Assessment of groundwater quality in Vavuniya and Mullaitivu, Sri Lanka using multivariate statistical techniques and a Water Quality Index

The exploitation of groundwater drastically increased in the 20th century, resulting in greater benefits to mankind. But it has triggered unpredicted changes in the state of groundwater systems and the rate of increasing water consumption is greater than double the population growth (van der Gun 2012). Thus, it has been contributing to two significant problems: a decrease in the quantity of available groundwater and a decline in the quality of that water. Numerous natural and man-made variables have an impact on groundwater quality. Some of the natural elements are weathering of rocks, mineral dissolution, aquifer depth, recharge rate, ion exchange, and evapotranspiration rate (Liu et al. 2003; Balasooriya et al. 2021; Subba Rao et al. 2021) and over-extraction, domestic wastes, septic tank leakages, excessive use of agrochemicals, and pesticides are only a few anthropogenic activities that have a negative impact on water quality (Balasooriya et al. 2021; Subba Rao et al. 2021).

The study area for this study was the Vavuniya and Mullaitivu districts of the Northern flat terrains of the dry zone of Sri Lanka which suffers from low rainfall, high temperature, and droughts. Since perennial rivers are not available throughout the region and limited availability of waterbodies, drinking water supply merely depends on the groundwater sources. The groundwater sources in the study area have faced severe risks over the past decade (Piyasiri & Senanayake 2016; Athapattu et al. 2018). Overexploitation due to large-scale infrastructure development projects, intensive resettlement, rapid urbanization, and extensive agricultural activities after the civil war had created serious issues such as depletion and deterioration of groundwater quantity and quality in Northern Province, Sri Lanka (Loganathan 2011; De Silva 2016; Ravi et al. 2016; Akther & Tharani 2017; Rajapakse et al. 2017; Shah et al. 2019; Gobalarajah et al. 2020).

Akther & Tharani (2017) assessed groundwater quality in Vengalacheddikulam Divisional Secretariat Division (DSD) in the Vavuniya district of Sri Lanka and found nearly 21.5% of the area is not suitable for human consumption. Rajapakse et al. (2017) assessed groundwater quality in the Sinnasippikulam area of Vavuniya and found dental fluorosis is a highly endemic problem in several areas of the Vavuniya District. Domestic wells located around the urban council limits of Vavuniya were contaminated severely with Fecal Coliform (Loganathan 2011; Ravi et al. 2016), and high nitrogen levels were observed in the city area and Thandikulam and Kurumankadu areas of Vavuniya district in Sri Lanka (Loganathan 2011; De Silva 2016). Piyasiri & Senanayake (2016) assessed the fluoride and hardness levels in groundwater in Vavuniya city, Sri Lanka, and found higher concentrations of fluoride and hardness in North Western part of the city whereas South Eastern part of the city indicated lower concentrations of fluoride and hardness. Athapattu et al. (2018) assessed the quality of drinking water sources in Vavuniya MOH division and found several parameters exceeded the maximum permissible levels of SLS 614:2013. A limited number of groundwater quality studies in Mullaitivu district can be found to date. Gobalarajah et al. (2020) assessed the impact of water quality on CKDu in Thunukkai area in Mullaitivu, Sri Lanka and found a significantly positive correlation (p < 0.05) between total dissolved solids (TDS) and arsenic.

Modern approaches like multivariate statistical analysis have been widely employed for differentiating natural or anthropogenic groundwater contamination sources, data reduction, and classification (Singh et al. 2004; Nosrati & van den Eeckhaut 2012; Machiwal & Jha 2015; Balasooriya et al. 2021). To determine the underlying reasons for poor groundwater quality, these methodologies provide information on the links between parameters and sampling sites as well as an idea of similarities and differences between parameters (Nosrati & van den Eeckhaut 2012; Noshadi & Ghafourian 2016; Balasooriya et al. 2021).

Tajmunnaher & Chowdhury (2017) evaluated the water quality parameters along the Kushiyara River in Bangladesh and found strong and moderate positive relationships among BOD, COD, TDS, TS, and SS. Liu et al. (2003) examined groundwater in the coastal Blackfoot disease area of Yun-Lin, Taiwan, and found strong positive correlations between EC, TDS, Cl⁻, SO₄²⁻, Na⁺, K⁺, and Mg²⁺. Strong positive relationships between EC, Ca²⁺, Mg²⁺, Na⁺, and TDS are found in Coimbatore city in India (Selvakumar et al. 2017).

Machiwal & Jha (2015) applied PCA for groundwater sources of the Udaipur district, India and the PCs explained 75–80% of the total variance. Nosrati & van den Eeckhaut (2012) applied PCA to assess the groundwater quality of Hashtgred plain in Iran. Liu et al. (2003) employed PCA and identified parameters influencing the geochemical processes in the aquifer system of Yun-Lin, Taiwan. Selvakumar et al. (2017) applied PCA for groundwater in Coimbatore, India. Balasooriya et al. (2021) analyzed principal in 25 districts of Sri Lanka and found >69% of variability by the first six significant components. Rajapakshe & Rathnayake (2018) applied PCA for groundwater sources located in the very urbanized Malabe area in Sri Lanka and found first four principal components (PCs) explained 79.8% of the total observed variance in the data.

Machiwal & Jha (2015) performed HCA and found two clusters from 53 sampling sites in the Udaipur district in India. Selvakumar et al. (2017) investigated hydrogeochemical characteristics and groundwater contamination in Coimbatore, India by employing HCA. Noshadi & Ghafourian (2016) investigated the quality of groundwater in Fars province, southern Iran by employing HCA and found three clusters of Ca-HCO₃ and Na-Cl types. Bencer et al. (2016) applied HCA for groundwater sources in Ain Djacer (Eastern Algeria). Belkhiri et al. (2011) applied HCA and found three sampling clusters in Ain Azel plain, Algeria. Balasooriya et al. (2021) employed HCA and found two clusters during the study conducted in all administrative districts of Sri Lanka.

The structure of the models, the parameters included and their weightings, as well as the techniques for sub-indexing and aggregation, have all been altered in several Water Quality Index (WQI) models (Sun et al. 2016; Uddin et al. 2021). Ravi et al. (2016) calculated the WQI using the weighted arithmetic index for groundwater in Vavuniya, Sri Lanka and found that the WQI of 4 GNDs fall under the ‘Poor and Very Poor’ water subclasses while only 6 GNDs fall under ‘Good and Excellent’ subclasses out of 10 GNDs. Mahagamage et al. (2006) used CCMEWQI to investigate the suitability of groundwater in the Kelani River basin for drinking, irrigation and livestock purpose. Sinha & Saxena (2006); Latha & Rao (2010); Harshan et al. (2017) have used the method proposed by Horton (1965) and modified by Tiwari & Mishra (1985) to calculate the WQI. Nevertheless, they have categorized water subclasses according to their regions and considered water quality guidelines. Cooray et al. (2019) also developed a WQI using the weighted arithmetic method without imposing an upper limit.

GIS is a very valuable and essential tool that is used widely all over the world by water-related environmental planning and management professionals (Tsihrintzis et al. 1996). Brhane (2018) developed the spatial distribution maps for several parameters in the Adigrat area in Tigray, northern Ethiopia. Akther & Tharani (2017) developed the spatial distribution maps over the Vengalacheddikulam DSD in Vavuniya, Sri Lanka. Piyasiri & Senanayake (2016) incorporated GIS maps to depict the spatial distribution of EC, fluoride, total hardness, and pH. Jeihouni et al. (2014) developed spatial distribution maps for sulphate, chloride, hardness, EC, pH, and WQI. Gobalarajah et al. (2020) developed spatial distribution maps and identified vulnerable areas in Thunukkai DSD of Mullaitivu, Sri Lanka. Spatial distribution maps were developed employing the Kriging method by Machiwal & Jha (2015) for a hard-rock aquifer system in Udaipur, Rajasthan in India.

The literature has ultimately led to the conclusion that groundwater quality and geochemical parameters vary widely depending on the temperature, topography, geological formations, hydrogeological conditions, and anthropogenic activities. Among the districts of the Northern province, Mullaitivu and Vavuniya have been considered ‘at risk’ for the occurrence of CKDu with nine other districts from North Central, Central, and Uva provinces of Sri Lanka (Kafle et al. 2019; Gobalarajah et al. 2020). Few groundwater-related studies are available for Vavuniya and Mullaitivu districts and are only limited to Vengalacheddikulam DSD, town area of Vavuniya and Thunukkai DSD, Mullaitivu (Loganathan 2011; De Silva 2016; Piyasiri & Senanayake 2016; Ravi et al. 2016; Akther & Tharani 2017; Athapattu et al. 2018; Shah et al. 2019; Gobalarajah et al. 2020). In such a backdrop, it is aimed to assess groundwater quality by incorporating individual wells scattered within Vavuniya and Mullaitivu districts during this study. By developing WQI and spatial distribution maps, and employing multivariate statistical approaches the ultimate aim is to enhance and safeguard the sustainability of groundwater resources in the Vavuniya and Mullaitivu districts as there is no such study carried out in the identified research area.

Geologically, Precambrian metamorphic hard rock, which belongs to Wanni Complex, dominates the Vavuniya district. Prominent rock types in the area are Chanokitic gneiss, Granitic Gneiss, and Hornblende biotite gneiss (Athapattu et al. 2018). In Mullaitivu district, in the upper reaches of Pali Aru and Parangi Aru basins existence of Vijayan rocks is predominant. Furthermore, along the streams, Quaternary Alluvium deposits exist. The shallow regolith aquifers of the metamorphic terrain are the dominant type of aquifer found in the research area.

The National Water Supply & Drainage Board (NWS&DB), Vavuniya, Sri Lanka, provided secondary data on groundwater wells that were gathered and tested between 2018 and 2020. All samples have been collected either from a dug well or a tube well which is being used regularly by the dwellers. Therefore, these data fairly denote the characteristic status of groundwater quality in the vicinity. There were altogether 122 number (62 and 60 samples in Mullaitivu and Vavuniya, respectively) samples including 15 NWS&DB intake wells have been tested for 10 physical and chemical parameters (color, turbidity, pH, TDSs, total alkalinity, total hardness, chlorides, fluorides, nitrates, and nitrites).

TDS and pH of the water samples have been measured using a conductivity meter (HACH EC 7) and pH meter (HACH pH1) respectively while turbidity has been tested by the Nephelometric method using a turbidity meter (HACH 2100N). Hardness and alkalinity of the samples have been tested by EDTA Titrimetric Method following APHA standard method 2340 and Acidimetric titration adhering to APHA standard method 2320 (APHA 2017) respectively while chloride has been tested by the silver nitrate titrimetric method (argentometric titration) adhering to APHA standard method 4500-Cl⁻ B (APHA 2017). Nitrate, nitrite, and fluoride have been tested using a spectrophotometer (HACH DR 5000) following (Hach method 8039) and (Hach Method 8029), respectively.

Uncorrelated PCs can be obtained by transforming actual variables (Nosrati & van den Eeckhaut 2012), and the variance of correlated variables and lowering data set dimensionality can be explained (Subba Rao et al. 2020) by applying PCA. Before PCA, the suitability of data for PCA was assessed by employing Kaiser–Meyer–Olkin (KMO) and Bartlett's tests. KMO values of 0.65 and a χ² value of 682.17 (p-value <0.0001) from Bartlett's test of sphericity show that the data set has a sufficient variance to be subjected to PCA or Factor Analysis (FA).

In this work, HCA was carried out utilizing a single linkage amalgamation algorithm and a Euclidean distance similarity measure on 1,220 data. Since there is no reliable method to establish the ideal number of clusters, the researcher must decide how many clusters to use. Hence, the only approach which can be adopted to choose the clusters in the dendrogram is visual inspection (Güler & Thyne 2002; Belkhiri et al. 2011) similar to (Subba Rao et al. 2021). Accordingly, clusters were identified by drawing a phenon line at linkage distance 15.

PCA, HCA, KMO, and Bartlett's tests were carried out by SPSS (SPSS version 26.0) software. By computing Pearson's correlation coefficient (r), the inter-relationship between physio-chemical characteristics was examined using Microsoft Excel 2013 software for 10 parameters. It produced the correlation matrix which depicts the correlation between any two parameters considered.

The WQI proposed by Horton (1965) and modified by Tiwari & Mishra (1985) was employed to compute the WQI considering 10 water quality parameters.

Obtained WQI values were classified into five subclasses operationally similar to Harshan et al. (2017).

Using the spatial interpolations feature in GIS software, values of characteristics at unsampled sites can be predicted by employing the already obtained values at identified sites (Akther & Tharani 2017). ArcGIS 10.4.1 was used to implement the ed (IDW) interpolation approach during the study.

Table 2 provides overall data on the water quality parameters that were assessed.

The color of groundwater samples ranged between 0 and 271 Pt/Co units. 85.2% of the groundwater samples were below 15 Pt/Co units. The groundwater samples' turbidity ranged from 0.1 to 33.4 NTU. 79.5% of the groundwater samples were below 2 NTU as per the SLS 614:2013 standards. pH values of 93.4% of samples were within the range of 6.5–8.5 as stipulated in Sri Lankan standards.

TDS values were in the range of 44–3,533 mg/l and with a mean of 588.74 mg/l while the maximum permissible level of 500 mg/l in SLS 614:2013 was exceeded in 59.8% of the samples. This represents a wide range of variance in the salinity of the water in terms of the numerous ions dissolved in it. Because of the chemical process of silicate weathering the dissolved ions are released into the groundwater body (Subba Rao 2021). Total hardness values ranged from 4 to 1,050 mg/l. 63.9% of samples were above the permissible limit of 250 mg/l in Sri Lankan standards for potable water. Total alkalinity ranged between 22 and 570 mg/l. 64.8% of samples exceeded the permissible limit of 200 mg/l in SLS 614:2013. The amount of hardness and alkalinity are the same when calcium and magnesium carbonates are present alone. Various dissolved ions, predominantly Ca²⁺ and Mg²⁺ cause the hardness in water. Most of the water in the regions of the dry zone in Sri Lanka is Ca–Mg-rich water (Rubasinghe et al. 2015). The dissolution of calcium and magnesium ions is the dominant factor for occurrence of harness in groundwater while dissolved alkali substances resulting total alkalinity in groundwater.

The observed nitrate concentrations were between 0 and 44 mg/l and none of the samples exceeded the maximum permissible limit of SLS 614:2013. Nitrite (NO₂⁻) is not usually present in significant concentrations. Observed nitrite values ranged from 0 to 0.15 mg/l and none of the samples exceeded the maximum permissible limit of 3 mg/l as specified in the SLS 614:2013. Nitrate is considered as a non-lithological pollutant and reaches groundwater from nitrate-fertilizers, human and animal excretes, septic tank leakages, domestic effluents and irrigation return flows (Zhang et al. 2018; Li et al. 2019; Subba Rao et al. 2019, 2021). A higher level of NO₃⁻, greater than 10 mg/l, indicates water contamination as a result of anthropogenic activities (Subba Rao 2021). Extensive agricultural activities using fertilizers and domestic wastes cause higher nitrate values in the area. Observed fluoride values ranged from 0–2.45 mg/l and 40% of samples were above 1 mg/l. Piyasiri & Senanayake (2016) also revealed higher fluoride concentrations in some areas of the Vavuniya district. During the study carried out in Thunukkai DSD, Mullaitivu by Gobalarajah et al. (2020), it was found that 39% of samples exceeded 1 mg/l while the mean fluoride value was found as 1.73 mg/l. According to Chandrajith et al. (2020), fluoride levels in Thunukkai were comparatively higher than those in Sri Lanka's dry zone. The presence of fluoride containing minerals in basement rocks, such as hornblende, biotite, and apatite, is the principal source of fluoride content in subterranean water, which degrades groundwater quality due to extended interaction of water with aquifer elements in an alkaline condition (Subba Rao 2021). Thus, leaching fluoride from fluoride-bearing minerals results in higher fluoride values in the study area. Observed chloride values ranged from 10 to 950 mg/l while 6.6% of samples exceeded 250 mg/l as stipulated in SLS 614:2013. Chloride is also considered as non-geogenic source and presence of higher chloride levels in groundwater is an indication of possible anthropogenic activities like septic tank leakages and extensive irrigation practices (Subba Rao et al. 2019; Subba Rao 2021). Higher evaporative conditions would also lead to increased chloride concentrations (Marghade et al. 2021), as the study area lies in the dry zone of Sri Lanka.

TDS showed solid positive links with chloride (r = 0.935) and total hardness (r = 0.847). Water hardness is mostly caused by the Calcium and Magnesium cations, whereas TDS is made up of inorganic salts that are dissolved in water and primarily include calcium, magnesium, sodium, potassium, bicarbonates, chlorides, and sulphates (NWS&DB 2020). Thus, TDS, chloride, and total hardness are caused by ions dissolved in the water, these parameters were strongly and positively correlated. TDS and hardness of groundwater mainly occurred due to geogenic processes like weathering of minerals (Marghade et al. 2021). Yet, TDS is strongly correlated with chloride, it indicates the possible anthropogenic contamination.

It is found strong positive correlations between turbidity with color (r = 0.756) and nitrite with chloride (r = 0.887) and TDS (r = 0.761). Nitrate and nitrite predominantly resulting due to man-made activities (Marghade et al. 2021). As nitrite is positively linked with chloride both ions have the same source, i.e. anthropogenic activities. It is revealed that fluoride has a moderate negative correlation (r = –0.729) with the pH value of the water as tabulated in Table 3. Fluoride absorption in soils reduced from humid to dry regions and from acidic to alkaline soils (Wang et al. 2002; Balasooriya et al. 2021). Furthermore, fluoride, pH, and total alkalinity occupied a separate parameter loading in PCA (Table 5). Thus, this explains the increased fluoride leaching from fluoride-bearing minerals in an acidic environment.

PCA demonstrated >77% of the variance of the measured variables by its four initial components (PCs) as tabulated in Table 4.

PCs elucidated >97% of variance in TDS; >96% in chloride; >88% in color and turbidity; >83% in total hardness, nitrate and nitrite; >68% in fluoride and pH; and >53% in total alkalinity as shown in Table 5.

The highest proportion (34%) of the total variance was explained by PC1, PC2, PC3, and PC4 explain approximately 18, 15, and 10% of the total variance by the component, respectively. Factor loadings were categorized according to Liu et al. (2003).

Chloride, TDS, and total hardness loadings on PC1 are strongly positive. The main contributors to total hardness are carbonate, calcium, and magnesium ions. Gathering the aforementioned ions in one factor typically reflects how groundwater develops naturally through groundwater-geological interaction (Nosrati & van den Eeckhaut 2012; Noshadi & Ghafourian 2016; Balasooriya et al. 2021). Therefore, it is hypothesized that PC1 represents groundwater and natural geological interactions.

Fluoride and total alkalinity do seem to have positive loadings, and pH has a significant positive loading in PC2. Fluoride-rich minerals readily leach fluoride. Thus, this could be the driving factor for presenting higher fluoride concentration in groundwater.

From wet to dry locations, and from acidic to alkaline soils, fluoride absorption in soils shows a decline (Wang et al. 2002). This may help in explaining the inter-relationship between fluoride, pH, and total alkalinity in groundwater. Therefore, PC2 also reflects the natural geological interactions with groundwater.

Strongly positive loadings on color and turbidity are present in PC3. Hence, PC3 explains the natural processes like high rainfall and a higher rate of infiltration. Nitrate and nitrite show significantly positive loadings on PC4. Shah et al. (2019) explained the wells that were dug nearby an agricultural field had a high nitrate content in Vavuniya. The usage of nitrogen-containing fertilizer for agricultural activities is the reason behind that. Hence, it is apparent that PC4 indicates the anthropogenic pollution sources.

Groundwater sampling sites were grouped using HCA so that the sites within a cluster had almost similar groundwater quality, but were distinct from those in other clusters. Ward's approach was used to perform HCA on the standardized data set (1,220 observations) by employing the SPSS Statistics Version 26.0 software package.

Sampling sites of cluster 1 are located in Maritimepattu and Puthukudiyirippu areas in Mullaitivu where deep confined aquifers of the sedimentary limestone and sandstone formations are available (Panabokke & Perera, 2005). Sampling sites of clusters 2 and 3 are mainly located in Oddusudan, Vavuniya South, and Vengalacheddikulam areas where hard rock aquifers are available (Panabokke & Perera, 2005). Total hardness, total alkalinity, TDS, and fluoride in groundwater mainly occur due to mineral dissolution by water–rock interactions (Rubasinghe et al. 2015; NWS&DB 2020; Subba Rao 2021). Geogenic processes of hard rock aquifers might be the possible causative factor for having higher average concentrations of total hardness, total alkalinity, TDS, and fluoride in clusters 2 and 3 than in cluster 1. Presence of higher chloride and nitrate values in groundwater is possibly due to anthropogenic pollution sources like septic tank leakages and extensive irrigation practices. Puthukudiyirippu and Maritimepattu areas have a lesser number of irrigable lands than Vavuniya South, Oddusudan, and Vengalacheddikulam DSDs as they are closer to the sea. Hence, it is apparent that the extensive usage of fertilizer and irrigation return flows are the possible causative factors for having higher average values of chloride and nitrate in clusters 2 and 3 than in cluster 1.

The WQI has varied considerably among the three clusters. In cluster 1, 24, 55, 17, and 4% of samples were in the Excellent, Good, Fair, and Marginal categories, respectively, while none of the samples were observed in the Poor category. In cluster 2, none of the samples were observed in the Excellent, good, and Fair categories while 18 and 82% of samples were observed in Marginal and Poor categories, respectively. In cluster 3, 2, 34, 35, 20, and 9% of samples were observed in the Excellent, Good, Fair, Marginal, and Poor categories, respectively.

Supplementary material, Table A1 illustrates the calculated WQI values and water subclass classification. Obtained WQI classes were classified into 5 water subclasses as shown below. According to Table 8, there are 7% of samples are in the ‘Excellent’ category while 36, 28, 16, and 13% of samples are in the ‘Good’, ‘Fair’, ‘Marginal’ and ‘Poor’ water subclass categories, respectively.

Poor groundwater quality was found in Olumadu, Mankulam GNDs of Oddusudan DSD in Mullaitivu, Nedunkerny North, Nedunkerny South, Mamadu GNDs of Vavuniya North DSD, Pampaimadu, Poomaduwa and Rankethgama GNDs of Vavuniya and Vavuniya South DSDs, respectively, and Sinnasippikulam, Neriyakulam, Maradanmaduwa, Pavatkulam unit 2,4,5 and 6, Awaranthulawa, Kurukkalputhukulam and Andiyapuliyankulam GNDs of Vengalacheddikulam DSD in Vavuniya. In addition to that 8 GNDs of Oddusudan DSD in Mullaitivu and 4,6 and 8 GNDs of Vavuniya North, Vengalacheddikulam and Vavuniya DSDs, respectively, in Vavuniya were observed under the Marginal category.

Altogether 26 GNDs of the study area fall under Poor and Marginal water subclasses. It reflects the potential health threats of directly consuming groundwater for drinking. The groundwater quality is predominantly controlled by the geology as most of the contaminants are present in groundwater due to the geogenic processes. Hard rock aquifers are available in the identified Poor and Marginal water quality areas. Higher mineral weathering is possible due to the higher residence time in hard rock aquifers. Thus, artificial groundwater recharging at household levels shall be effectively used to dilute the dissolved ion concentration and improve the WQI in individual wells. Furthermore, introducing proper sanitation facilities and practical regulations in agricultural practices can reduce anthropogenic pollution sources and improve the WQI in the vicinity.

This study provides an integrated strategy using multivariate statistical techniques and GIS-based spatial distribution maps, including the WQI, for the first time in Sri Lanka. CA revealed strong positive relationships among TDS, chloride, total hardness, and between turbidity, and color. PCA explained >77% of variability by the first four PCs in terms of measured groundwater quality parameters. HCA classified 122 sampling sites into three statistically significant clusters. Resulted WQI values with the combination of spatial distribution maps revealed significant deterioration of groundwater quality mainly in the 18 GNDs in study area. The integrated map emphasized that sampling sites of the above three clusters are parallel with the spatial distribution of water subclasses. Moreover, groundwater sources of Maritimepattu and Puthukudiirippu DSDs showed comparatively much lower average concentrations of measured water quality parameters than the rest of the region. Promoting artificial recharging at the household level, introducing proper sanitation facilities, and imposing regulations in agricultural practices shall be implemented to improve the WQI further. The findings can be used to develop policies for handling groundwater management issues in the vicinity and to make decisions on groundwater activities in the other parts of the dry zone to guarantee a safe and high-quality groundwater supply.

The National Water Supply & Drainage Board, Vavuniya is gratefully acknowledged by the authors for providing data on groundwater quality, and to the reviewers whose comments and suggestions have significantly contributed to the improvement of this manuscript.

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

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

The authors declare there is no conflict.