Reservoir Trap Efficiency Using Artificial Neural

International Journal of Ecology & Development Winter 2010; Vol. 15, No.W10 ; Int. J. Ecol. Dev. ISSN 0972-9984 ( Print ); ISSN 0973-7308 (Online) Copyright © 2010 IJED (CESER Publications)

Reservoir Trap Efficiency Using Artificial Neural Networks Vaibhav Garg1 and V. Jothiprakash 2 Department of Civil Engineering, Indian Institute of Technology Bombay, Powai, Mumbai – 400 076, India. [email protected], [email protected]

ABSTRACT The developmental activities, undergoing within the vicinity of reservoir watershed, result in discharge of large volumes of sediment into the reservoir, which in turn affects biodiversity of the reservoir and reservoir useful life period. These necessitate the reservoir sedimentation studies. The volume of sediment deposited or trapped in a reservoir can easily be quantified by the simple knowledge of its trap efficiency (Te), which is been estimated by various conventional empirical approaches till date. In the present study empirical methods proposed by Brown and Brune have been modified and adopted to estimate Te of Pong Reservoir (Beas Dam) on the Beas River in Kangra district of Himachal Pradesh, India. The major contribution of the study is incorporation of reservoir age in an empirical equation. It is found that the present study estimation of the Te is much better than any other conventional methods. Further, an attempt has been made to estimate ‘Te’ using Artificial Neural Networks (ANN). The study shows that ANN can successfully be applied to estimate not only the trend but also the magnitude of the Te better than the conventional models. KEYWORDS: Age of the reservoir, Brown method, Brune method, Gill method, Reservoir Sedimentation, Trap efficiency, Artificial Neural Networks. Mathematics Subject Classification Numbers: 92B20, 68W20, 68T10 JEL Classification: C67, Q25

INTRODUCTION The catchment erosion and reservoir sedimentation are a major concern for conservation biologists, due to their ecological consequences (Myers, 1984). Hitherto, Sediment deposition is a major inevitable phenomenon in all the reservoirs. From hydrological point of view, the sedimentation, gradually reduces the available storage and threatens the primary uses of the reservoirs in a time less than the expected useful lifespan. On the other hand, sedimentation alters biodiversity of lake in many ways. The reduction in penetration of light, due to suspended sediments, affects both plankton and benthos (Cairns, 1968; Fuller, 1974; Berwick and Faeth, 1988). It reduces photosynthetic rates also (Cohen, et al. 1993). The sediments in suspension may also interfere with the feeding apparatuses of filter feeding organisms (Cohen, et al. 1993). The settling sediments, affects algal communities directly by blanket them and indirectly reduce foraging efficiency of herbivorous fishes (Grobbelaar, 1985). The sediments may physically damage body surface of aquatic organisms by abrasion www.ceser.res.in/ijed.html www.ceserp.com/cp-jour/

International Journal of Ecology & Development

(Harman, 1974; Bruton, 1985). Finally, sediments potentially alter the nutrient dynamics of entire water body (Golterman et al., 1977; Sly, 1986; de Groot and Golterman, 1990; Graham, 1990; Golterman, 1991). In order to study the impact of sedimentation on ecology in the stored water bodies namely reservoirs, it is very important to have the temporal quantification of the sediment rate. Therefore, accurate estimation of the amount of sediment that will be trapped in a reservoir is very important and the present study aims the same. In this respect the trap efficiency (Te) of the reservoir may be an informative factor in the estimation of sediment distribution, storage loss and useful life (Heinemann, 1981; Verstraeten and Poesen, 2000; Yeoh, et al. 2004). The Te is the proportion of the incoming sediment that is trapped / deposited in a reservoir. It depends upon many parameters out of which the important are (1) capacity and shape of reservoir, (2) inflow rate of sediment and water (3) sediment size distribution and (4) type of reservoir operation (Annandale, 1987; Yang, 1996; Morris and Fan, 1998; Verstraeten and Poesen, 2000; Campos, 2001; Yang, 2003). The estimation of Te has been a subject of several empirical studies, by fitting the data collected from existing reservoir, since the 1940’s. Many empirical relationships Brown (1944), Churchill (1948), Brune (1953) and their modifications (Dendy, 1974; Gill, 1979; Heinemann, 1981; Jothiprakash and Garg, 2008) are available in the literature for estimating the Te of reservoirs. These relationships relate Te to capacity of reservoirs, annual inflow or catchment area of the reservoir. Even for the reservoirs, with totally different characteristics from the reservoirs those data have been used to derive these relationships, these are the most widely used methods to estimate Te. Most of these empirical relationships are developed only for normally ponded large reservoirs. Also, these methods are not suitable for extreme flow events (Verstraeten and Poesen, 2000). Many researchers have already questioned Brune (1953) method, which uses the capacity to inflow (C/I) ratio alone to describe the dynamics of sedimentation in reservoirs (Gottschalk, 1964 and Heinemann, 1981). In the present study, the adequacy of some of the available methods in predicting the Te has been examined using the available data for the Pong Reservoir on Beas River in Kangra district, Himachal Pradesh, India. With the scarcity in the availability of sediment data, the conventional methods might be complicated, time consuming and cumbersome estimation procedures. However, reservoir sedimentation estimation has never been an easy task due to complicated simultaneous processes involved such as sediment transport, erosion and deposition. In recent years, ANNs have shown exceptional performance as regression tool, especially when used for pattern recognition and function estimation. They are highly non-linear and can capture complex interactions among the input variables in a system without any prior knowledge about the nature of these interactions (Hammerstrom, 1993). Therefore, an attempt has also been made to estimate Te of the reservoir using ANN with the available data length.

The main advantage of ANNs is that one does not have to explicitly assume a model form, which is a prerequisite in conventional modeling approaches. Indeed, in ANNs the data points themselves generate a relationship of possibly complicated or orthodox shape. In comparison to the conventional

15

Int. J. Ecol. Dev.; Vol. 15, No. W10, Winter 2010

methods, ANNs tolerate imprecise or incomplete data, approximate results, and are less vulnerable to outliers (Haykin, 1999). They are highly parallel i.e., their numerous independent operations can be executed simultaneously. The estimation of the hydrologic data plays a significant role in solving water resources engineering problems. ANNs have been applied to many kinds of hydrologic data over the last two decades (ASCE Task Committee, 2000a, b). In the present study cause effect MultiLayer Perceptrons (MLPs) ANN model has been developed using the available measured data for estimation of Te of a large reservoir namely Pong Reservoir in India. The parameters; annual rainfall, annual inflow and age of the reservoir were decided as the input (on basis of parameters causing and affecting the sedimentation process). The reservoir Te was considered as the output parameter. It was found that the developed ANN model has captured the trend of Te well. A BRIEF REVIEW OF ADOPTED TECHNIQUES Conventional Empirical Methods One of the first researchers to derive an empirical relationship to estimate the reservoir Te using reservoir characteristics was Brown (1944). Brown developed a curve that relates Te (%) to the reservoir capacity (in acres/feet) and watershed area (in mi2) ratio based on data from 15 reservoirs. This curve can be represented by a general equation given below: (Gill, 1979; USACE, 1989; Campos, 2001) in which

N is a coefficient which varies from 0.046 to 1.0. 1

Te

According to USACE (1989),

N increases

1 C· § ¨1 N ¸ A¹ ©

(1)

with increase in retention time, increase in average grain

size and for reservoir operations that prevents release of sediment through sluicing or movement of sediment towards the outlets by pool elevation regulation. A value of the average conditions, and values of

N=

N = 0.1 was recommended for

1.0, 0.1 and 0.046 for the coarse, medium and fine

sediments, respectively (Gill, 1979). The major advantage of Brown’s method is that it uses only two parameters i.e., catchment area and the reservoir capacity. However, according to USACE (1989), the relationship proposed by Brune is considered to be more accurate than that of Brown, since it considers the annual inflow and the reservoir capacity. As stated earlier, Dendy (1974); Gill (1979); Heinemann (1981); Garg and Jothiprakash (2008) modified Brune’s method and developed algebraic best fit equations which provide a very close fit to the three curves proposed by Brune. Gill (1979), provided the following equations as a very close fit to the median curve proposed by Brune for medium grained sediments: Te

C / I ( 0.012 1.02C / I )

(2)

Garg and Jothiprakash (2008) developed the following equation for the relationships proposed by Brune for medium grained sediment (only for Pong Reservoir) by incorporating the age of the reservoir: 16


Te ,t

§ Ct · ¨¨ ¸¸ © It ¹ ª § Ct · Ct º «0.00025 0.01u ¨¨ ¸¸ 0.0000045 u t u » I t »¼ «¬ © It ¹

(3)

Where Te,t is the trap efficiency of the reservoir at time ‘t’, Ct is the capacity of the reservoir at time ‘t’, It is the inflow during the time ‘t’ and t is the age of the reservoir in years from its inception. Artificial Neural Networks Approach A limited application of ANN in the area of reservoir sedimentation has been reported in the literature. Most of work reported is on basin erosion prediction or estimation. Lee et al. (2006) conducted quantitative estimation of reservoir sedimentation for Shihmen Reservoir watershed in Taiwan. Temporal variations of water surface elevation, discharge, and concentration of suspended sediment were measured during three typhoon events (Rammason, Nakri and Sinlaku) on the field. A numerical model, Hydrological Simulation Program FORTRAN (HSPF), developed by the United States Environmental Protection Agency was adopted to simulate the sediment yield. However, the calibration/verification of such a model is complicated and tedious. Therefore, ANN model was proposed to estimate sediment yield. The synthetic sediment yield data generated from HSPF model has been used to develop Back Propagation (BP) ANN model due to scarcity of such data. The rainfall intensity and discharge were considered as input parameters against sediment yield as an output parameter.

Abrahart and White (2001) conducted some initial pathfinder experiments to assess the competence of a BP network to produce a combined model of sediment transfer occurring under different types of agriculture and land management conservation regimes. Cigizoglu (2004) forecasted daily suspended sediment in a stream by MLPs. Cigizoglu (2002a) used ANNs to forecast and estimate sediment concentration values. The sediment concentration estimation, on the other hand, using only observed river flow values and the previous sediment value in a nearby river as input, provided realistic approximations in terms of Mean Squared Error (MSE) and total sediment amount. Similarly, Cigizoglu (2002b) compared ANNs and sediment rating curves for two rivers with very similar catchment areas and characteristics in the north of England. According to Licznar and Nearing (2003), neural networks may provide a user-friendly alternative to the complex physically based models for soil erosion prediction. Sarangi and Bhattacharya (2005) compared the performance of ANNs model for sediment loss prediction with regression model for Banha watershed in India. Two ANN models, one geomorphology-based and another non-geomorphology-based for the prediction of sediment yield were developed and validated using the hydrographs and silt load data of 1995–1998 for the Banha watershed in the Upper Damodar Valley in Jharkhand state in India.

Sarangi et al. (2005) developed ANN and regression models using water-scale geomorphologic parameters to predict surface runoff and sediment loss of the St. Esprit Watershed, Quebec, Canada. Agarwal et al. (2005) developed feed-forward error BP ANN and Linear Transfer Functions sediment 17


yield models for the Vamsadhara River basin. Cigizoglu and Alp (2006) developed ANN model for river sediment yield using generalized regression algorithm. Generalized Regression Neural Networks does not require an iterative training procedure as in BP method. ANN models were developed by Raghuwanshi et al. (2006) to predict both runoff and sediment yield on a daily and weekly basis, for a Upper Siwane River watershed, India. A total of five models were developed for predicting runoff and sediment yield, of which three models were based on a daily interval and the other two were based on a weekly interval. It is found that all the above models were application of ANN in estimating the sediment yield from a catchment or sediment load in canals only. No application of ANN has been reported to estimate the Te of the reservoir directly. Thus, in the present study it was aimed to develop an ANN model to predict the Te of a reservoir. STUDY AREA In the present study, the Te has been estimated for the Pong Reservoir located on the Beas River in Kangra district of Himachal Pradesh, India. For this reservoir, Te data is available for the period 1980 to 2006 on the basis of hydrographic survey. The adequacy of available methods has been evaluated using these observed values of Te. This reservoir has an enormous water-spread of 260 km2 at full reservoir level and impounds 8578.99 x 106 m3 of water. The live storage of the reservoir is 7291.22 x 106 m3 and the dead storage is 1287.773 x 106 m3.

The catchment area of river Beas is fern shaped. The total catchment area above dam site is 12,562 km2 out of which only 777 km2 is under snow and rest of the area is covered by forests, degraded forests, cultivated land and uncultivated withered rocks (BBMB, 2006). The Beas catchment along with location of Pong dam is shown in Fig. 1. The annual yield of the catchment is about 10,720 x 106 m3. Pandoh dam, constructed about 140 km upstream of Pong Dam, diverts the normal flow of river to River Satluj (Beas-Satluj Link Project) and intercepts an area of 5278 km2. With the construction of Pandoh Dam, the sediment contribution of Beas River to this Pong was reduced to a large percent as major sediment load brought in the river up to Pandoh gets deposited in the Pandoh reservoir. Thus, the Pandoh Dam acts as a check dam for Pong Reservoir.

RESULTS AND DISCUSSION Bhakra Beas Management Board (BBMB) is carrying out the hydrographic survey annually to estimate the sediment volume along with the measurements of annual rainfall, annual inflow and sediment yield. The data collected from the BBMB for the period of 1980 to 2006 as given in the Table 1. The availability of such data has made present study suitable to consider the variation of Te with time (age of the reservoir). As an initial step, the suitability of some of the available methods in estimating the Te has been examined using the above data. In the present study, methods proposed by Brown (1944) and Brune (1953) have been applied to estimate Te of Pong Reservoir on the Beas River, India for medium grained sediments. Garg and Jothiprakash (2008) has mentioned that sediments incoming into this particular reservoir are mostly medium grained in nature. The results obtained from this analysis are presented in the Table 1. The value of 18

N = 0.022 in Brown’s method is


found to be a best fit coefficient for the present set of data. Table 1 further shows that the methods proposed by Brown and Gill yield fairly constant value of Te indicating no variation with age of the reservoir. However, the modified Brune’s median curve equation proposed by Garg and Jothiprakash (2008) takes into account this effect of age on the Te as shown in Table 1. From Table 1 it can be seen that the average observed Te is 97.23% and the average Te from the present study is 97.49%, other approaches are showing more deviation from the observed values. Thus present study shows that incorporation of age of reservoir in empirical equations improved the Te estimation. Further, an attempt has been made to apply ANN technique for Te estimation. There are no fixed rules for developing an ANN model, even though a general framework can be followed based on previous successful applications in engineering. Using available data of the study area, trial and error approach was employed in the present analysis to select the appropriate ANN architecture. In the present study, Multi Layer Perceptrons (MLP) ANN model architectures consisting of three layers to estimate Te of the reservoir was developed as shown in Figure 2.

The number of input parameters in the ANN was determined on basis of parameters causing and affecting the sedimentation process and is also easily measurable at the reservoir site. The annual rainfall, annual inflow, and age of the reservoir were considered as input nodes. The number of nodes in the hidden layer plays a significant role in ANN model performance; it has been determined by trialand-error procedure. By trial and error a single hidden layer with 4 numbers of hidden nodes were selected. The Sigmoid and Hyperbolic Tangent (tan h) transfer functions have been tried corresponding to a single output Te. The split-sample approach was used, in which part of the available data is used to develop a predictive relationship, which is then tested with the remaining data. For analysis, first 70% and remaining 30% of the data length has been chosen for training and testing respectively to avoid any parsimony network. It was found that changes in epoch size had no significant effect on ANN performance using validation datasets. For training purpose BP algorithm was used along with Momentum, Conjugate Gradient (CG) and Levenberg-Marquardt (LM) as learning rules.

The learning process is terminated when an optimum prediction statistics (MSE = 0.01) is obtained in relation to epoch size and validation results. Once the training process was satisfactorily completed, the network was saved, the test and validation datasets recalled, and values predicted by the model were compared with the observed values. If the prediction error statistics for these datasets are acceptable, then the neural network structure is considered to perform well for estimating the Te with different sets of data. The performance of the models was tested through the common statistical indicators such as coefficient of correlation (R), Mean Square Error (MSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and Nash-Sutcliff (1970) Efficiency (E).

19


The performance statistics for the testing period of few best ANN models developed are presented in Table 2. Based on coefficient of correlation (R), it was found that feed forward BP ANN model with a structure of 3-4-1 has captured the pattern in the Te very well with sigmoid activation function, momentum factor 0.7 and learning rate of 1.0. The network 3-4-1 is also a non-parsimonious network as number of connection weights are less than the number of dataset used for development. Thus, it may be concluded that based on the overall performance, the ANN model developed using 70% data for training and 30% data for testing performed the best.

The results in terms of trend of observed and estimated Te for conventional as well as ANN models are shown in Figure 3 for the testing period of ANN model. From Figure 3, it can be seen that the trend of Te is well captured by ANN model. From this Figure 3, it can be seen that the conventional methods estimated only the trend of Te, whereas, the developed ANN model has captured the magnitude too. The correlation analysis among various methods (presented in Table 3), shows that present study empirical equation with age of reservoir and ANN model performed equally better than other models in estimation of Te. CONCLUSIONS The sediment trap efficiency (Te) of Pong Reservoir on the river Beas in Kangra district, Himachal Pradesh, in foothill of the Himalaya, India has been estimated using Brown and Brune empirical methods. It was noticed that the constant

N=

0.022 instead of 0.1 in Brown’s method yields result

close to the observed values of Te for the Pong Reservoir. Both Brown and Gill’s method yielded the average trend of the Te. It was also observed that the developed regression equation with age of the reservoir is fairly applicable to this particular reservoir. For reservoirs with different characteristics, the existing Brown and Brune methods need modification. Further, an effort was made to employ the ANN approach for estimation of Te in this particular reservoir. The results of this study are encouraging and show that the above developed ANN models can be applied to estimate the Te with less effort and time. It may be concluded that such studies may be sometimes useful to prevent subsequent sedimentation damage to the underwater reserve ecosystem.

REFERENCES Abrahart, R.J., and White, S.M. 2001, Modeling Sediment Transfer in Malawi: Comparing Backpropagation Neural Network Solutions against a Multiple Linear Regression Benchmark Using Small Data Sets, Phys. Chem. Earth (B), 26(1), 19-24. Agarwal, A., Singh, R.D., Mishra, S.K., and Bhunya, P.K. 2005, ANN-based Sediment Yield Models for Vamsadhara River Basin (India), Water SA, 31(1), 95-100. Annandale, G.W., 1987, Reservoir Sedimentation, Elsevier Science Publishers, New York. ASCE (American Society of Civil Engineers) Task Committee on Application of the Artificial Neural Networks in Hydrology, 2000a, Artificial neural networks in hydrology I: Preliminary concepts, J. Hydrologic Engg., ASCE, 5 (2), 115-123.

20


ASCE (American Society of Civil Engineers) Task Committee on Application of the Artificial Neural Networks in Hydrology, 2000b, Artificial neural networks in hydrology II: Hydrologic Applications, J. Hydrologic Engg., ASCE, 5 (2): 124-137. BBMB, 2006, Pong Reservoir Sedimentation Studies, Sedimentation Survey Report 2005-2006, BBMB, Pong Dam Circle, Talwara, India. Berwick, N. and Faeth, P., 1988, Simulating the impacts of sewage disposal on coral reefs, Proceedings of the 6th International Coral Reef Symposium, Australia, 2, 353-361. Brown, C.B., 1944, Discussion of Sedimentation in Reservoirs, J. Witzig. Proceedings of the American Society of Civil Engineers, 69, 1493-1500. Brune, G.M., 1953, Trap Efficiency of Reservoirs, Trans. Am. Geophysical Union, 34(3), 407-418. Bruton, M.N., 1985, The effects of suspensoids on fish, Hydrobiologia, 125, 221-241. Cairns, J., 1968, Suspended solids standards for the protection of aquatic organisms, Purdue University England Bulletin, Part 1, 129, 16-27. Campos, R., 2001, Three-Dimensional Reservoir Sedimentation Model, PhD Thesis, Department of Civil Engineering, University of Newcastle, Newcastle. Churchill, M.A., 1948, Discussion of “Analysis and Use of Reservoir Sedimentation Data”, in L.C. Gottschalk (Ed.), Proceedings of Federal Interagency sedimentation Conference, Denver, 139-140. Cigizoglu, H.K., 2002a, Suspended Sediment Estimation and Forecasting using Artificial Neural Networks, Turkish J. Eng. Env. Sci., 26, 15-25. Cigizoglu, H.K., 2002b, Suspended Sediment Estimation for Rivers using Artificial Neural Networks and Sediment Rating Curves, Turkish J. Eng. Env. Sci., 26, 27-36. Cigizoglu, H.K., 2004. Estimation and Forecasting of Daily Suspended Sediment data by Multi-layer Preceptrons, Advances in Water resources, 27, 185-195. Cigizoglu, H.K. and Alp, M. 2006, Generalized Regression Neural Network in Modeling River Sediment Yield, Advances in Engineering Software, 37, 63-68. Cohen, A.S., Bills, R., Cocquyt, C.Z. and Caljon, A.G., 1993, The impact of sediment pollution on biodiversity in Lake Tanganyika, Conservation Biology, 7(3), 667-677. de Groot, C. and Golterman, H., 1990, Sequential fractionation of sediment phosphate, Hydrobiologia, 192, 143-148. Dendy, F.E., 1974, Sediment Trap Efficiency of Small Reservoirs, Trans. of ASAE, 17(5), 898-988. Fuller, S. 1974, Clams and mussels. In Hart, C. and Fuller, S. (Ed.) Pollution ecology of freshwater invertebrates. Academic Press, New York. Garg, V. and Jothiprakash, V. 2008, Trap Efficiency Estimation of a Large Reservoir, ISH Journal of Hydraulics. 14(2), pp. 88-101. Gill, M.A., 1979, Sedimentation and Useful Life of Reservoirs, Journal of Hydrology, 44, 89-95. Golterman, H., 1991, Influence of sediments on lake water quality, Proceedings of the International Conference on Land-Water Interactions, New Delhi, India. Golterman, H., Viner, A. and Lee, G. 1977, Sediment/Freshwater interaction. Developments in hydrobiology, 9. Junk, The Hague, Netherlands. Gottschalk, L.C. 1964, Reservoir Sedimentation, In V. T. Chow (Ed.), Handbook of Applied Hydrology, Section 17-1, New York, McGraw-Hill.

21


Graham, A., 1990, Siltation of stone-surface periphyton in rivers by clay-sized particles from low concentrations in suspension. Hydrobiologia, 199, 107-115. Grobbelaar, J., 1985, Phytoplankton productivity in turbid waters, Journal of Plankton Research, 7, 653-663. Hammerstrom, D., 1993, Working with neural networks, IEEE Spectrum, July, 46–53. Harman, W., 1974, Snails (Mollusca: Gastropoda). In C. Hart and S. Fuller (Ed.) Pollution ecology of freshwater invertebrates, Academic Press, New York, 275-313. Haykin, S., 1999, Neural Network: A Comprehensive Foundation, Prentice Hall—Upper Saddle River, New Jersey. Heinemann, H.G., 1981, A New Sediment Trap Efficiency Curve for Small Reservoirs, Water Resources Bulletin, 17(5), 825-830. Jothiprakash, V. and Garg, V. 2008, Re-look to conventional techniques for trapping efficiency estimation of a reservoir, International Journal of Sediment Research, 23(1), 76-84. Lee, H-Y., Lin, Y-T. and Chiu, Y-J.. 2006, Quantitative Estimation of Reservoir Sedimentation from Three Typhoon Events, J. Hydrologic Engg., ASCE, 11(4), 362-370. Licznar, P. and Nearing, M. A. 2003, Artificial Neural Networks of Soil erosion and Runoff Prediction at the Plot Scale, Catena, 51, 89-114. Morris, G.L. and Fan, J. 1998, Reservoir Sedimentation Handbook, McGraw Hill, New York, USA. Myers, N., 1984, The primary source, tropical forests and our future, Norton, New York. Nash J.E., and Sutcliffe J.V. 1970, River flow forecasting through conceptual models – 1: A discussion of principles, J. Hydrol., 10, 282–290. Raghuwanshi, N.S., Singh, R. and Reddy, L.S. 2006, Runoff and Sediment Yield Modeling Using Artificial Neural Networks: Upper Siwane River, India, J. Hydrologic Engg., ASCE, 11(1), 71-79. Sarangi, A. and Bhattacharya, A.K. 2005, Comparison of Artificial Neural Network and Regression Models for Sediment Loss Prediction from Banha Watershed in India, Agricultural Water Management, 78, 195-208. Sarangi, A., Madramootoo, C.A., Enright, P., Prasher, S.O. and Patel, R.M. 2005, Performance evaluation of ANN and geomorphology-based models for runoff and sediment yield prediction for a Canadian watershed, Current Science, 89(12), 2022-2033. Sly, P., 1986, Sediments and Water interactions. Springer Verlag, New York. USACE (U.S. Army Corps of Engineers) 1989, Engineering and design: Sedimentation investigations of rivers and reservoirs, Engineering Manual 1110-2-4000, Washington, D.C. Verstraeten, G. and Poesen, J. 2000, Estimating Trap Efficiency of Small Reservoirs and Ponds: Methods and Implications for the Assessment of Sediment Yield, Progress in Physical Geography, 24, 219-251. Yang, C.T., 1996, Sediment Transport: Theory and Practice, McGraw-Hill, New York. Yang, X., 2003, Manual on Sediment Management and Measurement, World Meteorological Organization, Operational Hydrology Report No. 47, WMO-No. 948, Secretariat of the World Meteorological Organization – Geneva, Switzerland. Yeoh, J.S., Loveless, J.H. and Siyam, A.M., 2004, New Approach in Determining Useful Life of Reservoirs, In Yazdandoost F. and Attari J. (Eds.), Proceeding of the Hydraulics of Dams and River Structures International Conference, Tehran, Iran, 26-28 April, 229-236.

22


Table 1 Comparison of Observed and Estimated Te

Year 1974-80 1980-82 1982-84 1984-86 1986-88 1988-89 1989-90 1990-91 1991-92 1992-93 1993-94 1994-96 1996-97 1997-98 1998-99 1999-00 2000-01 2001-02 2002-03 2003-04 2004-05 2005-06

Reservoir Age

Annual Inflow

Annual Rainfall

mm 106 m3 6 9320 78302.58 8 3832 18551.84 10 2581 15381.75 12 6586.89 3250 14 15258.4 3155 15 6722.58 2441 16 11730.6 1306 17 7326.99 2012 18 8807.19 1504 19 7647.7 1837 20 13469.8 1491 22 13075.1 3838 23 8967.55 1758 24 12840.7 1934 25 10583.4 1773 26 10756.1 1702 27 8091.76 1636 28 8252.12 1817 29 5723.44 1183 30 6932.16 1708 31 6562.11 1507 32 9658.15 1559 Coefficient of Correlation (R) Average

C/I Ratio

Observed Te

0.581 0.633 0.679 0.705 0.735 0.72 0.734 0.732 0.741 0.743 0.752 0.742 0.747 0.74 0.739 0.736 0.743 0.747 0.757 0.764 0.771 0.77

96.8 96.65 95.4 96.5 96.55 97.2 96.8 97.1 95.7 96.8 96.4 97.37 97.8 98.1 98.2 97.93 97.24 97.56 97.68 98.37 98.33 98.63 1.00 97.23

Trapping Efficiency (Te) Modified Gill Brown Brown N = 0.022 Eq. 2 Eq. 1 99.29 96.87 96.09 99.29 96.85 96.25 99.28 96.81 96.37 99.27 96.78 96.43 99.27 96.76 96.49 99.27 96.75 96.46 99.26 96.74 96.49 99.26 96.73 96.49 99.26 96.73 96.51 99.26 96.73 96.51 99.26 96.72 96.53 99.26 96.71 96.51 99.25 96.70 96.52 99.25 96.69 96.50 99.25 96.68 96.50 99.25 96.67 96.50 99.25 96.67 96.51 99.25 96.66 96.52 99.24 96.65 96.54 99.24 96.64 96.55 99.24 96.64 96.57 99.24 96.63 96.56 -0.74 -0.74 0.43 99.26 96.72 96.47

Table 2 Performance Statistics of the few best ANN Models during Validation Period

Performance Criteria 3-4-4-1 Sigmoid/Momentum 3-4-1 Sigmoid/LM 3-6-1 Sigmoid/Momentum 3-4-1 Sigmoid/Momentum

R

MSE (in 106 m3)

RMSE (in 106 m3)

MAE

E

0.807

1.153

1.074

1.00

-0.821

0.801

1.172

1.678

1.157

-0.32

0.971

0.548

0.740

0.63

0.133

0.997

0.539

0.734

0..553

0.147

23

Present Study Eq. 3 96.20 96.62 96.96 97.18 97.40 97.39 97.50 97.29 97.38 97.43 97.52 97.53 97.60 97.62 97.67 97.71 97.78 97.85 97.93 98.00 98.08 98.12 0.66 97.49


Table 3 Correlation Analysis of all the Methods for the validation period of ANN Model (i.e., from 2000-2001 to 2005-2006)

Observed

Modified Brown

Brown

Gill

Present Study Brune

Observed

1.00

Brown

-0.97

1.00

Modified Brown

-0.97

1.00

Gill

0.95

-0.97

-0.97

1.00

Present Study Brune

0.97

-0.99

-0.99

0.99

1.00

ANN

0.99

-0.97

-0.97

0.94

0.97

N

1.00

Himachal Pradesh

INDIA

Beas Dam

ANN

Pong Reservoir

FIG. 1 LOCATION OF PONG RESERVOIR ON BEAS RIVER

24

1.00


Fig. 2 ANN Architecture used for sediment Te estimation

100 Observed

Brown

Modified Brown

Gill

Present Study

ANN

Trap Efficiency, Te (%)

99

98

97

96 2000

2001

2002

Time (Year)

2003

2004

2005

Fig. 3 Comparative Plot of Estimated Te for Pong Reservoir for the Validation Period of ANN Model

25