An Integrated Assessment of Spatial Variability Mapping of Surface Water Properties and its Impacts on Drinking Water in Mahanadi River Basin, Odisha, India

1.

Introduction

Water is a resource that is necessary for all living things to survive. Thus, it is essential for human health, food security, and the environment. Surface water describes any body of liquid water that is present on Earth’s surface [1]. This includes rivers, lakes, and wetlands, which are the primary supply of water for residential, commercial, and industrial uses. As a result, they are the source of water for most needs, including those related to drinking, household use, research and industrial activities, agriculture and crop yields, floriculture, cattle breeding, and the strategic planning of aquatic life, including fish stocks and fisheries. However, overexploitation and inappropriate garbage disposal into surface water reservoirs are two examples of how anthropogenic activities are negatively affecting water availability and quality at an alarming rate [2]. A healthy river also has an impact on the aquatic ecology, humans, and animals that depend on the water for sustenance. River health is determined by water-quality indicators and the sources of that water. Current agricultural practices pose a major threat to human health, especially when it comes to the overuse of fertilizers, dirty conditions, and the dumping of sewage into surface water. Hence, water-quality status estimation is a challenging undertaking due to the abundance of potential physical, chemical, and biological indicators [3]. Researchers have conducted a great deal of study in support of maintaining and managing the quality of fresh surface water since water is essential to the sustainability of human civilization and the environment.

As a result, over the past 20 years, a significant amount of research has been conducted to profile the water quality for understanding the overall health of an ecosystem and the condition of surface water [4]. Optimizing the scale of use of these resources can be achieved by lowering the high cost of their production and application and by implementing suitable management techniques when utilizing the water resources that are currently available. The water quality index (WQI) method has gained popularity as an effective means of evaluating the impact of specific parameters on the total surface water quality (SWQ). Similarly, there is a defined limiting value for each water-quality parameter based on its intended use; samples that exceed this value are not appropriate for the intended use. The actual water sample from a river is made up of a constituent matrix, which is an accumulation of numerous water-quality indices [5]. Furthermore, it is not as simple to grade water-quality indicators because it is dependent on the relative toxicity rate and indicator concentration. Although it is a difficult process, the end result is a useful index value or number that can be used to relate the effects of the primary indicators in the constituent parameter matrix. The main drawback is that it raises questions about how even a slight alteration in weighting may impact how water quality is interpreted overall. Because of this, choosing appropriate weights for the various water-quality factors is a crucial step in the process.

To get around this problem, a large number of indices are created using a meaningful set of parameters, subindex values, and weights applied to the parameters [6]. Entropy-based weights have become a valuable method that leverages information entropy to provide water-quality criteria weights. This method is a group of correlation-based techniques that rely on analytical testing of the decision matrix to ascertain the existing data in the criteria used to calculate the weights of the criterion. The assessment of water quality and determination of precise correlations between the quality parameters and measuring locations have been conducted by multivariate statistical analysis (MSA). Numerous statistical formulations have been put forth previously, and each substitute index possesses some benefits and drawbacks [7]. Moreover, the challenge of assigning unknown samples and extracting important information from them becomes more difficult as the number of measurement parameter dimensions in samples increases [8–11]. Therefore, methods like self-organizing map (SOM) have also been extensively employed in the evaluation of SWQ due to their potent dimension reduction and classification capabilities. In addition, SOM is useful along with cluster analysis (CA) for analyzing intricate datasets. Effective river-water quality management requires evaluating the quality of water, determining the sources causing pollution, and comprehending the temporal and spatial fluctuations in the water quality that has been assessed [12].

Generally, these integrated methods often extract a limited number of components that capture significant patterns from complex multivariate data while minimizing information loss, and then display the result in a low-dimensional space. These methods use the Euclidean distance in conjunction with Ward’s approach to calculate a measure of similarity that yields intuitive relationships between each given sample and the full dataset, which can be represented by a tree diagram. Recently, the evaluation and routine monitoring of SWQ have been conducted by employing the geographic information system (GIS) model, combined with the inverse distance weighted (IDW) interpolation method, and has been shown to be an effective tool for assessing and analyzing geographic data related to water resources [13]. Moreover, IDW is broadly utilized to assess values at unfamiliar locations by relying on the available data points in the vicinity. It applies deterministic techniques that are most frequently used for multivariate interpolation, which utilizes the standard mixture of existing grid points to generate an unexpected grid value. It is a popular choice in the fields of spatial analysis and geostatistics to interpolate data points on a smooth surface. It can include both analytical and semi-empirical approaches for estimating and creating quantitative or qualitative water maps. This approach operates on the assumption that data points in close proximity to the target position exert greater impact on the predicted value than those that are far away. Thus, this is a technique that can be applied to quickly and cheaply turn huge datasets into a range of regional distribution maps and projections that illustrate the origins, patterns, and connections of pollutants [14]. Therefore, IDW has demonstrated superior performance in comparison to alternative methods that generate interpolation for the designated points with greater variability. Besides that, multicriteria decision-making (MCDM) approaches like weighted linear combination (WLC) have generated noteworthy findings that necessitate a methodological synthesis for a thorough understanding of the management and monitoring of SWQ for practical application. This requires the use of the WLC approach in order to prioritize decisions during times of emergency and to provide an overall assessment of the sites based on both physical and chemical factors [8–11]. This further involves having the ability to collaborate and develop an integrated or combined strategy that, on the one hand, makes sense and is within the purview of the MCDM format and, on the other hand, receives good feedback. During the COVID-19 epidemic breakout, the use of clean water and proper sanitation was the first line of defense and prevention. Surface water is also an important source of freshwater, which was crucial to humanity’s fight for survival. To this end, comparative studies should highlight the various water elements and how they affect health.

It is important to state that no study has been undertaken within the present study area of making use of these tools in predicting water-quality suitability for surface water. Even so, in numerous studies, only the modeling of SWQ has been dealt with, and there has not been much research done on validating the models and procedures employed. Therefore, the experiment aimed to fill the gap in the available research and at integrating the entropy WQIs with MCDM, GIS, and MSA, by utilizing SOM and CA, in the prediction of water suitability in the investigated study.

Specifically, the objectives are as follows: (1) assess the physicochemical composition of surface water; (2) assess the drinking suitability of surface water; (3) determine the relationship within drinking parameters; (4) predict drinking indices using entropy, MCDM, and SOM models; (5) make adequate suggestions and recommendations based on the findings. Thus, in the current experiment, significant datasets collected during the course of a seven-year (2016–2023) monitoring study were subjected to CA, and SOM was used to collect implicit data regarding the parallels and divergences between the monitoring locations. Twenty-one water-quality parameters collected from the Mahanadi River (Odisha, India) in pre-monsoon were evaluated using WQI, WLC, and MSA.

It is asserted that this method is helpful in evaluating the SWQ of a region, particularly when there is a large amount of data because it considers the integration of geographic and temporal fluctuations in addition to the overall quality by measured variables and specific spots. Furthermore, creating more accurate predictive models that enable the prediction of suitability of water for drinking and agriculture is an innovative concept that will aid subsequent studies in the area and the techniques that can be adapted to accommodate other localities worldwide.

The framework of the study includes the following. Surface water samples were obtained and subjected to an evaluation of their physicochemical properties encompassing major ions and cations. The application of optimization tools and GIS enabled the creation of WQI maps, providing valuable information for both drinking and irrigation purposes. Furthermore, the examination of surface water chemistry was conducted through the utilization of entropy and an SOM diagram. The estimation of surface water contamination sources was facilitated through the application of WLC analysis.

2.

Overview of study area

The river Mahanadi served as the study area. Mahanadi literally means “mighty river” or ‘great river’. This watershed is the people’s lifeblood in Chhattisgarh and Odisha. The location of this investigation lies between 19°30^′N and 22°30^′N latitude and 81°45^′E and 87°00^′E longitude at the chosen study site [8–11, 15]. Hydrographically, it is a major peninsular river that flows across the states of Chhattisgarh and Odisha with a total drainage area of 141,589 Km² (approximately 4.28% of the entire land area of India). The river flows through the majority of the districts in the states of Chhattisgarh and Odisha; in the state of Jharkhand, it covers approximately 0.45% and finally, in the state of Maharashtra, it covers roughly 0.07%. In addition, its area is approximately 141,589 Km² and it drains an area of about 65,580 km² in Odisha [16]. It exhibits comparable flood susceptibility, and its riverbank area is among India’s most densely populated.

Additionally, it supplies water to about 38,606,665 people who depend on it for survival and basic needs. Approximately, 80% of this comes from the south-west monsoon that occurs from July to September. The basin is distinguished by a hot, tropical climate, with summer peak temperatures reaching 45 °C and winter low temperatures hovering around 30 °C [17]. Afterwards, an average of 140 cm of precipitation is predicted for this river each year. This river is constantly filled with stagnant water from January to May of each year, which contributes to the buildup of pollutants in the surrounding area. The Eastern Ghats Mountain ranges also have a big impact on the pattern of rainfall in the area. The primary crops that are grown in this river basin are oilseeds, sugarcane, and rice. Therefore, the two primary land uses in the basin are agriculture and forestry, which are backed by extensive irrigation infrastructure created by major- and medium-sized projects [18]. The transportation network in the district of Cuttack is reliable and facilitates the interchange of these crops through the neighboring towns and cities. The local farmers who cultivate crops for sale in and around cities or towns as their primary source of income benefit from this exchange [19]. However, the amount of irrigation extraction per person in this river channel is approximately 686 m³, and it is estimated that there are 1.85 million hectares of net irrigated land in this river region. The Mahanadi basin is separated into three smaller regions, namely, the upper catchment constitutes about 21.34% of the cumulative area, 37.16% belongs to the middle region, and lower Mahanadi contains an area approximately 41.5%. Thus, the lower basin typically encompasses an area of approximately 57960 Km². Regarding this, a few features of the sampling locations are displayed in Figure 1.

3.

Sampling and data collection

The present study follows a systematic sampling design framed to portray the nature of surface water (river) suitability for drinking purposes in the river basin. The State Pollution Control Board, Odisha, provided the data for 19 water-quality monitoring sites, which included 40 water-quality parameters tracked monthly during a 7-year period (2016–2023) using a judgmental sampling technique to better reflect the hydrochemical parameters of water samples. To be employed in the current research, 21 of these 40 parameters were chosen based on their sampling continuity at each of the chosen monitoring locations. It should be highlighted that each experimental procedure and sample collection, preservation, and management received particular care throughout the evaluation period [8–11, 20]. Notably, sampling was conducted specifically during the dry season, characterized by reduced water levels, elevated cation concentrations, and increased stability of anions. An extensive evaluation of the following physicochemical parameters was conducted: temperature, pH, biochemical oxygen demand (BOD), dissolved oxygen (DO), total dissolved solids (TDS), alkalinity, chloride (Cl⁻), sulfate (SO), coliform/total coliform (TC), iron (Fe²⁺), fluoride (F⁻), boron (B⁺), total hardness (TH), total suspended solids (TSS), electrical conductivity (EC), chemical oxygen demand (COD), NH₃–N (ammoniacal nitrogen), free ammonia (free NH₃), total Kjeldahl nitrogen (TKN), sodium adsorption ratio (SAR), and total hardness (TH). Pre-calibration was performed on each instrument in accordance with the standards of practice before starting tests.

Subsequently, water-quality indicators such as temperature, EC, pH, TSS, and TDS were calculated at the sampling spots (model HQ40d, Hach, Co, USA), while F⁻ and SO concentrations were determined using the nephelometric method and were measured using a UV–visible colorimeter at 650 nm (model Jenway 6405). A handheld Global Positioning System was used to record the geographic position of each sampling point (eTrex 20, Garmin). Parameters like DO were estimated using a multiparametric probe with calibrated sensors during field measurement. The concentration of Cl⁻ was determined using Mohr’s method. The most probable number (MPN) method was used to determine the count of coliform bacteria in the water sample. In the laboratory, COD was analyzed through permanganate titration, and BOD was measured via the reduction of DO in the raw water samples after 5 days. Moreover, alkalinity and TH were tested using the complexometric method with EDTA titration. Then, NH₄–N was measured through Nessler’s reagent spectrophotometry, while free NH₃ and TKN were analyzed through distillation using a distillation apparatus (UDK 129 Kjeldahl distillation unit, Velp Scientifica). Ion-selective electrode techniques (model Orion 9796 BNOP, ThermoFisher Scientific) were used to analyze nitrate (NO). Two metals in all, namely Fe²⁺ and B⁺, were investigated by ion chromatography.

The average of all replicate measurements within a relative standard deviation of ±2.2% has been reported in this study. Regarding quality control and quality assurance, to maintain accuracy and precision in observations, all analytical equipment was pre-calibrated using standard solutions or in accordance with the company’s specifications [21]. According to the principle of electroneutrality, the concentrations of cations (positive ions) and anions (negative ions) must be equal, which is popularly known as ion balancing. Regarding the validation of surface water, ion balancing is an important tool. The accuracy of the laboratory analysis has been validated by charge balance error (%CBE) assessment computed using Equation (1). The observed CBE falls within the acceptable limit of ±5%:

In the study area, statistical analysis was performed using IBM SPSS Statistics 25 on a dataset comprising 21 physicochemical parameters, which were then subjected to SOM and CA analysis. The spatial distribution maps of various physicochemical parameters were generated using GIS techniques and contouring methods with ArcGIS 10.6. The spatial distribution maps for each parameter were prepared utilizing IDW interpolation techniques to illustrate variations across the study area. Furthermore, the model’s maps were elaborated upon based on these spatial distribution maps.

4.

Methodology

The SOM and CA were used to categorize the samples of surface water. In order to determine the causes of pollution, the grading implications of surface water samples and the variation in ion concentration for every water specimen were acquired to obtain the desired result. The water sample scores for water quality were then calculated statistically by applying the entropy approach to compute the WQ score of each water sample. The WQ examination and its estimation from pollutant sources were thus completed using the entropy approach in conjunction with the SOM and CA data. Therefore, for determining relative weights, an entropy-based approach has been utilized, which captures the degree of disorder and randomness in the data. The versatility of this approach has led to their widespread adoption across diverse domains, including image and speech recognition, natural language processing, financial forecasting, and medical diagnosis. Thus, their capability to handle intricate, non-linear relationships within data renders them invaluable tools for addressing complex real-world challenges. So, it acts as a intelligence research progresses, SOM continue to drive innovation, powering intelligent system that enhance operational efficiency, accuracy, and decision-making across various fields [22]. 

Hence, based on Shannon’s information theory, depending on the distribution of the data, the significance of the relative intensities (weights) of the criterion is interpreted [23]. This technique functions with the assumption that assigning a a greater weight to data appears to be more useful than a lower weight to data. Thus, it is a productive method for presenting probability and uncertainty [24]. The most significant advantage is that because human intervention has been eliminated, it produces trustworthy results from the weight allocations (S_i) to parameters. To determine the relative weights of WQ data with several samples, the algorithm of this technique can be presented in the form of a flow chart as in Figure 2.

This article has covered conventional entropy mathematical approaches, which can be used to approximate all types of water samples. Nevertheless, there are a few processes involved in calculating the final result. Because of this, chemometrics can be advantageous based on the statistical multivariate technique known as SOM. This approach is an algorithm for unsupervised neural networks developed by Kohonen [25]. The Kohonen map or an SOM algorithm is the most widely used artificial intelligence model. They create two-dimensional maps using unsupervised learning and combine the objectives of clustering and projection methods [26]. Based on the similarity of the data, this map may present complex, high-dimensional data in a consistently organized, low-dimensional format.

Consequently, it is a successful linear dimensionality reduction technique that is frequently applied in classification and data mining [27]. It functions as a model for searching and visualizing linear and nonlinear relationships found in datasets. The input layer and the output layer are its two constituent layers [28]. One of the primary objectives of the SOM approach is that hierarchical cluster analysis (HCA) is utilized with Ward’s linkage to obtain instructive and explanatory reference vectors, also termed weight, prototype, and codebook vectors. At present, these models are employed by various academics for modeling purposes and to investigate a broad range of hydrological issues, particularly in the areas of rock geochemistry and hyperspectral imaging [29]. Furthermore, this mechanism is highly suited for processing ecological data since it is based on unsupervised learning, which eliminates the need for prior information. These techniques can therefore replicate more accurate patterns and correlations between input and output parameters, and they are very good at extracting response data from the challenging objective [30]. Conversely, in hydrological models, it can be utilized to simulate both regression and classification. Therefore, coupling SOM and CA with GIS is suggested for simulating the surface water to search for vulnerable zones based on similarity, which decreases costs and time [31]. This makes it possible to zoom in on areas with a respectable degree of precision, and it can quickly communicate the findings in cluster-map style to all users, researchers, and managers. The steps for the application of SOM with entropy are illustrated in Figure 3.

On the other hand, WLC has been shown to be a useful technique for managing data when there is a significant degree of collinearity and noise in the input variable data and a large number of output variables. Hence, SOM was investigated as a substitute method for predicting indices in this field of study since it is quick, easy, and only requires a few steps to compute—especially for the WQIs. This improves on the current standard WQIs, resolves conflicts among several WQIs, and aids in online monitoring system optimization to increase the data obtained via the optimization technique [32]. Integrating WLC and GIS is a useful and reasonably priced method for handling geographic data. For many years, researchers have weighted various thematic layers to determine WQ zones using the WLC approach. When there is insufficient data for analysis, splitting up several substitutions into pairwise comparisons and merging the results is suggested. This gives rise to a mathematical framework that addresses ambiguity and uncertainty by admitting that statements may be partially true or incorrect and enabling the mathematical description of varying degrees of truth. Furthermore, this approach presents a state-of-the-art workspace for spatiotemporal data management, and analysis, integration, and visualization of digital spatial data from several sources.

Additionally, it integrates with GIS, which supports surface water management planning and decision-making through spatial analysis, modification, and visualization. This combination creates a flexible and user-friendly program. In a nutshell, after the obtained data has been normalized, this model may produce accurate ratings for survey sites. It also analyses the decision-making process to comprehend the influence of uncertainties, which are a frequent characteristic of water management issues. The steps to compute the final ranking are shown in Figure 4. Using ArcGIS software, the surface water potential map was produced by superimposing all of the criteria maps and applying the WLC technique in order to estimate the final score. The highest absolute value shows the best possible choice. Thus, when sorting the WLC data into groups, if the factor is less than zero, it shows unsuitable class; 0 to 112 signifies low suitability; 112–124 is moderately suitable; 124–133 indicates suitable class; between 133 and 154 is the highly suitable zone. In fine, this was adopted to identify pollution using physicochemical parameters so that the extent of pollution could be ascertained.

5.

Results and discussion

The findings from the statistical analysis of the surface water sample data during pre-monsoon are compared with that of the World Health Organization [33].

6.

Conclusion

The combination of Geographical Information System (GIS), Weighted Linear Combination (WLC), and Self-Organizing Map (SOM) models offers valuable insights into predicting the suitability of surface water for drinking in Mahanadi River Basin, Odisha. In the current work, the spatial distribution of the WQ pattern and its distribution over pre-monsoon was investigated using entropy, SOM, WLC, and GIS over a period of 7 years, with quality data from 19 water gauge stations of the Mahanadi basin. To examine the dispersion of main ions in the research area, GIS-based spatial analysis and the IDW interpolation approach were used. In further assessment of the water quality, the computed entropy readings range between 0.15 and 4, demonstrating that the river water is suitable for use only after treatment. The TC values also exhibit higher readings across the survey stations, whereas the NH₃ and alkalinity values are firmly within permissible limits. Hence, high TKN values can be linked to the rise in PO limits, which consequently elevates the river’s nutrient/eutrophication load. This is because of nutrient enrichment, which causes a rise in algal blooms. Further, this outcomes of the study showed a significant presence of chloride, TKN, coliform in the water, likely originating from sources such as dissolution of salts, sea-water intrusion, or anthropogenic activities. Approximately, 47.36% (n = 9) of locations demonstrate bad–worst WQ, which is mainly attributed to the direct disposal of stored and municipal garbage into rivers. Additionally, the presence of mineral dissolution suggests the dissolution of minerals from geological formations, contributing to the chemical composition of surface water. Also, mixing processes indicate the blending of water from different sources or aquifers, resulting in variations in ion concentrations. Hence, this index establishes that the river water is unfit for human consumption and household use at nine sites, and the water should be consumed only after proper treatment and disinfection. Furthermore, an SOM was also generated with the HCA via Ward’s linkage to have a deeper comprehension of water suitability throughout the study area. The subcategories of the WQ reviewed by SOM values are based on or rely on the majority of the research field that falls under three categories, namely, low, moderate, and high pollution. It comprises 5 parameters and 10 locations in Cluster I while in Cluster II, it contains five indicators and locations. Finally, 10 indicators and 4 stations are included in Cluster III. This shows that greater concentrations are seen along the river’s flow, which could be caused by home wastewater systems or sewage infiltration. In addition, the MCDM technique, namely, WLC has been incorporated into the work, which acts as a practical and effective instrument to compile and report monitoring data to decision makers so that they can comprehend the SWQ’s current state and have the opportunity to apply it more successfully in the future. The observed range is found to be 0.5–154. In fine, the value of 0.5 that was measured at station P-(9) is found to be the lowest, which indicates an unsuitable WQ zone. The WLC values of stations (2), (8), (10), and (19) are 115, 110, 117, and 100, respectively, which indicates low–moderate suitable WQ. The general suitable WLC value for the river is 52.63%, indicating that the water is suitable for drinking. Some stations show unsuitable/moderate WQ, and this could be caused by a variety of anthropogenic activities and natural phenomena that take place alongside the river. Therefore, the proposed study uses a comprehensive strategy towards SWQ monitoring and healthy management of water resources. The present study is limited to the effective identification and integration of criteria, which helps determine the suitability of surface water recharge sites and integrate them with the actual site. Thus, this approach can also be adapted to target water-starved locations where surface water degradation is an increasing concern. In these areas, artificial recharge techniques can be used with minor site-specific alterations. Therefore, the study advocates for integrating these advanced modelling approaches to enhance efficiency and  reduce analysis time and costs. However, variability in model performance across datasets and the need for robust training data remain challenges. The specific limitations of this study are to: (a)  Hydrogeochemical parameters can exhibit temporal fluctuations influenced by seasonal changes, anthropogenic activities, and climatic variations, which were not fully explored due to the study's duration, (b) Availability of comprehensive historical data on hydrogeochemical parameters could enhance long-term trend analysis and improve predictive modeling accuracy, but such data may not have been readily accessible or uniformly available, (c) The machine learning models utilized assume stationarity and linearity within the dataset, which may not fully capture nonlinear relationships or abrupt changes in water quality dynamics over time, and (d) Field sampling limitations, such as accessibility to certain locations or constraints in sampling frequency, could influence the representativeness of water quality assessments across the study area. Keeping this in mind, the major recommendations of this study are to: (1) Implement comprehensive, ongoing monitoring programs to track hydrogeochemical parameters across diverse geographical locations and over extended periods. This will provide a robust dataset for understanding temporal variations and trends in groundwater quality, (2) Expand the spatial coverage of groundwater quality assessments to encompass regions with varying hydrogeological characteristics. This approach will facilitate comparative analyses and improve the applicability of findings to broader environmental contexts, (3) Foster partnerships with local communities, governmental agencies, and agricultural stakeholders to promote awareness of groundwater quality issues and implement adaptive management strategies tailored to regional needs, (4) Further explore advanced machine learning models, such as deep learning algorithms and ensemble methods, to enhance  variables, and (5) Advocate for the adoption of climate-resilient agricultural practices that mitigate the impacts of groundwater salinity and alkali risks on soil health and crop productivity. This includes precision irrigation techniques and soil amendments tailored to local hydrogeochemical conditions.

Data availability

All data are underlying the result are available as part of the article and no additional source of data are required. On special cases, data will be made available on request.

Competing interest

The author declares that he has no known competing financial interests or personal relationships that could have influenced the work reported in this paper.

Author contributions

Abhijeet Das: conceptualization, data curation, formal analysis, investigation, methodology, resources, software, validation, project administration, visualization, fund acquisition, writing—original draft, writing—review, and editing.

Acknowledgments

The author gratefully acknowledges the editor’s and anonymous reviewers’ contributions to the improvement of this paper. The author would like to express his sincere gratitude to C. V. Raman Global University. The water-quality data utilized in this study was retrieved from the State Pollution Control Board, Odisha. Moreover, the author sincerely acknowledges the Research Group of Rani–Prativa Das for providing computational laboratory facilities to complete this research.