A generic framework to analyse the spatiotemporal variations of water quality data on a catchment scale

Highlights

• Propose a generic framework for spatiotemporal variation analysis of water quality.

• Inter- and intra-change of spatiotemporal variations of water quality were studied.

• Integrate data mining techniques to solve environmental problems.

• The proposed framework has the potential to be applied to other environmental data.

Abstract

Most spatiotemporal studies treat spatial and temporal analysis separately. However, spatial and temporal changes occur simultaneously and are correlated. In this study, we propose a generic framework to simultaneously analyse the spatial and temporal variations of water quality on a catchment scale. Specifically, we analyse the heterogeneity of temporal evolution of water quality data among different sampling sites, and the heterogeneity of spatial distribution of water quality data over different sampling times, respectively, by integrating the techniques of normalized mutual information, dynamic time wrapping and cluster analysis. To bring deep insight into the spatiotemporal variations, inter-change and intra-change are further defined and distinguished, respectively. Taking the Fuxi River catchment as a case study, results indicate that the proposed framework is intuitive and efficient. Beyond this, the generic framework can be expanded for other catchments and various environmental data.

Introduction

The water quality of a river can be attributed to both natural processes and anthropogenic activities including surface runoff (Abbaspour et al., 2007), climate change (Whitehead et al., 2009), geological structure (Bilgin and Konanç, 2016), land use (Sliva and Williams, 2001, Ding et al., 2015) and sewage discharge (Zhen and Zhu, 2016). Different atmospheric inputs, climatic conditions and anthropogenic inputs may result in spatial variation of water quality in rivers (Bricker and Jones, 1995). On the other hand, seasonal variation in precipitation may cause river discharge variations and subsequently affect the concentration of pollutants in river water (Vega et al., 1998). Therefore, a good understanding of temporal and spatial variations of water quality on a catchment scale is of great importance for water pollution control, aquatic ecosystem restoration and water management by regional communities (Bu et al., 2010, Iscen et al., 2008).

The methods for temporal and/or spatial analysis on water quality data have developed from univariate analysis to multivariate analysis. In the early stage, spatial and/or temporal analysis of water quality data mainly focused on a single argument (Laznik et al., 1999, Niu et al., 2004). This kind of analysis method cannot fit well to more and more complex water challenges. With the development of computer science, multivariate statistical analysis methods have grown, mainly including principal component analysis (PCA), principal factor analysis (PFA), factor analysis (FA), cluster analysis (CA) and discriminant analysis (DA) (Singh et al., 2004, Smeti and Golfinopoulos, 2016). These multivariate statistical techniques have been well-accepted and widely applied in water quality assessments (Shrestha and Kazama, 2007, Iscen et al., 2008, Wang et al., 2013), particularly for key parameter extraction and main pollution source identification.

For instance, applying CA and FA, Bu et al. (2010) grouped 12 sampling sites into three pollution level clusters (no pollution, moderate pollution and high pollution) and identified five factors of pollution sources for the Jinshui River of the South Qinling Mountains in China. Wang et al. (2013) applied CA and PCA/FA to evaluate temporal/spatial variations in water quality and identify latent sources of water pollution in the Songhua River Harbin region, China. Smeti and Golfinopoulos (2016) applied DA on surface water quality data of Yliki Lake to determine which variables were the most efficient in discriminating between clusters.

However, most existing spatiotemporal variation analytical works treat the temporal and spatial analysis of water quality data separately. For instance, Ouyang et al. (2006) assessed the seasonal variations in surface water quality in the lower St. Johns River. Wang et al. (2013) performed the temporal and spatial cluster analysis of water quality in the Songhua River Harbin region separately, classifying monitoring time into three periods and classifying monitoring stations into three groups. Chang (2008) conducted spatial analysis of water quality trends in the Han River basin, South Korea. However, the co-occurrence and correlation of temporal and spatial variations were not thoroughly considered during the previous spatiotemporal variation assessment.

Therefore, an approach, which allows to comprehensively analyse the temporal and spatial variations of water quality data simultaneously, is highly needed. Especially, in the context of a changing environment, it is essential to study the spatiotemporal variations of water quality data, so as to adapt to climate change and environment deterioration.

This paper aims to propose a generic framework to simultaneously analyse the spatiotemporal variations of water quality data on a catchment scale. The importance of this paper lies in teaching environmental engineers and scientists without a computational background to systemically analyse temporal and spatial variations of water quality data in a more effective, intuitive and easier way.

The objectives of this paper are as follows.

• to analyse how the spatial distributions of water quality data change over time;

• to analyse the spatial heterogeneity of temporal evolution of water quality data; and

• to demonstrate the procedure and verify the effectiveness of the proposed framework, taking a water quality dataset in Fuxi River catchment as a typical example.