5 Assessment of Groundwater Quality in Sri Lanka …
125
and average magnesium concentration in cluster 1 and cluster 2 was 42.470 mg/L
and 10.879 mg/L, respectively. Nitrate and arsenic concentrations exceeding Sri
Lankan standards (SLS 2013) were found in several districts. Agricultural activities
in sandy soils in coastal areas in Puttalam, Mannar, Jaffna, Vavuniya are associated
with excessive use of nitrogen fertilizer. Most of the above areas are located on the
soil type ‘Sandy regosols on recent beach and dune sands’ where higher arsenic
concentrations (>10 µg/L) were also reported.
5.3.2 Discriminant Analysis
Standard mode, forward stepwise mode and backward stepwise mode of DA were
used to obtain classification matrices as shown in Tables 5.2 and 5.3. Then, the Chisquare tests were applied for each mode, and results are shown in Table 5.4. The
eigenvalue >1.5, canonical correlation >0.75, Wilks’ Lambda <0.4 and Chi-square
transformation of Wilks’ Lambda show that the clusters are significantly different.
Finally, 14 parameters were remained in the forward stepwise mode, while it was
reduced to ten in the backward stepwise mode (Table 5.2).
In all analyses, classification matrices assigned 93% of the cases accurately
(Table 5.3). However, in the backward stepwise mode, data set was classified into
two clusters by using only ten water quality parameters (Tables 5.2 and 5.3). Hence,
fluoride, EC, sulphate, nitrate, sodium, magnesium, chromium, nickel, arsenic and
cadmium were identified as the most significant parameters to discriminate the two
clusters. Therefore, DA was able to substantially reduce the number of parameters
necessary for categorization. Box and whisker plots were constructed to evaluate
different patterns associated with spatial variations in groundwater quality in two
clusters (Fig. 5.3). As shown in the Fig. 5.3, Z scale transformed values (Y-axis)
clearly indicated the differences in mean values between two main clusters for each
parameter.
5.3.3 Factor Analysis
Factor analysis was able to explain >69% of variability by first six varimax factors
(VFs) in terms of measured groundwater quality parameters, in which eigenvalue
>1. They were able to explain >96% of variance in EC; >82% in chloride; 81% > in
lead; 80% > in aluminium; >70% in sodium, magnesium and iron; 60% pH, sulphate,
calcium and zinc; 50% in nitrate, chromium and arsenic; <50% in fluoride, nickel,
copper and cadmium (Table 5.5).
According to Liu, factor loadings can be classified as ‘strong’, ‘moderate’, ‘weak’
corresponding to absolute loading values; >0.75, 0.75–0.50, 0.50–0.30, respectively.
The highest proportion (25.18%) of the total variance was explained by VF1. It had
strong positive loadings on EC, chloride, sulphate, sodium, magnesium and calcium.
125
and average magnesium concentration in cluster 1 and cluster 2 was 42.470 mg/L
and 10.879 mg/L, respectively. Nitrate and arsenic concentrations exceeding Sri
Lankan standards (SLS 2013) were found in several districts. Agricultural activities
in sandy soils in coastal areas in Puttalam, Mannar, Jaffna, Vavuniya are associated
with excessive use of nitrogen fertilizer. Most of the above areas are located on the
soil type ‘Sandy regosols on recent beach and dune sands’ where higher arsenic
concentrations (>10 µg/L) were also reported.
5.3.2 Discriminant Analysis
Standard mode, forward stepwise mode and backward stepwise mode of DA were
used to obtain classification matrices as shown in Tables 5.2 and 5.3. Then, the Chisquare tests were applied for each mode, and results are shown in Table 5.4. The
eigenvalue >1.5, canonical correlation >0.75, Wilks’ Lambda <0.4 and Chi-square
transformation of Wilks’ Lambda show that the clusters are significantly different.
Finally, 14 parameters were remained in the forward stepwise mode, while it was
reduced to ten in the backward stepwise mode (Table 5.2).
In all analyses, classification matrices assigned 93% of the cases accurately
(Table 5.3). However, in the backward stepwise mode, data set was classified into
two clusters by using only ten water quality parameters (Tables 5.2 and 5.3). Hence,
fluoride, EC, sulphate, nitrate, sodium, magnesium, chromium, nickel, arsenic and
cadmium were identified as the most significant parameters to discriminate the two
clusters. Therefore, DA was able to substantially reduce the number of parameters
necessary for categorization. Box and whisker plots were constructed to evaluate
different patterns associated with spatial variations in groundwater quality in two
clusters (Fig. 5.3). As shown in the Fig. 5.3, Z scale transformed values (Y-axis)
clearly indicated the differences in mean values between two main clusters for each
parameter.
5.3.3 Factor Analysis
Factor analysis was able to explain >69% of variability by first six varimax factors
(VFs) in terms of measured groundwater quality parameters, in which eigenvalue
>1. They were able to explain >96% of variance in EC; >82% in chloride; 81% > in
lead; 80% > in aluminium; >70% in sodium, magnesium and iron; 60% pH, sulphate,
calcium and zinc; 50% in nitrate, chromium and arsenic; <50% in fluoride, nickel,
copper and cadmium (Table 5.5).
According to Liu, factor loadings can be classified as ‘strong’, ‘moderate’, ‘weak’
corresponding to absolute loading values; >0.75, 0.75–0.50, 0.50–0.30, respectively.
The highest proportion (25.18%) of the total variance was explained by VF1. It had
strong positive loadings on EC, chloride, sulphate, sodium, magnesium and calcium.