Determination of river pollution sources using source apportionment method: Ergene river

Abstract

Ergene Watershed in Thrace Region has been in the news about the river pollution problems. The provinces within the watershed, especially Tekirdağ, have intense industrial activity due to the geographical location. Uncontrolled industrialization and unplanned urbanization have increased rapidly since 1990s in the region, and caused numerous environmental problems. Ergene River and its tributaries supply water to the watershed. A significant part of the industrial facilities is concentrated in the Çorlu-Çerkezköy region close to Çorlu Stream in Tekirdağ Province, and the pollution in the river starts in this region, which is at the beginning of Ergene River. The population has been also increasing by the influence of the organized industrial districts and industrial activities. There are approximately 1 768 000 people living and 2 000 industries in the region. Additionally, agricultural activities are carried out intensively in the watershed as well as the industrial activities. Water quality is analyzed to estimate pollution level and determine the potential uses of the water source. Sampling and measurements are the important steps in watershed management along with the inventory of pollution sources to identify the problem. Environmental Quality Standards (EQSs) for specific and priority pollutants recently included in "Surface Water Quality Regulation". Measurements of these pollutants give more information than the routine monitoring for conventional pollutants to distinguish different pollution sources. Firstly, Ergene Watershed, boundaries of watersheds, and river were determined using ArcGIS software using a current high-resolution Digital Elevation Model (DEM) dataset. The current river was investigated using Google Earth to take into account the human interferences such as changes in riverbed and water irrigation channels. A total of 75 sampling locations were determined according to the number and spatial distribution of discharge points in the watershed. Measurements were evaluated and statistical analyses were performed to understand the characteristics of Ergene River pollution and contributing sources. In this study, three season (summer, fall, and winter) measurements of several micropollutants in addition to conventional pollutants and metals were used from 75 sampling locations in Ergene River. Pollution sources and contributions need to be identified to understand the state of the pollution and control pollution sources. Thus, source identification and apportionment using multivariate statistical analysis techniques namingly Cluster Analysis (CA), Principal Component Analysis (PCA), and Positive Matrix Factorization (PMF) were performed for this study. Out of 223 micropollutants, total of 132 in summer, 136 in fall, and 97 in winter were detected at least one sampling location. The multivariate statistical analyses were performed for the whole river, main river, and Çorlu Stream only. CA using Ward linkage method was performed with measured pollutants and sampling locations for three seasons with normalized concentration values. CA for sampling locations indicated two main clusters; first one with relatively clean locations and another one with polluted locations for all seasons. In addition, heatmaps were created. PCA was performed to the three seasons with pollutants with at least one measurement above Limit of Detection (LOD). The first 5-6 principal components usually explain more variation and 50% of the total variance was explained with the first seven components. PCA was also performed only with pollutants with more than 12.5% and 25% detection efficiency. For these cases, and main river, Çorlu Stream results, explained total variances were increased for all seasons. Çorlu Stream for three seasons with >25% detection of pollutants was identified approximately 100% with ten components. Fifty percent of the total variance was explained with the first three components Çorlu Stream with all measurements as well. PMF runs were performed for three to seven factors for three seasons (summer, fall, winter) and three cases (whole river, main river, and Çorlu Stream). Optimum number of factors were found as six factors for Ergene River, six factors for main river, and five factors for Çorlu Stream in summer; five factors for Ergene River, six factors for main river, and five factors for Çorlu Stream in fall; five factors for Ergene River, five factors for main river, and four factors for Çorlu Stream in winter. These findings were also consistent with the PCA results in which the same factors explained approximately 50%, 70%, and 70% of the variability in the dataset for Ergene River, main river, and Çorlu Stream, respectively. PMF runs were performed for three to seven factors for three seasons and three cases with pollutants greater than 25% detection frequency, and almost all of these pollutants were categorized as "Strong" according to their S/N ratio. According to the quality of fit parameter (Qtrue/Qexpected), optimum numbers of factors were determined as six factors for Ergene River, five factors for main river, and six factors for Çorlu Stream in summer; six factors for Ergene River, five factors for main river, and six factors for Çorlu Stream in fall; five factors for Ergene River, six factors for main river, and six factors for Çorlu Stream in winter. PMF results for the main river and Çorlu Stream with pollutants greater than 25% detection frequency were found to give the most stable solutions and used for identification of the sources. After determination the number of factors, uncertainty runs using Bootstrap (BS) and Displacement (DISP) methods were performed to understand the contributions of pollutants to source profiles. Factor fingerprints, factor contributions, and uncertainty methods error estimations of factor contributions for main river in three seasons were obtained. Major contributed pollutants for the factors and their contribution (%) in main river were combined, and contributions were indicated the important pollutants in each factor. According to pollutants with high factor profile percentages and low uncertainty, ranking was made. To understand the location of pollution sources, the factor contributions to the sampling locations of PMF were plotted against the distance from the beginning of the river (km) and the direction of flow. According to PMF results, Factor 1 was characterized by industrial pollutants, Factor 2 was by industrial, domestic, and agricultural pollutants, Factor 3 by industrial pollutants used in corrosion inhibitor, automotive and metal production, Factor 4 by pesticides, and Factor 5 strongly by metal pollutants in summer. In fall, Factor 1 was characterized by mixed industrial and domestic pollutants, Factor 2 by pesticides, insecticides, and herbicides, Factor 3 by industrial, domestic, and agricultural pollutants, Factor 4 mostly by domestic pollutants and Factor 5 by NO3 and pesticides. In winter, Factor 1 was characterized by specific industrial pollutants which can be identified as solvent and paint industry, Factor 2 by metal pollutants, Factor 3 by conventional pollutants and corrosion inhibitors, drug precursors, Factor by industrial pollutants and Factor 5 by metal pollutants, Factor 6 was strongly influenced by conventional and domestic pollutants in winter. According to PMF factor contributions for sampling locations, Factor 1 in summer, Factor 1 in fall, and Factors 1 and 4 in winter; Factor 2 in summer, Factor 3 in fall, and Factor 3 in winter; Factor 3 in summer, Factor 3 in fall, and Factor 4 in winter; Factor 4 in summer, Factor 5 in fall, and Factor 3 in winter; and Factor 5 in summer, Factor 4 in fall, and Factor 5 in winter were found similar contributed in main river. Factor 2 in winter was not related other factors in other seasons. According to PMF factor contributions for pollutants in main river, Factors 1 in summer, 1 in fall, and 4 in winter, Factors 2 in summer, and 3 fall, Factors 4 in summer, 5 in fall, and 4 in winter, Factors 5 in fall, 5 in winter were found similar.