Classification of hydrochemical facies using cluster analysis of groundwater hydrochemistry data : an example from Anseong area, Korea

다국어 초록 (Multilingual Abstract)

Graphical methods (e.g., Piper diagram) and statistical cluster analysis (e.g., hierarchical clustering) have been used to cluster water samples as an initial step to interpret hydrochemical processes (natural versus anthropogenic) controlling groundwater quality. However, these conventional approaches can be unsatisfactory for reasonable interpretation because 1) groundwater quality can be too complex to successfully analyze using conventional graphical methods, 2) hydrochemical properties of the natural groundwater system can vary continuously, and 3) statistically, the geometrical space of compositional data makes up the simplex and therefore the distance between two compositional observations should be measured by the Aitchison distance, whereas most of statistical clustering methods are based on Euclidean space. This study aims 1) to classify hydrochemical data (compositional data), which was difficult to classify by using hard clustering, through Fuzzy C-menas clustering after performing ilr transformation of hydrochemical data for using strictly statistical methods (fuzzy c-means clustering, PCA, modified piper diagram) in Anseong area. Additionally, 2) we suggest a new quantitative criteria named ANSI (Anthropogenic and Natural Source Identification) to distinguish whether groundwater quality is affected by natural or anthropogenic sources. For this study, hydrochemical data of 77 groundwater samples were collected in an agricultural area of Anseong, South Korea. The ilr (isometric log-ratio)-transformed data of the concentrations of ion (Na+K, Ca, Mg, HCO3, SO4, NO3+Cl, Si) were interpreted using the soft clustering method (i.e., Fuzzy C-means clustering). Then, the fuzziness of the samples belonging to specific groups was plotted on modified Piper diagram for the visualization of quantitative clustering results. In addition, the principal component analysis (PCA) was conducted to understand the factors determining hydrochemical facies. The results yielded three well-defined water types: 1) Ca-HCO3, 2) Ca-SO4, and 3) Ca-Cl-NO3 types. Each group is interpreted to represent distinct groundwater masses: 1) Ca-HCO3 type represents the groundwater quality that affected by natural processes and rarely affected by anthropogenic pollution. 2) Ca-SO4 type describes the groundwater quality that affected by human activities and shows especially highest SO4 concentration, and 3) Ca-Cl-NO3 type represents the groundwater quality that affected by anthropogenic pollutants such as agriculture practices and industrial effluents and indicates highest NO3 concentration. In order to understand the water quality, we plotted groundwater samples at modified piper diagram. We suggest a new quantitative criteria named ANSI (Anthropogenic and Natural Source Identification) to distinguish whether groundwater quality is affected by natural or anthropogenic sources. ANSI index is the ratio of HCO3 versus the geometric mean of SO4 and Cl+NO3. The average value of ANSI indices of overlapping range between natural groundwater (Cluster 1) and polluted groundwater (Cluster 2 and 3) is 1.55. Groundwater with ANSI index≥1.55 is regarded as non-polluted natural groundwater, while groundwater with ANSI index≤1.55 is regarded as polluted groundwater affected by anthropogenic activities. The spatial distribution of ANSI index is well explained by the land use patterns of the study area. Therefore, ANSI index can be proposed as a useful tool to identify the origin of natural and anthropogenic sources of groundwater.

목차 (Table of Contents)

• ABSTRACT IV
• CONTENTS VII
• List of Tables IX
• List of Figures X
• 1. INTRODUCTION 12
• 1.1. The importance of groundwater and necessity of water quality management 12
• 1.2. Cluster analysis: Fuzzy C-means clustering 14
• 1.3. Compositional data 16
• 1.4. Ovjectives of this study 17
• 2. STUDY AREA 18
• 3. METHODOLOGY 22
• 3.1. Sampling and chemistry analysis 22
• 3.1.1. Field measurement and sampling procedure 22
• 3.1.2. Chemical analysises 23
• 3.2. Compositional data and ilr transformation 24
• 3.3. Fuzzy C-means clustering and validity index 27
• 4. RESULTS AND DISCUSSIONS 31
• 4.1. Data preprocessing 31
• 4.2. General hydrochemistry 40
• 4.3. Dual isotopic compositions of groundwater nitrate 49
• 4.4. Fuzzy c-means clustering results 52
• 4.4.1. Cluster number validation 52
• 4.4.2. Description of hydrochemical characteristics of FCM-derived cluster using PCA 53
• 4.4.3. Suggestion of anthropogenic contamination index using modified Piper diagram 59
• 5. SUMMARY AND CONCLUSIONS 64
• References 66