Application of multivariate statistical methods to enhance the water quality monitoring system of Kashmir Valley with special emphasis on side-stream pollution

Surface water quality is a very important part of human life and is a major global problem due to its delicate characteristics. Since prehistoric times, riverine systems have been crucial to the growth of human civilizations because they provide a necessary and convenient source of fresh water for household, agricultural, and industrial uses. Unplanned industrialization and urbanization created for socioeconomic advancements today have greatly threatened the very existence of life on earth by polluting the water (Avtar et al. 2019). Both artificial and natural sources of pollution have an impact on the river's water quality (Singh et al. 2009). Changes in precipitation, surface runoff, erosion, and weathering are examples of natural processes that affect water quality (Khatri & Tyagi 2015). Human impacts including sewage, municipal waste, effluents, and irrigational operations also have an impact (Hanjra et al. 2012).

Humans must have access to clean water as a basic necessity for survival and well-being. Fresh water is essential in many regions of the world due to industrial activity's role in water contamination, but it will become even more scarce as a result of population growth, urbanization, and climate change (Rosegrant et al. 2009; Gude 2017). Anthropogenic activity-related declines in water quality are quite concerning. Deteriorations in water quality triggered the need to examine and evaluate surface water quality, particularly when taking into account its beneficial uses, such as drinking, industrial and agricultural operations, etc.

Since the dawn of time, nature has provided the large population of the Kashmir Valley with copious water supplies on which it depends in countless ways. Numerous direct and indirect activities connected to these water bodies provide people with a living. The rivers and lakes in Kashmir are treated as dumping sites of wastewater despite the enormous benefits they provide (Qayoom et al. 2022). This has led to significant degradation of water quality in the recent past. Furthermore, the valley's water resources’ biological and physicochemical characteristics have been significantly impacted by extensive changes in land use, spontaneous urbanization, forest cover degradation/deforestation, high tourism pressures, landform degradation, and uncontrolled use of fertilizers and pesticides (Mir & Gani 2019). A number of recent studies have concentrated on investigating the water quality of River Jhelum and its associated tributaries (Qadir et al. 2008; Mehmood et al. 2017; Bhat et al. 2021). For instance, Showqi et al. (2014) observed that changes in the land use/land cover (LULC), hydrometeorological climate, and anthropogenic effect had caused the River Jhelum to deteriorate. According to Ganie et al. (2021), the deterioration of Wular Lakes is a result of changes in the LULC and the hydrological changes that follow, such as decreased runoff, increased erosion, and sedimentation. However, it is worth mentioning that the water quality of major tributaries like River Sindh and Lidder has a significant effect on the water quality scenarios of River Jhelum and thereby on the hydro-geochemistry of Wular Lake (Khanday et al. 2021).

Making appropriate pollution prevention efforts requires knowledge of the river's degraded sections and the real sources of pollution along various river segments. Sedimentation is one of the biggest dangers faced by river ecosystems worldwide. The total dissolved solids (TDS) and total suspended solids (TSS) contents in river water have considerably increased. An analysis of the 145 largest rivers in the world with reliable long-term sediment records revealed that roughly 50% of them statistically had a trend of drastically reduced flow because of sedimentation (Butler et al. 2020). In recent years, it has been clear that one of the main causes of poor water quality is sediment transport in the water bodies. Different algorithms and models have been used by the researchers to estimate the dissolved sediment load (DSL) and suspended sediment load (SSL) carried by rivers (Sun et al. 2021; Zhao et al. 2021). Many direct and indirect factors influence the variations in sediment movement, with evapotranspiration being one of the potential factors. Accurate daily evapotranspiration calculations can give scientific direction for the formulation of sediment management plans and the rational use of water resources (He et al. 2022). Furthermore, the estimate of the longitudinal dispersion coefficient is a critical measure for evaluating and regulating pollution spread in rivers (Goliatt et al. 2021). In order to get relevant results from the examination of water quality data, multivariate statistical approaches like factor analysis (FA) and analysis of variance (ANOVA) have been widely used (Najar & Khan 2012; Aydin et al. 2021). Additionally, it has been frequently utilized to describe and assess the quality of water in order to analyze spatio-temporal fluctuations brought on by anthropogenic and natural processes, as well as heterogeneity between various river sections (Machiwal et al. 2018). The use of multivariate statistical methods enables the extraction of hidden information regarding potential environmental influences on water quality from the dataset.

In FA, relationships between observations are attempted to be explained in terms of underlying components that are not immediately visible. The first stage of FA is to derive the correlation matrix of parameters. It is used to take into consideration how much variability in each specific pair of water quality characteristics is shared by the other. The correlation matrix's eigenvalues and factor loadings are then calculated. The sets of variables that have a high degree of correlation with one another are identified by the eigenvectors that eigenvalues correspond to. Lower eigenvalues might not make much of an impact on the data's capacity for explanation. Most of the parameter variability can be explained by the first few components alone (Tipping & Bishop 1999). Factor loadings are used to gauge the degree of correlation between the variables and the factors once the correlation matrix and eigenvalues have been obtained (Garrido et al. 2013). To determine how the values of two categorical factors affect the mean of a quantitative variable and if there is a statistically significant difference between the means of three or more independent groups that have been divided on two factors, a two-way ANOVA is utilized.

Accurate information on the quality of the water is crucial for effective and efficient water management, since it helps to identify the causes of degradation as well as the state of rivers and the landscapes that surround them. Using this data, we can create restoration strategies, calculate the ecological risks connected to proposed land use in a watershed, or decide on already-existing development choices to reduce river degradation. In this study, water samples from different sites of River Jhelum and its major tributaries from north-eastern Himalayas were used as a preliminary survey on water contamination, with the following goals: (1) To access the water quality of major tributaries of north-eastern Himalayas of Kashmir and their impact on River Jhelum; (2) To detect the regional and temporal changes in water quality and potential sources of pollution using descriptive statistics (DS) and principal component analysis (PCA); and (3) To assess the variation of water quality measures at different stations using ANOVA. The findings of this study could help decision makers to manage water quality, stop pollution sources, and safeguard the water resources of Kashmir Valley.

The River Jhelum is the primary river that flows along the entirety of Kashmir Division's 140 km. The majority of the villages and towns are situated along its banks. It originates from a beautiful spring known as ‘Verinag’. The Lidder River, which is the largest of all the effluents and the source of the River Jhelum's head waters, meets on the right slope of the mountain at a distance of 2 km. The second-largest tributary, Nallah Sindh, combines with it at Shadipora on the right bank, along with a few smaller input sources. The river enters the Wular Lake near Banyari which also receives the Arin and Madhumati streams and finally leaves the lake at its southwest corner before flowing westward across the alluvial plain for 21 km till it reaches the Baramulla bridge.

In this study, the four seasons of spring, summer, autumn, and winter have been characterized by the data generated for the surface water samples gathered from 11 different sites during the year 2021. The sample vials were airtight and had a 3-l capacity, which was deemed adequate for sampling. The collected samples were examined at Tehkeek International's environmental laboratory and the State Pollution Control Board's (SPCB) water laboratory in Srinagar in accordance with the APHA standards (2017). While the other characteristics were determined in the laboratory, the parameters for transparency, pH, and conductivity were determined immediately near the sampling site. The various physicochemical parameters’ estimates using analytical methods are given in Table 2.

Different types of graphs have been used to rapidly and effectively present visual summaries of data that summarize the data's significant content and offer insight into the data (Tripathi & Singal 2019). Graphs make it easier to decide whether or not more intricate modeling is required (Ahmed et al. 2019). Line diagrams and box plots were employed in this study to summarize a dataset during exploratory data analysis. Data were evaluated using two-way ANOVA at a 0.05% level of significance with the aim of determining significant differences between the sites as well as seasons for all water quality measures. Furthermore, the multivariate technique known as PCA was applied to the stream water quality. By exploring groups and sets of variables with similar qualities, PCA may enable us to uncover the structure or patterns in the presence of chaotic or perplexing data, thus simplifying our explanation of observations (Molina et al. 2020). The SPSS (v. 26) software was used to conduct a typical statistical analysis. A data collection with many connected variables is reduced in dimension using FA, which does so by splitting the dataset into a new set of variables called principal components (PCs), which are orthogonal (noncorrelated) and are organized in decreasing order of importance. The PCA is a method of data reduction that specifies how many different types are crucial to understanding the observed variance in the data (Nguyen & Holmes 2019). When computing PCs mathematically, eigenvalues and eigenvectors are obtained using covariance or other cross-product matrices that depict the dispersion of the many observed parameters and the initial variables. With fewer variables, PCA can explain the same amount of variance as it does with more variables (PCs). Additionally, PCA makes an effort to explain the relationship between the observations in terms of underlying elements that are not immediately visible (Lawson et al. 2012).

It can be clearly seen that the water quality of River Jhelum shows significant variations along its length due to its instream pollution and abrupt changes at input sampling stations, namely, site 2 and site 5, which depict the water quality of Rivers Lidder and Sindh, respectively. Furthermore, the water quality of River Jhelum takes another shape when the river takes a pause in the Wular Lake due to other input sources of Wular Lake, namely Arin and Madhumati whose quality parameters are shown at site 8 and site 9, respectively. The water quality indicators of Wular Lake are shown at site 10 and site 11 which are quite different from the values shown at the starting section of River Jhelum. Hence, it can be evidently understood that the water quality of River Jhelum varies greatly along its length until it reaches the Wular Lake due to its various input sources in the form of its tributaries. Transparency, TDS, and TSS are at their peak values in Wular Lake due to higher levels of dissolved organic matter and sediment yield carried by its input water resources. Ammoniacal nitrogen and nitrate nitrogen levels are higher in all sampling stations of River Jhelum which may be due to residential and industrial cover adjacent to its banks contributing a greater amount of animal and human waste to the river.

Ironically, limited research works have been carried out in the past to access the water quality parameters of River Jhelum due to which comparison analysis of past studies was arduous. However, the available studies were compared with the results of this study as shown in Table 3. It can be clearly seen that there is significant surge in the amounts of TDS and TSS from 1982 to 2021 which hints toward excessive soil and nutrient loss from the adjoining areas of the River Jhelum in the last two decades. These findings indicate that the River Jhelum has become more eutrophic in contrast to prior water quality assessments, showing a continuing decline in the water quality.

To assess the variation of the parameters affecting water quality, a two-way ANOVA was used. At a probability of 5%, the value of parameters with significant F was compared between the stations as well as seasons. The findings indicate that there is a substantial difference between F and F-critical value for all sample values, and that the P-value is relatively small in comparison with alpha value (0.05) except for the phosphorus. Null Hypothesis (H₀) is rejected for almost all parameters showing there is a significant variation in parameter values across all sampling stations as well as among all the sampling seasons. The only parameter where Null Hypothesis is accepted across sampling stations is phosphorus showing there is no significant differentiating of values across 11 sampling sites. The results of the two-way ANOVA analysis are shown in Table 4.

FA is a multivariate statistical analysis technique for variables having numerous internal dependent interactions (Jehan et al. 2019). Its goal is to uncover the fundamental organization of the data and identify a small number of ‘abstract’ indicators that serve as the foundation (Auerswald & Moshagen 2019). These abstract indices can explain the observation of the interdependence between variables, naming these variables as factors; FA is about how to extract the information into a few factors with the least amount of loss. These abstract indices can represent a significant portion of information that is reflected by many of the original observation variables. Additionally, this approach can offer the scientific foundation for decision-making by analyzing and assessing in a reasonable and scientific manner (Sellbom & Tellegen 2019). PCA was used in this work to analyze 14 water quality indices at 11 study-area monitoring stations. First, the Kaiser-Meyer-Olkin (KMO) and Barlett tests were used to determine whether PCA was applicable. These tests were used to confirm, respectively, the sufficiency of the sample and the independence of each variable (Elsaman et al. 2022). KMO = 0.566 (>0.5) and Barlett test value = 0 (0.05) were the estimated findings, suggesting that the data are appropriate for PCA.

Using SPSS 26.0 software (IBM, Armonk, NY, USA), a correlation coefficient matrix was created by standardizing the original monitoring data. The DS obtained from the software is presented in Table 5 while the correlation matrix is shown in Table 6.

Table 7 displays the typical levels of FA. It includes initial common degrees which equals the value of 1 for all parameters. The communality of the variables following factor extraction is listed in the third column. We can have a look at the seven indices: pH, TDS, TSS, DO, alkalinity, calcium, and transparency, all of which have high common degrees (>0.80), with thorough information. Furthermore, the common degree of conductivity, CO₂, hardness, phosphorus, chloride, NH₃-N, and NO₃-N is low (<0.80), with insufficient information.

Table 8 shows that all four common factors have distinctive roots that are greater than one and their combined contribution rate to variation accounts for 79.234% of all variance. Furthermore, only the first four components are extracted and rotated. The variance explaining the original variables of the numerous factors was redistributed by the factor, bringing the variance of the factors closer together. It means that these four parameters essentially reflect the core elements of the original data.

The factor loading matrix, which provides the load each variable has on PCs, is shown in Table 9 before and after rotation. We can see that, following rotation, the load factor has been substantially polarized.

The first principal component (PC1) accounted for 35.23% of the overall variation and had a significant positive loading on TDS (0.877), and the second principal component (PC2) accounted for 18.51% of the overall variation and had a significant positive loading on TSS (0.929). It shows that a substantial sediment production as a result of bank erosion, agricultural runoff and watershed disintegration are to blame for the stream's worst pollution problem. For factor 3, the maximum factor loading values (>0.86) for DO and alkalinity indicate that soil erosion and bed decomposition (phosphates, limestone) are other significant environmental pollutants in rivers. The fourth major component, which accounted for 8.46% of the overall variation, significantly increased the positive loading on the pH index (0.877). The link between a river water pH and other water quality indices is complex. Toxic pollution from industrial manufacturing, however, may be to blame.

Table 10 represents the score matrix of PCs. The first and second component show the amount of water that has been contaminated by sewage and soil erosion, respectively, and represent inorganic and organic pollution factors. The third component is a complete pollution factor, and it primarily displays the amount of water pollution caused by the material's inherent structure in river beds; and the fourth component is metal pollution.

The relevant factor scores at different sampling stations can be obtained using the SPSS software.

The results obtained using the above formulas are shown in Table 11. The results in the table reveal that of the 11 sampling stations, site 5 has the poorest water quality and site 1 is the least affected sampling station. Comprehensive data indicate the 11 cross-sections’ pollution status, analyzed using evaluation function F in the following order: site 5 > site 10 > site 9 > site 11 > site 8 > site 2 > site 7 > site 9 > site 6 > site 3 > site 6 > site 3 > site 4 > site 1.

The key problem with conventional water quality monitoring systems is the rapid generation of massive physicochemical data matrices that demand an efficient data processing system in order to evaluate the results, associate variables, and draw conclusions that are relevant. The results of this research show that PCA, a potent multivariate statistical tool, decreases a dataset's dimensionality while keeping as much of the dataset's variability as possible and enables the assessment of relationships between variables. The environmental variables highlighted by PCA clearly define and explain the pollution gradients.

The purpose of this study was to determine the overall surface water quality scenario of the north-east Himalayan region of Kashmir Valley using standard testing procedures, and analyzing the results using multivariate statistical tools. Furthermore, the impact of side-streams, specifically Lidder, Sindh, Arin, and Madhumati, on the water quality of the River Jhelum and the Wular Lake was evaluated. Surface water quality testing was done at 11 sampling stations for the year 2021 by collecting the samples throughout the four seasons to check the spatial and temporal variability. Line diagrams and box plots show that certain parameters, as per WHO guidelines (Cotruvo 2017), are beyond the allowable limits at a specific site for the specified sampling season and are not acceptable for drinking, farming, fishing, or other household uses. The increasing amount of side-stream pollution was a sign of increased anthropogenic pressure in the watersheds of the north-east Himalayas.

The sampled parameters were subjected to a two-way ANOVA analysis to determine whether there was any seasonal or sectional divergence. The results showed that there was significant spatio-temporal variability. Furthermore, the most important indicator parameters impacting water quality and potential sources of pollution were extracted using the PCA method. On the basis of optimizing the retention of the original data information, PCA integrates and simplifies the high-dimensional variable system. Four significant PCs were extracted from the 14 water quality measures by PCA, which accounts for 79.23% of the variation in the initial dataset. PC1 (35.23%) and PC2 (18.51%) represented chemical pollutants, indicating the influence of bank erosion and other deposited sediments on water quality. PC3 (17.03%) provided a positive correlation with DO and alkalinity and represented pollution due to bed decomposition and industrial as well as residential wastes. PC4 (8.43%) has the positive loadings of pH representing toxic pollution from the municipal area surrounding the surface water resources.

In order to improve decision-making and the efficient management of water resources in the River Jhelum and its related tributaries, this study makes use of the technique's strengths in assessing and interpreting sources of pollutants as well as the fluctuation of the water quality. The findings of this study could prompt deeper and logical considerations and lead to an improvement in critical management for the ecology and environment of surface water resources in the north-eastern Himalayas.

The main limitation of the methodology used in this study is the difficulties faced in the interpretation of obtained results. The PCs are given by PCA. PCs are not as readable and interpretable as original parameters. PCs attempt to account for as much variance among the variables in a dataset as possible, but if the number of PCs is not carefully chosen, it may leave out some critical information from the original set of features. Secondly, field data from 2021 was used to determine the water quality status in the study watersheds. However, a long-term monitoring data would have been helpful in more accurate analysis to find the patterns of water quality degradation. In view of this, we recommend a comprehensive monitoring programme that would consider data over a longer period of time and for more water quality components, such as heavy metals and other contaminants, which might enter the River Jhelum through the use of pesticides in agricultural farms. Furthermore, it is suggested that the in-depth analysis of potential factors responsible for river water pollution may be carried out using optimized machine learning models.

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

The authors declare there is no conflict.