Background Concentration Assessment Techniques

Within the context of environmental quality researches, various approaches have been introduced to the literature depending on the features of available data and characteristics of the study area for the determination of background concentration.

Qualitative investigations and quantitative implementations (like modeling and statistical analysis) are two main approaches that are proposed for the determination of background concentration.

Runnells, Shepherd, & Angino (1992) stated that the qualitative approach is one of the most simple and useful ways of background assessment. This approach is based on a comparison between currently existing data and historical water quality data, which represents the status of an undisturbed natural environment (Runnells, Shepherd, & Angino, 1992). On the other hand, knowledge obtained from historical metal concentration data, which represents the natural background, goes under an alteration over time with geologic and climatic changes. Even though concentration values obtained from historical observations give purely natural results of past times, these values may not represent the current pristine status of the target environment.

Therefore, background concentration assessment based on a comparison of the measurement results from current and past datasets may not reflect correct background values.

Regarding quantitative approaches, modeling and statistical techniques are widely implemented procedures for the determination of background values. Carrasco-Cantos, Vadillo-Pérez, & Jiménez-Gavilán (2013) emphasized that hydrochemical modeling enables to investigate variations of water quality and the general transportation mechanism of chemicals. However, hydrochemical modeling studies require comprehensive information, including detailed data about the geological and hydrological characteristics of the study area. Moreover, the implementation of these modeling techniques is only applicable for small environments with a high number of parameters observed through long measurement periods (Urresti-Estala, Carrasco-Cantos, Vadillo-Pérez, & Jiménez-Gavilán, 2013). As a consequence, when

hydrodynamic properties, geology, complex transportation mechanism, biological and chemical interactions of metals with the contact environment are taken into account, the accuracy of modelling requires quite an extensive and detailed available database and also expert knowledge from different field of studies.

Compared to qualitative methods and modelling studies, statistical techniques have emerged as a reliable approach for determining background concentration. Ander et al. (2013) suggested that statistical methods provide to obtain robust background concentration results by reducing the impacts of anthropogenic point pressures on datasets. Implementation of analysis with a minimum number of assumptions is one the most substantial feature of a methodology to be followed in order to conduct accurate background analysis. With the advantages of relatively low subjectivity and strong assessment, statistical methods are accepted as a precursor in many studies performed recently in literature for background concentration determination purpose (Urresti-Estala, Carrasco-Cantos, Vadillo-Pérez, & Jiménez-Gavilán, 2013; Apitz, Degetto, & Cantaluppi, 2009; Masetti, Sterlacchini, Ballabio, Sorichetta, & Poli, 2009; Peh, Miko, & Hasan, 2010). For the assessment of background concentration, there are numerous different statistical methodologies recently implemented. The clean stream approach, erosion model, sediment approach, monitoring data approach, and their modifications are some of the widely used approaches for the determination of background concentration. In the clean stream approach, background concentration is assumed as the concentration determined in a pristine environment rather than the calculated value (Oste, Zwolsman, & Klein, 2012).

According to the background assessment studies performed by Oste et al. (2012), in order to obtain reliable results from the clean stream approach, it should be ensured that whether the target river is truly pristine or not. However, it is considerably complex to define a river environment as pristine due to the constant and complex interaction mechanism of the river bodies with the surrounding environment.

Therefore, the accuracy of the clean stream approach is questionable. In the sediment method, background concentrations are determined by using sediment partition

freshwater sources (Vijver, Spijker, Vink, & Posthuma, 2008; Oste, Zwolsman, &

Klein, 2012). This method can be reasonable when well-established Kp values and the data from undisturbed sediment are available. However, since the reliability of the method strictly depends on the value of Kp, this method is quite simple to implement, which creates uncertainty in the results.

As explained in the previous sections, low-level anthropogenic fractions are unlikely to be completely removed from the monitoring data. However, a statistical approach to be followed for the determination of NBCs should aim the elimination of major anthropogenic pressures and mitigation of diffuse anthropogenic inputs from the data as much as possible. For the implementation of this purpose, a low percentile analysis of the monitoring data was conducted in order to determine NBCs of the selected metals and metalloids. The percentile analysis can be conducted on a low-level basis (5^th, 10^th percentile) or high-level basis (50^th, 90^th percentile) depending on the quantity and quality of available data as well as the level of disruption by anthropogenic pressures. Peters et al. (2012) stated that high percentile analysis, like usage of 50^th and 90^th percentile, leads to misleading data interpretation since high percentile analysis is more suitable for the data obtained from the environment that is not subjected to any anthropogenic sources, including minor diffuse anthropogenic inputs. In other words, the study environment should be entirely pristine for the usage of high percentile analysis. In this respect, in this study, the low percentile analysis (5^th percentile) of monitoring data was performed in order to stick to a conservative approach for the determination of river basin specific NBCs of metals and metalloids. The implementation of low percentile analysis was carried out with the integration of different statistical tests and approaches. Within the context of this study, descriptive data analysis and outlier detection tests were also employed.

Furthermore, the data treatment strategy was developed for the observations below LOD values with the establishment of three different approaches. Besides these practices, the river basin specific EQS derivation methodology, which is described in detail in Section 2.3.4, was also addressed within the scope of this study.