Real time wave propagation simulation: model implementation for the Tagus River

Transcription

1 Real time wave propagation simulation: model implementation for the Tagus River P. A. Diogo, A. C. Rodrigues & A. Rodrigues Dep. of Environmental Sciences, Faculty, of Sciences and Tech., New Univ. of Lisbon, Quinta da Torre, 2825 Monte de Caparica; Water Institute, Lisbon. Abstract A wave propagation model is applied to Tagus river, built in the Flood Surveillance and Warning System (FSWS), developed by the Portuguese Water Institute (INAG). The purpose of this project, beyond simple wave propagation modelling, was to integrate modelling in the FSWS, minimising user interference: the modelling system should be able to maintain it automatically and overcome input data difficulties, minimising mathematical instability and providing an effective tool for flood forecast and warning. The model is set to run at regular time intervals, producing results accordingly. These time intervals may be changed depending on the number of simulations needed for effective flood prevention. As input data, real time information, obtained from automatic gauging stations and from some Portuguese and Spanish reservoirs, all registered and stored in the FSWS, is used. This paper will focus on the problems/solutions regarding real time implementation and will try to show how such a system may help flood situations managing. The system was first tested during the winter of 1996/97 but real time implementation was fully ready by the winter of 1997/98. Results showed that modelling can be a reliable tool for dealing with and managing flood situations and future developments include application to other rivers in the Portuguese territory, where flood situations are likely to occur. Keywords: Flood warning system, wave propagation modelling, real time implementation, Tagus river. 1 Introduction The need to control flow becomes more urgent as the demands on a river as a natural resource increase. With rivers regarded as multipurpose systems for water supply, transportation, drainage and

2 36 Hydraulic Engineering Software recreation, it is important to ensure that flows are properly and efficiently controlled and that damage due to the extreme events including flooding or pollution, are minimised [1]. The relatively low ratio of cost to benefit for a flood forecast and warning service makes it an ideal flood-protection measure in many areas where physical means cannot be economically justified. The soundest approach to the flood^ problem lies in a planned combination of water-control structures, floodplain zoning, insurance and adequate forecasting [2]. In the case of the Tagus basin, whenever very high precipitation occur, several dams located close to the Spanish-Portuguese border, storing high volumes of water, discharge significant flows above the capacity of the river channel at the downstream reach [3]. In less than 15 years (from 1978 to 1990) Tagus river suffered four severe flooding situations, as important areas at the downstream reach of the river (just before the Tejo estuary) were flooded [4]. During the winter seasons of 1996 and 1997, high flows were again registered and some villages isolated. Integrated in the Flood Surveillance and Warning System (FSWS), a one-dimensional model was implemented, using real time input data and producing results with regular time intervals. Methodology for real time model implementation and calibration are herein presented and results of the 1st year of application are evaluated. 2 Flood Surveillance and Warning System The Flood Surveillance and Warning System (FSWS) was first implemented on the Tagus basin during the floods of December 1995, using 3 automatic gauging stations (Tramagal, Almourol and Omnias) and data from Portuguese and Spanish reservoirs. Gauging stations information was updated every 5 or 10 minutes while reservoir information was collected hourly. Since then the system has been growing and several new automatic gauging stations have been installed along the Tagus and other important river basins, like Douro, Guadiana, Sado and Mondego. Nowadays the system includes 19 automatic gauging stations covering the Tagus basin and information from 5 Portuguese and 4 Spanish reservoirs.

3 Hydraulic Engineering Software 37 The surveillance system is centred at the Water Institute (INAG), located in Lisbon. A specific software application, developed by INAG, communicates with all the gauging stations via modem, stores and displays all the collected information, allowing the user to view river data from all the country river network, and graph and print the information. The system works in real time and may be made available to other remote users by using a modem. FSWS is permanently registering information although some of its functions and capabilities are relaxed during dry periods of the year. The information collected and processed is made available to all entities responsible for flood situation prevention and managing like fire departments, civil protection services and municipal authorities [3]. 3 The model Natural flood waves are considerably more complex than the simplified cases which yield to mathematical analysis, but theoretical treatment is specially useful in studies on surges in canals, impulse waves in still water, and waves released from dams [2]. The applied model simulates flow with variable regime in channels with no ramifications and simple topography. It is mainly useful for wave propagation study in rivers and estuaries, irrigation channels and basic testing of hydrodynamic waves. It's based on the Saint-Venant equations, solved by Preissman's method: Continuity: J a J Q + b. a // = n 0) <7 a <9 f momentum:,? g ^ r g^ /, K& ^ * Q = flow (nvvs) b = river width (m) ft = Boussinesq coefficient C =Chezy coefficient (m^/ s) h = water depth A = cross-section area (nf) I = reference level slope R = hydraulic radius (m) (= bottom slope) q = lateral inflow (nvvs) External boundary conditions consist of discharges from Cedilho and water level at V. F. de Xira. Internal boundary conditions are constituted by discharges released from two run-of-the-river type dams and the main Tagus tributaries (Ocreza and Zezere rivers).

4 38 Hydraulic Engineering Software River bathymetry is based on topographic surveys from the early seventies, although bottom sediment movements have made them partly out of date. For calculation, 380 cross-sections are used, 500 meters apart. Results of each simulation define stages and flow along the river, with predefined time and space intervals. 4 Implementation The model simulates wave propagation along the 190 km of the Tagus River Portuguese reach, from Cedilho reservoir (at the Spanish-Portuguese border) until V. F. de Xira, just before the Tagus estuarine area. Flood situations are often verified along this part of the river and several villages are affected. As output information, the model provides stage and flow values every 500 meters, but only 8 key cross-sections are available for the common user, although output information on all the other crosssections is also stored and can be accessed by the system administrator. These key cross-sections correspond to sites where some of the gauging stations are located and, therefore, where data comparison is possible. Results are produced hourly. The model was implemented in Microsoft Power Fortran, but several other algorithms had to be developed. Filtering input and output data, an important issue not only for model stability but also for output validation, was implemented in Turbo Pascal 6.0. To keep it independent from the FSWS, a modelling shell was developed, which may be set to active or inactive, according to the system administrator decision. In this way model processing errors and inadequate outputs don't FLOOD SURVEILLANCE AND WARNING interfere with the FSWS data aquisilion, performance asfilteringsoftware "tells" the main program that no I i storagcd A L modelling results are available. T T ER I 4.1 FSWS integration Figure 1 - Integration in FSWS The FSWS registers stage information (gauging stations) with predefined time intervals and reservoirs discharges hourly. These data can be used as an input to the model, which runs also with predefined time intervals. The model shell accesses these data directly from the database.

5 Hydraulic Engineering Software 39 As input data the model requires more data than can be provided by the FSWS. For every run an initial situation regarding stage and flow for each 500 meters and external boundary conditions are necessary (every cross-section of the river has to be described in terms of stage/flow). All information not available from the FSWS is obtained from output of the previous model run (figure 1). All the exchange of information between the model shell and the main part of FSWS is performed using ASCII data files. 4.2 Real time implementation's methodology Working with real time modelling and reducing user interference requires definition of an adequate methodology, possible to be implemented in an automatic way (and therefore programmed). Its final goal is to keep the model working and guarantee feasible modelling results. It is usually found that computerised methods for taking advantage of reported flows can became complex [5]. All the tasks must be automatically performed: initial and boundary conditions definition, model running and output results validation and storing. The implemented methodology consists of the following steps: A) Initial conditions file construction (named here as file A), containing flow and stage values for all cross-sections; B) Validation offilea; C) Validation offielddata, obtained from FSWS; D) File A correction by replacing the correspondent values; E) Model running, using 5 minutes time step and producing results for the next 24 hours; F) Output data validation and storing, using 3 datafiles:one with stage and flow simulation for the next 1/2 hour (named here as file B), the 2nd having output information to be displayed and the 3rd to be used as an historical record of the consecutive model runs. Steps A to D consist all together of input data processing and step F consists of output data processing. As to step E, 5 minutes time step was chosen by balancing model execution time and results resolution. 24 hours, with hourly results, was defined as an ideal forecast time lag for flood situations prevention.

6 40 Hydraulic Engineering Software Input data processing Input consists of data obtained from automatic gauging stations, reservoir discharges and output data from previous run of the model. Using this information all together, an ASCII file is prepared containing the initial conditions for each run. Main tasks of input data processing are: 1) Validation of previous run outputfile (file A); 2) Validation of gauging stations and reservoirs data; 3) File A copied tofileb, used as initial conditions for the next run; 4) When available, data from gauging stations and reservoir discharges is used to replace the correspondent values (flow and/or stage) in file B. During data validation, 3 problems can be expected: a) after periods of model inactivity, no output data from previous runs is available; b) previous run produced invalid results, and c) no field data is available. Situations a) and b) can be resumed to one case: no data from previous run is available. After some tests it was concluded that the use of predefined initial conditions, stored in backup files, induced mathematical instability. Instant data from gauging stations very seldom can be matched with an hypothetical flow situation. Better results where obtained by building a new file B, by interpolating gauging stations and discharge data along the river. Whenever interpolation is not possible then the use of predefined backupfilescan not be avoided Output data processing Outputs from the model are stored in 3 different files. The first is what was formerly calledfilea (section 4.2); the 2nd contains flow and stage results for the 8 key cross-sections of the river for the next 24 hours, and the 3rd is a daily file, which registers all modelling results each day. This last one is used to control model performance and results validation. Output from the model is used with two different objectives: as input data for the next run and as information on river flow conditions, to be displayed on screen. The first consists of input data processing and therefore explained in section Regarding the 2 * objective, only simple validation of data is performed: if no mathematical instability occurred then the results are assumed to be correct. As in any modelling study, it is up to the

7 Hydraulic Engineering Software 41 person who looks at the results to decide if they are to be totally trusted or not. 4.3 Displaying information When validated, output results area passed on to the FSWS main shell and made available o screen (Figure 2). Whenever flooding situations are forecasted, values are presented in a different colour and a report can be viewed and printed, Figure 2 - Results display informing about effects of water stage rising. For instance, within two hours the road number 100 will be closed due to flooding. Information about present stage and flow is also presented and the possibility of graphing stage evolution with time, both for field data and modelling results, helps the user understanding the evolution of the situation. 5 Results Implementing a real time wave propagation model adds difficulties to simple model calibration, because many situations have to be considered and the modelling inputs are not perfectly controlled. Using output data from previous runs together with real time field data without user interference can cause mathematical instabilities as field data may introduce flow/stage variations with which the model may not cope with. 5.1 Model calibration For model calibration gauging stations data from December 1996 and January (high river flow period), and February and March 1997 (regular river flow period) river flow was used. As some gauging stations were not yet implemented within those periods,

8 42 Hydraulic Engineering Software only 3 where used (Abrantes, Almourol and Omnias). Hourly dam operation data was available for both periods and for all reservoir within the modelling scope. As a calibration method, real time situations were simulated. By advancing the 5 computer's clock time and by * selecting data from stored gauging stations data, new input files are created every time the model runs. Sequential runs were Gdddata 8 hours 16 hours x 20 hoias ' - * 30 hours 40 touts Figure 3 - Stage at Abrantes performed for 48 hours, every hour (by advancing the clock), and all simulation results registered. This procedure allows simulation of the arriving of field data, as if it were in a real situation. Simulating 2 days of model functioning takes from 2 minutes (Pentium, 200 MHz processor) up to 2 hours (486, 50 MHz processor). Figures 3, 4 and 5 illustrate model performance for periods of relatively high flow. In every graph presented the x-axis represents hours passed since the model started to run as a standalone application. From the 48 simulations performed within each calibration period only partial results are 4000 Measured Mow 3500 H hoiars showed, as graphs would hours became unreadable hows Nevertheless all results obtained where verified and 1000 compared System performance hours Figure 4 - Flow at Almourol During system implementation several unexpected problems came up. Most of them had to do with incorrectfilteringof input data which led to frequent model crashes. After some weeks of testing, errors were minimised as newfilteringwas added. As the whole system is not dependent on the model, modelling errors do not interfere with FSWS performance but the model is dependent on the state of the FSWS: if insufficient data is available

9 Hydraulic Engineering Software 43 then the model performance is reduced and can lead not only to incorrect results but also to mathematical instability. Report production based on simulation results is easily available and is user friendly. This feature is not yet available for all users as the table containing Stage - Effect information has to be carefully verified. It's not acceptable to report flooding F^e 5 - Stage at Omnias situations just because this table is not updated. The real time modelling shell has been able to maintain it automatically and overcome data insufficiencies with minimum simulation results degradation. Nevertheless results quality differ as input data availability varies. This may constitute a problem as common users are not able to evaluate model performance and just have to decide to accept the results or not. Until now self validation of simulations is not implemented but automatic information on late model performance would be an interesting tool for all users. After mathematical instability or periods of inactivity, the system takes a few time to recover and to produce reliable results again. This time is dependent on the general state of the system (registered flows and data availability) and on the number of simulations performed per hour. This adaptation tofielddata can take up to 5 or 6 simulations. 6 Conclusions Simulations results show good stage and flow simulation was achieved. It was also evident that reliable results are always limited in time which can be defined as the wave propagation time from the most upstream section to the analysed section. This is due to the non-availability of field data, which was not yet registered. For example, model is not able to predict how dam discharges will vary and therefore simulations for the upstream river reach first 500 meters is only valid for short time period. Results have shown that modelling may be a useful flood prevention tool and have helped detecting areas where topographic surveys require updating. During the last two years, it proved to be

10 44 Hydraulic Engineering Software a useful tool for the flood management. In particular, it gave support to a coordinated action between Portuguese and Spanish authorities in regard to the dam operation during the flood period. Displaying immediate simulation data to decision makers may effectively help flood situations managing as preventing measures can be implemented much sooner. However it is necessary to create tools for evaluating results as most users are not familiar with modelling limitations. Calibration of a real time model application cannot be performed as a simple model application. Special attention has to be given to input data filtering and processing, as an automatic system must be programmed to carefully evaluate available data and be able to replace data gaps. This procedures must also be calibrated, as modelling performance is highly dependent on the options taken. Wave propagation time can be independently (off line) calibrated but all procedures should be tested by simulating real time situations. Future developments include an indication of model performance in previous model applications to other Portuguese rivers where flooding requires particular attention. 7 References [1] PRICE, R. K, "A mathematical model for river flows", I - Theoretical development, Report n. INT 127, Hydraulics Research Station, Wallington England, December 1975, revised September [2] KINSLEY, R. K. Jr., Max A. Kohler and Joseph L. H. Paulhus; Hydrology for Engineers, McGraw-Hill, London, [3] INAG, "O sistema de vigilancia e alerta de cheias", Direc^ao de Services de Recursos Hidricos, Institute da Agua, Lisboa, Mar^o [4] RODRIGUES, R. (1994a) "Algumas consideragoes sobre as cheias do Tejo em Portugal e a influencia das albufeiras em Espanha. In, T Congresso da Agua, Vol. 2, APRH, p.ii-9 a 11-19, Lisboa. [5] SITTER, W. T., and K. M. Krouse: Improvement of Hydrologic simulation by Utilising Observed Discharge as an indirect input (Computed Hydrograph Adjustment Technique - CHAT), NOAA Tech. Memo. NWS Hydro-38, February 1979.