An Approach for Flow Forecasting in Ungauged Catchments – A Case Study for Ho Ho reservoir catchment, Ngan Sau River, Central Vietnam

74 INTRODUCTION Reservoir management can be viewed as an important strategic perspective for the socio- economic development of a region, related to hydro-power development, water supply for irri- gation, flood control etc. However, the variability of hydro-meteorological forcing and economic activities within the river basin are expected to significantly impact its characteristic. In order to maximize the performance of a single-reservoir as well as multi-reservoir system, having accu- rate reservoir inflow forecasting with enough lead time poses a challenge to water managers. The development of model methods has been widely applied as an effective method for the res- ervoir inflow forecasting [An et al. 2012, Anh et al. 2015]. Coupled neural networks have recently become a well-known tool for better forecasting of the inflow hydrograph [Sanjeet et al. 2015, Krish- na 2014]. However, in this model, a large amount of hydrologic data is required to determine the adaptive weights, which is usually unavailable in data-scarce regions. Using the distributed models may be inappropriate for real-time flow forecast- ing since the simulation can be time-consuming [Hapuarachchi et al. 2008]. A lumped model, there- fore, is possibly the best choice for forecasting in an ungauged catchment. With the selected model, the accuracy of forecasting is prone to uncertainty depending on the forecasting meteorological data, forecasting scheme and the reliability of the used model. The aim of this study is to enhance the reliability of the model based on continuous op- timization method to specify updated parameter values for reservoir inflow forecasting. For that purpose, model validation and calibration are key steps in any forecast models and simulation study. Journal of Ecological Engineering Received: 2018.02.04 Accepted: 2018.03.15 Published: 2018.05.01Volume 19, Issue 3, May 2018, pages 74–79 https://doi.org/10.12911/22998993/85759 An Approach for Flow Forecasting in Ungauged Catchments – A Case Study for Ho Ho reservoir catchment, Ngan Sau River, Central Vietnam Dang Dinh Kha1*, Nguyen Y Nhu1, Tran Ngoc Anh1,2 1 Department of Hydrology and Water Resources, VNU University of Science, Vietnam National University, Hanoi, 334 Nguyen Trai, Thanh Xuan, Hanoi, Vietnam 2 Center For Environmental Fluid Dynamics, VNU University of Science, Vietnam National University, Hanoi, 334 Nguyen Trai, Thanh Xuan, Hanoi, Vietnam * Corresponding author’s e-mail: dangdinhkha@hus.edu.vn ABSTRACT Reservoir inflow forecasting with high reliability plays an important role in the operation and management of the reservoir for power generation, irrigation, flood prevention as well as ensuring the safety of the dam. However, the level of forecast accuracy is limited, since its performance depends on rainfall forecasting and hy- drological model. In order to increase the efficiency of forecasting, this study introduces the inflow forecasting method that integrates the real-time updating techniques with continuous optimization method of MIKE NAM model to specify the appropriate parameter set for forecasting time. The proposed forecasting method was tested for the Ho Ho reservoir, the area facing the scarcity of historical data for model calibration and verification. The analysis of the forecasting results for Ho Ho reservoir using transferred parameters from the stable calibrated parameter values at Hoa Duyet station (downstream of Ho Ho reservoir) and the results obtained using the adapted parameters by the proposed method shows that the adapted parameter values provides a more reliable forecast, which will better serve the decision making. Keywords: real time updating, flood, forecasting, ungauged catchment. 75 Journal of Ecological Engineering Vol. 19(3), 2018 However, it should be noted that there is a limited number of river basins with long enough observed discharge data to capture the coverage of the hy- drological event within the basin to both calibrate and validate the model, especially for the area with construction of new reservoir. Several studies “transfer” the parameter values of the model being specified for one catchment to the predicted catch- ment having similar characteristics [An et al. 2013, Bardossy 2007]. However, this procedure assumes that only one set of parameters is obtained, which possibly lead to a good model performance only for the events with similar conditions. In fact, due to the large complexity of the corresponding natu- ral phenomena in Vietnam as result of its tropical monsoon climate and its time variation, the model parameters need to be transferred for the predicted catchment and the weather conditions under study. Further, the socio-economic activities within the basin have an influence on the hydrological re- gime coupled with the impact of climate change which leads to an increase in the complexity of hydrological processes in a catchment. Conse- quently, the identification of a unique dataset providing a good model performance for any weather conditions is practically impossible. In such a case, a flow forecasting system incorporat- ing a real-time updating algorithm that adapts the model and catchment state can improve the fore- casting accuracy. On that basis, this study tested and discussed the inflow forecasting method that integrates the real-time updating techniques in- cluding three categories: updating (1) input vari- ables; (2) model parameters; and (3) output vari- ables and the continuous optimization method of MIKE NAM model to specify parameters on the basis of reservoir water balance equation for the Ho Ho reservoir inflow forecasting. MATERIALS AND METHODS Study area and dataset used The Ho Ho reservoir is constructed on the Ngan Sau River, a main tributary of La River located in Ha Tinh province, central Viet Nam. The area of Ho Ho reservoir catchment is 278.6 km2 with the storage capacity of the reservoir being 38 million m3. The Ho Ho reservoir offi- cially operates from the beginning of 2013. From 2017, three automatic rain gauges and an automatic water level recorder were installed upstream of the Ho Ho reservoir, these recorded data can be automatically updated for the data- base of reservoir management office. In addition, the observed daily rainfall for 7 years from Jan 01, 2010 to Dec 31, 2016 were collected from the National Hydro-Meteoro- logical Service, Vietnam for three stations Chu Figure 1. Rain gauge network in the study area Journal of Ecological Engineering Vol. 19(3), 2018 76 Le, Huong Khe and Hoa Duyet in the study area (Figure 1). The daily water level and discharge with the same period were collected from Hoa Duyet station with the catchment area of 1880 km2, located downstream of the Ho Ho reservoir (71 km from the site). These data are employed for calibration and validation of the model to serve the reservoir inflow forecasting. Methodology MIKE NAM, a lumped model, developed at DHI Water & Environment has been widely used in researches [Keskin et al. 2007, Liu et al. 2007, Kamel et al. 2008]. The model simulates the water movement in the land phase of the hydrological cycle by continuously accounting for the water content in four different and mutually interrelated storages, including snow storage, surface storage, lower or root zone storage and groundwater stor- age [DHI 2004]. The parameters of MIKE NAM model can be identified by either the manual trial- and-error method or automatic optimization meth- od. The auto-calibration is done to optimize two objective functions: (a) minimizing the water bal- ance error (%WBL) (b) minimizing the root mean square error (RMSE) [DHI 2004]. The auto-cali- brated method provides the good performance in simulating each hydrological event, but with vari- ous hydrological events, they require different pa- rameter sets to obtain good performance [Giang et al. 2010]. In recent years, MIKE NAM model has been considered as an efficient tool for forecasting and water resources assessment in Vietnam [An et al. 2013, Long et al. 2010, Giang et al. 2010]. However, most of the study focused on simulating the historical events with a single parameter set. In this paper, a different approach of integrating the real-time updating technique with optimization al- gorithm SCE (Shuffled Complex Evolution) with the RMSE objective function in MIKE NAM for parameter auto-calibration is applied. The cali- brated parameters will be linked and updated auto- matically into the runoff model to forecast the in- flow for the next time step. The rainfall and water level data from automatic recorder will be updated as input variables for parameters updating. The automatic water level recorders at the reservoir are used to estimate the reservoir inflow according to the reservoir water balance equation (1): (1) where: Q(t) – reservoir inflow [m3/s], qr(t) – reservoir outflow [m3/s], dV/dt – varying of storage capacity of the reservoir in time [m3/s]. Simplified, the trial-and-error method will be used to solve the equation (1) which is based on the characteristics of reservoir, including water level – storage capacity relationships Z ~ V, water level – structure outflow (turbine, spillway) Z~ q, and water level – surface area Z~ F. These relationships were developed based on the reservoir survey data. The daily rainfall data collected from three stations Chu Le, Huong Khe and Hoa Duyet in the study area will be used to estimate the basin aver- age rainfall with the Thiessen Polygon method. The model calibration and validation will be done by combining the trial-and-error with automatic optimization method to obtain the best result. The accuracy of simulation assessed by using the Nash-Sutcliffe coefficient, which were of 0.7 and 0.76 being obtained with the period of 2010-2013 and 2014-2016, respectively, for calibration and verification (Figure 2). This indicates that the ob- tained parameters were reasonably good for fore- casting purpose at Hoa Duyet station. In order to forecast the Ho Ho reservoir inflow, these parameters need to be transferred from Hoa Duyet to Ho Ho using regression relationships be- tween the catchment area defined by Ho Ho reser- voir (278.6 km2) and the one by Hoa Duyet station (1880 km2), referred to as parameter set 1. The pa- rameters of time constant for routing overland flow (CK1, 2) and baseflow (BF) will be revised since they present high sensitivity to the catchment size. Besides, the study integrates the real-time water level, rainfall updating algorithm and auto- calibrated method in MIKE NAM to optimize the parameter values. The algorithm was developed in the Matlab environment to auto-update the observed data from recorder, auto-calibrate and forecast the Ho Ho reservoir inflow via MIKE NAM model. In turn, when the water level and rainfall at the reservoir are updated at time t, the model will auto-forecast the reservoir inflow at t + Δt, the parameters of MIKE NAM model at subsequent time step, (t + 2Δt), can be obtained according to updating measured flow, water level and rainfall data at t + Δt. The updating observed input variables at t + Δt and the re-update param- eters for time t + 2Δt (parameter set 2) will be used to forecast the inflow at t + 2Δt. It means that the parameter set 2 will be updated continuously to correctly reflect the catchment attributes at the predicted time. 77 Journal of Ecological Engineering Vol. 19(3), 2018 RESULTS AND DISCUSSIONS The hourly rainfall data of flood event from July 15, 2017 to July 17, 2017 were collected for forecast testing and the results are presented in Figures 3 and 4. The Figure 3 indicates the forecasting result using the parameter set 1; it can be seen that un- der no rain conditions, water is kept in soil and water bodies on catchment to maintain the river, the model shows a good performance. However, under heavy-rainfall conditions, the result sug- gests that the model overestimated the simulat- ed discharge, the error of flood peak flows was around 80%. This indicates that reasonably good model parameters obtained during the calibration and validation at Hoa Duyet station, are not com- parable for the forecast of the Ho Ho reservoir inflow. This is possibly because the parameter set 1 does not correctly reflect the attributes of catch- ment at the predicted time (soil humidity, land’s water storage capacity etc). The Figure 4 shows the observed and forecast discharge using the parameter set 2 for different Figure 2. The observed and simulated discharge at Hoa Duyet station Figure 3. The observed and forecast discharge using the parameter set 1 Journal of Ecological Engineering Vol. 19(3), 2018 78 time steps. Similarly, under no rain conditions (from July 15, 2017 at 1:00:00 PM to July 16, 2017 at 1:00:00 PM) the model shows good per- formance as a result of stable inflow. At the initial time t when rain occurs (July 16, 2017 at 4:00:00 PM) the model performed relatively worse with the forecast error of 45% for the 12 hours ahead forecast cases. The possible reason of this result is to employ the parameters optimized from the previous time step (the no rain parameter set) for the forecast cases. In order to make prediction at t + 1 (July 16, 2017 at 7:00:00 PM), the model is updated with observed water level at t of the Ho Ho reservoir which means that this optimized parameters somehow better reflect the attributes of catchment. As a result, forecast values are improved significantly; the errors are 10% and 20% with the lead time of 6 hours and 12 hours, respectively. The error of peak discharge for a 3-h prediction is around 20%. For the subse- quent time steps (t + 2, t + 3,...) the forecast dis- charge hydrograph approaches ever closer to the observed discharge hydrograph. Even though the accuracy of inflow forecasting decreases when the lead time is increased because the forecast error is accumulated from the previous lead-time fore- casting; these results indicate that the proposed inflow forecasting method can increase the cred- ibility of forecast information in comparison with the forecast information using the transferred pa- rameter values from another catchment. CONCLUSION Accurate and reliable forecasting of reservoir inflow highly depends on the meteorological fore- casting (rainfall, temperature, etc.). As a result of using observed rainfall data, it somehow reduces the kind of error caused by input data. The analysis was carried out with (1) the pa- rameters transferred from the neighbour catch- ment, and (2) the parameters auto-optimized con- tinuously based on the SCE algorithm and real- time updating at predicted time indicated that the latter method provides more reliable results than the former one. It is possibly because the auto- calibrated parameters based on real-time updat- ing algorithm have a good reflection of the catch- ment states at predicted time – a region having Figure 4. The observed and forecast discharged hydrograph using the parameter set 2 79 Journal of Ecological Engineering Vol. 19(3), 2018 completely no data of catchment attributes (land humidity, land’s water storage capacity, etc.), the forecast shows good agreement with the observed flow. Using the proposed method for the inflow simulation is advantageous, because it can pro- vide credible forecast for the catchment without any historical discharge data. Nevertheless, the model parameters are capable of continually up- dating so that the forecasting at subsequent time steps can be operated with high accuracy; hence, the water level of the reservoir and observed rain- fall need to be updated continuously. This raising point can be done by the installation of automatic gauges. For the 6-h, 12-h ahead forecast cases, the forecast error is 10% and 20%, respectively. Although the accuracy of the forecast rainfall de- creases as the lead time increased, the results in- dicated that a real-time updating algorithm incor- porated into optimized auto-calibration method of the parameters of MIKE NAM model can enhance the efficiency of inflow forecasting for a data scarce region. Further research for more accurate and efficient prediction is still required. However, integrating the data assimilation algorithm into the forecast process to bridge the gap between the theory and practice can be a possible solution. Acknowledgements This research is funded by the VNU Univer- sity of Science under project number TN.17.16. The author is PhD/Doctoral Student under the 911 Program of VNU University of Science, Viet- nam National University. REFERENCES 1. An N.L., Ngoc N.T.B. 2012. Research on flood forecasting for reservoir in the Ba river. Journal of Water Resources & Environmental Engineering, No 38 (in Vietnamese). 2. An N.L., Hoa N.N. 2013. Research on flood fore- casting in Vu Gia – Thu Bon River Basin. Journal of Water Resources & Environmental Engineering, No 43 (in Vietnamese). 3. Anh L.T., Son N.T. 2015. Some experiences for ap- plying Hydrological, Hydraulic models to Hydro- logic Forecasting, Journal of Science VNU. Natural Sciences and Technology, 31(1S) (in Vietnamese) 4. Bardossy A. 2007. Calibration of hydrological model parameters for ungagged catchments. Hy- drol. Earth Syst, Scie., 11, 703-710. 5. DHI Water & Environment. 2004. MIKE 11 Refer- ence Manual. 6. Giang N.T. Phuong T.A. 2010. Calibration and verification of a hydrological model using event data, VNU Journal of Science, Earth Science, 26. (in Vietnamese). 7. Hapuarachchi H.A.P., Wang Q.J.A. 2008. Review of Methods and Systems Available for Flash Flood Forecasting; Report for the Bureau of Meteorol- ogy, Australia; Commonwealth Scientific and In- dustrial Research Organization (CSIRO): Dick- son, Australia. 8. Kamel A.H. 2008. Application of a hydrodynamic MIKE 11 model for the Euphrates River in Iraq. Slovak Journal of Civil Engineering, 2, 1-7. 9. Keskin F., Sensoy A.A., Sorman A. 2007. Applica- tion of MIKE11 Model for the Simulation of Snow- melt Runoff in Yuvacik Dam Basin, Turkey. Interna- tional Congress on River Basin Management, The role of general directorate of state Hydraulic works (DSI) in development of water resources of Turkey. 10. Krishna B. 2014. Comparison of wavelet based ANN and regression models for reservoir inflow forecasting. J Hydrol Eng, 19(7), 1385-1400. 11. Liu H.L., Chen X., Bao A.M., Ling Wang. 2007. Investigation of groundwater response to overland flow and topography using a coupled MIKE SHE/ MIKE 11 modeling system for an arid watershed. Journal of Hydrology, 347, 448-459. 12. Long V.D., Anh T.N., Binh H.T., Kha D.D. 2010. An introduction to flood forecast technology in Ben Hai and Thach Han river systems using MIKE 11 model. Vietnam National University Journal of Science, 26(3S), 397-404. 13. Sanjeet K., Mukesh K.T., Chandranath C., Ashok M. 2015. Reservoir Inflow Forecasting Using En- semble Models Based on Neural Networks, Wave- let Analysis and Bootstrap Method. Water Resour Manage, 29, 4863-4883, doi: 10.1007/s11269- 015-1095-7.