Research Journal of Applied Sciences, Engineering and Technology 4(21): 4284-4292, 2012 ISSN: 2040-7467 © Maxwell Scientific Organization, 2012 Submitted: February 02, 2012 Accepted: March 02, 2012 Published: November 01, 2012 Development of an ANN-Based Model for Forecasting River Kaduna Discharge J.O. Folorunsho, E.O. Iguisi, M.B. Mu’azu and S. Garba Department of Geography, Ahmadu Bello University, Zaria, Nigeria Abstract: Artificial Neural Networks (ANNs) provides a quick and flexible means of creating models for river discharge forecasting and has been shown to perform well in comparison with conventional methods. This paper presents a method of discharge prediction for River Kaduna by developing an ANN-based model. Given the major triple problems of unavailability, inconsistency and paucity of data, the water resources planning and development in any drainage basin always suffer a setback. Rainfall, temperature, relative humidity and the stage height (input variables) and discharge (target output) data were obtained for River Kaduna drainage basin for April-October 1975 to 2004. In order to develop the ANN model, the data set was partitioned into two parts of 24 months sets. 70% of the entire data was used as training data and 30% of the entire data used as the validation data. From the results obtained, the developed Artificial Neural Network (ANN) model developed in the PredictDemo NeuralWare Environment using the Neural Statistics shows a correlation value of 82%. Keywords: Artificial neural networks, discharge, forecasting, river Kaduna, water resources INTRODUCTION The application of Artificial Neural Network (ANN) forecasting tool for river discharge is the main focus of this research as a supposed tool for water resources management in the River Kaduna drainage basin. Thus, effective management of the highly variable water resources of any river in any environment requires an understanding of its flow characteristics such that, suitable quality water can be provided for the municipalities under various conditions within institutional constraints (Texas Water Resources Institute Report-165, 1994). Therefore, the study and understanding of the amount of water that would be discharged by a stream in the future is of crucial importance to the water resources development, planning and management of any area. This is so because, it affects directly the design and operations of many water resources structures (Kisi, 2004, 2005). Three major problems resulted in a great setback for water resources harnessing for effective planning and development in Nigeria inspite of its vast water resources endowment. First among these problems is the nonavailability of consistent streamflow data of Nigerian rivers with less than 30% of its rivers gauged (Oguntoyinbo, 1979). Secondly, is the vandalization of available manual (though obsolete) gauging equipment in few rivers in Nigeria and finally, is the prohibitive cost of gauging equipment. Apart from these ones, general knowledge about variation of river flows are often invalidated by the short time or unavailability of consistent and unreliable discharge data (Ward, 1968; Jagtap, 1995). It has also been established that these contributed noticeably to the inability to plan and manage the water resources assiduously (Nemec and Schaake, 1982; Cohen, 1987; Arnell et al., 1990; Lettenmaier and Gan, 1990; Saelthun et al., 1990; Vehvilainen and Lohvansuu, 1991; Krasovskaia and Gottenschalk, 1993; Krasovskaia et al., 1993; Wateren-de-Hoog, 1993; Samuel, 1993). In hydrological studies various methods or techniques have been used for stream flow forecasting as found in the literature, but they can all be categorized, following Ward (1975), as empirical methods, statistical methods, analytical methods and modeling methods. Similarly, whichever of these methods is chosen will depend on a number of factors including the purpose of the prediction or forecasting, the available data as well as the size and other characteristics of the basin (Ward, 1975). Most of these methods are based on the conventional statistical analysis of flow data which were measured in the past. Many of them offer very complex or too demanding tools for practical cases. Among them all, the most well employed method is regression analysis of the observed data. Regression of the observed data in a nearby river can give important information on the behavior of the targeted river. Unfortunately, the hydrological and topographical conditions of the two basins should be similar to each other (Kilinc et al., 2000). Recently, flow forecasting using ANN model has been accepted as a good alternative to forecasting with the erstwhile known hydrological and hydrodynamics models (ASCE, 2000a,b). According to Hsu et al. (1995), ANN approaches to rainfall-runoff modeling are more efficient than the conventional flow Corresponding Author: J.O. Folorunsho, Department of Geography, Ahmadu Bello University, Zaria, Nigeria 4284 Res. J. Appl. Sci. Eng. Technol., 4(21): 4284-4292, 2012 Fig. 1: The drainage basin of river kaduna source: Modified form drainage map of kaduns state forecasting models whenever explicit knowledge of the hydrological balance is not required and when the system may be treated as a black-box. However, in the hydrological forecasting context, recent experiments have reported that ANNs (particularly the MLP and the RBF network) may offer a promising alternative for rainfall-runoff modeling (Smith and Eli, 1995; Tokar and Johnson, 1999), streamflow prediction (Sivakumar et al., 2002; Kisi, 2004, 2005; Antar et al., 2006) and reservoir inflow forecasting (Jain et al., 1999) to mention just a few. The application ANNs to various aspects of hydrological modeling has undergone much investigation in recent years. This interest has been motivated by the complex nature of hydrological systems and the ability of ANNs to model non-linear relationships. Although deterministic models strive to account for all physical and chemical processes, their successful employment may be restricted by a need for catchmentspecific data and the simplifications involved in solving the governing equations. The use of time-series methods may therefore be complicated by non-stationarity and non-linearity in the data, requiring experience and expertise from the modeler and which are not always available especially in the developing environment but ANNs offer a relatively quick and flexible means of modeling and as such applications of ANN modeling are widely reported in hydrological literature (Raman and Sunilkumar, 1995; Maier and Dandy, 1996; Loke et al., 1997; Shamseldin et al., 1997; Zhang and Stanley, 1997; Brion and Lingireddy, 1999; Imrie et al., 2000). This study is is therefore aimed at developing an artificial neural network model in forecasting the discharge of river Kaduna. Study area: River Kaduna drainage basin lies between latitude 9º30!N and latitude 11º45!N; longitude 7º03!E and longitude 8º30!E with a total basin area of approximately 21,065 km2 (Fig. 1). The basin enclosed major rivers such as Kubanni, Galma, Tubo which are tributaries to the main River Kaduna and a greater part of the Kaduna metropolis. The basin lies on the High Plains of Northern Nigeria at altitude of about 670 m above sea level and situated within the Northern Guinea Savanna. Typical of the savanna climate, River Kaduna drainage basin experience distinct wet and dry seasons. The wet season (May-October) is characterized by conventional rainfall followed by intense lightning and thunderstorms. The mean annual rainfall can be as high as 2000 mm in wet years and as low as 500 mm in drought years, but with a long term average of 1000 mm (Folorunsho, 2004). The dry season (November-April) is characterized by a period of low temperature with harmattan season around December-February; and the hot dry season between March-April with temperatures as high as 32ºC. Relative Humidity is high only during the raining season, but drops during the dry season (Sawa and Buhari, 2011). MATERIALS AND METHODS Monthly averages of rainfall for Zaria, Kaduna and Jos (the drainage basin under study), temperature, relative humidity for Kaduna metropolis and stage height of River 4285 Res. J. Appl. Sci. Eng. Technol., 4(21): 4284-4292, 2012 Fig. 2: Showing the initial launch of the ANN model Fig. 3: Showing the ANN model building wizard dialog box Fig. 4: Showing the initial step of the ANN model building wizard 4286 Res. J. Appl. Sci. Eng. Technol., 4(21): 4284-4292, 2012 Fig. 5: Showing building ANN model (step 2 of 6) -location of input data Fig. 6: Showing building ANN model (step 3 of 6) -location of target output Fig. 7: Showing building ANN model (step 4 of 6) - data properties 4287 Res. J. Appl. Sci. Eng. Technol., 4(21): 4284-4292, 2012 Fig. 8: Showing building ANN model (step 5 of 6) - variable selection and network search Fig. 9: Showing data analysis and transformation dialog box Fig. 10: Showing the data partitioning dialog box 4288 35 30 25 20 15 10 5 0 1975 Apr Apr 1977 Apr Apr 1980 Apr Apr 1982 Apr Apr 1984 Apr Apr 1986 Apr Apr 1988 Apr Apr 1990 Apr Apr 1992 Apr Apr 1994 May 1996 Jun 1997 Jun Jun 1998 Jun Jun 2001 Jun Aug Kaduna (input variables) and discharge data (output) for River Kaduna was used for the forecast. The data covering a period of 30 years (1974-2004) and for the rainy months in the basin (Apr-Oct) were sourced from the Hydrology Department, Kaduna State Water Board, Kaduna and Nigerian Meteorology Agency, Oshodi, Lagos, Nigeria. The ANN-based model of forecasting for this study was developed in the NeuralWare PredictDemo environment using the 6 basic procedures such as launching the model and loading the data set partitioned into 70% of the data sets for training and 30% for the testing and validation; configuration of the model; training of the model using 4 sets of variables for input; model performance testing; model validation for reliability for forecasting the discharge and graphical plots of actual and forecasted discharge output in order to ability of the model to learn the trend, cycles and pattern Rainfall (mm) Res. J. Appl. Sci. Eng. Technol., 4(21): 4284-4292, 2012 (Apr-oct, 1975-2004) Fig. 13: Showing the rainfall plot for April-October, 1975-2004 of the discharge. In this research, the ANN-based model development using the NeuralWare PredictDemo interface to achieve the desired operations from the initial launch to the completed training era are enumerated below from Fig. 2 to 12. Fig. 11: Showing data transformation and input variable dialog bo Fig. 12: Showing the training completed dialog box 4289 0.30 0.25 0.20 0.15 0.10 0.05 1975 Apr Apr 1977 Apr Apr 1980 Apr Apr 1982 Apr Apr 1984 Apr Apr 1986 Apr Apr 1988 Apr Apr 1990 Apr Apr 1992 Apr Apr 1994 May 1996 Jun 1997 Jun Jun 1998 Jun Jun 2001 Jun Aug 0 1975 Apr May 1977 Jun July 1980 Aug Sept 1982 Oct Apr 1984 May Jun 1986 July Aug 1988 Sept Oct 1990 Apr May 1992 Jun July 1994 Aug Sept 1996 1997 Oct Apr 1998 May Jun 2001 July Sept o 90 80 70 60 50 40 30 20 10 0 Stage height (m) Temperature ( C) Res. J. Appl. Sci. Eng. Technol., 4(21): 4284-4292, 2012 (Apr-oct, 1975-2004) (Apr-oct, 1975-2004) Fig. 16: Showing the stage height plot for April-October, 19752004 1200 100 1000 Discharge (m3/S) 120 80 60 40 20 800 600 400 200 0 1975 Apr May 1977 Jun July 1980 Aug Sept 1982 Oct Apr 1984 May Jun 1986 July Aug 1988 Sept Oct 1990 Apr May 1992 Jun July 1994 Aug Sept 1996 1997 Oct Apr 1998 May Jun 2001 July Sept 0 1975 Apr May 1977 Jun July 1980 Aug Sept 1982 Oct Apr 1984 May Jun 1986 July Aug 1988 Sept Oct 1990 Apr May 1992 Jun July 1994 Aug Sept 1996 1997 Oct Apr 1998 May Jun 2001 July Sept Relative humidity (%) Fig. 14: Showing the temperature plot for April-October, 19752004 (Apr-oct, 1975-2004) (Apr-oct, 1975-2004) Fig. 15: Showing the relative humidity plot for April-October, 1975-2004 Fig. 17: Showing the discharge plot for April-October, 19752004 Fig. 18: Showing the ‘Test’ command dialog box for the validation data RESULTS AND DISCUSSION The plots of the monthly average of all the input variables and the target output used for this study are shown in Fig. 13-17. These variables were loaded for the ANN model development which was achieved using the procedures outlined in the NeuralWare PredictDemo interface. These processes were used in order to achieve the training of the model with the results displayed in Fig. 18. However, one of the major tests of reliability of a good ANN- based model is that the training set performance and that of the test set are fairly similar. The key indicator of such model is the performance of the model when subjected to the standard procedural and statistical tests such as outlined earlier. For this study, 4290 Res. J. Appl. Sci. Eng. Technol., 4(21): 4284-4292, 2012 Discharge (m3/S) 1200 the ANN-based model is a veritable tool in overcoming the discharge record paucity, inconsistency and unreliability hampering water resources development in the study area. Series 1 Series 2 1000 800 600 400 ACKNOWLEDGMENT 200 141 151 161 171 1 11 21 31 41 51 61 71 81 91 101 111 121 131 0 Month Fig. 19: Showing the actual discharge and the ANN predicted discharge (1975-2004) these values were found to be similar such that while the R for the trained data was 80%, the tested data was 82%, RMSE is 145, Accuracy is 82% and the Confidence Interval is 290 (Fig. 2). Considering the test result generated by the ANNbased model as displayed in Fig. 2, the model has learned both the trend and the cycle of the actual data. This has further confirmed the reliability of the model for purposeful water resources planning and development for the basin. However, in order to validate the model, the performance of the actual discharge and the ANN-based model in its ability to predict the River Kaduna discharge was tested. This was carried out by comparing the actual discharge output and the ANN-based model output and the result is shown in Fig. 19. From Fig. 19, it becomes clearer that the ANN-based model developed has been able to learn both the trend and the cycle of the River Kaduna discharge with a relatively high figure of merit as indicated by the 82% correlation value. This present a good effects on the reliability of the ANN model for forecasting discharge series that can enhance formidable policy that can stand the test of time for sustainable water resources planning for the study area. CONCLUSION The sustainable development of any region is of paramount importance to the planning, development and management of its water resource. As such, developing models to effectively predict the river discharge is quite apt. In this research work, ANN-based model was developed to forecast the discharge of River Kaduna depending on available variables (rainfall, temperature, relative humidity and stage height). The ANN-Based Model was developed in the NeuralWare PredictDemo environment. Hence based on the results obtained from this research, it can be concluded that ANN is a reliable modeling tool for predicting river discharge considering the strong, high and positive Correlation Coefficient the model displayed. From the foregoing, it can be concluded The author grateful to Dr. M.B. Muazu and Engr. Bashira both of Department of Electrical Engineering, Ahmadu Bello University, Zaria, Kaduna State for their effort to painstakingly taking me through several classes on how to use the ANN model toolbox. REFERENCES Antar, M.A., I. Elassiouti and M.N. Allam, 2006. Rainfall-runoff modelling using artificial neural network techniques: A Blue Nile catchment case study. Hydrol. Process., 20: 1201-1216. Arnell, N.W., R.P.C. Brown and N.S. Reynard, 1990. Impact of Climate Variability and Change on River flow regimes. In the U.K. Report, No. 107, Institute of Hydrology, Wallingford, U.K. ASCE, 2000a. Task committee on application of artificial neural networks in hydrology, II: Hydrologic application. J. Hydrol. Eng., 5: 124-136. ASCE, 2000b. Task committee on application of artificial neural networks in hydrology, artificial neural networks in hydrology, I: Preliminary concepts. J. Hydrol. Eng., 5: 115-123. Brion, G.M. and S. Lingireddy, 1999. A neural network approach to identifying non-point sourcesa of microbial contamination. Water Resour., 33: 3099. Cohen, S., 1987. Sensitivity of water resources in the great lakes region to changes temperature, precipitation, humidity and wind speed. The Influence of Climate Change and Climate Variability on the Hydrologic Regime and Water Resources, (Proceedings of the Vancouver Symposium, August 1987), IAHS Publ. No. 168, pp: 489-499. Folorunsho, J.O., 2004. An examination of some stream flow characteristics of river Kaduna state. Unpublished M.Sc. Thesis, Geography Dept., A.B.U., Zaria. Hsu, K., H.V. Gupta and S. Sorooshian, 1995. Artificial neural network modelling of the rainfall-runoff process. Water Resour. Res., 31: 2517-2530. Imrie, C.E., S. Durucan and A. Korre, 2000. River flow prediction using artificial neural networks, generalisation beyond calibration range. J. Hydrol., 233: 138-153. Jagtap, S.S., 1995. Change in annual, seasonal and monthly Rainfall in Nigeria, during 1961-1990 and consequences to agriculture. Acad. Sci., 7(4): 311426. 4291 Res. J. Appl. Sci. Eng. Technol., 4(21): 4284-4292, 2012 Jain, S.K., D. Das and D.K. Srivastava, 1999. Application of ANN for reservoir inflow prediction and operation. J. Water Resour. Plann. Mgmt. ASCE, 125(5): 263-271. Kilinc, I., H.K. Cigizoglu and A. Zuran, 2000. A comparison of three methods for the Prediction of Future Stream Flow Data. Technical and Documental Research of 14th Regional Directorate, State Hydraulic Works (DSI), Istanbul, Turkey. Kisi, O., 2004. River flow modelling using artificial neural networks. ASCE J. Hydrol. Eng., 9(1): 60-63. Kisi, O., 2005. Daily river flow forecasting using artificial neural networks. Turkey J. Eng. Environ. Sci., 29: 920. Krasovskaia, I. and I. Gottenschalk, 1993. Frequency of extremes runoff and its relation to climate fluctuations. Nordic Hydrol., 24: 1-12. Krasovskaia, I., N.W. Arnell and I. Gottenschalk, 1993. Development and application of procedures for classifying flow regimes in northern and Western Europe, In: FRIEND:-flow regimes from international experimental and network data. (Proceeding of 2nd International FRIEND Conference, UNESCO, Braunschweig, Germany, Oct., 1993), IAHS Publ. 6 No.221, pp: 185-193. Lettenmaier, D.P and T.Y. Gan, 1990. Hydrologic sensitivities of the Sacramento-san Joaquin river basin, California, to global warming. Water Resour. J., 26(1): 69-86. Loke, E., E.A. Warnaars, P. Jacobsen, F. Nelen and M. do Ce’u Almeida, 1997. Artificial neural networks as a tool in urban storm drainage. Water Sci. Technol., 36: 101-109. Maier, H.R. and G.C. Dandy, 1996. The use of artificial neural networks for the prediction of water quality parameters. Water Resour., 32: 1013-1022. Nemec, J. and J. Schaake, 1982. Sensitivity of water resources systems to climate variation. Hydrol. Sci. J., 27(3): 327-343. Oguntoyinbo, J.S., 1979. Climate in Geography of Nigerian Development. 2nd Edn., Heinemann Publisher, USA, pp: 14-39. Raman, H. and R. Sunilkumar, 1995. Multivariate modelling of water resources time-series using artificial neural networks. J. Hydrol. Sci., 40: 145163. Samuel, O., 1993. Analysis of climate variation and katsina-ala stream flow variation and its effects on water resources. M.Sc. Thesis, Benue State University, Makurdi, pp: 1-2. Saelthun, N.R, J. Bogen, M.H. Flood, T. Laumann, L.A. Roald, A.M. Tvede and B. World, 1990. Climate Change Impact on Norwegian Water Resources. NVE Publ. No. V42, Norwegian Water Resources and Energy Administration, Oslo, Norway. Sawa, B.A. and B. Buhari, 2010. Temperature variability and outbreak of meningitis and measles in Zaria, northern Nigeria. Resc. J. Appl. Sci. Egnr. Technol., 3(5): 399-402. Shamseldin, A.Y., 1997. Application of a neural network techniques to rainfall-runoff modelling. J. Hydrol., 199: 272-294. Sivakumar, B., A.W. Jayawardena and T.M.K.G. Fernando, 2002. River flow Forecasting: Use of phase space reconstruction and artificial neural networks Approaches. J. Hydrol., 265: 225-245. Smith, J. and R.N. Eli, 1995. Neural network models of rainfall-runoff process. J. Water Resour. Plan. Mgmt., 121: 499-508. Tokar, A.S. and P.A. Johnson, 1999. Rainfall-runoff modelling using artificial neural networks. J. Hydrol. Eng., 4(3): 232-239. Vehvilainen, B. and J. Lohvansuu, 1991. Effects of climate change on discharges and snow cover in Finland. J. Hydrol. Sci., 36: 109-121. Ward, R.C., 1968. Some runoff characteristics of British River. J. Hydrol., 6: 358-372. Ward, R.C., 1975. Principles of Hydrology. 2nd Edn., McGraw-Hill, London. Wateren-de-Hoog, B., 1993. The influence of variations of precipitation on discharge in the upper loire area as determined with flow duration curves. Instituut Voor Ruimtelijk Onderzoek, Report No GEOPRO 1993, 07, Utrecht, Netherlands. Zhang, Q. and S.J. Stanley, 1997. Forecasting raw-water quality parameters for the north saskatchewan river by neural network modelling. Water Resour., 31: 2340-2350 4292