s - Universiti Teknologi Malaysia
Transcription
s - Universiti Teknologi Malaysia
DEVELOPMENT OF FLOOD FORECASTING MODEL FOR SUNGAI KELANTAN (TANK MODEL) CHONG CHEW HOONG A report submitted in partial fulfillment of the requirements for the award of the degree of Bachelors Degree in Civil Engineering Faculty of Civil Engineering Universiti Teknologi Malaysia APRIL 2007 iii To the two men who have made me who I am today. iv ACKNOWLEDGEMENT My deepest gratitude goes to my supervisor, Associate Professor Dr Ahmad Khairi Abdul Wahab, for being consistently supportive and enthusiastic throughout the course of this project. His enthusiasm has been a strong motivation to the project team. I would also like to express my sincere gratitude to my father, Ir Chong Sun Fatt, and Mr. Cheok Hou Seng for giving me the inspiration to initiate this project, providing me with guidance and ideas to further enhance the Sungai Kelantan flood forecasting model. My sincere gratitude also goes to Ir. Low Koon Sing and all the personnel of the Jabatan Pengairan dan Saliran, Malaysia for their assistance in the data acquisition process, as well as providing me with valuable suggestions that helped to improve the features of the model. I would also like to thank Mr. Lee Jin Ming and Mr. Ang Zhili, who have collaborated enthusiastically in the development of the Sungai Kelantan flood forecasting model. Lastly I would like to thank my family and friends for their support and faith in me. v ABSTRACT The state of Kelantan is situated at the east coast of the Peninsular Malaysia, which is subjected to the north-east monsoon, and major flooding caused by the overflow of Sungai Kelantan and Sungai Golok has been occurring on a yearly basis typically during the months of November and December. A reliable flood forecasting system will be able to provide early warning to local authorities so that appropriate measures can be taken to protect lives and properties. With this objective, a flood forecasting model is developed for Sungai Kelantan in this study. The model is divided into two components: hydrological and hydraulic model. Tank model by M. Sugawara is adopted as the hydrological model and it is built using Microsoft Excel with Visual Basic for Application (VBA). After calibration and verification of model using four sets of storm data, the model is able to provide reliable forecast with an average model efficiency of 96 % and mean absolute error of 0.33 m for twelve hours forecasts. vi ABSTRAK Negeri Kelantan terletak di kawasan pantai timur semenanjung Malaysia yang tertakluk kepada pengaruh monsun timur-laut, dan setiap tahun berlakunya banjir utama yang disebabkan oleh kelimpahan Sungai Kelantan dan Sungai Golok, terutamanya semasa bulan November dan Disember. Dengan adanya satu model ramalan banjir yang dapat memberikan amaran awal, ia akan membantu pihak berkuasa tempatan dalam mengambil tindakan untuk menyelamatkan nyawa dan harta benda. Maka dalam kajian ini, satu model ramalan banjir telah dibina untuk Sungai Kelantan. Model ini terbahagi kepada dua komponen: model hidrologi dan hidraulik. Model tangki yang dicipta oleh M. Sugawara telah dipilih sebagai model hidrologi and ia dibina menggunakan MS Excel dan aturcara Visual Basic for Application (VBA). Selepas menjalani proses kalibrasi and verifikasi model dengan menggunakan empat set data hujan, adalah didapati bahawa model yang dibangunkan ini mampu memberikan ramalan paras air dengan keberkesanan model (ME) sebanyak 96 % dan purata perbezaan mutlak sebanayak 0.33 m untuk ramalan 12 jam. vii TABLE OF CONTENTS CHAPTER 1 2 TITLE PAGE DECLARATION ii DEDICATION iii ACKNOWLEDGEMENT iv ABSTRACT v ABSTRAK vi TABLE OF CONTENT vii LIST OF TABLES x LIST OF FIGURES xi LIST OF SYMBOLS/ABBREVIATIONS xiii LIST OF APPENDICES xiv INTRODUCTION 1.1 Background of Study 1 1.2 Statement of Problem 3 1.3 Objective of Study 4 1.4 Scope and Limitation of Study 4 1.5 Significance of Study 5 LITERATURE REVIEW 2.1 Introduction 6 2.2 Types of Model Commonly Used in 7 Malaysia 2.2.1 Unit Hydrograph 8 viii TABLE OF CONTENTS CHAPTER TITLE PAGE 2.2.2 Tank Model 9 2.2.3 10 Sacramento Soil Moisture Accounting and Routing 2.3 Hydrological Model Selection 11 2.4 Software Programming 12 2.5.1 Microsoft Excel 12 2.5.2 Microsoft Visual Basic for 12 Application 3 HYDROLOGICAL MODEL: TANK MODEL 14 3.1 Introduction 14 3.2 Computation Process within a Tank 17 3.3 Computation of Total Discharge and 18 Water Level 3.4 4 Autoregressive Model 18 RESEARCH METHODOLOGY 20 4.1 Research Design 20 4.2 Background of the Sungai Kelantan River 22 Basin 4.3 Data Acquisition and Analysis 24 4.3.1 27 Missing Rainfall Data 4.4 Structure of Model 28 4.5 Development of Tank Model 32 4.6 Model Calibration and Verification 34 4.6.1 34 Model Calibration Process ix TABLE OF CONTENTS CHAPTER TITLE 4.7 5 6 PAGE 4.6.2 Model Performance Evaluation 36 4.6.3 Model Verification 37 Model Integration 37 RESULT ANALYSIS 39 5.1 Tank Model Calibration Results 39 5.2 Significance of Autoregressive Model 42 5.3 Tank Model Verification 45 CONCLUSION 47 6.1 Model Performance 47 6.2 Limitations of the Model 48 6.2.1 Model Structure 49 6.2.2 Hydrological Data 49 6.3 Recommendations REFERENCES 50 52 x LIST OF TABLES TABLE NO. TITLE PAGE 2.1 Operational Flood Forecasting Models of JPS (2004) 7 3.1 Computation of AR coefficients and adjusted forecast 19 values 4.1 Rainfall stations with its corresponding replacement 27 stations 5.1 Calibrated parameters for the Sungai Kelantan Tank model 40 5.2 Result of model performance evaluation 40 5.3 Model efficiency for three historical storm events used for 43 model calibration 5.4 Mean absolute error for three historical storm events used 43 for model calibration 5.5 Tank model verification results using year 2003 storm 45 event (23/11/2003 – 27/12/2003) 5.6 Tank model verification results using year 2004 storm event (27/11/2004 – 27/12/2004) 45 xi LIST OF FIGURES FIGURE NO. 1.1 TITLE PAGE Flooding at Tambatan Diraja, Kelantan on 18 December 2 2005 2.1 Concept of the Sacramento model to simulate the 11 discharge hydrograph 3.1 Schematic Diagram of the Structure of Tank Model 15 4.1 Project Flow of the Development of the Flood 21 Forecasting Model for Sungai Kelantan 4.2 Main Tributaries of Sungai Kelantan 23 4.3 Thiessen Polygon for Sungai Kelantan Catchment 25 4.4 Rating curve for Sungai Kelantan at Guillemard Bridge 26 (1993) 4.5 Schematic diagram of the structure of the flood 28 forecasting model for Sungai Kelantan 4.6 Screen shot showing the Tank Model setup worksheet 30 4.7 Screen shot showing the rainfall data input and 31 processing worksheet 4.8 Screen shot showing the main forecasting worksheet 32 4.9 Procedure within the Tank model 33 4.10 Graph showing the comparison between individual tank 36 discharge and total discharge xii LIST OF FIGURES FIGURE NO. TITLE PAGE 5.1 Observed and forecasted discharge for 1988 storm event 40 5.2 Observed and forecasted discharge for 1993 storm event 41 5.3 Observed and forecasted discharge for 2001 storm event 41 5.4 Comparison between observed and forecasted discharge 43 (with and without correction) for 1988 storm event 5.5 Comparison between observed and forecasted discharge 44 (with and without correction) for 1993 storm event 5.6 Comparison between observed and forecasted discharge 44 (with and without correction) for 2001 storm event 5.7 Comparison between observed and forecasted discharge 46 (with and without correction) for 2003 storm event 5.8 Comparison between observed and forecasted discharge (with and without correction) for 2004 storm event 46 xiii LIST OF SYMBOLS/ABBREVIATIONS JPS - Jabatan Pengairan dan Saliran, Malaysia DOS - Disk Operating System PC - Personal Computer FFC - Flood Forecasting Center UH - Unit Hydrograph VBA - Visual Basic for Application RF - Rainfall WL - Water Level EV - Evaporation TS - Tank Storage Q - Discharge AR - Autoregressive ObsQ - Observed Discharge SimQ - Simulated Discharge Res - Error between observed and simulated discharge FCQ - Adjusted Forecasted Discharge GUI - Graphical User Interface TIDEDA - Time Dependant Data MS Excel - Microsoft Excel 1-D - One Dimensional ME - Model Efficiency MAE - Mean Absolute Error xiv LIST OF APPENDICES APPENDIX A TITLE Graphs showing comparison between individual PAGE 53 tank discharge and total discharge B Detail Calculation of the Tank Model in Simulating the Discharge and Water Level for the year 2004 Storm Event (27/11/2004 – 27/12/2004) 55 CHAPTER I INTRODUCTION 1.1 Background of Study Malaysia has a tropical climate and is endowed with abundant rainfall. Peninsular Malaysia receives an average of 2500 mm rainfall annually. With the heavy rainfall, flood has become the most severe natural disasters that causes great amount of socio-economic losses. An estimated 9 % of Malaysia’s land area, which amounts to 29,000 km2, is flood prone. The average annual flood damage was estimated to be RM 1 billion 1 (year 2000 pricing) and about 22 % of our nation’s population is affected by flood. The state of Kelantan covers a total land area of 15,022 km2, which is equivalent to 4.4 % of Malaysia’s land area. The population, according to the statistic revealed by the Department of Statistics Malaysia in the year 2000, is close to 1.313 million people. Kelantan is situated at the northeast corner of Peninsular Malaysia facing the South China Sea. It is subject to the north-east monsoon, and flooding has been occurring on a yearly basis typically during the months of November and December. For the river basin of Sungai Kelantan itself, a total of 1 http://agrolink.moa.my/did/river/r_fcom.html 2 RM 12.1 2 millions was reported as direct flood damage for year 2005, and three people were killed. Figure 1.1 Flooding at Tambatan Diraja, Kelantan on 18 December 2005 The severe losses have called for a need to develop an effective flood control strategy. The control measures suggested by the Flood Commission Committee are generally divided into categories of structural and non-structural measures. Proposed structural measures include the implementation of flood mitigation projects, such as construction of dams, river diversion, construction of river levee and embankment, widening and deepening of river channel, and flood retention using mining ponds and lakes. Meanwhile there are two types of non-structural measures being proposed; the development of planning control and the provision of flood forecasting and warning system. Flood forecasting and warning system is a cost-effective way to significantly reduce the damage of flood occurrence, especially to urban areas with high-density of population. Early warning will provide adequate lead time for the residents to be mobilized to safer places, help them to save lives and reduce damages to properties. 2 Laporan Banjir 2005/2006. Bahagian Hidrologi, Jabatan Pengairan dan Saliran Kelantan. 3 In the subject of this research, the river basin of Sungai Kelantan includes a few major population centres that are threatened by the overspills of Sungai Kelantan, such as Kota Bharu, Tanah Merah, Machang and Kuala Krai. A total of 3209 people were evacuated in these areas during the year 2005 floods. It was suggested in the flood report for year 2005, that there is a need to upgrade the current flood warning system. By installing more flood warning sirens, the authorities will be able to warn the public of possible flood threat. But in order for the siren system to provide early warning, the system will require a flood forecasting system that can accurately predicts water level in the near future. 1.2 Statement of Problem The current DOS version of Tank Model for Sungai Kelantan was developed by Jabatan Pengairan dan Saliran (JPS) for flood forecasting purposes in the early 1980s. After more than 25 years in operation, the performance of the Sg Kelantan Tank Model has deteriorated primarily because of changes in the catchment land-use and river characteristics. However its source code is not accessible to users and it does not allow model re-calibration. This research aims to develop a similar hydrological model, yet it is more user-friendly and allows re-calibration so that the model can be updated in order to respond to current developments in the catchment area. The current Tank Model can only provide flood forecast of Sg Kelantan at Guillemard Bridge. It can not provide forecast for important population centre such as Kota Bahru because it is subject to tidal effect. A hydraulic model is also being developed to consider the tidal effect so that flood forecast can be provided for tidal reach of the river. This part of the research is conducted concurrently by Lee J.M. (2007) and Ang (2007). 4 1.3 Objectives of Study The main objectives of the project are: 1. To develop and calibrate a hydrological model to simulate discharge hydrograph of Sungai Kelantan at Guillemard Bridge. 2. To develop a PC-based flood forecasting model, combining with the hydrological and hydraulic models to simulate the water level profile at Guillemard Bridge and Jeti Kastam (Kota Bahru) by taking into considerations the tidal effects. 1.4 Scope and Limitation of Study 1. To obtain and process relevant hydrological data for several historical storms which will be used for calibration and verification purposes of the hydrological model. 2. To develop, calibrate and verify the performance of the hydrological model using historical flood events. 3. To develop a flood forecasting model that integrates both the hydrological model and hydraulic model into a complete model. 4. The main limitation of the study is the availability of reliable data. The performance of the model is highly dependent upon reliable and timely data, and the availability of information on catchment and river characteristics. 5 1.5 Significance of Study The flood forecasting model adopted by JPS currently is a DOS version of Sugawara’s Tank Model. The model parameters could not be updated as the users are unable to edit the source code of the model. This has also caused inconvenience for the users to understand the underlying theory of the model. Therefore it is proposed in this study to re-develop the flood forecasting model using Microsoft Excel. Another significant improvement of the integrated model produced at the end of the study is the hydraulic model, which is a combination of St. Venant’s unsteady flow and Standard Step model that is able to include tidal effect to the prediction of water level profile at Jeti Kastam. The integration of both hydrological and hydraulic model is expected to be able to produce better forecast results. The end product of this study should be a cost-effective and user friendly flood forecasting model, that will aid the JPS officers to accurately predict the flood levels at designated points along the river. Using the predicted water level, the local authorities will be well informed of the possible time and severity of flood occurrence. CHAPTER II LITERATURE REVIEW 2.1 Introduction The literature review in this study is focused on two major components. It will be discussing on the fundamentals of the modeling technique and the programming technique applied in this study. Modeling technique refers to the hydrological and hydraulics modeling that will generate forecast of water level profile. Programming technique will discuss on how one can utilize several software that are readily available in most personal computers, in order to develop the flood forecasting system. This chapter serves two main purposes: one is to summarize the information gathered throughout the process of reviewing literature, including the various types of flood forecasting models, and to justify on the selection of modeling and programming techniques. Selecting and developing a model where its fundamental theories are popular among the flood forecasters can be done by understanding the flood modeling trend in Malaysia. During recent years, Japan has been working towards the uniformity of modeling techniques in order to minimize the effort needed in operating and maintaining the flood forecasting models. 7 2.2 Types of Model Commonly Used In Malaysia A flood forecasting model is a type of hydrologic model that is developed specifically to simulate catchment responses to precipitation, and generate forecasts of the water level and streamflow. Generally flood forecasting models can be categorized into three types: 1. Distributed physics-based models 2. Lumped conceptual models (e.g. Tank, Sacramento) 3. Black box models (e.g. Unit Hydrograph, Stage Regression) The Flood Forecasting Center (FFC) of JPS has developed operational flood forecasting models for a number of important river systems in the country. The models developed by FFC were handed over to the State JPS for operation. Currently operational forecasting models have been developed for 12 river basins providing water level prediction at 21 forecasting points (see Table 1). Table 2.1: Operational Flood Forecasting Models of JPS (2004) 3 No. 1 2 3 Muda River Perak River Muar River 4 5 6 7 Johor River Besut River Golok River Pahang River 8 9 Kuantan River Kelantan River 10 11 Sadong River Kinabatangan River Klang River 12 3 River Basin No. of Forecasting Model Forecasting Point 1 Stage Regression 2 Stage Regression 2 Discharge-Correlation Model & Tank Model 2 Regression Model 1 Stage Regression 1 Stage Regression 3 Linear Transfer Function and Stage Regression (back-up) 1 Tank Model 2 Tank Model and Stage Regression (back-up) 1 Linear Transfer Function 1 Linear Transfer Function 4 Flood Watch Ir. Chong Sun Fatt, Cheok Hou Seng, “Development of a Digital Database and Information System for the Six Selected Flood Forecasting Systems of JPS Malaysia (Final Report)”, 2006. 8 The Stage Regression model is being adopted as the flood forecasting model for Sungai Muda, Sungai Perak, Sungai Besut, and Sungai Golok. It features an equation that predicts the water level of a downstream point using the current water level of an upstream station. For example, the Stage Regression model for Sungai Golok uses the water level at Jenob to predict the water level at Rantau Panjang with an 8 hours lead time. This model is purely mathematical and statistical, using historical storm events to obtain the best fit equation. 2.2.1 Unit Hydrograph Unit Hydrograph method (UH) can be defined as the hydrograph resulting from unit depth of surface runoff generated by a storm of uniform intensity and specified duration. It is usually being categorized as a type of black box model. Unit hydrograph is generally derived from streamflow data and estimates of temporal distribution of rainfall excess, or by synthesis using catchment characteristics data. There are 2 important assumptions made in the UH model: (a) Catchment linearity – the proportion of inflow and outflow of the system remains constant, regardless of the volume of inflow. (b) Catchment as a lump system – the input rainfall excess is assumed to be uniform over the entire catchment area. The input rainfall excess for the model is a single rainfall excess hyetograph. Advantages of the Unit Hydrograph method can be summarized as follow: (a) The model is simple and easy to understand; (b) It is considerably accurate especially for storms with high intensities over short durations; (c) The assumption of linearity is acceptable for large events in most catchments; (d) The model is based on the integrated response of the catchment, and it does not require assumptions regarding the spatial variation of rainfall and losses, and of storage routing effects; 9 Disadvantages of the Unit Hydrograph method are as follows: (a) The model is not suitable for multiple period storms of long duration; (b) It is not suitable for catchments with non-linear behaviors, for example catchments with large reservoirs; (c) The model cannot provide accurate flood predictions when spatial distribution of rainfall excess has very low uniformity; 2.2.2 Tank Model The Tank model is categorized as one of the lumped conceptual models and it consists of several tanks that erected vertically on top of one another. Each tank represents a layer of soil, from the surface soil layer to the bottom soil layer above the impervious rock layer. Storage volume in each tank represents the moisture storage in each layer of the soil. The discharge of the tank outlets are equivalent to the discharge to the river flow contributed by the respective soil layer. Several advantages of the Tank model are: (a) The model is suitable for a great variety of catchment size; however the model efficiency does increase when the catchment area decreases4 . (b) It has a small number of parameters compared to other conceptual rainfall-runoff models, and the calibration process is relatively easy once the operator understands how the parameters affect the hydrograph simulation; (c) 4 The model calculation only involves simple equations; Ng, B.C. (2003). Ramalan Banjir Bagi Sungai Kuantan Dengan Menggunakan Model Tangki. Skudai: Universiti Teknologi Malaysia: Thesis Sarjana Muda. 10 (d) The model is able to provide accurate forecast results for a considerably longer lead time (for the current model for Sungai Kelantan, the lead time ranges from 6 to 24 hours); (e) 2.2.3 The users can easily understand the structure of the model. Sacramento Soil Moisture Accounting and Routing This model was originally adopted for the flood forecasting system of Sungai Kelantan, but it was being replaced by the Tank model in year 1981. The Sacramento model can be defined as a spatially-lumped continuous soil moisture accounting model. It is an ideal model to simulate large catchments that covers an area of more than 1000 km2. The model takes into account of all water entering, stored in and leaving the river basin. Though many parameters are used in the water balance accounting process, precipitation contributes the most impact to the runoff. The model consists of 3 main components: the rainfall generation component, the soil water mass balance component, and the kinematic routing component for channel flow propagation and attenuation. The model calibration can be done by adjusting the baseflow, tension water capacities and runoff simulation parameters. The model requires point or areal estimates of historical precipitation, potential evaporation and temperature as main input. Secondary input data include the catchment topography, soil characteristics, and location of important features such as reservoirs and river junctions. 11 Figure 2.1 Concept of the Sacramento model to simulate the discharge h hydrograph 2.3 Hydrological Model Selection Choosing the right type of model is essential to meet the objectives of the study. Therefore it is relevant to outline the criteria to be adopted in the selection process: (a) The model is able to generate discharge and water level forecast with reasonable accuracy and acceptable lead time; (b) The model should be simple and easy to programme; (c) The model is suitable for the type of catchment area that the Sungai Kelantan river basin is having; (d) The model can be adjusted and calibrated easily when more observed flood events are available in the future; (e) The operators can easily understand and comprehend the model. 12 The Tank model is chosen to be the hydrological model to simulate discharge in this study. Both Unit Hydrograph method and Tank model satisfy all the criteria as listed above. But, considering that the available hydrological data is not complete and reliable, adopting a conceptual model will be more suitable because it does not entirely depend on historical data. 2.4 Software Programming The development of the hydrological model as well as the complete flood forecasting system requires the utilization of various types of software developing and programming tools. The following sub-chapters will briefly discuss on the programming machines involved, its basic features and uniqueness that will contribute to the effectiveness and user-friendliness of the flood forecasting system. 2.4.1 Microsoft Excel Microsoft Excel is a spreadsheet program that is bundled with the Microsoft Office package, which is available in most personal computers that operate on the Windows Operating System. It features a useful collection of calculating modules as well as extensive graphing abilities. The advantages of using Microsoft Excel as the underlying programming tool are: a) The user interface in Microsoft Excel allows user to understand the logical progression and calculations involved in the model. Modules can be separated in different sheets, and data can be transferred from one sheet to another easily. 13 b) It allows user to re-calibrate the model parameters without the need to edit source codes. Instead of going through lines of source codes that requires the user to be well versed in computer programming, the user can perform the calibration by simply replacing the parameter values in cells. Instantaneous hydrograph plotting will allow user to compare between the forecasted and observed values and determine the new parameters’ performance. Thus the JPS officers will be able to maintain the system without relying on external IT expertise. c) Microsoft Excel is available on almost every computer that runs on the Microsoft Windows operating system. The user will not need to purchase additional software in order to operate the system. 2.4.2 Microsoft Visual Basic for Application Visual Basic for Application (VBA), which is also known as macro, is an event driven programming language that is built into most of the Microsoft Office applications, and partially implemented into several third party software such as AutoCAD and ESRI ArcGIS. It can be used to control almost every aspect of the host application, such as manipulation of user interface, customization of user forms and dialog boxes. VBA is also able to control one application from another, using the Object Linking and Embedding (OLE) Automation. CHAPTER III HYDROLOGICAL MODEL: TANK MODEL 3.1 Introduction The Tank model was founded by M. Sugawara in year 1979, and the model was widely used in Japan, Korea and other Asian countries for purposes of flood forecasting, reservoir and catchment flow simulation. In Malaysia, the Tank model has been adopted as the flood forecasting model for Sungai Kelantan (1981) and Sungai Kuantan (2004). When rainfall occurs, the precipitation enters directly into the top tank. Most of its volume will be discharged into the river, causing the water level to rise in direct respond to the rainfall. Part of the volume in the top tank will then infiltrate to the second tank, just as how surface runoff would infiltrate through the soil layer in reality. Another part of it will be evaporated back into the atmosphere. In the second tank, water will vertically infiltrate into the third tank, and at the same time horizontally infiltrate and form discharge that goes into the river. The discharge from the base tank represents base flow. 15 RF EV C1 TS1 X1 Tank 1 X2 C2 Q1 = C1 * (TS1 – X1) Q2 = C2 * (TS1 – X2) C3 TS2 Tank 2 C4 Q3 = C4 * (TS2 – X3) C6 Q4 = C6 * (TS3 - X4) X3 C5 TS3 Tank 3 X4 C7 C8 TS4 Figure 3.1 Tank 4 X5 C9 Q5 = C8 * (TS4 - X5) Q6 = C9 * (TS4) Schematic Diagram of the Structure of Tank Model There are four types of model parameters in a Tank Model, as shown in Figure 2.2. They are: a) Tank Storage (TS) Indicates the amount of water stored in the tank in mm. Its initial value for each tank is obtained from the calibration process, and it changes along the modeling process. The initial tank storage indicates the moisture content of the soil layers at the beginning point for the forecast. The value increases when water enters the tank in the form of rainfall or infiltration, and 16 decreases when water leaves the tank through infiltration to lower tank or discharge through the outlet. b) Outlet Coefficient (C) This parameter indicates the outlet efficiency in letting water to flow through it. Thus outlets (including discharge and infiltration) with higher coefficient value will generate more discharge or infiltration, depending on the location of the outlet. This set of parameter is obtained from calibration, and it will remain constant throughout the modeling process, until it is being modified in future re-calibrations. c) Outlet Height (X) This is the vertical distance between the outlet and the base of tank. It may be interpreted as the storage capacity of the soil layers. The higher the outlet height is, the higher the tank storage is needed to generate discharge. d) Evaporation (EV) This parameter represents the amount of water in mm that is being evaporated from the top tank. In this study, the lower tanks are assumed to have no evaporation as the evaporation rate in the lower soil layers is negligible. 17 3.2 Computation Process within a Tank The tank starts off with its initial tank storage value which is obtained from the trial and error calibration process. Computation will start by adding rainfall (RF) and deducting the evaporation (EV) from the initial tank storage (TS), in order to obtain a new tank storage value, named as TS(a). Step 1: TS(a) = Initial TS + RF – EV The next event to consider will be the infiltration (Infil), followed by the discharge of the lower outlet (Q2), and lastly the upper outlet (Q1). Water is assumed to infiltrate first, then only being discharged by the outlets; though in reality, these movements occur simultaneously. Step 2: Infil = C3 * TS(a) Then deduct the amount of water being infiltrated from the tank storage. Step 3: TS(b) = TS(a) – Infil Next is to compute the discharge of the lower outlet. Step 4: Q2 = C2 * (TS(b) – X2) Deduct the amount of water being discharged from the tank storage. Step 5: TS(c) = TS(b) – Q2 Then compute the discharge of the upper outlet, if TS(c) is greater than X1. Step 6: Q1 = C1 * (TS(c) – X1) Finally calculate the final tank storage. This value will be used as the initial tank storage for the next time step. Step 7: TS(d) = TS(c) – Q1 18 3.3 Computation of Total Discharge and Water Level The summation of all the tank outlet discharges forms the total discharge Q (in mm), which is a depth value that symbolizes a thin layer of water covering the entire catchment area. In order to convert this symbolical layer of water into volume of discharge, the following equation is used: ( ) Total Discharge m 3 / s = Q(mm )× Catchment Area(km 2 ) 3600(s ) (3.1) Conversion to water level is done by interpolating the stage-discharge rating curve to obtain the corresponding water level for the discharge volume. Otherwise the conversion can also be done using the rating curve equation derived from the data. 3.4 Autoregressive Model Autoregressive (AR) is a correctional model that is included in the Sungai Kelantan Tank model in order to improve the forecasted discharge and water level simulation results. AR model is a group of linear prediction formulas used to predict an output of a system which only depends on the previous outputs. The name autoregressive comes from the fact that predicted value is regressed on the past value of itself. In this case, the value that is being predicted using the AR model is the error, not the discharge itself. The predicted error is then used to adjust the forecasted discharge value. Detailed calculation is demonstrated in Table 3.1, where the bold numbers in the table are the adjusted forecast values of water level. Table 3.1 Time ObsQi SimQi Resi Computation of AR coefficients and adjusted forecast values Lag1 Lag2 Lag3 Lag4 Step, i 1 9.30 11.06 -1.76 2 9.30 11.14 -1.84 -1.76 3 9.30 11.22 -1.92 -1.84 -1.76 4 9.30 11.30 -2.00 -1.92 -1.84 -1.76 5 9.30 11.30 -2.00 -2.00 -1.92 -1.84 -2.00 -2.00 -2.00 6 11.30 7 11.22 8 11.22 9 11.30 j FCQ1+i FCQ2+i FCQ3+i FCQ4+i (AR1) (AR2) (AR3) (AR4) 9.38 Plot graph Res against Lag. The slope is the ARi coefficient. 9.38 9.47 9.38 9.47 9.55 -1.76 9.30 9.39 9.48 9.57 -1.92 -1.84 9.30 9.31 9.40 9.49 -2.00 -1.92 9.23 9.24 9.33 -2.00 -2.00 9.24 9.25 -2.00 = Number of time steps ahead of i 9.33 AR coefficients used ObsQi = Observed discharge value for time step i in this example: SimQi = Simulated discharge value for time step i AR1 = 0.9988 Resi = Error between observed and simulated discharge = ObsQi - SimQi AR2 = 0.9963 ARj = Least square correlation (slope) between Res and Lagj (AR Coefficient) AR3 = 0.9918 FCQi+j = Adjusted forecasted discharge value (refer to blue boxes) AR4 = 0.9855 = SimQi+j + (Resi * ARj) = SimQi+j + (Lagi+j * ARj) 19 CHAPTER IV RESEARCH METHODOLOGY 4.1 Research Design The development of the flood forecasting model for Sungai Kelantan involves hydrological and hydraulics modeling, as well as software programming works. In this study, only the hydrological model is covered while the hydraulic model is jointly developed by Lee J.M. (2007) and Ang (2007). The hydrological model is built to generate simulation of discharge based on rainfall data for the upstream area of the interest point, while the hydraulic model simulates the tidal effect towards the water level profile of the river. The model integration process is a joint effort to produce a complete flood forecasting model. The process of developing the system for Sungai Kelantan can be divided into several stages so that the project progress can be monitored. The general flow for the project is shown in Figure 3.1, and the proposed work schedule is attached in Appendix A. The project is generally divided into four major phases: model development, calibration, model integration and verification. First phase of the project involved tasks such as hydrological data collection and processing, catchment characteristics identification, and development of the Tank model. The second phase, which is the calibration process, was the effort of fine tuning the model parameters for optimum 21 performance. In this phase, three past storm events are identified as the basis of calibration, and the model’s performance is evaluated using mean absolute error and model efficiency. In the third phase, both hydrological and hydraulic model are integrated to form a complete flood forecasting model that will consider effects of both the rainfall-runoff and tidal effect. Model integration focused mostly on programming and development of graphical user interface (GUI). The fourth phase, model verification, involved the evaluation of the Tank model by itself as well as the complete model. The latest storm event is used for the verification and the model performance will be evaluated. Data Collection Development of Hydrological Model Development of Hydraulic Model Model Calibration and Verification Model Integration Test Run Figure 4.1 Project Flow of the Development of the Flood Forecasting Model for Sungai Kelantan 22 4.2 Background of the Sungai Kelantan River Basin Sg Kelantan is the third largest river system in Peninsular Malaysia. It originates from mountain ranges with elevation exceeding +2,000m. At Guillemard Bridge, the point of interest for the hydrological model development in this study, the catchment area is about 12,000 sq km5 . There are two main tributaries contributing to Sungai Kelantan: Sungai Galas and Sungai Lebir. Sungai Kelantan itself does not have any major tributary, but Sungai Galas and Sungai Lebir are fed by many tributaries and their flow volume has significant impact to the flow of Sungai Kelantan. Among the major tributaries are Sungai Nenggiri, Sungai Aring, Sungai Chiku and Sungai Pergau, where a hydroelectric dam is built. The major tributaries as well as the main river of Sungai Kelantan are as shown in Figure 4.2. As the tributaries are denoted in different colours, the significance of discharge contribution by every tributary can be estimated by observing the area covered by the network of river branches. The river flows past several major towns, such as Pasir Mas, Kuala Krai and Kota Bharu. Sungai Kelantan overspills on a regular basis, especially during months of November till February, where the north-east monsoon brings along heavy downpour. For the year of 2004, JPS has recorded a total of 10476 evacuees and a reported flood damage of RM 14.3 millions 6 . 5 Sepakat Setia Perunding Sdn Bhd and SMHB Sdn Bhd. (1999). Final Report: Kelantan River Flood Mitigation Plan Feasibility Study. 6 Laporan Banjir 2004/2005. Bahagian Hidrologi, Jabatan Pengairan dan Saliran Kelantan. 23 Figure 4.2 Main Tributaries of Sungai Kelantan 24 4.3 Data Acquisition and Analysis The performance of the flood forecasting model is significantly limited by the availability of hydrological data in terms of its quality as well as quantity. Data have been acquired from JPS, which included rainfall and water level data for all stations situated in the state of Kelantan. The data are then filtered, where stations situated outside of the Sungai Kelantan catchment area will be eliminated. In this case, most of the stations situated outside of Sungai Kelantan catchment belong to the Sungai Golok catchment. Stations with unreliable data and significant amount of missing data are also being discarded. The water level data used in this model is taken from the Sungai Kelantan at Guillemard Bridge station (No. 5721442). Several rainfall stations are then selected to be the contributing rainfall gauges for the Tank model. Selection was based on data availability for the selected storm events, percentage of missing data, and the location of stations for accurate representation of the entire catchment area. These selected stations are all telemetric stations except for JPS Machang, Gob and Blau. The rainfall stations that are required to provide input to the Sungai Kelantan Tank model ought to be telemetric stations so that real-time data could be acquired for forecasting operations. Based on preliminary evaluation on data availability, eight rainfall stations that are strategically scattered around the catchment area are selected to represent the entire catchment area. These stations are: 1. JPS Machang (5722057) 2. JPS Kuala Krai (5522047) 3. Dabong (5320038) 4. Gob (5216001) 5. Kg Aring (4923001) 6. Gua Musang (4819027) 7. Gunung Gagau (4726001) 8. Blau (4717001) 25 Processed data are stored in an MS Excel file that has a predefined format. This is to ensure that the data extraction macro will be able to identify the start and end of the data correctly. Users are required to do some preliminary data editing, such as to convert the TIDEDA data from Notepad file into an MS Excel workbook, and also to copy and paste the data for the storm duration needed into a formatted sheet with a designated starting cell. Both rainfall and water level data used in this model are hourly data. It is crucial that the users are aware that the storm data used in the model must share the same start and end time. Once the complete storm data is ready, an average rainfall data set is calculated using the Thiessen Polygon method. The polygon is as shown in Figure 3.2. Figure 4.3 Thiessen Polygon for Sungai Kelantan Catchment 26 Evaporation rate is a necessary parameter for model calibration and operation. The daily evaporation value was obtained from JPS Kelantan. In the Sungai Kelantan Tank model, an average value of 4.8 mm/day for non-rainy day has been adopted. Evaporation rate is assumed to be half when the hourly rainfall is greater than 0.5 mm. Besides that, the stage-discharge rating curve for Sungai Kelantan at Guillemard Bridge is another necessary input for the model. Conversion between water level and discharge occurred in the model in both ways. Rating Curve for Sg Kelantan at Guillemard Bridge 25.0 Water Level (m) 20.0 15.0 10.0 4 3 2 y = -4E-14x + 5E-10x - 2E-06x + 0.0049x + 7.6842 5.0 0.0 0.0 1000.0 2000.0 3000.0 4000.0 5000.0 6000.0 3 Discharge (m /s) Figure 4.4 Rating curve for Sungai Kelantan at Guillemard Bridge (1993) 27 4.3.1 Missing Rainfall Data The occurrence of missing data during flood is very common, as telecommunication system used to transmit the data or even the recording station itself was being affected by the flood or the stormy weather. Therefore a sound logical system must be used in order to fill in the missing data with the best representation of what may be the actual situation. For the Sungai Kelantan Tank model, users have the freedom to choose either to fill in the missing data manually, or allow the model to do it for them. The missing data module contained in this model will first try to replace the data with the ones recorded by the nearest station. The replacement stations corresponding to each station are as shown in Table 3.1. For example if Station A has a missing data for a certain time step, Station B is assigned to provide replacement data. In the case where the Station B also missed its data for that time step, Station A will be ignored in the calculation of average rainfall, meaning that the catchment area does not include the area covered by Station A. It is not advisable to carry on the forecast if less than 6 stations contain data. Table 4.1 Rainfall stations with its corresponding replacement stations. Rainfall Station Replacement Station(s) Blau (4717001) Gua Musang (4819027) Gua Musang (4819027) Blau (4717001) Kg Aring (4923001) Kg Aring (4923001) Gua Musang (4819027) Gunung Gagau (4726001) Kg Aring (4923001) JPS Machang (5722057) JPS Kuala Krai (5522047) JPS Kuala Krai (5522047) JPS Machang (5722057) Dabong (5320038) Dabong (5320038) JPS Kuala Krai (5522047) Gob (5216001) Gob (5216001) Dabong (5320038) 28 4.4 Structure of Model The complete flood forecasting model for Sungai Kelantan is a combination of the hydrological Tank model, 1-D hydrodynamic (hydraulic) model and also the tidal model. Tank model requires input of rainfall and water level data to simulate the discharge at Guillemard Bridge; while the 1-D hydrodynamic model consists of the Standard Step 7 module and St. Venant’s unsteady flow 8 module. The 1-D hydrodynamic model requires river channel cross section data during the model setup stage; it uses the discharge value at Guillemard Bridge generated by Tank model as the upstream boundary, and the tidal levels as the downstream boundary to simulate the water profile at Jeti Kastam. In this study, only the Tank model is covered. A schematic diagram of the model structure is shown in Figure 4.4. Input: Rainfall, water level, rating curve, model parameters Tank Model Input: River cross section data Output: Simulated Discharge and Water Level at Guillemard Bridge Tide Model 1-D Hydrodynamic Model Output: Tide Level Standard Step Standard Step & St. Venant’s Output: Simulated Discharge and Water Level at Jeti Kastam Figure 4.5 Schematic diagram of the structure of the flood forecasting model for Sungai Kelantan 7 Ang, Zhili (2007). Development of Flood Forecasting Hydraulic Model for Sungai Johor. Lee, J.M. (2007). Development of Flood Forecasting Model for Sungai Kelantan (1-D Hydrodynamic Model). 8 29 Most of the worksheets (or modules) are hidden from user’s access and only seven worksheets are accessible; and they can generally be categorized into three types: (a) Model setup These worksheets require extensive information input at the initial setup stage of the model. Once all the required information has been inserted, update is only required when new information is available (e.g. new river survey data, new set of harmonic equation constituents for tidal model) or when re-calibration of Tank model is required. The worksheets placed under this category are Tank model, hydraulic model (1-D Hydrodynamic model), tidal model8 and also the rating curve for Sungai Kelantan. (b) Data input The worksheets grouped under this category are the rainfall and water level input sheets and they need to be updated for every storm event. It is recommended that at least fifteen days worth of data should be provided prior to the start of forecasting and the data will be kept for use for the entire storm event. For Sungai Kelantan’s case, most probably these data will be only need to be updated once a year. (c) Forecasting This is the main forecasting worksheet which user will be dealing with every time they wish to generate the water level forecast. This worksheet requires user to provide the latest rainfall and water level data for every time step, which for this model the time step used is one hour. Forecasted hourly rainfall values for the next twelve hours are also required for the water level simulation. Results of the simulation are displayed within this worksheet in both tabular and graphical forms. 30 Some of the main worksheets are shown below along with brief explanation of its functions. The Tank model setup worksheet (Figure 4.6) is designed for the purpose of Tank model setup and calibration. The model parameters are to be modified here. In order to aid the users in the calibration process, this page also shows the model performance evaluation results (ME and MAE), as well as the following graphs: (d) Observed and forecasted discharge (Figure 5.2) (e) Individual tank discharge and total discharge (Figure 3.4) Model Parameters Model Performance Evaluation Results Figure 4.6 Screen shot showing the Tank Model setup worksheet. 31 Thiessen Polygon Weights Figure 4.7 Screen shot showing the rainfall data input and processing worksheet The rainfall data input and processing worksheet (Figure 4.7) is where the user determines the rainfall stations to be used in the model and the Thiessen polygon weights for each station; it also contains a macro that calculates the average rainfall. Currently the model is capable of handling a maximum of eight rainfall stations and it can be customized to suit other rivers by changing the Thiessen weights. The main forecasting worksheet (Figure 4.8) is the main worksheet that the user will be working on for flood forecasting purposes. The users need to insert the current water level data as well as the forecasted rainfall data for the next twelve hours, and these data will become the input to the Tank model to generate the water level forecasts. Forecasted water level will be displayed here along with a graph that shows the projected water level as well as the water level profile for the immediate past three days. 32 Rainfall and Water Level Data Input Result of Water Level Simulation Figure 4.8 4.5 Screen shot showing the main forecasting worksheet Development of Tank Model The model has been built using MS Excel, while Visual Basic for Application (VBA) macros are embedded to execute the operations of the model. All the operations such as computation and data ingestion are executed by the macros. The MS Excel spreadsheets are only used to display the data as well as the outcome of the computation, which means that the cells only contain numbers without any formulation. This is to minimize the risk of any unintended changes being made by users when they are using the model. Another advantage of building the model as such is that the model will be able to work with any amount of data as required by the user, which also means that the model is capable of handling various storm durations as well as various combinations of outlets. Figure 4.9 shows the sequence of procedures within the model. 33 Data Ingest Tank Discharge Calculation (Refer to Section 3.2) Total Discharge in mm and m3/s Convert to water level using rating curve - Autoregressive model (refer to Section 3.4) correction on the simulated water level conversion to corrected discharge Model Performance Evaluation Figure 4.9 Procedure within the Tank model The data ingestion macro is the first step in the Tank model. This macro serves two purposes: to extract rainfall and water level data from its respective worksheet, and to convert the water level data into discharge through interpolation of the rating curve. The model then proceeds on with the determination of evaporation rate for every time step. With this, all the inputs for the model are ready and the next steps are the computation of discharge and the Autoregressive (AR) model. After running the model, the forecast results simulated are then used to evaluate the model performance and both the results of simulation and evaluation are displayed in its respective worksheets. 34 4.6 Model Calibration and Verification The Tank model calibration is a process of trial and error to obtain the best fitting set of model parameters. It is done by comparing the simulated outcome to the actual observed water level through visual inspection with both hydrographs plotted together, and the process involves all selected storm events simultaneously. Apart from that, the model performance is also evaluated through its mean absolute error (MAE) and model efficiency (ME). Then finally the model will be verified using the latest recorded storms (year 2003 and 2004). 4.6.1 Model Calibration Process The model calibration involves the repetitive trial and error process in order to obtain a set of parameters which gives a best fit by comparing between the graphs of simulated and observed discharge for Sungai Kelantan at Guillemard Bridge, as shown in Figure 5.1. For a large catchment such as the one for Sungai Kelantan, its hydrograph are typically very smooth with gradual rising and falling limb, and it generally reacts slower to the rainfall hyetograph compared to smaller catchments. From the graph, we can observe that there is a 2 days gap between the peak rainfall and the peak discharge. We can also observe that the current set of parameters is able to behave as the actual catchment (shape of the hydrograph is similar), but it is yet to be able to provide accurate prediction of the peak flow in terms of the timing as well as the magnitude of flow. In order to assist the calibration process, another graph showing the contribution of each tank towards the total discharge was plotted, as shown in Figure 4.10. From this graph, the main contributing tank at a certain point of time can be identified. And thus we can make suitable adjustments to the relevant parameters to generate better forecast results. For the Sungai Kelantan Tank model, the main 35 contributing tanks are tank No.3 (Q4) and tank No.4 (Q5). Any tank can be selected as the main contributing tank by adjusting the outlet coefficient for the discharge outlet of that particular tank. In this case, Q4 is selected as the main contributing discharge as it generates a discharge profile that is the most similar to the observed hydrograph of the river. Q4 represents the discharge from lower layer of soils while Q5 represents the baseflow. The influence caused by each type of the tank parameters are as explained here: (a) Evaporation (EV) The rate of evaporation will affect on the amount of discharge generated by the first tank as the evaporation rate for lower layers of soil is assumed to be negligible. (b) Tank Storage (TS) The initial values of tank storage represent the moisture content of each layer of soil at the starting of forecast. This value will influence the tank’s discharge at the early stage, as in the later stage, the discharge will be governed by the amount of rainfall (for top tank) or infiltration (other tanks apart of the top tank) entering the tank. (c) Outlet Height (X) This parameter will determine when will the tank start or stop generating discharge. The tank will only generate discharge when the tank storage exceeded the outlet height. When the tank storage falls below the outlet height, the remaining water in the tank will only be leaving the tank through infiltration, until the tank becomes empty. Therefore a higher outlet height will cause a tank to generate more infiltration compared to discharge, while a lower outlet height will cause the tank to react fast to flow entering the tank. 36 (d) Outlet Coefficient (C) This coefficient will determine which tanks contribute more to the flow while other tanks contribute less. In this model, the coefficient for Q4 is much bigger compared to other outlets. Individual Tank Discharge and Total Discharge for 1993 Storm Event (17/12/1993 - 6/1/1994) 1.80 6000.0 Q1 1.60 Q2 Q3 5000.0 Q4 Q5 1.20 Total Q 1.00 4000.0 Q6 Obs Q 0.80 3000.0 0.60 2000.0 0.40 0.20 1000.0 0.00 12/16/1993 0:00 -0.20 12/21/1993 0:00 12/26/1993 0:00 12/31/1993 0:00 1/5/1994 0:00 0.0 Date Time Figure 4.10 Graph showing the comparison between individual tank discharge and total discharge. 4.6.2 Model Performance Evaluation The Tank model’s performance is evaluated quantitatively using two approaches: mean absolute error (MAE) and model efficiency (ME). During the process of calibration, these performance indicators are able to signify whether the calibration is progressing towards the right direction or the other way round. The MAE indicates the average of absolute differences between each point of the simulated and observed hydrograph, while the ME compares the deviation between simulated and observed discharge to the average discharge for the storm event. Total Discharge (m3/s) Tank Discharge (mm) 1.40 37 MAE = 1 1 ∑ QObs − QSim n n ∑ (Q ME = 1 − ∑ (Q Obs − QSim ) Obs 4.6.3 (4.1) −Q 2 ) 2 (4.2) Model Verification Model verification is divided into two stages: the Tank model verification is done using the observed water level at Guillemard Bridge; while the the complete model is tested using data recorded at Jeti Kastam. The most recent storm events (year 2003 and 2004) will be used for verification, and the model efficiency and mean absolute error obtained from this process will be concluded as the result of the study. In this report only the result of the Tank model verification will be covered. 4.7 Model Integration The completed Tank model will be integrated with the 1-D hydrodynamic model, which consists of the Standard Step model and the St. Venant’s unsteady flow model, to form a complete flood forecasting model for Sungai Kelantan that will generate water level forecasts for Guillemard Bridge and Jeti Kastam. The model integration focused mostly on efforts to establish the structure of the entire model and the sequence of procedures, debugging the macros as all the worksheets from the four independent models (Tank model, Standard Step, St. Venant’s, Tide) are being combined into one workbook, and developing the user interface. 38 For the simulation of water level at Jeti Kastam, the user has the freedom to choose either to run the simulation using only the Standard Step model, by using the combination of Standard Step and St. Venant’s unsteady flow model. In the latter’s case, the Standard Step model generates the initial routing condition by using the discharge and water level generated by Tank model as the upstream boundary, and the tidal level generated by the tide model as the downstream boundary; discharge is assumed to be constant along the channel. The St. Venant’s model then extracts the output of Standard Step model as its initial input. Therefore another major task in the model integration stage is to establish the connections between the models, so that the output of a model can be transferred as the input of another model (refer to Figure 4.5). CHAPTER V RESULT ANALYSIS 5.1 Tank Model Calibration Results The Sungai Kelantan Tank model is calibrated using three storm events (years 1988, 1993 and 2001). The year 1993 storm event was used for the initial calibration, while further adjustment of the parameters was done using the three storms simultaneously. After going through numerous trial and errors, an optimum set of parameter has been obtained and it is shown in Table 5.1. This set of parameter will be used to simulate the water level profile in the future. But the user should be aware that the model requires re-calibration after several years of use or when the river experiences changes or developments. The observed and forecasted discharge generated using the set of model parameters as listed in Table 5.1 for all three storm events (years 1988, 1993 and 2001) are shown in Figure 5.1, Figure 5.2 and Figure 5.3. The model performance is evaluated and the results are shown in Table 5.2. The average model efficiency is 0.83 while the average mean absolute error is 0.97 m for twelve hours forecast. In order to generate better simulation results, the Autoregressive model is used and its significance in improving the model’s performance will be discussed in the following section. 40 Table 5.1 Calibrated parameters for the Sungai Kelantan Tank model Parameter Value Parameter Value (mm) EV 0.2 mm/hr TS1 15 C1 0.001 TS2 20 C2 0.001 TS3 30 C3 0.08 TS4 20 C4 0.002 X1 10 C5 0.09 X2 0 C6 0.0065 X3 0 C7 0.01 X4 0 C8 0.002 X5 20 C9 0.0008 X6 0 Table 5.2 Result of model performance evaluation Storm Event Model Efficiency Mean Absolute Error (m) 1. 17/1/1988 – 15/12/1988 0.86 0.98 2. 17/12/1993 – 6/1/1994 0.91 0.82 3. 10/12/2001 – 10/1/2002 0.72 1.12 Observed and Forecasted Discharge for 1988 Storm Event (17/11/1988 - 15/12/1988) 6000.0 25.00 Observed Q 5000.0 Forecasted Q 20.00 4000.0 15.00 3000.0 10.00 2000.0 5.00 1000.0 0.00 0.0 11/15/1988 0:00 11/20/1988 0:00 11/25/1988 0:00 11/30/1988 0:00 12/5/1988 0:00 12/10/1988 0:00 12/15/1988 0:00 -5.00 Date Tim e Figure 5.1 Observed and forecasted discharge for 1988 storm event Rainfall (mm) Discharge (m 3/s) Rainfall 41 Observed and Forecasted Discharge for 1993 Storm Event (17/12/1993 - 6/1/1994) 6000.0 16.00 Observed Q 5000.0 12.00 Rainfall 4000.0 10.00 8.00 3000.0 6.00 2000.0 Rainfall (mm) Discharge (m 3/s) 14.00 Forecasted Q 4.00 2.00 1000.0 0.00 0.0 12/15/1993 0:00 12/20/1993 0:00 12/25/1993 0:00 12/30/1993 0:00 -2.00 1/9/1994 0:00 1/4/1994 0:00 Date Tim e Figure 5.2 Observed and forecasted discharge for 1993 storm event Observed and Forecasted Discharge for 2001 Storm Event (10/12/2001 - 10/1/2002) 16.00 4000.0 Observed Q 3500.0 Forecasted Q 12.00 Rainfall 10.00 2500.0 8.00 2000.0 6.00 1500.0 Rainfall (mm) Discharge (m 3/s) 3000.0 14.00 4.00 1000.0 2.00 500.0 0.0 12/7/2001 0:00 0.00 -2.00 12/12/2001 0:00 12/17/2001 0:00 12/22/2001 0:00 12/27/2001 0:00 1/1/2002 0:00 1/6/2002 0:00 1/11/2002 0:00 Date Tim e Figure 5.3 Observed and forecasted discharge for 2001 storm event The graphs show that the model has limited ability of imitating the actual catchment behavior. The overall pattern for both simulated and observed hydrographs shows certain similarity; hence the model is able to react to rainfall in a similar manner as the actual catchment in terms of timing. But the magnitude of discharge values simulated by the model has a significant difference from the observed values. 42 5.2 Significance of Autoregressive Model Autoregressive (AR) model is used to improve the forecast results generated by the Tank model (refer to Section 3.4 for details of AR model). The adjusted hydrograph for all three storm events (years 1988, 1993 and 2001) are shown in Figure 5.4, Figure 5.5 and Figure 5.6. Forecasted Q represents the original simulation results before any adjustment, while the AR12 represents the adjusted simulation results for twelve hours forecast. The model performance is re-evaluated based on the adjusted forecast of discharge and water level and the model evaluation results are as shown in Table 5.3 and Table 5.4. The average model efficiency has increased by 0.13, from 0.83 to 0.96 for twelve hours forecast; while the average mean absolute error has been improved by 60.8 %, from 0.97 m to 0.38 m for twelve hours forecast. It can also be observed from the graphs that the peak flow simulation has improved in terms of timing, where the simulated peak occurred almost at the same time as the observed peak, but the magnitude of the simulated peak seems to be always lower than the observed peak flow. 43 Table 5.3 Model efficiency for three historical storm events used for model calibration. No. Storm Event Model Efficiency With AR Correction Without Correction 3 hours 6 hours 9 hours 12 hours 1. 1988 0.86 1.00 0.99 0.98 0.97 2. 1993 0.91 1.00 0.99 0.98 0.97 3. 2001 0.72 0.99 0.98 0.96 0.93 Table 5.4 Mean absolute error for three historical storm events used for model calibration. No. Storm Event Mean Absolute Error (m) With AR Correction Without Correction 3 hours 6 hours 9 hours 12 hours 1. 1988 0.98 0.12 0.21 0.30 0.38 2. 1993 0.82 0.12 0.20 0.28 0.35 3. 2001 1.12 0.13 0.23 0.32 0.41 Observed and Forecasted Discharge for 1988 Storm Event (17/11/1988 - 15/12/1988) 6000.0 Observed Q 5000.0 Forecasted Q Discharge (m 3/s) AR12 4000.0 3000.0 2000.0 1000.0 0.0 11/14/1988 0:00 11/19/1988 0:00 11/24/1988 0:00 11/29/1988 0:00 12/4/1988 0:00 12/9/1988 0:00 12/14/1988 0:00 12/19/1988 0:00 Date Tim e Figure 5.4 Comparison between observed and forecasted discharge (with and without correction) for 1988 storm event 44 Observed and Simulated Discharge for 1993 Storm Event (17/12/1993 - 6/1/1994) 6000.0 Observed Q 5000.0 Simulated Q Discharge (m 3/s) AR12 4000.0 3000.0 2000.0 1000.0 0.0 12/16/1993 0:00 Figure 5.5 12/21/1993 0:00 12/26/1993 0:00 12/31/1993 0:00 Date Tim e 1/5/1994 0:00 Comparison between observed and forecasted discharge (with and without correction) for 1993 storm event Observed and Forecasted Discharge for 2001 Storm Event (10/12/2001 - 10/1/2002) 4000.0 Observed Q 3500.0 Forecasted Q Discharge (m 3/s) 3000.0 AR12 2500.0 2000.0 1500.0 1000.0 500.0 0.0 12/7/2001 0:00 Figure 5.6 12/12/2001 0:00 12/17/2001 0:00 12/22/2001 0:00 12/27/2001 0:00 Date Tim e 1/1/2002 0:00 1/6/2002 0:00 1/11/2002 0:00 Comparison between observed and forecasted discharge (with and without correction) for 2001 storm event 45 5.3 Tank Model Verification The Tank model is verified using the year 2003 (23/11/2003 – 27/12/2003) and 2004 (27/11/2004 – 27/12/2004) storm events. The model verification results are as shown in Figure 5.7 and Figure 5.8 (AR12 indicates the twelve hours forecast results), and the model evaluation results are shown in Table 5.5 and Table 5.6. For twelve hours forecast, the average model efficiency is 0.97 and the average mean absolute error is 0.26 m. This result shows that the Sungai Kelantan Tank model is able to generate reliable forecast for up to twelve hours ahead. Table 5.5 Tank model verification results using year 2003 storm event (23/11/2003 – 27/12/2003) Without With AR Correction Correction 3 hours 6 hours 9 hours 12 hours Model Efficiency 0.78 1.00 0.99 0.97 0.96 Mean Absolute Error (m) 0.88 0.09 0.16 0.22 0.28 Table 5.6 Tank model verification results using year 2004 storm event (27/11/2004 – 27/12/2004) Without With AR Correction Correction 3 hours 6 hours 9 hours 12 hours Model Efficiency 0.76 1.00 0.99 0.99 0.98 Mean Absolute Error (m) 1.38 0.08 0.14 0.19 0.24 46 Comparison Between Observed and Forecasted Discharge 4000.0 Observed Q 3500.0 Forecasted Q Discharge (m 3/s) 3000.0 AR12 2500.0 2000.0 1500.0 1000.0 500.0 0.0 11/20/2003 0:00 11/25/2003 0:00 11/30/2003 0:00 12/5/2003 0:00 12/10/2003 0:00 12/15/2003 0:00 12/20/2003 0:00 12/25/2003 0:00 12/30/2003 0:00 1/4/2004 0:00 Date Tim e Figure 5.7 Comparison between observed and forecasted discharge (with and without correction) for 2003 storm event Observed and Forecasted Discharge for 2004 Storm Event (27/11/2004 - 27/12/2004) 6000.0 Observed Q 5000.0 Forecasted Q Discharge (m 3/s) AR12 4000.0 3000.0 2000.0 1000.0 0.0 11/24/2004 0:00 11/29/2004 0:00 12/4/2004 0:00 12/9/2004 0:00 12/14/2004 0:00 12/19/2004 0:00 12/24/2004 0:00 12/29/2004 0:00 1/3/2005 0:00 Date Tim e Figure 5.8 Comparison between observed and forecasted discharge (with and without correction) for 2004 storm event CHAPTER VI CONCLUSIONS 6.1 Model Performance Conclusions that can be drawn from this project are as follows: 1. After the calibration and verification processes for the Sungai Kelantan Tank model, it is clearly demonstrated by the evaluation results that the model is capable of generating reliable water level forecasts for Sungai Kelantan at Guillemard Bridge, with an average model efficiency of 96 % and mean absolute error of 0.35 m for twelve hours forecast. 2. The Sungai Kelantan Tank model is very user-friendly as it is built within MS Excel with minimal user interfaces, with the hidden operation code that safeguards the system from unintended alteration by the users. Most of the operations are executed under the control of a minimal number of buttons. Display of results such as the tabulated forecast values and the hydrographs are all programmed to be updated automatically. The model is also costeffective as it only requires presence of MS Excel, which is available in most computers that are using MS Windows as the operating system. 48 3. The Sungai Kelantan Tank model has high flexibility as it is capable of handling any storm duration as defined by user, and it can be calibrated to simulate water level profile for other rivers as well. 4. The model integration that combines the Tank model, 1-D Hydrodynamic model (Standard Step and St. Venant’s) and the tide model has been accomplished successfully. Currently the complete flood forecasting model is able to generate water level forecast at Jeti Kastam with a lead time of five hours, and the simulated hydrograph pattern is very similar to the observed hydrograph. 6.2 Limitations of the Model The actual hydrological processes that occur within a catchment involve a series of complex interaction among various components such as precipitation, evaporation, infiltration, surface detention and runoff, and many others. These processes take place across the entire catchment simultaneously and it is impossible to estimate the amount of water involved accurately. Modeling techniques that are available currently, such as the Tank model that is being adopted in this study, are mostly based on simplified representative hydrological processes in order to imitate the actual catchment behavior and provide simulation with the slightest possible error. The limitations of the model as recognized in this study are: 1. Though the model is able to simulate the peak magnitude accurately, but the simulated peak usually occurs around six to twelve hours before the actual peak. The model’s performance is limited by the model’s structure itself. 2. The performance of the model is greatly limited by the completeness and reliability of the hydrological data (rainfall, water level and rating curve) that serves as the input for the model. 49 3. Forecasted rainfall for the next twelve hours is needed as the model input in order to generate forecast of discharge and water level. Thus the accuracy of the forecasted rainfall also becomes a limitation to the accuracy of forecasted water level. Further explanations on how the model structure and hydrological data are limiting the Tank model’s performance are discussed in Section 6.2.1 and Section 6.2.2 respectively; while recommendations on how to overcome these limitations and improve the model’s performance are discussed in Section 6.3. 6.2.1 Model Structure In this model, there is only one series of tank with a minimal number of outlets. Each of the outlets is able to generate a unique hydrograph and the summation of these outlet hydrographs yields the total simulated discharge. By referring to Figure 4.8 we can observe that the hydrograph patterns generated by the tanks are very limited, which makes the model have less ability to imitate the actual behavior of the catchment. Therefore the performance of the model can be enhanced by adding more outlets or by adding another series of tank which is parallel to the existing tank. 6.2.2 Hydrological Data The performance of the Tank Model is significantly limited by the availability, completeness and reliability of hydrological data such as rainfall and water level data. Generally Kelantan’s data collection is considerably complete and 50 reliable, and the eight rainfall stations chosen to be used in this model are the main stations in the state that provide good data. The accuracy of rating curve is also another issue to be concerned of. River survey is a costly process and it is not being conducted on a frequent basis. The rating curve used in this model was developed at the year 2002. The channel geometry may have experienced changes in this past five years, and this is one of the major limitations of the model if we consider its performance without the Autoregressive adjustments. Water level data is crucial for the model calibration process. Currently this model does not include the feature of filling in the missing water level data. If the gap of data is considerably short, for example a few hours, interpolation may be done to fill in the gap. But if the recorded data has a gap which exceeds twelve hours, interpolation may not be able to provide reliable data substitution. 6.3 Recommendations The Tank model’s performance can be further improved in terms of the accuracy of its simulation results and its user friendliness. The recommended solutions are suggested as follows: 1. An additional parallel series tank can be added to the existing tank model structure to retain water in tanks for longer. This will create a lag effect which will help to resolve the problem of the early arrival of simulated peak. 2. By using longer storm duration, the effect of missing data will be reduced as the large amount of data will be able to stabilize the tank storage. For this project it is recommended to use at least one month of data prior to the start of forecast. 51 3. Forecasted rainfall values are usually generated through educated guesses with the aid of weather forecast reports provided by the Malaysia Meteorological Department. However better estimation can be made by using the radar to forecast rainfall. 4. The model should be re-calibrated after several years of use, or when the river experiences significant changes or developments. Re-calibration will allow the model to take into account the changes of discharge that may be caused by sedimentation along river which would make the river bed shallower, or developments along the river side 5. Currently there are three rainfall stations used in the model that are automatic recording stations, they are: JPS Machang, Gob and Blau. These stations need to be upgraded to telemetric stations so that real-time data can be acquired for forecasting purposes. 6. Automatic web query can add value to the model as it will be able to obtain data automatically from the JPS’s online hydrological database Infobanjir for real-time flood forecasting. 52 REFERENCES 1. Ng, B.C. (2003). Ramalan Banjir Bagi Sungai Kuantan Dengan Menggunakan Model Tangki. Skudai: Universiti Teknologi Malaysia: Thesis Sarjana Muda. 2. Cheok, H.S. (2005). Development of a PC-Based Tank Model Real-time Flood Forecasting System. Skudai: Universiti Teknologi Malaysia: Thesis Sarjana. 3. Dr. Nik & Associates. (2005). Development of a Rainfall-Flood Correlation Model cum Flood Message Dissemination System for a Pilot Area in the Klang River Basin. Kuala Lumpur: Final Report. 4. Chong, S.F. and Cheok, H.S. (2006). Development of a Digital Database and Information System for the Six Selected Flood Forecasting Systems of JPS Malaysia. Kuala Lumpur: Final Report. 5. Kraijenhoff, D.A. and Moll, J.R. (1986). River Flow Modelling and Forecasting. Netherlands: D. Reidel Publishing Company. 6. Sugawara, M. (1979). The Flood Forecasting by a Series Storage Type Model. Tokyo, Japan: National Research Center for Disaster Prevention. 7. Viessman, W., Lewis, G. L. (2003). Introduction to Hydrology. New Jersey: Prentice Hall, Pearson Education, Inc. 8. Sepakat Setia Perunding Sdn Bhd and SMHB Sdn Bhd. (1999). Final Report: Kelantan River Flood Mitigation Plan Feasibility Study. 9. Lee, J.M. (2007). Development of Flood Forecasting Model for Sungai Kelantan (1-D Hydrodynamic Model). Skudai: Universiti Teknologi Malaysia: Thesis Sarjana Muda. 10. Ang, Zhili (2007). Development of Flood Forecasting Hydraulic Model for Sungai Johor. Skudai: Universiti Teknologi Malaysia: Thesis Sarjana Muda. 53 APPENDIX A Graphs showing comparison between individual tank discharge and total discharge Individual Tank Discharge and Total Discharge for 1988 Storm Event (17/11/1988 - 15/12/1988) 1.80 6000.0 Q1 1.60 Q2 Q3 1.40 5000.0 Q5 4000.0 Total Q 1.00 Q6 Obs Q 0.80 3000.0 0.60 2000.0 0.40 0.20 Total discharge (m3/s) Tank Discharge (mm) Q4 1.20 1000.0 0.00 11/15/1988 0:00 11/20/1988 0:00 11/25/1988 0:00 11/30/1988 0:00 -0.20 12/5/1988 0:00 12/10/1988 0:00 12/15/1988 0:00 0.0 Date Time Individual Tank Discharge and Total Discharge for 2001 Storm Event (10/12/2001 - 10/1/2002) 4000.0 1.20 Q1 Q2 1.00 3500.0 Q3 3000.0 Q5 Total Q 0.60 2500.0 Q6 Obs Q 2000.0 0.40 1500.0 0.20 0.00 12/7/2001 0:00 -0.20 1000.0 500.0 12/12/2001 0:00 12/17/2001 0:00 12/22/2001 0:00 12/27/2001 0:00 1/1/2002 0:00 1/6/2002 0:00 1/11/2002 0:00 0.0 Date Time Total Discharge (m3/s) Tank Discharge (mm) Q4 0.80 54 Individual Tank Discharge and Total Discharge for 2003 Storm Event (23/11/2003 - 27/12/2003) 4000.0 1.00 Q1 0.90 Q3 Q4 3000.0 Q5 0.60 Total Q 2500.0 Q6 0.50 Obs Q 2000.0 0.40 1500.0 0.30 0.20 Total Discharge (m 3/s) 0.70 Tank Discharge (mm) 3500.0 Q2 0.80 1000.0 0.10 0.00 11/20/2003 -0.100:00 500.0 11/25/2003 0:00 11/30/2003 0:00 12/5/2003 0:00 12/10/2003 12/15/2003 0:00 0:00 Date Tim e 12/20/2003 0:00 12/25/2003 0:00 12/30/2003 0:000.0 Individual Tank Discharge and Total Discharge for 2004 Storm Event (27/11/2004 - 27/12/2004) 6000.0 1.60 Q1 1.40 Q2 Q4 Q5 1.00 4000.0 Total Q Q6 0.80 Obs Q 3000.0 0.60 0.40 2000.0 0.20 1000.0 0.00 11/24/2004 -0.200:00 11/29/2004 0:00 12/4/2004 0:00 12/9/2004 0:00 12/14/2004 0:00 Date Tim e 12/19/2004 0:00 12/24/2004 0:00 12/29/2004 0:00 1/3/2005 0:00 0.0 Total Discharge (m 3/s) Tank Discharge (mm) 5000.0 Q3 1.20 APPENDIX B Detail Calculation of the Tank Model in Simulating the Discharge and Water Level for the year 2004 Storm Event (27/11/2004 – 27/12/2004) 55 APPENDIX B - continue Detail Calculation of the Tank Model in Simulating the Discharge and Water Level for the year 2004 Storm Event (27/11/2004 – 27/12/2004) 56 APPENDIX B - continue Detail Calculation of the Tank Model in Simulating the Discharge and Water Level for the year 2004 Storm Event (27/11/2004 – 27/12/2004) 57 APPENDIX B - continue Detail Calculation of the Tank Model in Simulating the Discharge and Water Level for the year 2004 Storm Event (27/11/2004 – 27/12/2004) 58 APPENDIX B - continue Detail Calculation of the Tank Model in Simulating the Discharge and Water Level for the year 2004 Storm Event (27/11/2004 – 27/12/2004) 59