squarestatisticis calculatedto selectthefinal model.Thedistributionwith ahigherp-value,namely,the distribution with the better fit is selectedas the finalstatisticalmodelof the AHO.
2) SGTmodeling: The genesistime of a storm is the firstfix data of that storm. SGT modeling aims to model thegenesistime of the hurricanes.This is achieved by modelingthe numberof hoursbetweenthe genesisof a storm in six-hourresolutionandthestartof its hurricaneseasonratherthandirectlymodelingthestormgenesistime.A stormseasonstartsfrom May 1st of oneyearandendsat April 30th of the nextyear. After modelingthe numberof stormsusing AHO fromthe historicaldata,the SGT modelcanbe usedto predict thetime intervals amongstorms,andthusthe stormgenesistimeof eachstormcanbe predictedaswell.
Let & datasamplesin the user-specifieddatasetretrievedfrom the databasebe denotedby '( �*) �+�,�%� -�.�/�0��� ������&1 where & is the numberof hurricanesin the datasetand thetime associatedwith the
�#"%$hurricanein the selecteddataset
be donotedby) �
. The statisticalmodelof SGT is built usingthe data samples ' . Specifically, a nonparametricapproachis applied to estimatethe Cumulative Distribution Function(CDF) of the time intervals . Let the randomvariable timedenotedby 2 . First theempiricalCDF 3 �%4 for 2 is calculatedfrom thedatasamples' . Thenthesmoothestimatorof 3 ��4 is
calculatedbasedon empiricalCDF usingEpanechnikov kernelwhich5 servesas the final statisticalmodelof SGT.
C. Simulation/Projection
Based on the statistical model built from the historicalhurricanedata,the systemcan run a simulation/projectioninresponseto the user’s requestbasedon the desirednumberofyearsfor the simulationhe/shespecifies.After the simulationprocedure,the simulationresultsare storedinto the databasevia the “databaseJavaBean”.
Let 6 denote the number of years the user specifies.First, 6 randomnumbers7 �8�%� 9�.�/�0��� ���:��6; are generatedfrom the AHO model, namely either the estimatedPoissondistribution or Negative Binomial distribution. Each randomnumber, 7 � , meansthe numberof hurricanesoccurredin anindividual year. The total numberof hurricanessimulatedis<
=> �@?BA 7 � . Then,<
randomnumbers C0D �FE G�.������� ���:� < are generatedfrom the distribution in the SGT model. Eachrandomnumber, C D , denotesthe interval associatedwith theE."%$
simulated hurricane. Therefore, for theE�"%$
simulatedhurricane,its genesistime is projected,which is the first dayof the hurricaneseasonplus the interval C D .
VI . DATABASE COMPONENT
Dataanalysisandmodelingis a vital aspectof thedatabasecomponent.In our system,an object-relationaldesignpatternis applied to model hurricanedata. Object-relationalmodelcan assist the reuseof the databaseobjects. The Oracle9idatabaseis incorporatedin the system as the informationstorehouse,which storesdatarecordsfor all stormsoccuringin theAtlantic Basinsince1851.An object-relationaldatabaseschema is designed to facilitate the data reusability andmanageability. The major advantagebrought by the object-relational conceptsis the ability to incorporatehigher levelsof abstractioninto our data models,while current relationaldatabasesareusuallyhighly normalizedmodelsbut with littleabstraction.The overall view of the hurricanedataschemaisdepictedin Figure5 .
VI I . CONCLUSION
In this paper, a web-baseddistributed systemfor the pro-jection of hurricane occurrencesis presented.It integratesa group of individual applicationsby combining hurricanedataacquisition,storage,retrieval, andanalysisfunctions.Thesystemexhibits a modular, extensible,and scalablearchitec-ture that makes it possibleto adaptto more complex tasks,suchasstormtracksimulationandwind field generation.Thewell-establishedthree-tierarchitectureis exploitedto build thesystem.A varietyof advancedtechniquessuchasJSP, JNI andJDBC areusedin the designanddevelopmentof the applica-tion. Both theOracleDatabaseandthe ApplicationServer aredeployed to integrate the systemcoherently. The completedimplementationis easyand convenientto use.In addition, it
Fig. 5. Databaseschema
is accessibleto any userwho is ableto connectto the Internetandhasinterestin hurricanepredictioninformation.
ACKNOWLEDGMENT
This work is partially supportedby Florida DepartmentofInsuranceunder“FIU/IHRC Public HurricaneRisk and LossModel Project.” While the project is funded by the FloridaDepartmentof Insurance(DOI), the DOI is not responsiblefor this papercontent.
REFERENCES
[1] M. Xue, K. Droegemeir, and D. Wang, “The Advanced RegionalPredictionSystem(ARPS)- A MultiscaleNonhydrostaticAtmosphericSimulationand PredictionTool. Part I: Model Dynamicsand Verifica-tion,” Meteor. Atmos.Physics, vol. 75, pp. 161–193,2000.
