Appropriate and acceptable prediction of bed load being carried by streams is vitally important for water resources quantity and quality studies. Although measuring the rate of bed load in situ is the most consistent method, it is very expensive and cannot be conducted for as many streams as the measurement of suspended sediment load. Therefore, in this study the role of suspended load on bedload prediction was examined by using sensitivity analysis. On the other hand, conventional sediment rating curves and equations can not predict sedi- ment load accurately so recently the usage of machine learning algorithms in- crease rapidly. Accordingly, soft computational methods are used in the study. These are; artificial neural network (ANN), support vector machine (SVM) mod- els and a decision tree (CHAID) model that is not used before in sediment studies. Some particular parameters are frequently used in these soft computational methods to form input sets. Hence, well known and commonly used three input sets and a new generated set are used as inputs to predict bedload and then the suspended load variable is added in these input sets. The performances of models with respect to input sets are compared to each other. To generate the results and to push the limits of models a very skewed and heterogeneous data is col- lected from distributed locations. The results indicate that the performance of ANN and CHAID tree models are good when compared to SVM models. The us- age of a suspended load as an additional input for the models boosts the model performances and the suspended load has significant contributions to all models.
Keywords: sediment prediction, bed load, suspended load, artificial neural net- works, support vector machines, CHAID tree models
1. Introduction
Modeling and predicting the amount of bed load and suspended load is an extremely important side of planning and handling the water resources projects. The sediment load transported by the streams may cause a decrease in a useful storage of a dam (Nakato, 1990; McBean and Al-Nassri, 1988). The
transportation of sediment also changes ecologic and hydraulic equilibrium of the river bed. Furthermore, the design of steady channels, estimation of bedding and degradation at platform piers and abutments, estimation of sand and grav- el mining effects, and the analysis of the ecological impact evaluation are also dependent to sediment load transport.
The sediment load of a stream is commonly determined from direct measure- ments or otherwise calculated indirectly by using sediment transport formulas. Even though direct measurement of sediment transport rate is more trustwor- thy, it is unfeasible and uneconomical to establish gauging stations at all desired locations and acquire data for a satisfactory long period of time. Furthermore, measurement of bed load is more expensive and complex than measurement of suspended load.
In the literature there are many sediment rate transportation models. These models have been proposed in different forms for many parameters as a function of the river and sediment characteristics. Some of them are obtained in a labora- tory environment, while others are developed using in-situ data or theoretical methods. On the other hand, most of the sediment transport equations need com- prehensive information on the channel, flow and sediment characteristics (Öztürk et al., 2001; Yang and Wan, 1991). With respect to the conditions under which the data are gathered, the same formulation could yield dissimilar scores of accuracy, and usually do not fit with the observed data. Therefore, none of such equations have achieved universal acceptance (Vanoni, 1971; Yang, 1996). Because of these facts it can be asserted that the assumptions stated in the derivation of these specific equations is only valid under certain situations and also is not to be re- garded as a general rule (Yang, 1972). Because of the encountered difficulties, the researchers strive to search easy methods to estimate sediment load. Initially, such relationships have obtained by using regression analysis and usually these models are called as sediment rating curves (e.g. Jain, 2001; Cigˇizogˇlu, 2002a, b; Öztürk et al., 2001). But in this technique the interior uncertainties are not con- sidered explicitly while determining the sediment yield with water discharge (Şen and Altunkaynak, 2003). Additionally, sediment rating curves does not contribute much on the insight of the physical meaning of used parameters and so, do not improve understanding of sediment transport processes (Yang, 1996). As known, the regression techniques can not determine the non linear relationships or it is only suitable to present simple non linearities after basic transformations. Re- cently, because of these problems, researchers are looking for simpler, cheaper and easier methods to predict sediment load, and they are beginning to use non- linear models such as neural networks to solve nonlinear problems.
There are many implementations of artificial neural networks (ANN) at almost all branches of science. The method is famous for its capacity to model the nonlinear relationships and high predictive accuracy. Motivated by success- ful applications of ANNs have been applied in hydrological engineering problems. In hydrology the method has been emerged as a strong application for planning
GEOFIZIKA, VOL. 32, NO. 1, 2015, 27–46 29
studies and management purposes. ANNs have been used for rainfall-runoff modeling, flow predictions, flow/pollution simulation, parameter identification, and modeling nonlinear/input-output time series (ASCE, 2000).
Jain (2001) used the ANN models to build up an integrated stage-discharge- sediment concentration relation for two watersheds of the Mississippi River. Cigˇizogˇlu (2002a, 2002b) used ANNs to analyze suspended sediment concentra- tions and made an assessment between ANNs and sediment rating curves for two catchments in the Northern England. He asserted that the results of ANN model are superior to classical sediment rating curve method. Nagy et al. (2002) made sediment discharge predictions that and concluded that the ANN model gives better results when compared to different widely used formulas of sediment discharge. They used Multi layer perceptrons (MLP) in their ANN model and indicated that MLP could capture the complex nonlinear behavior of the sedi- mentary series relatively better than the conventional models. Tayfur and Gul- dal (2005) computed the daily total suspended sediment in natural rivers by ANN and using a two dimensional unit sediment graph theory (2D-USGT) from pre- cipitation data. The evaluation of results demonstrated that the ANN model has better performance than the 2D-USGT. Raghuwanshi et al. (2006) designed an ANN model to estimate both runoff and sediment yield in daily and weekly time frame, for a small watershed. When compared to ANN applications, the other machine learning algorithm implementations on sediment modeling is scarce and new. Bhattacharya et al. (2007) compared the ANN and decision tree (DT) model performances on the bed load transport dataset and concluded that both machine learning algorithms give sufficient results but the ANN model is supe- rior to DT model. Oehler et al. (2012) used a boosted form of regression trees to to predict suspended-sediment reference (near-bed) concentration in six shelf areas of New Zeland. Kisi et al. (2012) compared the genetic programming results with ANN, adaptive neuro-fuzzy inference system (ANFIS) and support vector machine (SVM) results on the suspended load prediction. Misra et al. (2009) established SVM models to predict the daily, weekly and monthly discharge and sediment yield of an Indian watershed.
Since bed load observations are labor extensive and expensive, the sediment studies have been focused on total or suspended load models. A comprehensive study about bed load transportation is carried out by Sasal et al. (2009). The researchers used a large dataset and concluded that the developed ANN model gives satisfying predictive performance on bed load transport model studies. Yu et al. (2009) observed bed load sediment transportation rates of Diaoga River in China then they investigates the relations of bedload transport and some wide- ly used non dimensional parameters such shear stress and stream power. Gao (2011) derived a power formula to predict maximum bed load transport rates by using nonlinear regression models.
There are three main purposes in this article; the first of them is evaluating
the performances of widely-used machine learning algorithms while predicting
the bed load sediment. For this purpose ANN, SVM and chi-squared automatic interaction detection (CHAID) Tree model is used and compared in the study. The second aim is to assess the performances of well known and commonly used input sets of bed load sediment models. The third aim is examining the role of suspended load observations while predicting the bed load so Suspended load parameter is added all built models and sensitivity analysis are performed. The results indicate that the performance of neural networks and CHAID models are good when compared to SVM models. The usage of a suspended load variable as an input for the models boosts the model performances and has a significant contribution on model accuracy. The input sets have a similar predictive perfor- mance but only the Pektaş (2015) input set has given sufficient results for all applied models.
CS 340 Milestone One Guidelines and Rubric Overview: For this assignment, you will implement the fundamental operations of create, read, update,
Retail Transaction Programming Project Project Requirements: Develop a program to emulate a purchase transaction at a retail store. This
7COM1028 Secure Systems Programming Referral Coursework: Secure
Create a GUI program that:Accepts the following from a user:Item NameItem QuantityItem PriceAllows the user to create a file to store the sales receip
CS 340 Final Project Guidelines and Rubric Overview The final project will encompass developing a web service using a software stack and impleme
