The Seman River is one of the seven most important rivers in Albania, which flows from the mountainous heights in the east and descend to the west where the coastal lowlands have created good quality land for agricultural production. It consists of two important tributaries: Devoll and Osum with a length of 281 km, basin area of 5649 km² and an average flow of 96 m³/sec (Figure 1). The physio-geography classifies the study area as a mountainous region (>700 m above sea level) in the upstream east, with ultrabasic rocks at the source and a small population density (<50 inhabitants/km²), and a hilly-plain region downstream mainly composed of geologically limestone, with a high population density (>200 inhabitants/km²) and with intensive agricultural and industrial development. The Seman basin provides potable water for about 300,000 inhabitants, and irrigation water for about 55,000 hectares of land with high intensity of agricultural production. High seasonal variation of the flow in the summer period leads to lack of water for irrigation by increasing pollution pressure on the of drinking water quality. The Seman River water quality is defined as average to poor, especially due to the TSS (Total Suspended Solids) and the organic materials contained [27] [28]. The quality of water for irrigation and drinking is often questionable due to the discharge of untreated sewage from the four cities that Semani River runs through, namely Gramsh, Corovoda, Berat and Fier.

Due to the complex geomorphology, six monitoring stations were defined along the middle course of the Seman River: two at the discharge of its tributaries

(Devoll and Osum), two in the upper stream prior to reaching the city of Fier and two further downstream, passed the city of Fier. Water sampling were performed in four seasons (spring, summer, autumn and winter) for a period of five years beginning from 2014 to 2018. The sampling stations were selected based on the expectation that water quality was modified whenever there were significant impacts through point pollution—especially from industry and untreated urban wastewater—and diffuse pollution, mainly from agriculture and erosion.

Six water samples (pursuant the preselected sampling stations) were taken in four seasons, namely winter (January), spring (April), summer (July) and autumn (November) in 2014, 2015, 2016, 2017 and 2018, to determine the following parameters (pH, DO, EC, TSS, Cl⁻, NO 3 − , Total–N, Total–P, BOD₅, Cu²⁺, Ni²⁺, Pb²⁺, Cd²⁺ and Temperature ˚C). Samples were taken with plastic bottles of 1 L volume, at a depth of about 20 cm below the water surface. The bottles were stocked in freezer boxes at about 4˚C during the transport, and were taken to the laboratory where they were placed in the refrigerator at −20˚C for further analysis. Parameters such as temperature (Temp. ˚C), pH, DO (dissolved oxygen), EC (electrical conductivity), salinity and turbidity were measured in situ using a HANA multiprobe instrument. Analysis of the mineral forms of nitrogen and phosphorus ( NO 3 − , NH 4 + and PO 4 3 − ) were conducted based on the ultraviolet spectrophotometric screening method [29]. Determination of orthophosphate was performed according to the ammonium molybdate spectrometric method specified in standard ISO 6878:2004. Determination of mineral nitrogen forms dissolved in water, was conducted based on the manual spectrometric method ISO 7150: 1984 (for ammoniacal forms) and on the 2,6-dimethylphenol spectrometric method ISO 7890-1:1986 (for nitrates and nitrites). For the determination of DO (Dissolved Oxygen) and BOD (Biological Oxygen Demand) were used the standard methods of the American Public Health Association [30] and for the determination of heavy metals (Cu²⁺, Ni²⁺, Pb²⁺, Cd²⁺) was used a Flame Atomic Absorption Spectrometer type AA350.

The Water Quality Index (WQI) represents a numerical expression which has the purpose of establishing the ecological state of the water body. The formulation of the CCME WQI is described in the Canadian Water Quality Index 1.0 - Technical Report Principally, the model consists of three measures of variance from selected water quality objectives (Scope, Frequency, Amplitude). The “Scope (F1)” represents the extent of water quality guideline non-compliance over the time period of interest. The “Frequency (F2)” represents the percentage of individual tests that do not meet objectives. The “Amplitude (F3)” represents the amount by which failed tests do not meet their objectives. These three factors combine to produce a value between 0 and 100 that represents the overall water quality [31].

For the calculation of this index, we have considered the following parameters: temperature, pH, EC, DO, TSS, BOD, nitrogenous, nitrites, ammonium ( NH 4 + ), chlorine and heavy metals (Pb²⁺, Cd²⁺, Ni²⁺, Zn²⁺). The WQI calculator used by us categorized water as: excellent (95 - 100), very good (89 - 94), good (80 - 88), fair (65 - 79), marginal (45 - 64), and poor (0 - 44) [19]. Cluster analysis (CA) was used to find the internal structure of a data grouped in advance based on their similarities [32]. Hierarchical Cluster is the most common approach, where clusters are formed sequentially starting by the most similar pair of objects and forming higher clusters in a step-by-step pattern, according to JUMP SAS 11 software. Principal Component Analysis (PCA) is a powerful technique used for recognition of patterns which can explain the variance of large data sets of inter-correlated variables and transform them into smaller sets of independent variables (principal components) [14]. Software used is: JMP, 11, 2014, John P. Sall, it is a trial version.

For all seasons in five years, the value of WQI fluctuates between 55% and 86% with variations between years and seasons (Figure 3). Except for the autumn 2014, all downstream values were higher compared to the upstream ones due to the impact of untreated sewage and industries present in Fier city, the largest one amongst the cities where Seman River flows. However, the largest differences between the average values of WQI in the upper and downstream were observed in the spring and winter seasons, which are rainy seasons affecting the transport of urban and industrial wastes to the river, the erosion of agricultural land, and the transport of agricultural fertilizers, especially Total-N and Total-P.

The average values of downstream stations based on WQI can lead to a possible explanation of the water degradation, visible mainly in the deteriorated values of TSS, BOD₅, DO and EC. All these parameters are determined mainly by human activity due to the release into the river of solids of mainly organic nature, which is evidenced mainly by the high biological demand as well as by the low values of dissolved oxygen. Cluster analysis weight groups of parameters by revealing the internal structure of a parameter compared to other parameters by grouping these according to the “weight” they have in water quality, always compared to the threshold values. Below are the dendograms of the four seasons, where the clusters (Figure 4) present the groups of the monitored mean values and relationships between them for the following parameters pH, DO, EC, TSS, Cl⁻, NO 3 − , Total-N, Total-P, BOD₅, Cu²⁺, Ni²⁺, Pb²⁺, Cd²⁺ and Temp. ˚C. At Cluster 1 during the Autumn seasons (2014-2018) (Figure 4(a)), the parameters are classified in eight groups, where the most important ones are ranked as follows: pH linked with DO, EC, NO 3 − , Cl⁻; the second group NO 3 − linked with Total-N and Total-P; the third group Total-P linked with Cu²⁺ and Cd²⁺; and the fifth group BOD_5, TSS and EC linked with Temp. ˚C, heavy metals and macronutrients. At Cluster

2 during the Spring seasons (2014-2018) (Figure 4(b)), the parameters are classified in seven groups, starting with pH which is linked to TSS, DO, EC, NO 3 − and Cl⁻, then with the second group NO 3 − which is influenced by Total-N, Total-P and Cu²⁺. A link between Cd²⁺ and Ni²⁺ is noted in the group of five average values. The creation of two large groups in the cluster shows that EC, Temp. ˚C, macro and microelements together are influenced by TSS as a physical indicator of the presence of suspensions in the Seman water system. Cluster 3 during the Summer seasons (2014-2018) (Figure 4(c)) the parameters are classified in seven groups starting with pH which is linked with DO, EC, NO 3 − , Cl⁻, then with the second group NO 3 − which is linked with Total-N and Total-P, continuing with the third group where is a link between Total-P with Ni²⁺, Cu²⁺ and Cd²⁺. The groups and links are similar to the autumn season and more distanced to the spring season, with a high intensity of rain showers and with a stream flux similar to autumn. There is a similarity to autumn season with respect to the division of two major groups, where TSS and EC influence the groups of macro and microelements, as well as the pH in the Seman River water system. At Cluster 4 during the Winter seasons (2014-2018) (Figure 4(d)) the parameters are classified in four groups, starting with the first one that includes the pH which is linked with DO, NO 3 − , Pb²⁺, Ni²⁺, Cu²⁺ and Cd²⁺, Total-N, Total-P and Temp. ˚C; in the second group the first group elements are linked with Cl⁻ and BOD₅; and in the third group there is a link between second group parameter results correlated with_. EC, TSS. The two major groups in the cluster indicates that there is a similarity to all cluster results—EC, macro and microelements, are altogether influenced by TSS, as a physical indicator of the presence of suspended solids in the Seman River waterbody.

As a result of the hierarchical CA applied in the analysis of seasonally ranked data as an average of 5 years (2014-2018), the dendrogram (Figure 4), refers that in the autumn and summer seasons, the relationship between Electrical Conductivity (EC) and Total Soluble Solids (TSS) is close and matched into one category. It is different for winter-spring seasons where TSS forms a category and influences all other measured physic-chemical parameters, due to the high presence of sediments through precipitation in this period and as a result of redox reactions that occur, influenced also by water temperature. Based on the results, it can be concluded that during wet seasons such as winter and spring, there are more sediments which influence other physico-chemical parameters, while during dry seasons (summer and autumn) there are more decomposition reactions of elements released by sediments and influenced by temperature.

The Principal Component Analysis (PCA) is a technique for explaining changes in data sets and the interrelationship of variables and their significance across the classified groups by “pressuring” the least influential variables. In other words, PCA reduces the impact of some variables and identifies other variables that are unobservable, hypothetical, or latent. The analyzed diagrams highlight a clear pattern regarding how parameters or variables affect each other. PCA was applied by processing the data of each season in each year studied. The PCA method is based on the study of correlation groups using Pearson distribution with the Varimax rotation method which consists of linking and directing factors to the original variables for easier interpretation. Figure 5(a) shows the results obtained from the PCA analysis of the measurements performed during the spring seasons from 2014-2018 which are grouped in Factor 1 which includes Cl⁻, BOD₅, Cu²⁺, Ni²⁺, Pb²⁺, Cd²⁺, which do not correlate with CCME WQI. This factor does not show a significant correlation and is seen to be opposite to the vertical axis (Factor 1). The second group consists of TSS, Cu²⁺, DO which do not correlate either with CCME WQI or Ni²⁺, BOD₅, Total-P, Total-N, NO 3 − , (Factor 2) as they are located in different axes and are not close to each other to be called as influential of each other. In the third group the strong connection of CCME WQI with pH and temperature is noticed (Factor 3). In a general view we can say that CCME WQI also links to metals like Ni^{2 +}, Pb^{2 +} and Cl⁻, but it has an extremely small bond with BOD₅, Cd²⁺, DO, Total-P, Total-N, and NO 3 − . All three factors explain over 56% of the total variation. The results of PCA analysis in Figure 5(b) in the summer season shows three groups with respective links between them. The first group includes DO, Temp. ˚C, Cl⁻, NO 3 − , Total-N, Total-P and BOD₅ which do not correlate with CCME WQI. The second group includes TSS, pH which correlate strongly with CCME WQI and the third group includes mainly metals Cu²⁺, Ni²⁺, Pb²⁺, Cd²⁺ which do not correlate with CCME WQI. From the comparison of F1 vs. F3 the formation of three groups is observed where the first group include DO, Temp. ˚C, NO 3 − , Total-N, Cu ²⁺ and Total-P, which do not correlate with CCME WQI, the second group include pH, TSS, Cl⁻ which correlate closely with CCME WQI and the third group is composed by heavy metals such as Ni²⁺, Pb²⁺, Cd²⁺ and BOD₅. From the comparison between F2 and F1 are observed two main groups such as NO 3 − , Total–N, DO, Temp. ˚C which correlate with Total-P, BOD₅ and Cu²⁺; and Ni²⁺, Pb²⁺, Cd²⁺, pH, TSS which correlate with CCME WQI except Cu²⁺ which does not correlate with any of the groups. All three factors explain over 56% of the total variation. Unlike the data recorded during the spring and summer seasons, the data collected during the five autumns of monitoring activities illustrate slightly different correlations (Figure 5(c)). PCA analysis in F1 vs. F2 pointed out that there is a correlation between CCME WQI and chemical parameters such as: Total Total-N, NO 3 − , pH, BOD₅, but a weak correlation with the other two groups formed on both sides of the Y axis. In the second group there is a correlation between Temp. ˚C, Cl⁻, Cd²⁺ and in the third group is a correlation between Ni²⁺, DO, Total-P, TSS, Cu²⁺, Ni²⁺, Pb²⁺. The comparison between F1 and F3 highlighted two main groups where in the first group there is a correlation between DO, TSS, NO 3 − , Total-N, Total-P, Cu², Pb²⁺, Ni²⁺, in the second group there is a correlation between Cd²⁺, Temp. ˚C, Cl⁻ with CCME WQI. Parameters such as pH and BOD₅, do not correlate in this comparison with any of the groups standing on the Y-axis. The comparison between F2 and F3 indicate the formation of four groups of parameters where in the first group are BOD₅ with Cd²⁺, in the second group there is a correlation of CCME WQI with NO 3 − and Total-N, in the third group there is a correlation of Temp. ˚C with Cl⁻, Ni²⁺, pH and Cu²⁺, and in the fourth group there is a correlation between Total-P, DO, TSS, Pb²⁺. All three factors explain over 64% of the total variation. The winter period results in different “weight” relationships (Figure 5(d)) compared to the three seasons above. When comparing the factors F1 and F2 two main groups have established correlative links. In the first group there is a correlation between CCME WQI with pH, Cd²⁺, Temp. ˚C, Cl⁻, and in the second group indicates a link of Cu²⁺, Ni²⁺ with DO and weaker with TSS, Pb²⁺ and Total-P. Three parameters do not correlate with any of the groups and they are NO 3 − , Total-N and BOD₅, thus forming a separate group. The comparison between F1 and F3 factors highlight the formation of correlation links in the first group where are included NO 3 − , Total-N, DO, TSS, Cu²⁺, Ni²⁺, Pb²⁺. The formation of the second group indicates a link between CCME WQI and Cd²⁺, Temp. ˚C, Cl⁻. However, it can be noted that Total-P, BOD₅, and pH do not correlate with any of the groups. The comparison between F2 and F3 factors indicates weak correlation between parameters, and the closest link with CCME WQI is from NO 3 − and

Total-N, while the other parameters have weaker links with the group of parameters compared above. This is also concurred by the low value (about 40%) of the variability of factors with each other.