Quantification of pollution sources contribution on surface water resources of the Abaya-Chamo Lake Basin, Ethiopia: an integrated multivariate statistical analysis and a receptor model approach Open Access

Surface water pollution is nowadays a critical environmental issue, resulting from various anthropogenic and natural sources. Excess nutrients from fertilizer applications are causing eutrophication, ultimately triggering algal blooms and depleting oxygen levels, endangering aquatic life (Fan et al. 2010; Madjar et al. 2024). Besides, heavy metals from mining (Obasi & Akudinobi 2020) and industrial activities (Oladimeji et al. 2024) enter into water bodies posing severe health risks. Furthermore, bacterial contamination from untreated human waste, particularly in developing regions, is increasing water-borne disease prevalence (Haldar et al. 2022).

Despite the increased complexity of environmental contaminants, integrated multivariate statistical analysis (MSA) and receptor models provided a clear framework for identifying and pinpointing pollution sources based on the available water quality datasets (Ouyang et al. 2006). The assessment of surface water quality is thus becoming more important for decision making and MSA techniques such as the application of principal component analysis (PCA) and factor analysis (FA) have gained advantages in this regard where they can manage large sets of water quality data (Shrestha & Kazama 2007; Özdemir 2016; Hajigholizadeh & Melesse 2017). These statistical methods, when integrated with environmental multivariate receptor models (MRM) such as positive matrix factorization (PMF), absolute principal component score-multiple linear regression (APCS-MLR) and chemical mass balance (CMB), enhance pollution assessment and management capability, thereby quantifying their contribution to the water resource pollution (Gholizadeh et al. 2016; Zhang et al. 2020). The MRMs have been extensively utilized in surface water studies due to their ability to decompose complex pollutant mixtures without prior knowledge of source profiles (Chen et al. 2022). While the CMB model depends on predefined source profiles, PMF assigns pollution sources based on their characteristic chemical signatures.

Many studies reveal that the impact of pollution mainly from nonpoint sources such as agricultural and urban land uses in a watershed is heavily influenced by the type of land use and the planning and management approaches (Wang et al. 2005; Lee et al. 2012; Kim et al. 2018). As a result, the type, nature and planning of land use are critical factors in river pollution management (Ding et al. 2015; Chiang et al. 2021). According to Angello et al. (2020), land use planning and management in Ethiopia is poor, and stream corridor restoration evaluation is limited and often omitted from all river restoration programs, resulting in significant river pollution from various land uses. Consequently, many surface water resources in the country are becoming heavily nutrient-polluted, intensifying eutrophication in rivers (Menberu et al. 2021; Kassaye et al. 2022). River pollution in Ethiopia hence is becoming a critical issue, primarily due to land use changes causing erosion and runoff (Moges et al. 2018).

In Ethiopia, land use change is very fast which is partly due to factors such as high population pressure, urbanization and resettlement (Negesse et al. 2024). On the other hand, agricultural practices and unsustainable practices and depletion of natural resources are also intensifying the land use change predominantly in the rural areas (Megersa & Tullu 2018). As a result, the land use modification has greatly caused various hydrologic and water quality disruptions. The Abaya-Chamo Basin (ACB) is best characterized by its diversified animal and plant species composition. However, recent studies suggest the basin is under threat mainly due to population pressure. Yohannes et al. (2018) reported human-induced factors, such as increasing demand for fuelwood, intensification of construction using wood, agricultural land expansion and irrigation activities and increased bush burning activity for pasture. This land use alteration is highly impacting the water resources of the basin. Therefore, the combined analysis of both water quality and quantity due to land use modification has nowadays become more and more important for the environment and the people depending on it (Kebede et al. 2014). This study is, therefore, aimed at investigating, assessing and characterizing the spatial river pollution, determining their potential pollution sources and quantifying the contribution and composition of each pollution source for individual water quality constituents using integrated MSA and MRM.

The rainfall pattern in the river basin has two annual peaks, usually occurring in April/May and August/September. The ACB climate is further characterized by a high rate of evaporation. The area is also characterized by high topographic variability ranging from 1,080 to 3,500 m above sea level, the lowest at the lake shore and the highest at the Guge Mountain.

Geologically, the volcanic rocks are predominantly found in the area evolving from the tertiary and quaternary periods. The distribution of the geological formations in the basin of the lowland part of the upper land has a recent quaternary period Aden series with basaltic flows and related spatter cones, whereas the parts have varying formations of tertiary period trap series alkali basalt and trachyte and quaternary period alluvium deposits (Tiruneh 2005). The basin has many water resources, some of which are the Hare River, Kulfo River, Sile River, Bilate River, Gidabo River, Baso River, Sego River, Abaya and Chamo Lake.

Water samples were collected for 17 physico-chemical cations and anions (Table 1) in the selected surface water sources (lakes and rivers) of the ACB at 24 monitoring stations.

All the analytical methods for the constituents were conducted by (APHA 1999) standard procedures. The monitoring stations were selected in consideration of the spatial/geographic location, available water sources, prevalence of anthropogenic influence and economic and social importance of the water resources.

According to the land use classification of Ayele et al. (2019), the ACB is dominated by grass & shrub land (33.49%) followed by farm land (28.63%). Other types of land uses are barren land (17.12%), vegetation (9.35%), Lake Abaya (5.91%), wood land (2.79%) and Lake Chamo (1.65%).

Development of better and efficient water quality management systems in a water resource is the main challenge these days. Despite some limitations, MSA techniques have become successful at filling the gap (Noori et al. 2010). However, availability of data with reasonable quality plays a key role in the effective interpretation of the water quality by MSA. Several MSA techniques are used nowadays to interpret water quality variables such as PCA, FA and CA. Statistical techniques are used to characterize and interpret river water quality variables by minimizing the bulk datasets without losing much information contained in the original data (Silva et al. 2017).

In the ACB, FA was used to determine the possible pollution sources contributing to the pollution of the surface water resources qualitatively. Prior to statistical analysis and result interpretation, the suitability of the collected samples for FA was checked using the parameters suitability test: Kaiser–Meyer–Olkin (KMO), a measure of sampling adequacy and Bartlett's test of sphericity which examines whether the available data are independent or not. A KMO value close to one would generally mean the correlations are compacted and hence the sampling and the samples are highly suitable for FA, whereas smaller values would generally mean that the variables in consideration have very little in common. Although KMO greater than 0.5 is often considered adequate (Ogwueleka 2015), higher values are usually recommended. On the other hand, CA was used to group all the monitoring stations according to their spatial similarity where Hierarchical Cluster Analysis (HCA) was used to classify the monitoring stations based on the similarity between constituents through Ward's method. All the statistical analyses are conducted using SPSS (version 26).

During FA, the most common extraction method is the principal component-based correlation matrix determination where eigenvalue greater than unity is considered. For maximizing the variance and extraction of the determining factors, Kaiser normalization is used (Angello et al. 2021). Each principal component loading determines the weight (impact) of the respective constituents and generally loadings greater than 0.75 are often taken as strong (Kilonzo et al. 2014), whereas the components loadings 0.5–0.75 and 0.3–0.5 are classified as moderate and weak, respectively (Cid et al. 2011). Similarly, the CA analysis was used to group the water quality monitoring station based on their chemical similarity, represented by a dendrogram. HCA is a commonly used technique to cluster the water samples according to their chemical composition and similarity.

Multivariate statistical techniques are nowadays gaining popularity in determining the types of pollution sources contributing to water quality deterioration. However, quantification of the contribution and composition of each source are difficult. In this regard, with the multivariate receptor models (MRM) such as EPA's UNMIX (Norris et al. 2007), PMF (Huang et al. 2018) and Absolute Principal Component Score/Multiple Linear Regression, APCS/MLR (Gulgundi & Shetty 2016) the contribution of the pollution sources to individual water quality constituents can be quantified and decision-making has been simplified. The UNMIX model is one of the simplified tools developed to analyze large sets of water, air and soil/sediment samples and has become effective in recent times (Sun 2017). The UNMIX assumes the given water quality data are linearly related to the unknown number of sources with an unknown composition and contribution (Norris et al. 2007).

Detailed spatial water quality variation in the Abaya-Chamo Lake Basin is given in Table 2. Accordingly, the physico-chemical characterization in the rivers and lakes revealed that the pH has shown a nearly constant trend and no significant variation was observed across the monitoring stations. A mean of 7.9 ± 0.4 was recorded on the ACB monitoring station where few irregularities were observed between monitoring stations. The observed pH generally ranged from 7.2 to 8.7, indicating the reduced impact of acidity in the water resources. Similarly, the DO variation in the surface water resources of the ACB has shown nearly a uniform trend in concentration and all the monitoring stations have met the minimum requirement oxygen concentration on surface water resources as set by the Ethiopian water quality standard >4.5 mg/L (EEPA 2003). Furthermore, a strong negative correlation was observed between DO and TDS (r = −0.75), EC (−0.76) and bicarbonate (−0.86) at a significance level of p < 0.01. The high negative correlation between DO and TDS/EC could be attributed to the high presence of dissolved salt concentration that reduced the dissolvability of oxygen.

The TDS in the ACB has shown a significant spatial variation across the monitoring stations where a maximum concentration of 1,205.7 mg/L was recorded at LCH, whereas a minimum (28.9 mg/L) was observed at HNB, the earlier monitoring station being a dumping site for leftover fish products and public recreation areas (Figure 3). Besides, high TDS standard deviation was calculated on almost all the monitoring stations mainly due to the high spatial variation and high variation in anthropogenic activities (Angello et al. 2021). On the other hand, the EC, which measures the proxy salt concentration in the surface water resources, was also found significantly high at monitoring stations with high TDS concentration (LCH). The excess presence of EC in the water resources is a good indication of significant quantity of dissolved salt (Barakat et al. 2016). The high EC level in LCH could be attributed to the increased anthropogenic and agricultural practice impact on the water resource. Accordingly, high and low EC concentrations were recorded at monitoring stations LCH and HNB corresponding to the Lake Chamo and the Bridge on Hare River, which is the tributary for Lake Chamo and characterized by intensive agriculture (Figure 2). They earlier receive a large number of sediments mainly from agricultural lands. The conductivity in the ACB has shown a strong positive correlation with bicarbonate (r = 0.9), chloride (0.86) and sodium (0.93). Similarly, the water temperature is the major factor that causes significant environmental impacts and controls various water quality variables (Silva et al. 2017). However, in the ACB surface water samples, the water temperature ranging from 22.1 to 31.4 °C with a mean temperature of 26.7 °C is under the national (EEPA 2003) and WHO (WHO 2011) guideline standards and showed a weak to moderate correlation with all parameters.

Excess presence of nutrients can seriously impact the well-being of aquatic life (Barakat et al. 2016). In ACB, however, the NO₃-N concentration was found from 3.3 to 31 mg/L with a mean concentration of 13 mg/L (Table 3). All the monitoring stations have recorded NO₃-N within the permissible limit, whereas at some monitoring stations such as BSF and BRB, the NO₃-N concentration has recorded slightly higher concentrations that could be attributed to the leaching of agricultural land saturated with fertilizers. More than 70% of the monitoring stations in ACB have recorded a NO₃-N concentration of <10 mg/L. On the other hand, the PO₄-P concentration at some of the monitoring stations has deviated from the national guideline standard for surface water requirements. More than 20% of the monitoring station water samples had a PO₄-P concentration of >1 mg/L, making the stations at the critical stage of eutrophication. The correlation analysis using Pearson correlation showed that PO₄-P had a strong negative correlation with pH (r = −0.77) and a strong positive correlation with total iron (r = 0.83) and NO₂-N (r = 0.75).

Like physico-chemical water quality constituents in ACB, the anion and cation concentrations have shown a slight variation across all monitoring stations. The Ca²⁺ concentration was found to be within the national guideline standard with a mean concentration of 16 mg/L (Table 3). Similarly, an observed mean concentration of 23.7 mg/L Mg²⁺ was observed in the ACB making it within the limit standard. However, at some monitoring stations, such as WOR and CER, the concentration was slightly higher than the standard limit, making the water resources unsuitable for public uses. Very strong and positive correlation was found between Mg²⁺ and total hardness (TH) (r = 0.99), indicating Mg²⁺ was the major cause of the water hardness.

The TH in the ACB was significantly high mainly attributed to the presence of these Ca²⁺ and Mg²⁺ that varied from 11.3–27.3 mg/L and 1.9–57 mg/L, respectively. On the other hand, the high bicarbonate concentration in most of the water resources of ACB was possibly attributed to large cation-exchange capabilities of clay mineral and the high ground-surface water interaction that allows easy contact with sandstone and carbonate rocks. This can easily be interpreted by the high correlation between TH and Na⁺ (0.78) and Cl⁻ (0.78). Similarly, the analyzed water quality sample in the surface water resource of the ACB also indicated that the Cl⁻ concentration was high at some of the monitoring stations, although they all are within the guideline standard limit having a maximum concentration of 160.3 mg/L at LCH (Lake Chamo) monitoring station. The presence of Cl⁻ in excess of 200–300 mg/L in drinking water may cause a salty taste to the water and beverage consumers (WHO 2011). On the other hand, a very strong positive correlation between Cl⁻ and Na⁺ (r = 0.96) and K⁺ (r = 0.81) was observed, which could have possibly accounted for the high concentration of TDS and EC in the ACB, especially at LCH monitoring stations (Gebresilasie et al. 2021).

The pollution source apportionment in ACB was based on the FA where the suitability of the collected water samples was checked by two factors: KMO and Bartlett's test of sphericity. Accordingly, the KMO in ACB was 0.56 > 0.5, indicating the FA was more suitable for the analysis and interpretation of water quality variables in the ACB.

The FA in ACB extracted four principal components explaining a total variance of 86.9% with an eigen value >1. Accordingly, factor 1 explained 34.4% of the variance and had very strong loading for chloride (0.98), sodium (0.98), TDS (0.94), EC (0.93), bicarbonate (0.87) and strong loading for potassium (0.77). The presence of high component loading on sodium, bicarbonate and chloride is a good indication of the excess presence of dissolved salt in the water resources. It can also be interpreted by the presence of the impact of mineral constituents in the water resources (Yawar et al. 2017). Moreover, the high loading on EC, TDS and potassium could be attributed to the increased land use alteration that triggered sediment transport and the high ground water intrusion in the watershed. The second principal component that explained 26.42% of the variance has a strong component loading for phosphate, nitrite, nitrate and sulphate with a loading of 0.84, 0.84, 0.81 and 0.78, respectively. This could be interpreted by the high contribution of mixed pollution sources, mainly originating from agricultural lands and households (Rahman et al. 2021).

The third principal component had explained 15.6% of the variance with a very strong component loading on magnesium and TH having a calculated component loading of 0.89 and 0.91, respectively. The high loading on the constituents could be interpreted by the significant impact of the water resources by metal hardness. These impurities might be contributed by ground water and limestone originating from the ground surface. The fourth principal component extracted by FA in ACB had a strong positive loading for water temperature (0.88) and negative loading for DO (-0.74) that explained 10.3% of total variation. The component can be explained by the impact of seasonal variation on the water quality of the ACB water resources. From all the loadings, it can be deduced that the role of nonpoint sources originating mainly from agricultural land was evident and the impact is more pronounced on the two ACB lakes: Lake Chamo and Lake Abaya.

Cluster 1 is represented by the Chamo Lake (LCH) monitoring station. The station is characterized by a relatively high pollution level compared to other monitoring stations in the ACB. The high number of pollution sources and increased anthropogenic impact in the lake could be a potential factor behind the high pollution level in the lake (Wang et al. 2012). Cluster 2 is composed of eight monitoring stations including BAB, BIR, MMA, BNR, AMB, BSF, AFB and BRB where urban land use impacts are prevailing (Figure 2). An internal analysis in this cluster revealed that the monitoring stations BAB and BIR have shown similar features. Monitoring stations in this cluster are characterized to have nearly similar characteristics and are affected by similar pollution sources. Similarly, cluster 3 is composed of eight monitoring stations, whereas cluster 4 consists of seven monitoring stations (Figure 4).

Before model interpretation, the UNMIX model in this study was checked for the minimum requirement of Noise to Signal ratio (N/S) and a coefficient of determination (R²) where both parameters were above the minimum requirement with 3.38 > 2 and 0.88, respectively. Accordingly, more than 88% of the total variance of each water quality constituent or their contribution can be explained using the predefined pollution profile (agricultural, ground water intrusion and mineral dissolution), leaving the remainder to be contributed by unknown (undefined) sources.

The UNMIX model predicted and assigned the pollution source contribution with a mean predicted to a measured ratio of 0.96, showing that the model has slightly underestimated the assignment of the contribution of the sources to the individual constituents. The mean absolute error was 4.18%, revealing the assignment of the pollution source contribution was quite good. The model performance for each parameter had good to very good range with an R² ranging from 0.58 to 0.98 (TDS, Cl⁻ and Na⁺). The model was relatively weak on TDS to assign the contribution of identified sources with an error of about 10.18%. Table 4 shows that source 1 (mineral dissolution) has the highest contribution for water temperature, DO, Ca+ and pH followed by ground water intrusion and agriculture, whereas it has a relatively low contribution for others. On the other hand, the contribution of source 2 for Cl- and Na⁺ was relatively higher, whereas the contribution of source 3 for TI, nitrate, nitrite and phosphate is high. Details of all the contributions of the pollution sources for selected water quality parameters are summarized in Table 4.

In this study, the water quality of the rift valley water resources of ACB obtained from 24 monitoring stations has been evaluated based on the application of MSA and MRM. The FA and hierarchical clustering in CA has been used to spatially analyze and represent the water quality variation where four meaningful clusters have been identified, which could be used as a representative location for future water quality intervention programmes. The MSA result interpretation revealed that intensive agricultural practices are the dominant pollution sources in the watershed, followed by groundwater contribution and mineral dissolution. On the other hand, MSA based on the hierarchical analysis in ACB water quality monitoring stations revealed that cluster 4, which constitutes the Chamo Lake, is on the verge of water quality deterioration due to the uncontrolled controlled anthropogenic activity including illegal fish market in the lake's peripheries. The MRM analysis showed that the contribution of agriculture to each water quality constituent, specifically nutrient pollutants, was quite high. Hence, integrated MSA and MRM could be a feasible option for water quality management and pollution control in the study area.

The authors are very thankful for the Arba Minch University, Institute of Water Technology, Water Resource Research Center for providing some water quality data.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.