Source identification and contribution of land uses to the observed values of heavy metals in soil samples of the border between the Northern Ireland and Republic of Ireland by receptor models and redundancy analysis

Highlights

• PMF was more specific in source apportionment of HMs in soil samples.

• Three sources were identified by PCA-APCS compared to four sources detected by PMF.

• Hg was the only element influenced by LOI regarding the results of RDA.

• Cu and Zn were influenced by fertilizer application based on RDA.

Abstract

The main objectives of the current research were source identification and quantification of the relationship between land use pattern and heavy metals (HMs) (Cr, Ni, Cd, Hg, Pb, Co, Zn, Cu, As) in soil samples collected in the border of Republic of Ireland and Northern Ireland. For the first goal, positive matrix factorization (PMF), principal component analysis with absolute principal component scores (PCA/APCS) and Unmix were utilized whereas, for the second objective, redundancy analysis (RDA) was employed. The results of source apportionment indicated that the geological formations (e.g. parent rocks), mineral explorations along with application of fertilizers in agriculture were the most influential contributing factors for the elevated levels of HMs. In this context, PCA/APCS and Unmix identified 3 sources compared to 4 sources detected by PMF with R² values larger than 0.7, except for As and Hg, indicating the reasonable accuracy of these receptor models for source identification. Among the 9 HMs considered, the performance of both PMF and PCA/APCS for As and Hg were poor with R² values equal to 0.23 and 0.51 for PMF versus 0.71 and 0.48 yielded by PCA-APCS. According to the findings of RDA; Cr, Co, As, Ni and Cu appeared to be the primary elements having strong correlations with pH and land use types. Additionally, the results of RDA demonstrated that Zn and Cu are the most probable elements that may be influenced by the amount of phosphorus in soil whereas Hg, Pb, Cr, Co and Ni are less likely to be affected.

Introduction

Among environmental contaminants, heavy metals (HMs) have attracted great attention of researchers due to toxic properties of some species besides their difficult degradation (Zhong et al., 2014, Liang et al., 2017). In this context, heavy metals refer to any chemical element with atomic density greater than 4 g/cm³ (Duruibe et al., 2007) whereas, trace elements imply those elements that occur in small amounts in natural environment (Blum et al., 2009) and when present in excessive bioavailable concentrations are known as toxic elements. Nonetheless, the term HM has become less popular in environmental sciences so, it was replaced with potentially toxic elements (PTEs) in some recent researches (Pourret and Hursthouse 2019).

In this respect, the traditional method for source identification of HMs is multivariate statistical methods, but application of receptor models has become more popular among researchers. These models were originally developed for dealing with air contaminants however, since then they have been widely applied for source identification of pollutants in soil (Lv and Wang 2019), surface water (Shi et al., 2012) and groundwater (Meng et al., 2018) ecosystems.

Some receptor models such as positive matrix factorization (PMF) (Zhang et al., 2018a, Wu et al., 2019, Men et al., 2019) and Unmix (Jain et al., 2017, Gulgundi and Shetty, 2019) are able to estimate the uncertainty of extracted pollution sources which lack in conventional methods of source apportionment like multivariate statistical methods including factor analysis, principal component analysis and cluster analysis. In this respect, the most common receptor models include PMF (Zhang et al., 2018b, Wu et al., 2019), principal component analysis/absolute principal component scores (PCA/APCS) (Singh et al., 2008), Unmix (Jain et al., 2018, Guan et al., 2019) and chemical mass balance (CMB) (Feng et al., 2019, Cheng et al., 2020). To this extent, despite lots of past and recent studies conducted about source apportionment of HMs through receptor models (Yang et al., 2013, Salim et al., 2019, Guan et al., 2019, Lv, 2019, Wang et al., 2020b), there is not any common conclusion regarding the efficiency of these approaches among researchers. For instance, Salim et al. (2019) employed spatial storm water runoff data in order to identify the contributing pollution sources using PMF and PCA/APCS models and concluded that PMF, as the superior model, is able to reveal a larger number of sources. In particular, in a recent study, Guan et al. (2019) compared the efficiency of grouped principal component analysis/absolute principal component scores (GPCA/APCS), PMF and Unmix for identification the contribution of different sources to heavy metal contamination in soil samples of Wuwei, China and concluded that GPCA/APCS was the optimal method. Since there is a dearth of information regarding the performance of receptor models for source identification of soil pollution so, further research is required to unveil the efficiency of these models.

On the other hand, from the management point of view, it is important to understand whether or not there is any difference between the levels of HMs in various types of land uses. Such data will provide the basis for prioritization of management goals and their efficient control in this regard. Different methods were employed to investigate the impacts of land uses on degradation the quality of environmental media such as pollution indices (Trujillo-González et al., 2016) and multivariate statistical methods including ensemble of multiple linear regression and constrained least squares (Kang et al., 2010), cluster analysis (Kändler et al., 2017), geographically weighted regression (Wu et al., 2017) and redundancy analysis (RDA) (Gabarrón et al., 2019, Zhang et al., 2019b). Among these techniques, RDA proved to be one of the most efficient methods and has been applied widely in different parts of the world, recently. Despite the fact that receptor models can help to identify the primary sources of contamination in an area of study however, more in depth analysis of these sources is not possible by only relying on receptor models. The joint application of receptor models and redundancy analysis would enhance our understanding of the exact land uses and contributing factors for contamination of soil and prioritization of management goals for their control. This study is the first joint application of the three most applied receptor models, redundancy analysis together with multivariate statistical methods for identification the sources of HMs accumulation in soil and analysis the contribution of land uses to the observed concentrations of HMs, to the best of our knowledge. The only recent application of a receptor model and redundancy analysis was the study of Jia et al. (2020) focusing on source apportionment by the combined method of absolute principal component score/multiple linear regression (APCS/MLR) and redundancy analysis for this purpose. Accordingly, the main objectives of this study were (i) source apportionment of HMs (e.g. Cr, Ni, Co, Zn, Cu, Hg, Cd, As, Pb) in soil samples of the border of Republic of Ireland and Northern Ireland using receptor models including PMF, Unmix and PCA/APCS (ii) uncertainty analysis of identified sources using both PMF and Unmix models (iii) assessment the relationship between land use types, pH together with soil organic carbon and levels of HMs.