Surface Water Quality Assessment of the Arkavathi Reservoir Catchment and Command Area, India, through Multivariate Analysis: A Study in Seasonal and Sub-Watershed Variations

Abstract

The Arkavathi River, one of the major tributaries of the Cauvery River in southern India, is a major source of drinking water and agricultural irrigation to villages and townships in the region. Surface water quality distribution and characteristics of the Arkavathi Reservoir catchment and command area were evaluated using multivariate statistical analysis on 29 water quality parameters collected across 30 monitoring stations over a two-year, three-season period. Factor analysis (FA), agglomerative hierarchical clustering (AHC), analysis of variance (ANOVA) and t-tests were used to reveal strong links between parameters and to reveal significant variations in their concentration levels with respect to monsoon seasons and sampling locations across the sub-watersheds. Results from factor analysis showed strong groupings of specific parameters across seasons, while results from clustering revealed distinct clusters of sampling points around the river, upstream from the reservoir (where human activity is high), in the command area downstream from the reservoir (where irrigation activity is similar), in hilly regions towards the northeast of the study area and in the scrubland regions. Based on multivariate analysis findings, specific recommendations are made for water quality improvement in the reservoir catchment and command area.

1. Introduction

Water, being one of the vital components of life, is under more scrutiny in recent decades due to urbanization, industrial and agricultural activities, and improper management at various levels causes deterioration in water quality, in turn affecting its use for all human activities [1,2]. Deterioration of surface water quality in reservoirs and lakes is of particular significance when they are used for crop irrigation and more-so for drinking purposes, as evident in the present case of the Arkavathi River, which feeds into the Cauvery River, which is one of the main sources of drinking water for the region. For an effective water management strategy, it is therefore imperative to collect and analyze reliable data for spatio–temporal variations of surface water contaminants across monsoons and sub-watersheds in reservoir catchment and command areas, which in turn can be used to identify pollution sources and propose remedial measures to improve water quality.

1.1. Multivariate Analysis of Surface Water

Various contaminants, along with sediments, through runoff are carried through the tributaries of the Arkavathi River (total flow length of about 210 kms) at various locations and further into the Cauvery River, which is the main source of water supply to many townships in the southern Indian state of Karnataka. The quality of vegetation in the command area of the Arkavathi Reservoir (also known as the Harobele Reservoir) is affected by the water quality due to various stressors such as untreated domestic sewage, effluents from agricultural runoff, etc. Surface water samples were collected to analyze various characteristics of water quality in the six sub-watershed areas (Section 2.1) and were analyzed with the aid of appropriate multivariate statistical tools (Section 3) to derive meaningful conclusions about various parameters within each of these datasets and to propose remedial solutions.

One of the most-applied multivariate methods in watershed studies is principal components analysis (PCA), which uses correlations among multiple water quality constituents to effectively reduce the number of parameters [3]. For example, in [4], the potential pollution sources affecting the Jinsha River watershed in western China from 2016 to 2018 were investigated using an improved method in combination with correlation analysis and absolute principal component score multiple linear regression receptor modelling. The multivariate statistical techniques presented good adaptability for the analysis of pollution sources in this river watershed, and the results were useful for the protection and management of the watershed eco-environment. To determine relationships between physical and chemical water quality parameters, PCA and CCA (canonical correlation analysis) have been widely used in studies such as those in restored wetlands in northeast Denmark [5] and in the Llobregat River in northeast Spain [6]. Though PCA has been widely used in such studies, factor analysis (FA) has been found to be better-suited if the main objective is to identify parameters of water quality that are related to each other and to separate them from each other. FA has been applied to surface water quality data, for example, for a river basin in Turkey during two different hydrological periods of low- and high-flow periods [7], where the source of pollution changed from agricultural uses to land use during high-flow periods.

Cluster analysis (CA) has also been widely used, sometimes in conjunction with PCA/FA for assessing variation of contaminants across either locations or seasons. In a longitudinal analysis study, pollution profiles of 40 rivers in India were created based on the levels of biochemical oxygen demand (BOD) [8]. CA was used to represent groups of rivers with similar levels of pollutants. Another similar application of CA was to evaluate dam water quality, for example that of northcentral Algeria [9], where clusters of sampling sites reflected differences between water quality at different locations. There have been studies demonstrating the usefulness of multivariate statistical approaches for analyzing temporal variations in water quality for effective river water quality management [10,11,12,13,14]. In [13], the authors assessed the surface water quality data for 16 physical and chemical parameters collected from 22 monitoring sites during the years from 1998 to 2001. PCA was used to extract the parameters that were most important in assessing seasonal variations of river water quality. The analysis showed that parameters that are most important in contributing to water quality variation during one season may not be important during another season, except for dissolved organic carbon and electrical conductance, which were always the most important parameters in contributing to water quality variations over all four seasons.

One-way and two-way ANOVA have been used to extract temporal and spatial variations of water quality [15], mainly to show the reliability of results from CA and FA [16]. However, two-way, and particularly three-way, ANOVA have not been widely applied in this context. To obtain a comprehensive view of variations across years, seasons and locations, we need to combine all the above tools in a meaningful way. In this study, we have used a combination of factor analysis, clustering, two-way and three-way ANOVA and t-tests to reveal interesting annual spatio–seasonal variations and groupings of different parameters in order to identify potential sources of pollution and remedial measures for effective water management in the catchment and command area.

Multivariate statistical analyses of surface water quality datasets in different environmental settings have been proven to be effective for deriving meaningful associations, relationships between various quantities and environmental factors and in assessing spatio–temporal variations, as seen from some of the relevant studies discussed above. In our present study, FA is used comprehensively to derive relationships between parameters across seasons and to reconfirm the clustering of sampling locations, which are derived through AHC. Spatial variations are not only confirmed through AHC but also through three-way ANOVA, which includes, apart from spatial, seasonal and annual variations in parameter values. Most of the literature has focused on individual use of these tools, which provides a limited view of the relationships between surface water parameters. Additionally, most articles invariably focus on either one or the dual aspects of seasonal and locational variations. In this article, we combine annual, seasonal and spatial variations in a significant study area that involves reservoir catchment and command areas. The above methodology of FA–AHC–ANOVA and t-tests highlights some very interesting relationships amongst parameters and spatio–temporal variations across the study area. Additionally, the articles in the literature that focus on seasonal aspects do so in a monthly or a traditional seasonal approach [17,18]. However, we have collected samples and analyzed them based on monsoon seasons typical for this part of the world. This gives insight into the monsoon effects with respect to concentration levels and groupings between physico–chemical and biological parameters, which is vital for water management and water usage monitoring programs in this part of the world.

1.2. Geogenic Contamination Analysis of Groundwater

The quality and quantity of surface and groundwater is influenced by geogenic and anthropogenic factors. Pollution of groundwater primarily results from substances that naturally occur in groundwater and the mineral environment and/or by all types of point and diffuse sources of pollution [19]; therefore, groundwater requires regular monitoring. In a study related to global water pollution and human health [20], the authors discussed water pollution issues related to various pollutant source classifications such as agriculture, geogenic, biogenic, mining, hazardous waste, and urban wastewater in industrial, developing, and emerging countries. One of the important conclusions of this study was that geogenic contaminants act as diffuse sources of toxic elements at regional scales, inflicting chronic diseases on large populations of all continents. To solve this problem, the authors suggested using geochemical modeling of hydrogeochemical data and spatial analysis to improve the geogenic problem. Monitoring of the quality of groundwater is based on water chemistry analysis to identify the nature of pollutants. In one study [21], reaction path modelling was used to investigate the evolution of water chemistry in shallow to deep crystalline aquifers, with a special focus on fluoride. The water–rock interaction led to Ca-enrichment due to the dissolution of Ca phases originating from the Ca-HCO₃ water type. Magnesium was mainly contributed to the aqueous solution through biotite dissolution, whereas chlorite dissolution played a subordinate role throughout the whole rock-dissolution process. This study became a valuable tool for environmental applications to understand the geochemical processes occurring in aquifers of interest and to predict the fate of pollutants in different geological settings. In another study [22], geochemical modelling was used to understand the release and fate of pollutants in crystalline aquifers. It was concluded that the geochemical behavior of groundwater depends on several geogenic processes that cause hazardous enrichment of natural waters, even in remote areas far from anthropogenic sources. In this work, arsenic pollution was addressed by studying water–rock interactions and applying reaction-path modelling as a tool to understand the rock-to-water release of arsenic and the fate of this natural pollutant in crystalline aquifers. comparison of theoretical trends and experimental data showed that SO₄, Fe and As groundwater concentrations in which no arsenic anomalies were observed were explained by the dissolution of the considered granitic rock.

A multivariate, nonparametric approach has been successfully applied for estimating the probability of exceeding the local natural background level of arsenic in the aquifers of the Calabria region (southern Italy) [23]. In this study, the concept of the natural background level was used to distinguish between natural and anthropogenic contributions to concentrations of specific contaminants as a groundwater management and protection tool. It was concluded that, based on the triangular plots of major anions and major cations, 165 of the 337 groundwater samples from crystalline-metamorphic water groups could be attributed to the Ca-HCO₃ chemical type, followed by Na-HCO₃ (110 samples), Na-Cl (33 samples) and Mg-HCO₃ (23 samples) hydro–chemical facies. These compositions reflected the typical groundwater evolution from shallow to deep crystalline-metamorphic aquifers [23]. In a recent study [24], the hydrogeological features, main sources of geogenic hazardous substances, dominant hydrogeochemical processes and key factors controlling the occurrence of high arsenic (As ≥ 10 μg/L), high fluoride (F  ≥  1 mg/L) and high iodine (I  ≥  100 μg/L) in groundwater were discussed. Four basic genetic types of geogenic contaminated groundwaters (GCGs) were proposed by summarizing the characteristics of the distribution and major hydrogeochemical processes, namely leaching–enrichment, burial–dissolution, compaction–release and evaporation–concentration types. The complex genetic mechanisms of GCGs were integrated into a new theoretical framework to analyze their genesis and predict their spatial and temporal distribution.

In our present work, the following methodology was followed to analyze groundwater samples. The total study area was divided into six sub-watersheds for surface water sampling based on the drainage features of the area. The monitoring/sampling stations were located based on practical considerations. A total of 33 groundwater samples were collected for each of the three monsoon seasons over a period of two years. The samples were collected and subjected to various physical, chemical and biological characteristics as per the Bureau of Indian Standards [25] and as explained briefly in the Supplementary Material. Piper Trilinear diagrams were generated to interpret the groundwater chemistry, and the conclusions are noted in Section 3.4.

1.3. Significance of the Present Study

This work highlights the significance of multivariate analysis in analyzing surface water quality of reservoir catchment and command areas, with the focus on variations across sub-watershed locations, monsoons and years. This study is significant especially due to the type of region we analyzed, as it has a good variety of landcover, including forest, scrubland and irrigated land and a variety of land uses, such as domestic, agricultural and quarrying activities, as discussed in Section 2.3. The geological features, as given in Section 2.2, and the chemical analysis of groundwater, as given in Section 3.4, make this region interesting for analyzing the quality of both surface water and groundwater. This work brings out important relationships amongst water quality parameters and valuable information on spatio–seasonal variations for effective watershed management, thereby improving water quality for agricultural and drinking purposes in this significant region of southern India.

2. Materials and Methods

2.1. Study Area

The Arkavathi River, with a total flow length of about 210 kms, originates in the Nandidurga Hills (13°22′11″ N and 77°41′5″ E) of Karnataka state at an altitude of 1467 m amsl (above mean sea level) and flows through various parts of the region before draining into the Cauvery River, which is one of the most prominent rivers in southern India. The study area covered approximately 1572 km² with a total perimeter 245.72 kms and is located between 12°20′ to 12°54′ N and 77°15′ to 77°44′ E (Figure 1, Figure 2, Figure 3, Figure 4 and Figure 5). The flow system of the main Arkavathi River from the point of its entry into the study area at Manchanabele Reservoir (12°52′19″ N and 77°20′5″ E) till its exit with a waterfall at Chunchi village (12°21′5″ N and 77°26′ 46″ E) is geologically structurally controlled. The Arkavathi River has a flow length of 65 kms from Manchanabele (720 m amsl) up to the Chunchi waterfalls (540 m amsl) and has a total fall in gradient of 180 m, i.e., an average fall in gradient of 2.8 m per km of flow length. The flow path of the subsidiary drainages is also linear. A total area of 62.32 km² (cultivable command area), which is part of sub-watershed SW6, is irrigated with Arkavathi Reservoir water. The study area is demarcated into six sub-watersheds, which are named accordingly (

Table 1. Details of sub-watersheds.

Sub-Watershed	Local Name of Sub-Watershed	Area (km2)
SW1	Ramanagara sub-watershed	348.69
SW2	Suvarnamukhi sub-watershed	420.25
SW3	Mavathurkere sub-watershed	369.70
SW4	Kodihalli sub-watershed	175.06
SW5	Kanakapura sub-watershed	168.96
SW6	Harobele sub-watershed	89.39
	Total	1572

) based on the drainage features of the area. The geology map of the study area is given in Figure 2. Contour maps, a digital elevation model map (prepared using RS and GIS), a slope map and a drainage map are given as online Supplementary Materials.

2.2. Geology and Geomorphological Features

A large extent of the Arkavathi Reservoir catchment and part of its command area, particularly the western parts, with a geomorphic display of abundant hills, hill ranges, inselbergs, intermittent valleys, rock-cut valleys, etc., is a type of area generally referred to as ‘Closepet Granites’. The Geology map of the study area (Figure 2) and a soil erosion map (Figure 3) are given here for reference. The hills at the eastern peripheral parts of the area are the southern extension of the granite belt [26]. This long but narrow intrusive rock body is composed of diorites, granodiorites and grey granites, and forms a divide between the Ponnaiyar River catchment to the east and the Vrishabhavati stream system of the Arkavathi sub-basin to the west. Suvarnamukhi, a subsidiary drainage of the Vrishabhavati stream, is born in the western part of this granite belt. The gneissic rocks of the younger gneissic complex are drained mainly by the Vrishabhavati stream system in the northcentral part of the area, which is bound by the Closepet Granites. The western part of the area, parallel to the southern-flowing Vrishabhavati stream, forms a longitudinal contact between the massive Closepet Granite body to the west and the young Gneissic group of rocks to the east. The Closepet Granites are of both grey and pink varieties and are coarse-grained and porphyritic. Granites are poorly jointed and are weathered to an average depth of 15 m in the valley zone. The depth of weathering thins out as we move towards the pediment part of the terrain.

Granite to the west and southern extension of the granite belt forms a moderately undulating topography. A small extent of the Arkavathi Reservoir command area also forms part of the gneissic terrain. Most of the gneissic terrain forms a pediplain with considerably weathered rock. The depth of weathering ranges up to 30 m. The gneissic rocks are easterly dipping, and the dip angle of the mineral foliation ranges from between 65° and 80°. Dip joints are more prominent. These gneissic rocks are fractured at the contact with granites, and the presence of pegmatites and quartz veins in gneisses are common.

A large part, excluding the part constituting hills and the hill range, forms the pediment zone. The presence of pediplains becomes transitionally significant from third-order streams and onwards. The thickness of soil is dependent upon the type of landscape. Red gravelly to loamy soils 0.3 to 1.5 m thick are commonly seen in the pediment part of the terrain, whereas the thickness ranges up to 2.5 m in the pediplain. While the hill range in the granite belt to the east of the southerly flowing Arkavathi River is N–S oriented, the hills west of the river (i.e., right bank catchment part) are E–W to ESE–WSW oriented. The height of the hills in the granite terrain ranges between 680 and 1043 m above mean sea level (amsl). These hills are massive and steep, and many are monoliths, with most of them attaining a height of more than 950 m amsl. The entire study area is a highly tectonically disturbed zone. South of 13° latitude, the Closepet Granite is flanked on its east by a ‘broad gravity low’ with its center near the town of Kanakapura (latitude 12°32′36″ N and longitude 77°25′2″ E).

2.3. Land Use and Landcover of the Study Area

The study area is mainly composed of fertile land where kharif and rabi crops are grown (Figure 4). It also comprises scrub forest, land with scrub, agricultural plantations and an equal area of barren rocky sheetrock. A small portion of the area is occupied by moist, dry deciduous, dense and open forest. Though not significant, mining/industrial wasteland forms a minor part of the study area. Vegetation forms a major portion of the study area, along with a few villages and small towns. Sub-watershed areas SW1 and SW2 have townships along the Arkavathi River, where some cultivation takes places using the river water. This is true to some extent in SW5 as well. Human activity is increased in these three areas. SW3 and SW4 are predominantly forest and hilly areas with less human activity. However, SW4 has a significant number of quarries. SW6 is the command area and is rich in nutrients since it is mostly agricultural/irrigation land. A total area of 62.32 km² (cultivable command area), which is part of sub-watershed (SW6), is irrigated with Arkavathi Reservoir water. Further details regarding land use and landcover are given in

Table 2. Details of land use and landcover of the study area.

Description	Area (km2)	Percentage of Area
Agricultural Plantation	111.26	7.07
Barren Rocky/Stony Waste/Sheet-Rock Area	95.26	6.06
Degraded Forest	22.45	1.42
Fallow Land	12.53	0.79
Forest Plantations	4.72	0.3
Gully/Ravine Land	0.73	0.04
Industrial Area	6.46	0.41
Kharif + Rabi (Double Crop)	264.89	16.85
Kharif Crop	547.91	34.85
Land With Scrub	139.72	8.88
Land Without Scrub	0.82	0.05
Mining/Industrial Wasteland	11.03	0.7
Mixed Vegetation	10.72	0.68
Moist and Dry Deciduous Dense Forest	58.06	3.69
Moist and Dry Deciduous Open Forest	20.90	1.32
Rabi Crop	0.31	0.02
Scrub Forest	169.40	10.77
Tree Groves	11.46	0.72
Town/Cities	17.81	1.13
Village	25.87	1.64
River/Stream	11.54	0.73
Lakes/Tanks	28.06	1.78
Total area	1572	100

below.

2.4. Sample Collection and Testing

Water samples were collected from 30 monitoring points over a study period of two years covering three prominent seasons of Pre-Monsoon, Monsoon and Post-Monsoon (Figure 5). To analyze the surface water quality of the area, we identified the sub-watersheds using Survey of India (SOI) Topographic Maps at 1:50,000 scale—57 H5, 57 H6, 57 H7, 57 H9, 57 H10 and 57 H11. Surface water samples within these sub-watersheds were then collected for further analysis. Two-liter polyethylene bottles/cans were thoroughly cleaned and were used to collect samples from the sampling locations. To avoid impurities on the surface of streams, the samples were drawn away from turbulence due to the banks and few centimeters below the water surface. For microbiological analysis of water samples, glass containers were used that could withstood 160 °C sterilization and would not produce or release any harmful chemicals that would inhibit biological activity, induce mortality or encourage growth. Samples for bacteriological examination were collected in clean, sterilized, narrow-mouthed, neutral glass bottles of 250 mL, 500 mL or 1000 mL capacity. The bottles were sterilized in a hot air oven at 100 °C for one hour. The sampling bottles were not opened except at the time of sampling. The detailed sampling (BIS, Part 1: Sampling [25]) and testing procedures for each of the contaminants along with the full dataset are given in the online Supplementary Materials section.

The samples were subjected to various physical, chemical and bacteriological tests as per the procedure given in [25] (Bureau of Indian Standards: Public Safety Standards of the Republic of India: Chemical: Environmental Protection and Waste Management: Part 1 to 63), as explained briefly in the Supplementary Materials. For example, to test for dissolved oxygen (DO), sampling and sample preservation were done as prescribed in BIS [25], Part 38: Dissolved Oxygen. The min–max ranges of the concentration levels of these parameters are provided in

Table 3. Seasonal variation of surface water quality parameters (min–max values) in the study area.

Parameters	Min–Max Range for Year 1			Min–Max Range for Year 2
	Pre-Monsoon	Monsoon	Post-Monsoon	Pre-Monsoon	Monsoon	Post-Monsoon
pH	5.7–8.3	6.9–8.7	7.1–8.5	6.1–8.4	7.7–9.2	7.4–8.5
Temp (°C)	28- 29	27	24–24	27	27	24–26
DO (mg/L)	3.8–6.4	3.4–6.8	4–6.3	2.2–6.14	3–6.6	2.7–4.1
BOD5 (mg/L)	2–15	1.6–6	3.2–9.1	2.1–15	2.2–12.8	5.5–16
COD (mg/L)	8–36	4.8–10.5	5.9–16.9	6.8–40	3.7–22	10.2–30
Total Suspended Solids (TSS) (mg/L)	2–13.9	1.6–8.2	3.5–9.2	1.9–15	3–11.2	6–11.5
Turbidity (NTU)	0.5–1.4	0.4–2.2	0.5–1.5	0.4–2.3	0.4–1.3	1–1.7
Total Dissolved Solids (TDS) (mg/L)	99–841	0.81–791	301–799	101–860	99–850	498–879
Conductivity EC (µmhos/cm)	220–1241	125–1217	465–1230	148–1250	399–1327	766–1353
Sodium Na+ (mg/L)	25–122	28–110	35–126	28–159	51–120	71–143
Potassium K+ (mg/L)	2.1–22.8	3–21	3.5–25	3.9–27	3–25	4.7–26
Calcium Ca2+ (mg/L)	9–119	10–90	14–92	18–111	10–101	59–96
Magnesium Mg2+ (mg/L)	3.7–34	1.7–33.6	1.9–31	2–36	6–74	8–29
Total Hardness as CaCO3 (mg/L)	81–439	42–430	209–432	89–450	55–428	275–541
Chlorides Cl− (mg/L)	31–205	28–182	35–210	33–305	35–221	130–275
Bicarbonate ${\mathrm{HCO}}_3^{-}$ (mg/L)	90–275	80–281	95–276	92–305	90–285	161–281
Flouride ${\mathrm{F}}^{-}$ (mg/L)	0.01–0.35	0.002–0.008	0.03–0.2	0.012–0.45	0.01–0.25	0.06–0.3
Nitrate ${\mathrm{NO}}_3^{-}$ (mg/L)	1.0–3.9	0.8–3.2	1.4–15.7	1.1–8.1	0.9–8.6	2.7–18.2
Phosphate ${\left[{\mathrm{PO}}_4\right]}^{3-}$ (mg/L)	0.01–0.31	0.02–0.52	0.02–0.9	0.01–0.48	0.05–0.45	0.1–0.28
Sulphate ${\mathrm{SO}}_4^{2-}$ (mg/L)	8.5–49	6–28	13–44	6.8–67	11–26	12–46.7
Hexavalent Chromium Cr+6 (mg/L)	Nil	0–0.008	0.003–0.007	0.006–0.008	Nil–0.006	0.005–0.09
Iron Fe2+ (mg/L)	0.02–0.22	0.01–0.4	0.06–0.44	0.006–0.45	0.03–0.01	0.07–0.43
Copper Cu (mg/L)	Nil–0.001	0–0.005	0.003–0.006	0.002–0.004	Nil–0.004	0.003–0.008
Lead Pb (mg/L)	Nil	0–0	0.03–0.1	0.002–0.002	Nil–0.004	0.02–0.1
Nickel Ni (mg/L)	Nil–0.001	0–0.004	0.001–0.004	0.001–0.002	Nil–0.004	0.001–0.004
Zinc Zn2+ (mg/L)	0.001–0.1	0.01–0.09	0.007–0.1	0.003–0.12	0.02–0.14	0.008–0.19
Total Alkalinity as CaCO3 (mg/L)	87–315	80–310	225–445	98–354	90–295	228–476
Total Coliform/100 mL	59–301	14–156	96–190	45–298	18–214	97–199
Fecal Coliform/100 mL	11–81	0–12	9–45	9–84	2–33	12–56

for both years of the study.

2.5. Data Suitability

The KMO (Kaiser–Meyer–Olkin) (Kaiser, 1974) measure of sampling adequacy tests how well-suited the data are to perform FA. It measures the proportion of variance among parameters that might be common variances. KMO [27] returns a value between 0 and 1. If the value is closer to 1, it means that the data are better qualified for FA. In most of our analyses, we got a KMO value in the range of 0.6–0.7. Bartlett’s test of sphericity compares the correlation matrix to an identity matrix, i.e., it checks for redundancy between parameters. In this case, Bartlett’s test was very highly significant (<0.001), and, therefore, FA was appropriate. FA, clustering, ANOVA and t-tests were performed using the IBM SPSS statistical software [28] and XLSTAT Excel add-on data analysis package [29].

2.6. Multivariate Analysis

2.6.1. Factor Analysis

Principal factor analysis (PFA) is a dimension-reduction technique and was used as an extraction method [30,31,32] to reduce the large number of parameters while retaining most of the information contained in the original dataset. Factor analysis (FA) is a model-based statistical technique to bring out relationships between measured parameters and latent factors, which are unobserved parameters that are believed to be the cause of the measured values of observed parameters. The assumption is that correlations between observed parameters are due to the causal influence of one or more factors (latent parameters). Each principal component is identified with its eigenvalue and eigenvector. The eigenvalue for each component (or factor, in the case of FA) measures the variance in all the parameters that are accounted for by that component. Factors can be better interpreted with rotation, a process which increases the loading of different parameters on the extracted factors while reducing their loading on the rejected factors. There are different methods of rotation, which can mainly be divided into two categories. Orthogonal rotations such as varimax are useful if factors are expected to be unrelated, while oblique rotations such as oblimin are more applicable if some correlation between the extracted factors is expected. Further details on the outcome of FA are given in the results (Section 3.1 below).

2.6.2. Cluster Analysis

Cluster analysis has been used with good success in various studies of water quality around the world. It is usually used in conjunction with other multivariate analyses as a spatial clustering tool [33] or for assessing seasonal and spatial variability of water quality [34]. In this study, the distance between datapoints was calculated by the Euclidean distance in m-dimensional space, and agglomerative hierarchical clustering (AHC) was used for clustering. An agglomeration criterion such as unweighted pair–group average was used to minimize and form clusters. Here, we clustered the dataset based on observations, i.e., the points at which samples were collected; this was done across all the parameters.

2.6.3. Two-Way and Three-Way ANOVA and t-Tests

Analysis of variance (ANOVA) is a statistical tool that tests the hypothesis of whether there are any differences in the means of a group of datasets or not [35,36]. The null hypothesis in ANOVA is that the means of subjects are the same in different groups, and the alternative hypothesis is that there is a statistically significant difference in the means—either between one group compared to all others or between multiple groups within the study. On the other hand, for comparing whether the means of only two groups are related or not, the t-test is used. Since sampling was conducted across three seasons over two years, a total of six datasets for each parameter were expected for analysis. Attention was drawn to some of the prominent parameters discovered in FA to avoid repetition of analysis of similar parameters or similar groups of parameters. The repeated measures considered in two-way ANOVA were the two factors of “season” and “year” together, while three-way ANOVA was used to consider the additional aspect of “sampling location”. The repeated measures method was used because the data from the same set of samples were used but across different seasons and different years.

3. Results and Discussion

3.1. Factor Groupings

The total variance explained for the dataset of the Pre-Monsoon season of the first year in terms of variability (%) per factor is 41.70, 13.50, 9.76, 6.39 and 3.25 for the first five factors, whose eigenvalues were >1. The cumulative variance for these first 5 factors is 74.6%, and for the first 12 factors it is 82.34%, as shown in the scree plot in Figure 6. We used varimax rotation to increase the loading of different parameters on the extracted factors. We then found that five new rotated factors (D1 through D5) accounted for a cumulative variation of approximately 75%, as shown in

Table 4. Percentage of variation after varimax rotation.

	D1	D2	D3	D4	D5
Variability (%)	23.901	21.003	13.103	8.453	8.139
Cumulative %	23.901	44.904	58.007	66.460	74.599

.

Table 5. Factor pattern or the rotated component matrix for the Year 1 Pre-Monsoon dataset.

Parameter	Rotated Factor
	D1	D2	D3	D4	D5
pH	−0.023	−0.564	0.185	−0.054	−0.123
DO	−0.166	−0.402	−0.334	−0.450	0.081
BOD5	0.160	0.915	0.199	0.210	0.097
COD	0.326	0.813	0.149	−0.022	0.065
TSS	0.082	−0.155	0.794	−0.092	−0.193
Turbidity	−0.080	0.257	0.286	0.782	0.228
TDS	0.800	0.205	0.320	0.017	0.261
Conductivity	0.833	0.092	0.302	0.157	0.157
Na+	0.558	0.393	0.049	0.105	0.717
K+	0.389	0.174	0.045	−0.191	0.387
Ca2+	0.737	0.239	0.325	−0.117	−0.139
Mg2+	0.591	0.418	0.338	−0.022	0.273
Total hardness as CaCO3	0.693	0.387	0.440	0.065	0.048
${\mathrm{Cl}}^{-}$	0.582	0.364	0.379	0.018	0.427
${\mathrm{HCO}}_3^{-}$	0.895	0.230	0.039	−0.166	0.213
${\mathrm{F}}^{-}$	0.264	0.917	−0.018	0.020	0.171
${\mathrm{NO}}_3^{-}$	0.355	0.152	0.716	0.352	0.038
${\left[{\mathrm{PO}}_4\right]}^{3-}$	−0.541	0.033	0.262	0.019	0.042
${\mathrm{SO}}_4^{2-}$	0.355	0.867	0.138	−0.049	0.127
Fe2+	−0.219	−0.055	−0.131	0.869	0.048
Zn2+	0.210	0.207	0.735	0.053	0.014
Total alkalinity as CaCO3	0.877	0.113	0.072	−0.136	0.031
Total coli form/100 mL	−0.056	0.671	0.403	0.221	0.255
Fecal coliform/100 mL	0.098	0.472	0.460	0.330	0.365
SAR	0.037	0.182	−0.247	0.251	0.761

Note: * Values in bold for each parameter correspond to the factor for which the squared cosine is the largest [30].

gives the component matrix or factor loadings for each of the five rotated factors for the Pre-Monsoon season. The corresponding matrices for the monsoon and Post-Monsoon seasons for Year 1 are given in

Table A1. Factor pattern or the rotated component matrix for Monsoon and Post-Monsoon seasons of Year 1.

	Monsoon					Post-Monsoon
	D1	D2	D3	D4	D5	D1	D2	D3	D4	D5
pH	−0.015	−0.379	0.225	−0.031	−0.134	−0.011	−0.212	−0.030	−0.089	−0.509
DO	0.008	−0.523	−0.428	0.157	−0.383	−0.267	−0.103	0.747	−0.003	0.017
BOD5	0.190	0.325	0.269	0.164	0.739	0.477	0.073	−0.012	0.169	0.811
COD	0.057	0.052	−0.410	−0.003	0.694	0.509	0.063	−0.006	0.182	0.783
TSS	0.046	−0.621	−0.044	0.032	0.105	0.695	0.333	−0.379	0.105	0.291
Turbidity	−0.039	−0.117	0.965	0.084	0.161	0.860	0.258	−0.043	0.186	−0.070
TDS	0.889	−0.047	0.059	−0.076	0.213	0.272	0.296	0.029	0.888	0.201
Conductivity	0.895	−0.057	0.065	−0.080	0.209	0.274	0.297	0.022	0.888	0.198
Na+	0.518	0.726	−0.001	−0.142	0.283	0.436	0.744	0.297	0.125	0.307
K+	0.640	0.164	−0.219	−0.054	−0.080	−0.088	0.523	−0.144	0.411	0.349
Ca2+	0.863	−0.186	−0.003	−0.199	0.205	0.107	0.699	−0.269	0.126	−0.217
Mg2+	0.697	0.158	−0.162	0.264	−0.331	0.242	0.769	0.028	0.176	0.144
Total hardness as CaCO3	0.935	−0.067	−0.006	0.011	−0.085	0.762	0.386	−0.074	0.287	0.219
${\mathrm{Cl}}^{-}$	0.796	0.446	−0.093	0.076	0.237	0.322	0.849	0.042	0.092	0.055
${\mathrm{HCO}}_3^{-}$	0.775	0.239	−0.270	−0.395	0.051	0.035	0.896	−0.045	0.190	0.166
${\mathrm{F}}^{-}$	0.203	0.479	−0.249	−0.413	0.318	0.765	0.279	−0.150	0.082	0.243
${\mathrm{NO}}_3^{-}$	0.523	−0.347	0.135	0.039	0.286	0.601	0.156	0.150	0.158	0.673
${\left[{\mathrm{PO}}_4\right]}^{3-}$	−0.275	−0.362	0.350	0.174	0.022	0.715	0.027	0.053	0.011	0.325
${\mathrm{SO}}_4^{2-}$	0.668	−0.364	0.071	0.045	−0.308	0.760	0.443	−0.045	−0.031	0.109
Fe2+	−0.172	0.026	0.793	0.090	−0.089	0.690	−0.154	0.186	0.096	0.618
Zn2+	0.120	0.097	0.551	0.340	−0.157	0.794	0.277	−0.329	0.109	0.310
Total alkalinity as CaCO3	0.734	0.082	−0.234	−0.430	0.035	0.769	0.340	0.023	0.321	0.095
Total coliform/100 mL	−0.113	−0.180	0.172	0.872	−0.014	0.742	−0.133	0.033	0.172	0.378
Fecal coliform/100 mL	−0.094	0.039	0.082	0.952	0.144	0.713	−0.062	0.268	0.185	0.575
SAR	−0.282	0.775	0.110	0.005	0.103	0.352	0.168	0.544	0.058	0.399

Note: Values in bold for each parameter correspond to the factor for which the squared cosine is the largest [30].

in Appendix A. The values in bold in Table 5, when read down through each column, indicate the groupings of the parameters that influenced the corresponding factor to a similar extent. During the Pre-Monsoon season, Factor 1 (D1) was influenced by parameter groupings TDS–EC, Cations K⁺, Ca²⁺, Mg²⁺, Total Hardness, ${\mathrm{Cl}}^{-}$, ${\mathrm{HCO}}_3^{-}$ and Total Alkalinity. These parameters loaded strongly on D1 and hence are related to each other and can be grouped together. This grouping was largely carried forward during the monsoon season, but during the Post-Monsoon period, TDS and EC formed a separate grouping on D4, while Na⁺, K⁺, Ca²⁺, Mg²⁺, ${\mathrm{Cl}}^{-}$ and ${\mathrm{HCO}}_3^{-}$ formed a distinct group based on their loading on D2. For both years, TDS and EC consistently loaded together on the same factor across all seasons. The more salts dissolved in the water, the higher the value of EC. Hence, TDS and EC grouped together. Similar is the case with BOD₅–COD and Total Coliform–Fecal Coliform and the group of Cations (Na⁺, K⁺, Ca²⁺, Mg²⁺), Total Hardness, ${\mathrm{HCO}}_3^{-}$ and Total Alkalinity. The granitic material of the study area contains mineral feldspar, which is composed of sodium, potassium and calcium, hence the grouping of these ions.

BOD₅ (for 5 days) and COD were consistently grouped together across all the seasons. This is because BOD is the oxygen demand of organic matter in water, whereas COD is the oxygen demand of organic and inorganic matter. We also noticed that pH and DO loaded negatively with BOD₅ and COD during Pre-Monsoon, and this negative linkage incrementally changed from the Pre- to Post-Monsoon seasons. High COD concentrations indicate organic and inorganic pollution [37]. Increased nutrient concentration leads to increased organic matter concentration, and the consequent respiration and degradation of organic matter reduces DO concentration [38]. There was another interesting grouping of BOD₅ and COD with Total and Fecal Coliforms in Pre-Monsoon, which reduced significantly during the Monsoon season and again became slightly more positively linked during the Post-Monsoon season. This was due to the population density of Fecal Coliform in the river being directly proportional to the degree of sewage pollution, which was reflected by BOD₅ levels [39]. Though heavy metals were largely nonexistent due to a smaller number of industrial sites, Fe²⁺ and Zn²⁺ formed a group and loaded together during the Monsoon and Post-Monsoon seasons. Turbidity and Fe²⁺ loaded together on the same factor consistently across seasons since iron content in the form of colloidal particles contributes to turbidity [40]. FA not only allows us to track the evolution of these groupings/relationships through the season, but, as shown in Figure 7, it also validates the clustering of datapoints based on the sampling locations across the study area. It confirms the clusters formulated through AHC (Section 3.2, Figure 8) by plotting the observation points onto the rotated factor axes D1, D2 as shown in Figure 7.

3.2. Clustering

While samples from sub-watersheds SW1 and SW6 formed clear clusters, the clustering is varied in other sub-watersheds. Sampling locations W9, W10, W14, W17, W18 and W19 formed a distinct cluster during Pre-Monsoon and Monsoon seasons since these sampling points are located in hilly areas in sub-watersheds SW3, SW4, SW5 and partially in SW2. Dendrogram representations revealed, as shown in Figure 8 for Pre-Monsoon season of Year 1 and Figure A1 and Figure A2 in Appendix A for the Monsoon and Post-Monsoon seasons, respectively, that the locations W1, W2, W3, W4, W5, W7 and W20 formed a subcluster not only during the Pre-Monsoon season of Year 1, but also during the Post-Monsoon season due to these samples being upstream from the reservoir and around the main Arkavathi River. W21, W22 and W23 formed smaller subclusters over all the seasons. W26, W27, W28, W29 and W30 formed a subcluster due to their location in the command area downstream from the reservoir (sub-watershed SW6), with irrigation activities being similar in these locations. Another interesting subcluster formation was of W6, W11 and, W12 and W8, W13, W15 and W16 (due to these locations being covered with scrub) across all three seasons. This is confirmed further by the distance to the cluster centroids, where points W6, W11 and W12 have similar distances to the centroids, while W8, W13, W15 and W16 have different distances to the same cluster centroid. For Year 2 observations, the clusters largely remain the same. For example, W9, W10, W14, W17, W18 and W19 form a distinct cluster during the Pre-Monsoon and Monsoon seasons. However, W9 and W10 form a separate cluster during the Post-Monsoon season. The other subclusters also largely follow patterns similar to those of Year 1.

3.3. Two-Way and Three-Way ANOVA: Effects due to Years, Seasons and Locations

For comparing differences between years, two-way ANOVA [35,36] was applied to compare the means between seasons and across the two years, as shown in Figure 9 below. The mean concentration levels across seasons and years and the corresponding F-statistics are given in

Table A2. Mean concentration levels of a few parameters across seasons and years, and the corresponding F-statistic.

	Pre-Monsoon	Monsoon	Post-Monsoon	F-Statistic for Seasons	Year 1	Year 2	F-Statistic for Years	F-Statistic for Interaction between Seasons and Years
Parameter	Mean concentration level (mg/L)	Mean concentration level (mg/L)	Mean concentration level (mg/L)	(F(2,58) at p < 0.001)	Mean concentration level (mg/L)	Mean concentration level (mg/L)	(F(1,29) at p < 0.001)	(F(2,58) at p < 0.001)
TDS	571.27	523.72	666.15	14.58	563.48	610.61	29.22	20.59
TSS	5.065	4.230	7.72	36.033	5.11	6.23	101.73	24.56
${\mathrm{NO}}_3^{-}$	2.804	1.987	9.58	52.8	4.28	5.31	79.25	22.61
Na+	90.22	80.82	96.83	22.22	83.53	95.04	37.37	8.676 (p = 0.001)
Total Hardness	242.017	201.688	347.83	47.04	249.29	278.4	53.02	31.02
Mg2+	15.397	11.74	15.74	9.2 (p = 0.01)	13.32	15.26	27.01	7.437
${\mathrm{F}}^{-}$	0.112	0.038	0.105	12.21 (p = 0.01)	0.073	0.097	22.6	13.97
Ca2+	61.73	62.03	69.68	4.93 (p = 0.01)	59.05	69.91	76.73	22.35
${\mathrm{HCO}}_3^{-}$	199.88	190.78	209.43	4.91 (p = 0.01)	193.156	206.91	28.34	9.48
K+	11.14	10.44	14.17	24.55 (p = 0.01)	10.52	13.31	47.17	18.16

in Appendix A. For DO, the F-statistic was F(1,29) = 69.16 for the variation between years (considered over all seasons), which was very highly significant at p < 0.001, while the seasonal variation (considered across both years) had F(2,58) = 56.51, which was also very highly significant with p < 0.001. This showed that when the effects of both seasonal and yearly variation were considered, the DO levels were significantly different across seasons and years. Similarly, the mean BOD₅ levels were different across seasons (5.069 mg/L, 3.275 mg/L and 8.467 mg/L) with F(1,29) = 139.997 and p < 0.001 and across the two years (4.474 mg/L and 6.733 mg/L) with F(2,58) = 55.608 and p < 0.001. Similarly, mean levels of other parameters such as TDS, TSS, ${\mathrm{NO}}_3^{-}$, Na⁺ and Total Hardness were statistically different across seasons and years. We found that most of the physical and chemical parameters were diluted and displayed lower concentrations during the Monsoon season versus during Pre- or Post-Monsoon seasons. Additionally, in cases such as that of Mg²⁺ and ${\mathrm{F}}^{-}$, results showed that, while Monsoon levels were substantially different to that of Pre- and Post-Monsoon levels, there was no difference in concentration levels of ${\mathrm{F}}^{-}$ between Pre- and Post-Monsoon levels in the case of both years. The mean Total Coliform levels were different across seasons (148.517 MPN/100 mL, 95.45 MPN/100 mL and 152.65 MPN/100 mL) with F(2,58) = 21.97 and p = 0.01, and across the two years (128.04 MPN/100 mL and 136.37 MPN/100 mL) with F(1,29) = 38.44 and p < 0.001; also, the effect of interaction between seasons and years was strong with regard to the mean levels of Total Coliform, with F(2,58) = 4.25 and p = 0.019. In the case of Total Coliform levels, however, paired t-tests between pairs of seasons for both years showed that, while Monsoon levels were substantially different from those of Pre- and Post-Monsoon levels, there were no difference in concentration levels of Total Coliform between Pre- and Post-Monsoon season in the case of both years.

On applying three-way ANOVA (to consider the effect of sampling locations), we observed that there were some differences in mean levels of most of the parameters across different sub-watersheds. BOD₅, Na⁺, ${\mathrm{NO}}_3^{-}$, Fe²⁺, ${\mathrm{HCO}}_3^{-}$, K⁺, ${\mathrm{F}}^{-}$, SAR and COD all showed significant (p < 0.05) differences between concentration levels across the six sub-watershed locations (Figure 10,

Table A3. Tests of between-subject effects of surface water when location is considered.

Source	Measure	Type III Sum of Squares	df	Mean Square	F	Sig.	Partial Eta Squared
Location	DO	13.500	5	2.700	1.348	0.279	0.219
	BOD5	465.845	5	93.169	7.239	0.000	0.601
	TDS	806,893.156	5	161,378.631	1.105	0.384	0.187
	TSS	92.440	5	18.488	1.680	0.178	0.259
	Na+	54,851.996	5	10,970.399	6.212	0.001	0.564
	Total Hardness	229,868.853	5	45,973.771	1.436	0.247	0.230
	${\mathrm{NO}}_3^{-}$	558.440	5	111.688	9.626	0.000	0.667
	Fe2+	0.796	5	0.159	9.461	0.000	0.663
	${\mathrm{HCO}}_3^{-}$	196,681.589	5	39,336.318	3.328	0.020	0.409
	K+	2739.124	5	547.825	6.748	0.000	0.584
	Ca2+	17,471.293	5	3494.259	1.581	0.203	0.248
	${\mathrm{F}}^{-}$	0.397	5	0.079	5.260	0.002	0.523
	Mg2+	2795.772	5	559.154	2.026	0.111	0.297
	Total Coliform	95,860.353	5	19,172.071	2.136	0.096	0.308
	SAR	38.724	5	7.745	6.947	0.000	0.591
	COD	1677.707	5	335.541	7.483	0.000	0.609
Error	DO	48.084	24	2.003
	BOD5	308.906	24	12.871
	TDS	3,505,192.488	24	146,049.687
	TSS	264.095	24	11.004
	Na+	42,387.315	24	1766.138
	Total Hardness	768231.299	24	32,009.637
	${\mathrm{NO}}_3^{-}$	278.471	24	11.603
	Fe2+	0.404	24	0.017
	${\mathrm{HCO}}_3^{-}$	283,709.877	24	11,821.245
	K+	1948.345	24	81.181
	Ca2+	53,035.943	24	2209.831
	${\mathrm{F}}^{-}$	0.362	24	0.015
	Mg2+	6623.993	24	276.000
	Total Coliform	215,430.875	24	8976.286
	SAR	26.758	24	1.115
	COD	1076.217	24	44.842

in Appendix A). Further, the parameters BOD₅, Total Hardness, ${\mathrm{NO}}_3^{-}$, K⁺, Ca²⁺, F⁻, SAR and COD showed statistically strong differences in mean values when interactions between year, location and seasons were considered together. Since organic-pollution-loading in water bodies increases from upstream to downstream from the reservoir, BOD₅ and COD of water samples increased, while DO decreased as we moved along from SW1 to SW6 due to addition of sewage and other domestic activities along the river course. Since nitrate is the representation of oxidized organic matter in water and is one of the main components of fertilizers used in irrigation, and its presence increases BOD₅, nitrates varied similar to BOD₅ with respect to location, especially during the Post-Monsoon season. Due to run-off, TDS was especially high in SW6, indicating potential soil erosion in upper sub-watersheds. This is supported by the soil erosion map (see online Supplementary Materials), which indicates high soil erosion in SW3, SW4 and SW5. The presence of iron in natural waters can be attributed to the dissolution of rock and minerals containing biotite as part of the weathering process. We observed that iron content was higher in the forest and hilly areas SW3, SW4 and SW5. Similarly, Mg was high in SW4 and SW6.

3.4. Groundwater Quality of the Study Area—Piper Trilinear Diagrams

The objective of groundwater analysis is to understand the groundwater chemistry, the impact of land use and landcover changes on quality of groundwater, and to suggest suitable measures to improve groundwater quality. This is important as local populations depend on groundwater to supplement surface water. A total of 33 groundwater samples were collected for each of the three monsoon seasons over a period of two years. Based on the results of the analysis, Piper Trilinear diagrams were generated to interpret the groundwater chemistry and the conclusions were noted. Graphical interpretation of groundwater quality of the study area is necessary since it contains minerals that are to be assessed for the irrigation suitability. A Piper Trilinear diagram is useful for bringing out the chemical relationships between groundwater samples in more definite terms than available with other plotting methods [41]. This plot helps in understanding and identifying the water composition in different classes. The triangular plots of major anions and major cations have been successfully used in the interpretation of results of groundwater from aquifers of the Calabria region (southern Italy) [23]. Piper Trilinear diagrams generated for groundwater for the three seasons and over the two years of study are presented in Figure 11a–f. They reveal the analogies, dissimilarities and different types of water in the study area as identified and listed in

Table 6. Characterization of groundwater of the study area based on Piper Trilinear diagrams.

Subdivision of the Diamond	Characteristics of Corresponding Subdivision of Diamond-Shaped Fields	Percentage of Samples in Each Category
		Pre- Monsoon (Year 1)	Monsoon (Year 1)	Post-Monsoon (Year 1)	Pre- Monsoon (Year 2)	Monsoon (Year 2)	Post-Monsoon (Year 2)
1	Alkaline earth (Ca+ Mg)	100	93.93	100	100	96.96	100
2	Alkalies exceed alkaline earths	0	6.01	0	0	3.33	0
3	Weak acids (CO3 + HCO3) exceed strong acids (SO4 + Cl)	78.78	81.81	100	81.81	75.75	100
4	Strong acids exceed weak acids	21.21	18.18	0	18.18	24.24	0
5	Magnesium bicarbonate type	78.78	75.75	100	81.81	72.72	100
6	Calcium-chloride type	0	0	0	0	0	0
7	Sodium-chloride type	0	0	0	0	0	0
8	Sodium-bicarbonate type	0	0	0	0	0	0
9	Mixed type (No cation–anion exceeds 50%)	21.21	24.24	0	18.18	27.27	0

. It can be concluded from the below results that during the Pre-Monsoon and Monsoon seasons, most of the samples consist of calcium and magnesium bicarbonate waters, which is typically shallow, fresh groundwater, and a few samples are of calcium and magnesium sulfate waters, which is typically gypsum groundwater. During the Post-Monsoon season of both the years, all samples are of calcium and magnesium bicarbonate waters, which is typically shallow, fresh groundwater. This indicates that Ca and Mg are dominant in the groundwater of the study area during the study period.

4. Conclusions

In this study, we carried out a comprehensive multivariate statistical analysis of the concentration levels of physical, chemical and biological parameters in surface water samples collected from six sub-watersheds of the Arkavathi Reservoir catchment and command area over a period of two years (2013 and 2014) and three seasons of monsoon (Pre-Monsoon, Monsoon and Post-Monsoon). We also conducted chemical analysis of groundwater of the study area and classified the groundwater based on Piper Trilinear diagrams. The present study demonstrates the applicability of FA–AHC–ANOVA as a significant analysis tool for effective watershed management to improve water quality in rivers and reservoir catchment and command areas. This work forms a basis for further exploration of variations of such surface water contaminants in different terrains and watersheds worldwide. A few specific, significant recommendations follow:

• Considering the many townships along the Arkavathi River in SW1, domestic sewage needs to be treated effectively at the border region of SW1 and SW5 and within SW6, especially during the Post-Monsoon season.
• Usage of fertilizers, especially in the agricultural lands in command area SW6, should be closely monitored and controlled.
• Erosion control plans need to be put in place in SW3, SW4 and SW5, as indicated by high TDS in SW6.
• Quarry activities in SW4 and SW3 need to be monitored for potential contamination of smaller streams.