Integrated GIS and multivariate statistical approach for spatial and temporal variability analysis for lake water quality index

Abstract

It is critical to monitor water quality to keep water bodies ecologically healthy and facilitate the sustainable development of Kenyir Lake. Water quality differs temporally and spatially and is affected by several factors. Typically, water quality inspection systems are cost- and labour-intensive depending on water quality indicator count and sampling frequency. Optimising the frequency and location of water quality sampling is crucial. This study focused on collecting water samples from 22 locations in Kenyir Lake during different seasons (normal, dry, and wet). The study aimed to assess the spatial and temporal variations in the water quality of Kenyir Lake based on multivariate statistical methods. In this study, the following water quality parameters were selected for analysis: temperature, dissolved oxygen (DO), pH, biochemical oxygen demand (BOD), chemical oxygen demand (COD), total suspended solids (TSS), and ammoniacal nitrogen (NH₃-N). In addition, a water quality index was also calculated. GIS software was used to assess water quality data, and various multivariate statistical methods like cluster analysis (CA), discriminant analysis (DA), and principal component analysis (PCA) were employed. The outcome shows minor spatial differences concerning Kenyir Lake; however, the temporal variations were noteworthy during this study duration. Cluster analysis divided the locations into 3 clusters with TSS being key parameter affecting the spatial differences in water quality. Stepwise discriminant analysis based on three parameters, pH, temperature, and TSS, produced the associated classification matrix that correctly estimated 69.7% of the input. NH₃-N and TSS were found to be the two critical aspects that affect water quality during dry, wet, or normal climatic conditions.

Keywords:

• water quality
• Kenyir Lake
• cluster analysis
• discriminant analysis
• principal component analysis

1. Introduction

Lakes and rivers are crucial for human well-being as they offer numerous benefits, including irrigation, potable water network, transportation, hydropower production, leisure, and fishing (Jiang et al., 2023; Varadharajan et al., 2022). However, as human activities, including urbanization and industrialization, in the surrounding catchment areas of these water bodies increase, they can become contaminated by a range of pollutants (H. Zhang et al., 2022), which negatively impact the lake ecosystem in various ways. Lakes often experience a decline in water quality due to point-source runoff inputs, including domestic and industrial wastewater (Wang et al., 2016), and non-point-source runoff originating from agricultural and urban activity (Ongley et al., 2010). Moreover, long-range atmospheric movement of pollutants also affects water quality. Excessive human activities can have a range of negative impacts on lakes, including eutrophication (Hobæk et al., 2012) caused by high levels of nutrients and organic matter, siltation due to poor erosion control in agricultural, construction, logging, and mining activities, the introduction of exotic species, acidification from atmospheric sources, and contamination by toxic and organic compounds.

Water quality monitoring is crucial to handle freshwater resources. Inspecting a water body concerning water quality allows water management authorities to (i) determine if water fulfils the water quality recommendations; (ii) to ascertain contaminants and pollution sources; (iii) to observe water quality patterns, and (iv) evaluate for impairment. Typically, water quality inspection approaches comprise selecting sampling stations, identifying necessary water quality indicators, ascertaining water sampling frequency, and accounting for labour and material costs. Water quality indicators typically comprise physical, chemical, and biological properties (Alobaidy et al., 2010). Monitoring processes also encompass subsequent phenomena like laboratory assessment, data processing, and analysis to gather the required insights.

Water quality can be studied spatially and temporally. Pollutant levels differ based on the time scale, such as daily, seasonal and interannual variations (Arheimer & Lidén, 2000; Larned et al., 2004; Saraceno et al., 2009). Spatial water quality differences in a water body are typically associated with source pollutants, land use and catchment characteristics. It is critical to comprehend the spatiotemporal changes in water quality to be able to devise suitable strategies to enhance and manage stream and lake water quality.

Kenyir Lake, current case study, in Terengganu, Malaysia is famous for ecotourism. Increasing annual tourism in Kenyir has further impacted the water quality in Kenyir Lake. Hence, inspecting the lake’s water quality is critical to protect the ecosystem and ensure the sustainability of Kenyir Lake. The temporal and spatial water quality differences are typically needed for inspecting and determining the physical, chemical, and biological properties of receiving water. It creates a sophisticated data matrix having numerous physicochemical parameters (physical, nutrient, inorganic, and biological characteristics) that might be challenging to assess and comprehend owing to the latent associations between variables and monitoring locations (Koklu et al., 2010; Q. Zhang et al., 2009; Zhou et al., 2021). Hence, it is critical to derive useful information from such data sets without skipping vital aspects (Q. Zhang et al., 2009; Zhou et al., 2021). To build a suitable framework to determine lake water quality, this research aimed to evaluate water quality in Kenyir Lake using multivariate statistical methods to determine spatiotemporal patterns, ascertain critical parameters, and categorise pollution sources and relative quantitative contributions to water quality problems.

2. Sources of pollution

2.1. Point source

Point source pollution is referred to as pollution from a known point of discharge or fixed outlet and can be released into water bodies in pipes or man-made drainage (Gyawali et al., 2013). Some examples of point source pollution are pipe discharges, industrial outflows, tributaries, industrial or municipal wastewater treatment plant outflows. Since possible contaminants from a point source can be easily monitored by measuring discharge and pollutant levels from an identified discharge point, its impact is easy to define and regulate. The focus over the previous years of research work was to address point source pollution through managing the known point of discharges, such as urban wastewater effluent, as described by 2010) and Perona et al. (1999) which was found to have been successfully under pollution control and management.

2.2. Non-point source

Non-point source (NPS) pollution is referred to as pollution from unknown sources or its origin cannot be guaranteed from a singular source, rather a combination of sources of different natures, which are often difficult to identify at a fixed locality. Generally, NPS pollution originates from stormwater runoffs from agricultural and anthropogenic activities at urban catchment surfaces. For example, the pollutants from a nearby agricultural land use can be entered into the river by rain-runoff or infiltrate through the soil layers into the groundwater aquifers. Pollutants include but not limited to excess fertilizers from agricultural lands, toxic chemicals from urban runoff, sediment from improperly managed construction sites, and nutrients from livestock, pet wastes, and faulty septic systems. These pollutants will deteriorate the water quality of receiving waters if no control measures are taken.

Non-point source pollution is a major problem for surface waters because it is often challenging to identify the source of the pollution as its origin can be diffused from different unknown sources. Land use surveys and groundwater or surface water quality samples are the only ways of identifying possible locations of nonpoint sources. Therefore, NPS pollution has become the major cause of water quality pollution (Gyawali et al., 2013; Li et al., 2009; Zampella et al., 2007). Recent studies focus on determining the relationship between land use with water quality at different scales of operation and management.

3. Spatial and temporal differences in water quality

Spatial and temporal differences concerning water quality are primarily due to three key processes (1) source: pollution source within the catchment area, (2) mobilisation: the separation of these pollutants from their source due to erosion, weathering, or biogeochemical processes, and (3) deliver: the movement of separated constituents from the catchment region to receiving areas (Granger et al., 2010).

Typically, the aspects that influence spatial differences in water quality include human activity in catchment areas: vegetation characteristics, land use and handling, and natural aspects like climate, geology, soil type, topography, and hydrology (Onderka et al., 2012). In contrast, temporal water quality differences may be affected by aspects like streamflow (Mellander et al., 2015; Sharpley et al., 2002) and rainfall and air temperature (Lecce et al., 2006; Robson, 2014). Atmospheric aspects may impact the mobilisation and movement of constituents from the catchment to receiving areas. Water temperature may regulate biogeochemical processes concerning nutrients in streams (Roberts & Mulholland, 2007). Moreover, the levels of substances in the catchment (source) are affected by antecedent dry periods (Arheimer & Lidén, 2000; Lecce et al., 2006), vegetation cover (Kaushal et al., 2014; Ouyang et al., 2010), and seasonal differences concerning anthropogenic activities (Stutter et al., 2008).

4. Methodology

4.1. Study area

This study was performed at Kenyir Lake, about 40 km inland from Kuala Terengganu, Terengganu, Malaysia. Its coordinates are 5º 1‘20“North and 102º 54’ 30” East. Kelantan and Pahang are situated west and south of Kenyir Lake. The lake was formed due to hydroelectric dam built for power and flood control in 1986. Kenyir Lake has total surface and catchment areas of 369 km² and 2,612 km², respectively. At the Full Supply Level (FSL), the lake stores a gross volume of 13.6 billion m³. The average and maximum depths are 37 and 145 metres, respectively. The reservoir is fed by six primary tributaries, namely Sg. Terengganu (primary river), Sg. Terengan, Sg. Cacing, Sg. Petang, Sg. Tembat and Sg. Petuang. The lake supports rich biodiversity and thousands of species with over 8000 flowers, 2500 plants, 8000 orchids, 300 fungi, 370 birds, and 300 freshwater fish species (Izharuddin Shah Kamaruddin et al., 2011; Rouf et al., 2009). Kenyir Lake in Terengganu is renowned as an ecotourism hotspot, with numerous resorts and tourist activities located in its catchment area. However, the growing number of visitors to Kenyir each year has contributed to the decline in the lake’s water quality. To preserve its ecosystem and ensure its long-term sustainability, it is essential to regularly monitor the water quality of Kenyir Lake.

4.2. Water sample and field data collection

Twenty-two sampling points around Kenyir Lake were identified, and water was sampled. Figure 1(b) displays the sampling location around Kenyir Lake, with their corresponding location name presented in Table 1. The primary datasets correspond to the three seasons: dry (August 2018), normal (April 2019) and wet (December 2018). In this study, the sample collection procedure followed the protocol described in the Environmental Protection (Water) Policy 2009 - Monitoring and Sampling Manual for surface water samples. Sampling was performed manually, utilizing 1.5-liter HDPE bottles, by collecting the water at a depth of 0.3 meters below the surface to exclude any scum or debris. In-situ evaluation of water samples was performed by employing the YSI Pro handheld multi-parameter meter. The evaluated in-situ aspects are water temperature, dissolved oxygen (DO), conductivity, salinity, total dissolved solids (TDS) and pH. Subsequently, the samples were immediately placed in an ice box at 4° C to ensure preservation. They were then moved to the environmental laboratory at the Universiti Tenaga Nasional (UNITEN), Malaysia.

Figure 1. Sampling points (S1-S22) in Kenyir Lake.

Table 1. Sampling points with corresponding location name in Kenyir reservoir

No	Location	No	Location
S1	Lake Land Resort	S12	Pulau Bayas
S2	Sungai.Terengganu	S13	Sungai.Petang
S3	Sungai.Tembat	S14	Sungai.Petang(border)
S4	Musang Kenyir Reso	S15	Petang Island Resort
S5	Sungai.Petuan	S16	Dam (MIDDLE)
S6	Suria Rakit Resort	S17	Dam (LEFT)
S7	Sungai.Mandak	S18	Dam (RIGHT)
S8	Taman Tropik	S19	Sungai.Kenyir
S9	Taman Orkid	S20	Sungai Comor
S10	Eco Resort	S21	Eco Snapper Farm
S11	Butterfly Island	S22	Herb Garden

4.3. Laboratory evaluation of water samples

This study evaluates several water quality aspects like Biochemical Oxygen Demand (BOD₅), Chemical Oxygen Demand (COD), Total Suspended Solid (TSS), Ammonia Nitrogen (NH₃-N), pH and Dissolved Oxygen (DO). These are the basic parameters for determining the water quality index (WQI) of a river/reservoir body which is adopted by The Department of Environment (DOE) in Malaysia. All laboratory evaluations were based on the standard water and wastewater examination techniques (American Public Health Association (APHA), 2017). The methods used for measurement of these parameters are presented in Table 2, which also contains their limiting values and units. The limiting values are approximate values obtained from the classification of different classes in Malaysian National Water Quality Standards (EQR2006) (Table 3).

Table 2. Summary of water quality parameters

Water Quality Parameter	Unit	Limit	Measurement Method
TSS	mg/L	25–300	Gravimetric analysis
BOD	mg/L	1–12	Dilution Method
COD	mg/L	10–100	Standard Method 5220 processes recommended by APHA
DO	mg/L	1–7	YSI Pro handheld multi-parameter meter
NH3-N	mg/L	0.1–2.7	USEPA Nessler technique
pH	–	5–7	YSI Pro handheld multi-parameter meter

Table 3. Malaysian national water quality standards (EQR2006)

Class Parameter	I	II	III	IV	V
BOD	<1	1–3	3–6	6–12	>12
COD	<10	10–25	25–50	50–100	>100
NH3-N	<0.1	0.1–0.3	0.3–0.9	0.9–2.7	>2.7
CO	>7	5–7	3–5	1–3	<1
pH	>7	6–7	5–6	<5	<5
SS	<25	25–50	50–150	150–300	>300
WQI	>92.7	76.5–92.7	51.9–76.5	31.0–51.9	<31.0

4.4. Water Quality Index (WQI)

The Malaysian Department of Environment (DOE) employs Water Quality Index (WQI) to assess water quality. The fundamental aspects for computing WQI comprise DO, BOD, COD, NH₃-N, TSS and pH, and it is calculated using the following expression:

$\mathit{WQI}=\left(0.22\ast \mathit{SIDO}\right)+\left(0.19\ast \mathit{SIBOD}\right)+\left(0.16\ast \mathit{SICOD}\right)+\left(0.15\ast \mathit{SIAN}\right)+\left(0.16\ast \mathit{SISS}\right)+\left(0.12\ast \mathit{SIpH}\right)$

where;

SIDO = Subindex of DO (% saturation)

SIBOD = Subindex of BOD

SICOD = Subindex of COD

SIAN = Subindex of NH₃-N

SISS = Subindex of SS

SipH = Subindex of pH

Subindices needed for calculating WQI are ascertained based on these parameters. WQI is measured in the 0–100 range, and contamination levels are designated as: 81–100 – clean, 60–80 – Slightly Polluted, and 0–59 – Polluted. Additional WQI categorisation and other aspects into categories as specified by National Water Quality Standards are listed in Table 3.

4.5. GIS-based water quality mapping

A geographical information system (GIS) is a database intended to handle all varieties of geographically referenced information. This study used GIS software (ArcGIS 9) to build visual maps to depict water quality outcomes in Kenyir Lake. The coordinates of sampling locations and water quality data pertaining to the three seasons were input into the GIS dataset, including soil maps and topographic data. The results section depicts the visual map pertaining to every WQI variable.

4.6. Multivariate statistical methods

4.6.1. Cluster analysis

Cluster analysis comprises several multivariate methods with the primary objective to categorise objects using their inherent characteristics. Cluster analysis provides object classification, allowing other intra-cluster objects to be similar based on a prespecified selection rule. The resulting object clusters must demonstrate high internal (intra-cluster) homogeneity and high external (inter-cluster) heterogeneity. Hierarchical agglomerative clustering is the most widely used technique that offers intuitive similarity associations concerning one set and the complete set, usually depicted as a dendrogram. This work uses cluster analysis to categorise inspection sites using several parameters: temperature, DO, pH, BOD, COD, TSS, ammonia nitrogen and water quality index. Evaluation is conducted on the parameter means for every station. A dendrogram was prepared, consolidating the sampling sites into three statistically significant clusters. The outcomes categorise the distinct subsets (clusters) comprising the ascertained values.

4.6.2. Principal component analysis

PCA is intended to change the initial variables into fresh, uncorrelated variables (axes), referred to as principal components, representing linear combinations of the original variables. The new axes are aligned along the maximum variance direction. PCA was used with normalised data to contrast the compositional sequence comprising the assessed water samples and to ascertain the aspects affecting each sample. PCA-specific axis rotation created new factor sets comprising a subset of the initial variables categorised across groups. The raw data were treated using PCA to process all water quality variables and observations to determine the aspects associated with contamination sources. PCA offers information concerning the critical aspects representative of the entire dataset with respect to interpretation, reduction, and summarisation of the statistical association concerning water constituents with little information loss.

Due to improperly chosen variables, PCA is affected by missing data points, outliers, and inadequate linear correlation among variables. Hence, a comprehensive dataset pre-processing must be implemented to obtain an accurate image representing complex data. It is a widely used method for recognising patterns to account for the variance concerning a massive set of inter-correlated variables and reducing them into relatively small independent (uncorrelated) variables (principal components).

In practice, CA and PCA can fail to preserve the similarities within the clusters even while they successfully preserve the global data structure by creating well-separated clusters. As a result, because CA and PCA rely on the correlation of the variables for their analysis, a high sample size is typically necessary for a reliable result. It is possible to express the sample size as absolute numbers or as ratios susceptible to change. For PCA, it is advised that the absolute sample size be at least 50, or at least 10 or 5 times the number of variables. A recommended sample size scale, on the other hand, suggests that a sample size of 300 is fine and over 1000 is outstanding. As a general rule, 50 samples minimum (or more is preferable) would be sufficient for CA and PCA analysis (Jolliffe & Cadima, 2016).

4.6.3. Discriminant analysis

Discriminant analysis (DA) is employed for segregating observations into category-specific values, typically a dichotomy. Considering that discriminant analysis is suitable for a dataset, the produced classification of correct and incorrect predictions is expected to have a high correct percentage. This study aims to use discriminant analysis to determine the most critical variables concerning the spatial variability among the clusters. Water quality variables are predictors, while the clusters represent the predicted variable. The water quality indicators used for this analysis are temperature, DO, pH, BOD, COD, TSS, and ammonia nitrogen. DA uses several quantitative aspects to differentiate two or more naturally present datasets. Temporal changes concerning water quality are estimated using DA, which uses several quantitative indicators for differentiation. DA offers a statistical categorisation of observations and is implemented using prior information concerning the category of objects and the associated clusters. DA consolidates observations with shared properties. It creates a discriminant expression representing every group by processing raw data.

5. Results and discussion

5.1. General properties of spatial water quality in Kenyir Lake

The overall water quality outcomes were assessed using 22 sampling locations in Kenyir Lake across three seasons. Figure 2(a-h) depicts the mean values of temperature, pH, DO, BOD, TSS, COD, NH-N and WQI, respectively. The outcomes indicate that the observed temperature and pH levels are similar for all sampled areas in Kenyir Lake. The average water surface temperature in the lake is typically around 30.5 ºC. Further, the average DO and pH concerning the lake’s surface are 7.21 mg/L and 6.8, respectively. Considering research conducted by Kamaruddin et al. (2010), the outcomes indicated average temperature, DO, and pH levels as 29.76 ºC, 6.18 mg/L and 6.91, respectively, for the Pengkalan Gawi-Pulau Dula area in Kenyir Lake. Earlier work by Yusoff and Ambak (1999) concerning Kenyir Lake indicate temperature, pH, and DO ranges of 24.2 to 30.6 ºC, 6.72 to 7.61, and 3.50 to 8.90 mg/L, respectively. Moreover, Shuhaimi-Othman and Lim (2006) indicated that the Chini Lake in Pahang had an average temperature, DO, and pH levels of 29.73 ± 0.44 °C, 6.08 ± 0.88 mg/L and 6.63 ± 0.24, respectively. The mean values for these variables were calculated over the sampling period. The similarity between the two lakes might be attributed to similar ecological characteristics. Azmi and Geok (2016) studied the pH level of Kenyir Lake, indicating a range 6.63–6.91 to be considered as suitable because most aquatic insects can survive in water having a 5 to 8 pH range. The authors indicated that houseboat wastewater release is the primary aspect causing water quality degradation. However, the wastewater is diluted owing to the relatively high quantities of reservoir water.

Figure 2a. Average temperature profile across all sampling points in Kenyir Lake.

Figure 2b. Average pH profile across all sampling points in Kenyir Lake.

Figure 2c. Average DO profile across all sampling points in Kenyir Lake

Figure 2d. Average BOD profile across all sampling points in Kenyir Lake.

Figure 2e. Average TSS profile across all sampling points in Kenyir Lake.

Figure 2f. Average COD profile across all sampling points in Kenyir Lake. para.

Figure 2g. Average NH₃ profile across all sampling points in Kenyir Lake.

Figure 2h. Average WQI profile across all sampling points in Kenyir Lake.

The average BOD and COD levels for the Kenyir Lake across all sampling locations are recorded as 4.7 ± 1.9 mg/L and 8.0 ± 1.6 mg/L, respectively. Nevertheless, the BOD and COD values vary significantly across the sampling locations. It is attributed to wastewater disposal differences at the sampling locations. Areas supporting recreational activities pollute more than other sources, like rivers. The corresponding BOD and COD averages are between 1.1 to 8.5 mg/L and 5.3 to 12.3 mg/L, respectively. In contrast, the mean TSS and NH₃-N levels aggregated across 22 sampling locations in Kenyir Lake were around 81.8 mg/L and 0.07 mg/L, respectively. TSS levels are between 53.3 mg/L and 120 mg/L, while NH₃-N ranges between 0.03 mg/L and 0.40 mg/L. Sampling locations’ water quality is affected by proximal land use patterns, affecting BOD, COD, TSS and NH₃-N concerning the Kenyir Lake. However, WQI levels range between 80.0 and 91.0, providing an 85.4 average WQI, classified as Class II per the NWQS.

5.2. Cluster analysis

In order to ascertain several similar inspection points and evaluate the spatial heterogeneity of water quality, cluster analysis was performed on the water quality dataset. Cluster analysis produced a dendrogram that consolidates the 22 sampling locations into three separate clusters, as listed in Table 4. Figure 3 depicts the resulting dendrogram concerning sample site clustering using water quality properties. The first cluster comprises Lake land resort, Sg. Terengganu, Sg. Tembat, Suria Rakit Resort, Sg. Mandak, Taman Orkid, Butterfly Island, Sg.Petang (border) and Sg.Kenyir. The second cluster comprises Musang Kenyir Resort, Taman Tropika, Sg.Petang, Petang Island Resort, Dam (LEFT), Dam (RIGHT), Sg. Comor, Eco Snapper Farm and Herb Garden, while the third comprises the Sg. Petuang, Eco Resort, Pulau Bayas and Dam (MIDDLE) sampling locations.

Figure 3. Dendrogram for sampling site clustering based on water quality characteristics.

Table 4. Cluster information with location and mean values

Cluster	Location	Temp (ºC)	DO (mg/L)	pH	BOD (mg/L)	COD (mg/L)	TSS (mg/L)	NH3-N (mg/L)	WQI
1	Lake Land Resort SG. Terengganu SG. Tembat Suria Rakit Resort SG. Mandak Taman Orkid Butterfly Island SG. Petang (border) SG. Kenyir	30.5	7.11	6.82	4.80	7.85	59.26	0.05	86.93
2	Musang Kenyir Resort Taman Tropika SG. Petang Petang Island Resort Dam (LEFT) Dam (RIGHT) SG. Comor Eco Snapper Farm Herb Garden	30.5	7.46	6.84	5.03	8.48	108.57	0.11	83.24
3	SG. Petuang Eco Resort Pulau Bayas Dam (MIDDLE)	30.6	7.05	6.67	4.19	7.50	84.44	0.06	85.78

Figure 4 depicts the water quality boxplots for every cluster. These plots are compared inter-cluster to ascertain spatial differences concerning the water quality of Kenyir Lake. Figure 4.10 depicts boxplot outcomes, suggesting cluster 2 has higher BOD, COD, TSS and NH₃-N levels; however, its WQI is superior to clusters 1 and 3. It indicates that the sampled areas in cluster 2 comprise relatively more polluted water than the other two. The average observations concerning temperature, DO, pH, BOD, COD, TSS, NH₃-N and WQI for cluster 1 are 30.5 ºC, 7.11 mg/L, 6.82, 4.80 mg/L, 7.85 mg/L, 59.26 mg/L, 0.05 mg/L and 86.93, respectively. Further, the cluster 2 has average temperature, DO, pH, BOD, COD, TSS, NH₃-N and WQI readings of 30.5 ºC, 7.46 mg/L, 6.84, 5.03 mg/L, 8.48 mg/L, 108.57 mg/L, 0.11 mg/L and 83.24, respectively. Lastly, average readings concerning temperature, DO, pH, BOD, COD, TSS, NH₃-N and WQI at cluster 3 are 30.6 ºC, 7.05 mg/L, 6.67, 4.19 mg/L, 7.50 mg/L, 84.44 mg/L, 0.06 mg/L and 85.78, respectively, for cluster 3.

Figure 4. Boxplots indicating spatial variation of water quality in Kenyir Lake.

To ascertain the inter-cluster significance, a one-way ANOVA analysis was performed using SPSS version 17.0. The ANOVA outcomes indicate that TSS and WQI levels have a significant difference between the three clusters. Hence, it is understood that TSS is a key parameter affecting the spatial differences concerning water quality.

TSS is a metric to categorise the sampling points into five classes: Class I, II, III, IV and V. TSS measurements have a significant difference concerning the three clusters, suggesting that human factors impact the varying TSS levels in Kenyir Lake. High TSS levels were found near recreational and agricultural land, such as Herb Garden, Fish Farms and resorts. Decaying animal and plant matter produces organic matter that increases TSS. This research established that anthropogenic activity has a noteworthy impact on TSS, which in turn impacts water quality.

All three clusters are designated class II on WQI, based on the Malaysian DOE water quality classification. Nevertheless, cluster 2 has relatively higher pollution levels than the other two clusters basis WQI outcomes.

Several recreational and agriculture parcels are located in the second cluster: Herb Garden, Tropical Garden, Eco Snapper Farm and resorts. Hence, cluster 2 WQI was higher. Typically, water quality indicators like BOD and TSS are class III for all clusters. Further, pH is the only aspect designated class II. Moreover, indicators like DO, COD and NH₃-N are class I for all clusters.

5.3. General properties concerning temporal water quality in Kenyir Lake

2013) and Syed Jamaludin et al. (2010) suggest that May to August is the period when the southwest monsoon (SWM) is active, while the northeast monsoon (NEM) is active between November and February. Moreover, September to October and March to April are considered inter-monsoon (IM) months. S. Abdullah and Ismail (2019) suggested that NEM leads to a maximum rainfall of 370 mm, leading to a 26.5 ºC minimum temperature. SWM causes overall rainfall of 117.8 mm and a temperature of 27.6 ºC. Additionally, 178.6 mm and 27.3 ºC rainfall and temperature, respectively, are attributed to the IM. The authors indicated that Kenyir Lake is relatively dry during SWM.

Ambak and Jalal (1998) asserted that Kenyir Lake typically witnesses heavy rains between November and March during NEM. The dry, hot season typically occurs between May and October. Water quality differences concerning the normal, dry, and wet seasons are evaluated using ANOVA, suggesting that water temperature, DO, pH, COD, TSS and WQI have significant differences. The temporal changes concerning overall water quality are depicted in box plots (Figure 5).

Figure 5. Water quality boxplots result in dry, normal, and wet seasons.

5.4. Principal Component Analysis

This study used principal component analysis (PCA) on water quality datasets representing seven aspects for the 22 sampling sites. PCA aims to process the datasets, reduce dimensions, and ascertain latent factors impacting water quality. The eigenvalue criterion and scree plot were used to ascertain the count of significant principal components (PCs). Components having>1 were considered principal components. In this research, PCA determined two powerful PCs with eigenvalue exceeding one; the scree plot in Figure 6 accounts for about 57% of the overall variance in the associated water quality datasets. The first and second PCs (Table 5) account for 37.8% and 19.3% of the total variance, respectively. Varimax rotation was implemented on the determined PCs to enhance the inference of factor analysis outcomes. The first and second rotated PCs explain about 30.3% and 26.8% of the overall variance, respectively (Table 6). Factor loading designations are strong for values exceeding 0.75, moderate between 0.5 and 0.75, and weak between 0.30 and 0.50. This study considered principal components with strong factor loadings (>0.70). The first rotated PC exhibits a strong correlation with DO, pH, and temperature. The second PC strongly correlates with NH₃-N and TSS. The first PC may be inferred as being related to aspects affected by natural or weather factors. DO correlates positively with pH; however, these two factors have a negative correlation with temperature. This outcome helps us understand that temperature varies inversely with DO and pH. It suggests that higher temperatures reduce DO levels in Kenyir Lake. The second PC correlates strongly with NH₃-N and TSS, suggesting that this contamination source is typically due to anthropogenic activities. NH₃-N is produced from agricultural and domestic activities, typically fertilisers and organic or faecal matter. TSS and NH₃-N levels were high in Herb Garden, Tropical Garden, Eco Snapper Farm, and resorts. Fertilisers used in Herb Garden and Tropical Garden are responsible for the NH₃-N and TSS levels. PCA/FA could not produce significant data reduction because it uses five parameters (71 % of the original 7) needed to account for the 57.1 % of the overall water quality data variance. However, additional water quality indicators concerning physical, chemical, biological, nutrient, and heavy metal characteristics of Kenyir Lake should be gathered to enhance PCA outcomes concerning the better explanation of total water quality variance.

Figure 6. Screen plot for principal components.

Table 5. Total variance explained by significant PCs for water quality datasets

Component	Initial Eigen values			Extraction Sums of Squared Loadings
	Total	% Variance	Cumulative %	Total	% Variance	Cumulative %
1	2.645	37.789	37.789	2.645	37.789	37.789
2	1.351	19.300	57.089	1.351	19.300	57.089
3	.987	14.101	71.190
4	.849	12.129	83.319
5	.489	6.991	90.310
6	.397	5.671	95.980
7	.281	4.020	100.000

Table 6. Total variance explained by significant rotated PCs for water quality datasets

Component	Rotation Sums of Squared Loadings
	Total	% Variance	Cumulative %
1	2.120	30.289	30.289
2	1.876	26.800	57.089

PCA is conducted further on seasonal water quality (dry, normal, and wet seasons). The eigenvalue limits indicate two significant PCs concerning the seasonal water quality datasets, as in Table 7. The PCs explain 50.5%, 48.9% and 51.0% variance concerning the normal, dry, and wet seasons, respectively. The initial VF outcome concerning normal season water quality indicates that temperature and DO are strongly and positively correlated. The second VF shows a positive correlation with TSS and NH₃-N. Likewise, the first VFs concerning the dry and wet seasons also correlate positively with TSS and NH₃-N. The second VF concerning the dry season exhibits a strong and positive temperature-specific loading. Concerning the normal season, VF1 shares a strong positive loading with temperature and DO, suggesting that these variables are common for monthly water quality inspection concerning Kenyir Lake. NH₃-N and TSS are the two significant indicators that affect water quality regardless of the season.

Table 7. Total variance explained by PCs for normal, dry, and wet seasons

Season	Component	Rotation Sums of Squared Loadings
		Total	% Variance	Cumulative %
Normal	1	1.832	26.166	26.166
	2	1.701	24.294	50.459
Dry	1	1.944	27.775	27.775
	2	1.477	21.100	48.875
Wet	1	1.919	27.418	27.418
	2	1.652	23.596	51.014

The significant positive correlation concerning these two variables may account for their recreational and agricultural origin around areas near Kenyir Lake. These variables are considered outcomes of human activity close to the lake: Eco Snapper Farm, Herb Garden and Tropical Garden. Organic and inorganic material pertaining to NH₃-N enters the lake through surface runoff.

5.5. Discriminant analysis

Spatial discriminant analysis was performed using water quality outcomes concerning the three clusters obtained after cluster analysis. Discriminant analysis aims to ascertain the major variables connected to the spatial variability between the three clusters. The clusters comprise the predicted variable; every other water quality indicator is a predictor. Discriminant function coefficients (DFs) and classification matrices (CMs) produced from the conventional and stepwise discriminant analysis approaches are listed in Tables 8, 9 and 10. The determined categorisation function coefficients indicate that some indicators like COD and NH₃-N have negative values. Conventional discriminant analysis produces the associated classification having 74.2% accuracy. Further, the stepwise approach depends on three indicators: pH, temperature and TSS; its classification matrix has a 69.7% accuracy. Discriminant analysis implemented concerning spatial water quality variables suggested that significant variables strongly affect water quality for different areas of Kenyir Lake. The outcomes offer better insights concerning water catchment management and devising better inspection approaches to safeguard Kenyir Lake’s water quality.

Table 8. Classification function coefficients

Parameters	Cluster
	1	2	3
pH	82.19	84.95	82.23
Temp	17.15	16.26	16.65
DO	0.78	0.87	0.81
BOD	4.54	4.52	4.30
COD	−1.08	−0.86	−1.18
TSS	0.27	0.48	0.36
NH3	−82.10	−88.11	−84.89
(Constant)	−589.39	−607.67	−582.40

Table 9. Classification results (standard mode)

	Cluster	Predicted Group Membership			Total
		1	2	3
Count	1	24	0	3	27
	2	0	20	1	21
	3	6	7	5	18
Percent	1	88.9	.0	11.1	100.0
	2	.0	95.2	4.8	100.0
	3	33.3	38.9	27.8	100.0
74.2% of original grouped cases correctly classified

Table 10. Classification results (stepwise mode)

	Cluster	Predicted Group Membership			Total
		1	2	3
Count	1	21	0	6	27
	2	0	21	0	21
	3	7	7	4	18
Percent	1	77.8	.0	22.2	100.0
	2	.0	100.0	.0	100.0
	3	38.9	38.9	22.2	100.0
69.7% of original grouped cases correctly classified

5.6. Geographical Information System (GIS)

GIS software helped assess the spatial variability concerning Kenyir Lake’s water quality. This section comprises the changes in water quality indicators and a map depicting spatial differences.

Low BOD levels suggest high water quality and vice versa. Figure 7 indicates that BOD peaked at 9 mg/L, designated Class IV. This BOD peak was detected in the dry season from Sungai Petang (border) and Sungai Comor. Reduced water flow might be responsible for speedier organic matter breakdown during the dry season (P. Abdullah & Khalik, 2012). If the water supply comprises excessive organic waste, the bacterial presence is expected to be high, leading to faster decomposition. Consequently, oxygen requirement will be high (owing to bacterial decomposition), leading to higher BOD levels.

Figure 7. Spatial distribution of Biochemical Oxygen Demand during dry, normal, and wet seasons.

Chemical oxygen demand (COD) indicates the levels at which water takes up oxygen as the organic matter breaks down and inorganic substances oxidise. Figure 8 indicates a 13 mg/L peak COD level pertaining to the Musang Kenyir Resort and Herb Garden during the normal and dry seasons. Extensive application of organic fertiliser and sewage release impact COD concentrations. Agricultural runoff and household water are responsible for high COD. Nevertheless, the peak COD level was good (Class II) as per the National Water Quality Standards.

Figure 8. Spatial distribution of Chemical Oxygen Demand during normal, dry, and wet seasons.

Figure 9 indicates that the peak DO level was associated with the Herb Garden during the wet season. In contrast, the lowest DO level was associated with the Dam (left) during the dry season. Human activity might result in DO levels falling below the essential range. Water levels reduce during the dry period as river flow reduces. Water flow and rainfall affect DO levels, which are expected to be high with higher water discharge and velocity.

Figure 9. Spatial distribution of Dissolved Oxygen during normal, wet, and dry seasons.

Ammonia is among the significant contaminants since it is widespread; however, it is harmful and reduces reproduction and development, or it might cause death. Unionised neutral ammonia (NH₃) affects aquatic beings adversely. Ammonia peaks during the dry season (0.9 mg/L) at the Eco Snapper Farm, resulting in a Class IV level (Figure 10). Ammonia levels increase with water temperature. Ammonia levels rise due to unconsumed food breakdown (Begum et al., 2014). Excess food flowing into the lake reduces water quality.

Figure 10. Spatial distribution of Ammonia Nitrogen during normal, dry, and wet seasons.

Total Suspended Solids comprise different materials like silt, decaying animal and plant residue, and sewage. Figure 11 indicates a peak TSS level of 120 mg/L at the Eco Snapper Farm during the dry season. These levels are adverse (Class IV). The Eco Snapper Farm cultures red tilapia, and excess waste from fish leads to massive areas loaded with green slime on the water’s surface.

Figure 11. Spatial distribution of Total Suspended Solids during normal, dry, and wet seasons.

Water acidity or basicity is determined using pH. Water having relatively high free hydrogen ions is acidic, while higher hydroxyl ion levels make water basic. Water pH is impacted by chemicals, and it is a critical indicator of chemical changes in water. Typically, this research found acceptable pH (Figure 12) and water quality. The average pH levels correspond to the acceptable regulatory range. The wet season leads to pH reduction, owing to the mildly acidic nature of rainwater (pH about 5.0). Hence, pH is expected to be low during the wet season (Kamaruddin et al., 2010).

Figure 12. Spatial distributions of pH during normal, dry, and wet seasons.

Temperature changes trigger variations in water quality. Warm water is associated with reduced dissolved oxygen and may not have sufficient levels to support several aquatic species. Several chemicals are relatively toxic to aquatic beings at elevated temperatures (Kamaruddin et al., 2010). Figure 13 depicts the spatial variability in temperature during dry, normal, and wet seasons at Kenyir Lake.

Figure 13. Spatial distribution of temperature during normal, dry, and wet seasons.

Figure 14 indicates water quality falling in Class II for every season, indicating suitability for recreational purposes. During the dry season, domestic sewage, industrial discharge, and saltwater intrusion primarily trigger water quality variations. In contrast, water pollution during the wet season is caused by drainage from cultivated land and farms, including other contamination sources, affecting water quality.

Figure 14. Water Quality Index during normal, dry, and wet seasons.

6. Conclusion

An assessment of Kenyir Lake’s water quality indicators highlights more temporal than spatial variations for the studied period. Total suspended solid (TSS) is the single parameter that has a major difference (p < 0.05) across sampling sites. Further, water quality tests concerning normal, dry, and wet seasons based on ANOVA indicated that water temperature, DO, pH, COD, TSS and WQI had significant differences. Kenyir Lake has Class II overall water quality, indicating suitability for supply use, sensitive aquatic beings, recreational use, and human contact, based on the Malaysian National Water Quality Standards. Cluster analysis split all sampling locations into three groups. Stepwise discriminant analysis suggests pH, TSS, and temperature as the critical variables affecting spatial water quality variability concerning the 22 sampling sites. PCA count did not reduce data significantly because it generated 5 parameters (i.e., 71 % of the initial 7), accounting for about 57.1 % of the overall water quality variance. NH₃-N and TSS are significant indicators that affect water quality regardless of the season.

Authors Contributions

Material preparation, data collection and analysis were performed by Poornasuthra AP Subramaniam, Chow Ming Fai, Marlinda Abdul Malek, Ali Najah Ahmed and Ahmed Elshafie. The first draft of the manuscript was written by Pavitra Kumar. Manuscript was revised by Ali Najah Ahmed, Mohsen Sherif and Ahmed Elshafie.

Availability of data and materials

This research is based on manually collected data which will be made available upon request.

Ethical approval

NA

Consent to Participate:

No human or animal participants involved in this research

Consent to Publish:

All authors have read and approved the final manuscript for publication.

Disclosure statement

No potential conflict of interest was reported by the authors.