Journal of Radioanalytical and Nuclear Chemistry, Vol. 269, No.1 (2006) 63–68 0236–5731/USD 20.00 Akadémiai Kiadó, Budapest © 2006 Akadémiai Kiadó, Budapest Springer, Dordrecht Prediction of the radioactivity in Hazar Lake (Sivrice, Turkey) by artificial neural networks F. Kulahci, 1, * A. B. Özer, 2 M. Doğru 1 1 Fırat University, Faculty of Arts and Science, Department of Physics, Elazığ, Turkey 2 Fırat University, Faculty of Engineering, Department of Computer Engineering, Elazığ, Turkey (Received December 13, 2005) This paper presents an Artificial Neural Network (ANN) model for determining the total radioactivity in Hazar Lake (Sivrice, Turkey). In order to cope with complex calculations and experiments required for the determination of total radioctivity. The proposed ANN system employs the individual training strategy with fixed-weight and supervised models. The simulation demonstrate the feasibility of the neural based model. Compared to the classical methods, the proposed ANN-based model makes the processes much easier. Introduction Many environmental radioactivity determination methods have been studied. 1–4 In these studies, the environmental radioactivity have been determined and the effects of these on humans and on the environment have been investigated. The obtained data have been envisaged by mathematical models. In these studies, complex mathematical methods and difficult experimental procedures were used, 5 whereas we obtain the same result using an easier method. Determination of the certain chemical parameters, such as pH, electrical conductivity (EC), total hardness (TH) and the water lake’s depth lead to the conditions that make possible a quantitative estimate of the total beta- and total alpha- radioactivity. 6 A relationship between the total alpha-, total beta- radioactivity and the total hardness was found by various scientific researches. Waters with high TH have high levels of alpha- and beta-radioactivity. 6 For the case of the total alpha-activity as a function of total hardness, a linear behavior could only be established in waters from granitic and quartzitic lithological types. Mostly, these structures are also seen in Hazar Lake. 7 Another factor which seems to have a decisive influence on the radioactivity of the water is the pH. A good linear behavior was found between pH and the total beta-radioactivity. In all cases the obtained regression coefficient is negative. Briefly, the ratio between the activity of the radioactive isotopes and the mass of the stable isotopes present in water decreases as the pH rises. 8 In this study, an ANN based model was applied to determine the total radioactivity in Hazar Lake. Since ANN can approximate a wide range of non linear functions to any desired degree of accuracy, recently many researches have used neural networks on different applications for identification and control of dynamic * E-mail: fatihkulahci@firat.edu.tr systems. However, the neural network is a static mapping. Moreover, it has the advantages of extremely fast parallel computation and fault tolerance characteristics. Experimental Description of the area Hazar Lake, located in the east of Turkey, is a tectonic lake within East Anatolian Fault Zone. The long axis is of about 20 km in the east-southeast and west- southwest direction. The altitude is 1248 m above sea level, and the surface area is 81 km 2 . The geologic form of the mountains is suitable to absorb and hold the radioactive minerals. The position of the lake and its aerial picture are given in Fig. 1. Determination of the chemical and physical parameters An Orion 230A digital pH meter is used for pH measurements. The EC measurements have been done by using Jenway 4070 digital conductometer. The TH measurements were made by a titrimetric method. Data set The radioactivity data set is taken from our previous work recorded between 1998–2001 for the Hazar Lake. 9 The average values of chemical measurements in the same stations during 3 years are shown in Table 1. We have obtained 45 chemical samples from all over the lake, 10 of them were from the surface and 35 samples from 12 different stations. The samples were taken from surface, mid-level and bottom of the lake. The symbols a, b and c were used for surface, mid-level and deep sample of the same station, respectively. The ANN calculations were made by using Matlab. 10