Hydroinformatic Web Application and Web Service for Real-Time Water Level Presentation and Short-Term Prediction

Hydroinformatic Web Application and Web Service for Real-time Water Level Presentation and Short-term Prediction

Bunchingiv Bazartseren, Gerald Hildebrandt, K.-Peter Holz1

Abstract Modern urbanization tends to cause a fast response time between a heavy precipitation and consequent runoff processes in a river basin. Therefore, it is even more essential nowadays to enable citizens to have a rapid and flexible access to information on the prevention or restoration measures in cases of flood, on the basis of the cutting edge advances of the Information and Communication Technology (ICT). The paper contains a description of a real-time Web application and service for water level observation, processing, presentation and a short term prediction of a river water level in the area of interest. The Artificial Neural Networks (ANN) are implemented for cost-effective water level prediction for a short horizon. On the whole, the Web applications and services should form a part of a publicly accessible Web based flood crisis management system.

1. Introduction A prediction of high water situation is one of the most essential hydrological tasks for a river basin management and is mainly performed by means of traditional conceptual and deterministic models using predicted precipitation. Due to ever increasing urbanization and the consequential short hydrological response in the river basin, the information on the prevention or restoration measures in case of flood ought to be even faster and flexibly accessible to general public. In this study, it was attempted to develop a Web-application and service for water level observation, processing, presentation and short term prediction as a part of the flood disaster management system. The study area forms a part of Oder River, east of Germany. The water level presentations, as well as the prediction models are implemented as Web application using Java Servlet technology. A cost-effective and rapid-responding on-line tool by the Artificial Neural Network (ANN) models has been implemented at two different cities along the Oder River. The current flood

1 Lehrstuhl für Bauinformatik, BTU Cottbus, Universitätsplatz 3-4, 03044 Cottbus, Germany www.bauinf.tu-cottbus.de 605

disaster management system is being developed within the framework of the European Commission funded project OSIRIS (Operational Solutions for the management of Inundation Risks in the Information Society). The sections 2 and 3 of the paper describe the study area, applied methodology and technology. The section 4 deals with the actual implementation of Web application and service, followed by the conclusion.

2. Study area and data The area under investigation is the Oder River on the eastern border of Germany. The sixth biggest confluence to the Baltic Sea, Oder is a trans-boundary river. The water level observations used for a real-time presentation and prediction for cities of Frankfurt/Oder and Schwedt are allocated real-time at a sampling interval of 15 min from a public data server on the Internet, provided by a relevant authority and stored into a local database. To predict the water level at two biggest cities in the study area, the observations from two consecutive gauging stations located 11 to 30 km upstream have been used, considering the travel time. For an off-line setup of neural networks, approximately one year of observation has been used.

3. Methodology and techniques The Artificial Neural Networks (ANN) are applied for water level prediction. The water level presentations, as well as the prediction models are implemented as Web application using the Java Servlet technology. The presentation of the collected and predicted water level data is implemented in Scalable Vector Graphics (SVG). The Simple Object Access Protocol (SOAP) has been used to deploy the water level prediction as a Web service.

3.1 Web application and Web service techniques 3.1.1 Java Web Services The Java Web Services Developer Pack (JWSDP) is applied as an integrated toolkit that in conjunction with the Java platform allows building, testing and deploying XML applications, Web applications, and Web services. The Java WSDP provides Java standard implementations of existing key Web services standards including WSDL and SOAP, as well as important Java standard implementations for Web application development. These Java standards allow sending and receiving SOAP messages, and quickly build and deploy Web applications based on the latest Servlet/JSP specification. The Apache Tomcat Servlet container is used as official reference implementation for the Java Servlet technology.

29.08.02, BazartserenHildebrandtHolz.doc

3.1.2 Scalable Vector Graphics (SVG) SVG is the description of a two-dimensional vector graphic as an XML application. Any program such as a Web browser that recognizes XML can display the image using the information provided in the SVG format. Vector graphics is the expression of an image using mathematical statements rather than bit-pattern description. Scalable emphasizes that vector graphic images can easily be made scalable. Thus, the SVG format enables the viewing of an image on a computer display of any size and resolution, whether a tiny LCD screen in a cell phone or a large CRT display in a workstation.

3.1.3 Simple Object Access Protocol (SOAP) The primary use of SOAP is for different programs, possibly written in different languages and running on different platforms, to communicate with each other. SOAP is a Remote Procedure Call (RPC) protocol that uses standard Internet protocols for transport - either HTTP for synchronous calls or SMTP for asynchronous calls. SOAP uses XML for the envelope (i.e. the format of data transmitted). Since Web protocols are installed and available for use by all major operating system platforms, HTTP and SOAP provide an already at-hand solution to link disparate systems within and external to a corporate network. SOAP specifies exactly how to encode a HTTP header and an XML file so that an application in one computer can call an application in another computer and pass it information. It also specifies how the called program can return a response.

3.2 Prediction technique - Artificial neural networks The ANN is a set of computational units, which aimed to emulate the learning behavior of a human brain. So called Multi-Layer Perceptron (MLP) networks with one hidden layer are applied in this study. Each layer consists of artificial neuron units. In the hidden and output layer of the MLP, the weighted sum of the input vector is passed to linear or non-linear activation function to get an output from a node. Upon optimizing certain weight or connection strength between those neurons to the minimum error between the target and computed values, the ANN can learn or generalize the input/output relations on the basis of the existing observation data set. Thus, accuracy of the ANN solution highly depends on the quality and quantity of data set. In the current study, the ANN models have been trained to provide the predicted water level on the basis of the hydrological conditions upstream.

29.08.02, BazartserenHildebrandtHolz.doc

4. Water level presentation & prediction 4.1 Off-line prediction of water level Carefully elaborated off-line study on water level prediction at the cities of Frankfurt/Oder and Schwedt by the ANN models resulted in a satisfactory performance for a short prediction horizon. The inputs to the prediction models for both locations are water level observations at two consecutive gauging stations upstream. The travel time has been taken into account considering various prediction horizons. The performance indices for the prediction models are the Root Mean Squared Error (RMSE). The prediction models have resulted in a high accuracy for different lead time (see summary in Table 1).

Table 1. The summary of prediction performance

t+1 t+6 t+12 training verification training verification training verification Frankfurt(O) 0.981 2.765 1.235 3.149 2.274 4.388 Schwedt 0.684 1.015 0.580 1.005 0.713 2.273

The ANN models trained off-line predict the water stage on the basis of easily available information, which are only hydrological conditions upstream. The on-line water level predictor can be used as an alternative or complement to conventional modeling approaches. The disadvantage of such a prediction approach is that the ANN models need to be occasionally trained over, whenever the new extreme values of water stage are observed.

4.2 Web application A real-time dynamic and interactive Web application for water level presentation has been developed with functionalities for data collection from a public data server (www.elwis.de), saving into a local database, processing the user’s requests and presenting it in a tabular and graphical view (www.ist-osiris.org:8092/op). Upon the user’s request on gauging station, time interval and time slice for presentation, the Java Servlet generates simultaneously a HTML page, which produces a tabular and SVG view of the observed water level. The water level observation can be requested from 1 hour up to 1 year for presentation (see Figure 1). The trained ANN models are implemented as an on-line water level predictor and integrated into a real-time water level presentation, from which the inputs to ANN model are allocated. The water level presentation gives user a possibility to compare

29.08.02, BazartserenHildebrandtHolz.doc

the predicted and observed values in the past for different prediction horizon and period, as well. The water level prediction is made for 1 to 12 hours of time span. Since, the implementation was not aimed the prediction models to be trained or setup on-line, the neural network models have to be trained over from time to time, when new extreme values are observed.

Fig. 1: Water level presentation & prediction, Web application

4.3 Web service The Web application for waterlevel presentation & prediction became a Web service by adding a SOAP interface to the existing application. Writing an RPC based SOAP service is a very trivial undertaking by either creating a new code artifact in ones preferred programming language or just using existing code, as done for OSIRIS. The code artifact does not have to know anything about SOAP, as either a method or a script function that exists has to be exposed within the artifact. For the waterlevel presentation & prediction application an existing Java class, that has a method called getWaterLevelPrediction(), just has been exposed as a SOAP service on a RPC router. The Web Service urn:OSIRIS-ICMS-WaterLevelPrediction simulates a waterlevel prediction at the gauging stations Frankfurt/Oder and Schwedt for a leadtime of 1, 6 or 12 hours on the basis of the implemented ANN models. The exposed method getWaterLevelPrediction() expects the name of the gauging station and the leadtime as parameter and returns the predicted waterlevel (see Figure 2).

29.08.02, BazartserenHildebrandtHolz.doc

The browser based interface of the Web application and the corresponding Java Servlet allows the method invocation only by a manual input. With the added SOAP interfaces it is possible to invoke the exposed method directly from any other application written in different languages and running on different platforms.

Fig. 2: Water level prediction, Web service summary and generic SOAP client

5. Conclusions There is a need for a rapidly accessible on-line system to help general public, as well as the decision makers to plan emergency and restoration measures during high water situations. A Web application for a real-time water level presentation was developed as a part of a flood disaster management system. The system should enable an average citizen with an Internet access to monitor the flooding events likely to occur in their area of interest and take the means of prevention measures suggested by the relevant authorities. A cost-effective and rapid-responding on-line tool by the ANN models has been implemented for water level prediction for a short horizon and integrated into the above Web application and service. The prediction based only on hydrological conditions upstream has produced a satisfactory accuracy. However,

29.08.02, BazartserenHildebrandtHolz.doc

the data from further upstream are needed to enable predictions for a longer time span. It is expected that an increased use of Web services will be seen in the near future. Web services will be a valuable solution to use a variety of system environments and languages. SOAP will be used as glue in connecting a Web service with the application calling it. The SOAP can solve most inter-operability problems connected with the Web services. It provides a way of working at the same layer with different technologies and alleviates maintenance and communication. Moreover, it enables to potentially reach everybody with a device that can receive Web services, such as a Web browser or a cell phone.

Bibliography

Apache Software Foundation: Apache XML Project, Apache SOAP implementation, xml.apache.org/soap Apache Software Foundation: Apache XML Project, Batik - Java based toolkit for Scalable Vector Graphics (SVG), xml.apache.org/batik Apache Software Foundation: Jakarta Project, Apache Tomcat 4 Servlet/JSP container, jakarta.apache.org/tomcat Bazartseren, B., Holz, K.P. (2001): Flood prediction using neural networks and neuro-fuzzy approach, Proceedings of the 7th International Conference on EANN, Cagliary, Italy pp.70-78 Bundesanstalt fuer Gewaesserkunde: Elektronisches Wasserstrassen-Informationssystem ELWIS, www.elwis.de (in German) Erlich, M.: Operational Solutions for the management of Inundation Risks in the Information Society (OSIRIS), www.ist-osiris.org Haykin, S. (1999): Neural networks: A comprehensive foundation, 2nd edition, Prentice Hall New Jersey Bazartseren, B., Hildebrandt, G., and Schoenefeldt, K.: OSIRIS Demonstrator for Frankfurt/Oder, www.ist-osiris.org/ffoder Bazartseren, B., Hildebrandt, G.: OSIRIS-ICMS Waterlevel presentation & prediction (Web application), www.ist-osiris.org:8092/op Hildebrandt, G.: OSIRIS-ICMS Waterlevel Prediction (Web service), www.ist-osiris.org/ffoder/op/opSOAP.html Hildebrandt, G.: Web Services Management Platform, dcms.bauinf.tu-cottbus.de/wsm Hildebrandt, G.: SOAP RPC Router, dcms.bauinf.tu-cottbus.de:8090/soap/servlet/rpcrouter Sun Microsystems: Java Technology & Web Services, java.sun.com/webservices Wilson, G., Khondker, M.-H., (2000): Data selection for a flood forecasting neural network, in: Proc. 4th ICHI’00, Iowa, IA, USA World Wide Web Consortium (W3C): Scalable Vector Graphics (SVG), www.w3.org/Graphics/SVG World Wide Web Consortium (W3C): Simple Object Access Protocol (SOAP), www.w3.org/TR/SOAP

29.08.02, BazartserenHildebrandtHolz.doc