Estimation of suspended sediment concentration from turbidity measurements using artificial neural networks

Abstract

Suspended sediment concentration (SSC) is generally determined from the direct measurement of sediment concentration of river or from sediment transport equations. Direct measurement is very costly and cannot be conducted for all river gauge stations. Therefore, correct estimation of suspended sediment amount carried by a river is very important in terms of water pollution, channel navigability, reservoir filling, fish habitat, river aesthetics and scientific interests. This study investigates the feasibility of using turbidity as a surrogate for SSC as in situ turbidity meters are being increasingly used to generate continuous records of SSC in rivers. For this reason, regression analysis (RA) and artificial neural networks (ANNs) were employed to estimate SSC based on in situ turbidity measurements. The SSC was firstly experimentally determined for the surface water samples collected from the six monitoring stations along the main branch of the stream Harsit, Eastern Black Sea Basin, Turkey. There were 144 data for each variable obtained on a fortnightly basis during March 2009 and February 2010. In the ANN method, the used data for training, testing and validation sets are 108, 24 and 12 of total 144 data, respectively. As the results of analyses, the smallest mean absolute error (MAE) and root mean square error (RMSE) values for validation set were obtained from the ANN method with 11.40 and 17.87, respectively. However these were 19.12 and 25.09 for RA. It was concluded that turbidity could be a surrogate for SSC in the streams, and the ANNs method used for the estimation of SSC provided acceptable results.

Prediction of suspended sediment concentration from water quality variables

“…3). The Eastern Black Sea Basin consists of 17 sub-watersheds, and the stream Harsit watershed having a main branch length of 143 km, and catchment area of 3,280 km 2 has a place in these subwatersheds [24,[27][28][29][30][31][32].…”

“…The streams Arzular, Korum, Ikisu, Cit, Kurtun, and Gavraz are important tributaries of it. There are four HEPPs in operation for the time being on Harsit, namely Akkoy I HEPP, Dogankent HEPP, Kurtun Dam and HEPP, Torul Dam and HEPP [24,[27][28][29][30][31][32]. Many dams and HEEPs are also under construction or planned, namely Akkoy II, Aladerecam, Aslancik, Avluca, Cileklikaya, Duzoren, Elmali, Gocen, Gokcebel, Koru, Kovacik, Kuletas, Tasoba, Tirebolu, Sogukpinar, Yasmakli, and so on.…”

“…One-liter surface water was sampled from approximately mid-depth of flow. High-density polyethylene bottles having 500 mL volume were rinsed with the stream water, shaken vigorously, and emptied before refilling for the analysis [24,33]. Sampling, preservation, and transportation of the stream water samples to the laboratory were as per standard methods [34].…”

“…Mount and Abrahart [23] reported a comprehensive set of single-input single-output neural network suspended sediment modeling experiments performed on two catchments in Puerto Rico. Bayram et al [24] evaluated whether the turbidity can produce a satisfactory prediction of the SSC, and improve an ANN method estimating the SSC based on in situ turbidity measurements in the stream Harsit, Eastern Black Sea Basin, Turkey. Demirci and Baltaci [25] proposed a fuzzy logic approach to estimate SSC from streamflow.…”

Wavelet and adaptive neuro-fuzzy inference system conjunction model for groundwater level predicting in a coastal aquifer

“…Under data-scarce conditions that are a common scenario in most of the developing countries, physical based models are highly restricted to be used [4]. In recent years, as alternative traditional methods, Artificial Intelligence (AI) techniques such as artificial neural networks (ANN) and adaptive neuro-fuzzy inference system (ANFIS) have been successfully used in the hydrological sciences [5][6][7][8].…”