Spatial Variation of Metal Concentrations in Watercourses of an Urban River Basin in Southeastern Brazil

ABSTRACT Spatial patterns of water quality at 29 sites, in a mixed land use watershed located in southeastern Brazil, were examined for eight metals, sampled over nine years —Arsenic, Cadmium, Copper, Lead, Mercury, Nickel, Selenium, and Zinc. Data analysis included delineation of the area of influence of each monitoring station, based on GIS analysis of Shuttle Radar Topography Mission (SRTM) images, estimation of the upper prediction limit 95% (UPL95) with censored data by the Kaplan-Meier technique, hierarchical cluster analysis (CA), Kruskal-Wallis and Friedman test. The locations of the groups generated by CA agreed with land and soil use and impact of anthropogenic activities. Use of UPL95 as entry data in CA allowed better use and interpretation of monitoring data. Areas with natural background metal-concentration levels in the drainage basin and areas of concern were identified.

Spatial Variation of Metal Concentrations in

Watercourses of an Urban River Basin in Southeastern Brazil

Cristiano CHRISTOFARO, Mônica M.M.D LEÃO

Department of Environmental and Sanitary Engineering, Federal University of Minas Gerais,

Escola de Engenharia, - Bloco II sala 4627 Av Antônio Carlos, 6627 – Campus Pampulha Belo Horizonte-MG CEP 31270-901 Brazil

ABSTRACT

Spatial patterns of water quality at 29 sites, in a mixed land use watershed located in southeastern Brazil, were examined for eight metals, sampled over nine years —Arsenic, Cadmium, Copper, Lead, Mercury, Nickel, Selenium, and Zinc Data analysis included delineation of the area of influence of each monitoring station, based on GIS analysis of Shuttle Radar Topography Mission (SRTM) images, estimation of the upper prediction limit 95% (UPL95) with censored data by the Kaplan-Meier technique, hierarchical cluster analysis (CA), Kruskal-Wallis and Friedman test The locations of the groups generated by CA agreed with land and soil use and impact of anthropogenic activities Use of UPL95 as entry data in CA allowed better use and interpretation of monitoring data Areas with natural background metal-concentration levels in the drainage basin and areas of concern were identified

Keywords: Water Quality; Urban basin; Censored data; Cluster Analysis; Heavy metals

INTRODUCTION

Water quality of rivers reflects the interaction between natural (lithology, weathering, precipitation, and erosion) and anthropogenic agents (urbanization, industrial and

agricultural activities, and water consumption) in the drainage area (Simeonov et al.,

2003; Shrestha and Kazama, 2007) Metals and metalloids can originate from natural and anthropogenic sources and are important parameters in water-quality assessment They occur in the environment in a variety of forms, and their speciation and, hence their bioavailability, is controlled by conditions such as pH and ligands, which can be

highly impaired by human activities (Chapman and Wang, 2000; Mendiguchia et al.,

2007)

Because of spatial and temporal variations in the concentrations of chemical species, a monitoring program must provide reliable and representative estimation of the water quality A program of this kind must include water samples obtained from many sites at

a regular frequency during a long interval of time, thus resulting in a large data matrix (Chapman, 1992) These datasets are often difficult to interpret, and the identification of the possible factors that can influence the occurrence and concentration of a specific

metal is reasonably difficult (Simeonov et al., 2003) It becomes even harder when

different metals are considered and other quality parameters are included in the monitoring program The use of multivariate techniques and data reduction is almost

mandatory to achieve satisfactory results (Wuderlin et al., 2001)

Multivariate techniques include factor analysis, principal component analysis, cluster analysis (CA), discriminant analysis, and neural networks Each technique presents

different features and is accordingly used for achieving specific objectives (Wuderlin et

Address correspondence to Cristiano Christofaro,

al., 2001) CA is a group of multivariate techniques used to assemble objects based on

the characteristics they possess Hierarchical agglomerative clustering is the most common approach, which provides intuitive similarity relationships between any of the samples and the entire dataset, typically illustrated by a dendrogram (McKenna, 2003) Hierarchical CAs are widely used in ecological work, and many applications in ground-

and surface-water quality assessment have been reported (Wuderlin et al., 2001; Astel et

al., 2007; Hussain et al., 2008) CA is also carried out to identify any analogous

behavior among different sampling stations or among measured variables in a dataset

from a monitoring program (Mendiguchía et al., 2007; Shrestha and Kazama, 2007)

Multivariate methods can also be used to optimize the number and the respective

locations of monitoring sites, thus reducing datasets and costs (Simeonov et al., 2003;

Shrestha and Kazama, 2007), and even to outline metalloregions (Fairbrother and Mclaughlin, 2002)

Identification of the spatial variation patterns of surface-water quality must be considered when establishing pollutant load–reduction goals and water-quality management strategies In this study, data from the Velhas River basin monitoring program have been analyzed through hierarchical CA to (i) classify different areas of the basin according to the concentration of metals and metalloids and (ii) establish natural background levels for the basin

MATERIALS AND METHODS

Study area

The Velhas River basin monitored is located at the central area of the Brazilian state of Minas Gerais, between the coordinates 17°15’ and 20º25’, latitude south, and between 43º25’ and 44º50’, longitude west (Fig 1), and is the largest tributary of the São Francisco River The São Francisco basin is the largest river basin entirely contained within Brazilian territory The area of the Velhas River basin is 29.173 km², and the length of the main river, which runs in a south-north direction, is 802 km The basin area includes 51 counties, with a population of nearly 4.8 million inhabitants (Fig 1) (Camargos, 2004)

Fig 1 – Location of the study area and monitoring stations Population size of adjacent

cities in the year 2000 is indicated by different colors (IBGE, 2002)

The upper Velhas river basin in the south includes Belo Horizonte, the sixth largest Brazilian city (2.4 million people) Major economic activities are industry and mining Significant pollution loads are discharged into the tributary rivers, mainly from domestic sewage and mining effluents In the middle and lower sectors of the basin, population density is lower, and agricultural activities and livestock production are prevalent (Camargos, 2004)

Monitoring

The monitoring program, carried out by Minas Gerais Environmental Authority (Instituto Mineiro de Gestão das Águas or IGAM), has 29 water-quality monitoring stations in the Velhas river basin Sampling, with frequency varied from quarterly to half-yearly, was conducted in two periods (wet season and dry season), from 1998 to

2006, to account for the seasonal variation in the Velhas basin The metals selected for the present study were arsenic, cadmium, lead, copper, mercury, nickel, selenium, and

zinc Sampling, preservation, and analysis were carried out according to APHA (1995) (Table 1)

Table 1 - Metals, units and methods used for water quality analysis of the Velhas River

basin, 1998 to 2006 (IGAM, 2007)

Metal Unit Determination Analytical method Detection limit

As mg/L Atomic absorption spectroscopy – hydride generator APHA 3114B 0.0003

Cd mg/L Atomic absorption spectroscopy – graphite furnace APHA 3113B 0.0005

Pb mg/L Atomic absorption spectroscopy – graphite furnace APHA 3113B 0.005

Cu mg/L Atomic absorption spectroscopy – plasma APHA 3120B 0.007

Hg mg/L Atomic absorption spectroscopy – cold vapor APHA 3112B 0.0002

Ni mg/L Atomic absorption spectroscopy – graphite furnace APHA 3113B 0.004

Se mg/L Atomic absorption spectroscopy – hydride generator APHA 3114B 0.0005

Zn mg/L Atomic absorption spectroscopy – plasma APHA 3120B 0.002

Samples were collected using one-liter plastic bottles that had been cleaned with detergents, soaked in 10% nitric acid and previously rinsed several times with distilled water Water samples were preserved by acidification with concentrated nitric acid to

pH < 2 (for Cd, Cu, Ni, Pb and Zn), potassium dichromate and sulphuric acid (Hg) or at 4ºC (As and Se) and stored in polythene bottles Sampling bottles were kept in large airtight plastic ice-cold containers at 4°C and transported to the laboratory within 6 h from sampling for further processing

In the laboratory, 10 ml samples were pipetted into a previously cleaned labeled tube The sampled aliquots and the standard solutions were digested with nitric acid (for Cd,

Pb, Ni, Zn, and Cu), acidic potassium persulphate (for As and Se), or sulphuric acid, potassium permanganate, and potassium persulphate (for Hg) The digested samples were then diluted to the initial volume of 10 ml with metal-free water and stored at 4°C The concentration of metals in water was analyzed using an atomic absorption spectrometer The quantification of metals was based upon calibration curves of standard solutions of each metal The precision of the analytical procedure, expressed as the relative standard deviation (RSD), ranged from 1 to 10% (APHA, 1995)

Data analysis

Estimation of the 95% upper prediction limit

Censored data present a serious interpretation problem for statistical analysis, unless previous manipulation is carried out (Helsel, 2005) Several methods to estimate the population mean and the standard deviation based on left-censored datasets exist in environment-related literature The influence of censored data on statistical summary has been discussed and evaluated in the studies by Helsel (1990, 2005), Singh and

Nocerino (2002), and Singh et al (2006)

In this study, the upper prediction limit of 95% (UPL95) was used for each metal at each monitoring point UPL95 was chosen as the entry level for metal data in CA to allow the inclusion of the worst water-quality conditions and to minimize the influence

of censored data UPL95 should approximately provide the 95% probability interval of future values of a set of samples and can be used to estimate the exposure-point contamination of the areas of concern, determine the achievement of cleanup standards, and finally estimate the natural background level for contaminant concentrations (Singh

et al., 2007)

The UPL95 for datasets with censored values was estimated using the Kaplan-Meier

method and resampling bootstrap procedure using PROUCL 4.0 software (Singh et al.,

2007) Detection limit was considered as UPL95 when 100% of the values were censored and the UPL95 was estimated from the student’s t-distribution when all values were greater than the detection limit

CA of monitoring sites

CA was carried out with the software Minitab 15.0 Dendrograms were constructed using the Euclidean distance, to measure similarity between samples, and by the Ward method, to establish different clusters The Ward method uses the analysis of variance approach to evaluate the distance between the clusters CA was applied after the data was normalized to zero mean and unit variance (standardized data) in order to avoid misclassifications arising from the different orders of magnitude of both numerical

value and variance of the parameters analyzed (Wunderlin et al., 2001; Simeonov et al.,

2003)

Data were classified and grouped according to their Euclidean distances, and the UPL95 values from each metal-monitoring site were depicted in box-and-whisker plots A Kruskal-Wallis and a Friedman test, a multiple comparison test between treatments (Conover, 1999), were carried out using the R package 'agricolae' (R Development Core Team, 2008; Mendiburo, 2009) to test for significant differences between groups (p < 0.05) The groups originated from the CA were then plotted in the basin map according

to incremental areas of influence of the monitoring stations

Delineation of the areas of influence of monitoring stations

Delineation of the area of influence of each monitoring station was based on the basin relief Relief can be digitally represented as a pixel matrix with topographic values for each cell This matrix is known as terrain numeric model and can be obtained by satellite and radar images (Burrough and Mcdonnell, 1998) In this study, images from the Shuttle Radar Topographic Mission (SRTM) were used For South America, SRTM images have a resolution of 90 meters (Miranda, 2005)

Georeferenced water-monitoring stations, the Velhas River basin hydrographic network, and the limiting areas were coupled with SRTM images Influence areas were delineated automatically by image treatment (Tarboton, 2005) Sub-basin boundaries were established by the software MapWindow GIS 4.5

RESULTS AND DISCUSSION

Table 2 presents the estimated UPL95 values for metals at each monitoring station of the Velhas River basin, along with the corresponding World Health Organization drinking water guidelines (WHO, 2004) for each metal

Table 2 – Kaplan-Meier UPL95 for metal concentration (mg/L) in watercourses of the Velhas river basin including the World Health Organization drinking water standard for

each metal (WHO, 2004)

Monitoring

BV013 0.015 0.0005 0.054 0.016 0.00020 0.030 0.0005 0.070

BV035 0.055 0.0005 0.053 0.060 0.00050 0.041 0.0007 0.235

BV037 0.085 0.0005 0.033 0.061 0.00027 0.046 0.0005 0.176

BV062 0.208 0.0005 0.017 0.065 0.00041 0.055 0.0005 0.181

BV063 0.135 0.0012 0.048 0.076 0.00020 0.093 0.0005 0.328

BV067 0.031 0.0005 0.019 0.058 0.00020 0.051 0.0005 0.113

BV076 0.022 0.0005 0.014 0.020 0.00020 0.031 0.0005 0.072

BV083 0.025 0.0258 0.263 0.041 0.00020 0.035 0.0005 0.203

BV105 0.040 0.0037 0.092 0.050 0.00020 0.051 0.0005 0.204*

BV130 0.018 0.0005 0.056 0.046 0.00020 0.028 0.0005 0.238

BV135 0.007 0.0005 0.026 0.010 0.00020 0.022 0.0005 0.072

BV137 0.080* 0.0005 0.042 0.057 0.00041 0.030 0.0005 0.278

BV139 0.037 0.0005 0.022 0.057 0.00020 0.033 0.0005 0.147

BV140 0.006 0.0005 0.010 0.011 0.00020 0.012 0.0005 0.128

BV141 0.156 0.0005 0.029 0.047 0.00026 0.029 0.0005 0.210

BV142 0.122 0.0005 0.043 0.064 0.00038 0.036 0.0005 0.244

BV143 0.003 0.0005 0.018 0.018 0.00020 0.013 0.0005 0.092

BV146 0.068 0.0005 0.020 0.021 0.00020 0.019 0.0005 0.140

BV147 0.026 0.0005 0.012 0.012 0.00020 0.017 0.0005 0.090

BV148 0.064 0.0005 0.018 0.020 0.00020 0.029 0.0005 0.102

BV149 0.062 0.0010 0.023 0.046 0.00020 0.030 0.0005 0.130

BV152 0.075 0.0005 0.022 0.024 0.00020 0.026 0.0005 0.152

BV153 0.141 0.0005 0.026 0.056 0.00020 0.041 0.0005 0.256

BV154 0.009 0.0245 0.051 0.027 0.00020 0.027 0.0005 0.170*

BV155 0.007 0.0005 0.102 0.050 0.00021 0.034 0.0014 0.317

BV156 0.196 0.0005 0.025 0.032 0.00026 0.036 0.0005 0.181

BV160 0.011 0.0005 0.014 0.012 0.00020 0.014 0.0005 0.108*

BV161 0.008 0.0017 0.011 0.026 0.00029 0.011 0.0005 0.127

BV162 0.002 0.0005 0.007 0.007 0.00020 0.010 0.0005 0.080

WHO drinking

water stds 0.01 0.003 0.01 2 0.006 0.07 0.01 3

*UPL95-t

Selenium, cadmium, and mercury showed the highest frequencies of censored data, corresponding to 26, 23, and 20 monitoring stations, respectively, with 100% censored data For zinc and arsenic, non censored data were obtained, and some of the UPL95 values were calculated using the t-distribution Copper, nickel, and lead presented a mean of 51.3%, 52.8%, and 61.1% respectively, of censored data The lowest mean frequencies of censored data were observed for arsenic (25.4%), and zinc (20.0%) The UPL95 calculated for arsenic ranged from 0.002 to 0.208 mg/L, and only seven

monitoring stations (BV135, BV140, BV143, BV154, BV155, BV161, and BV162) were below the WHO limit of 0.01 mg/L The UPL95 for cadmium varied between 0.0005 and 0.0258 mg/L, and three monitoring stations (BV083, BV105, and BV154) were above the WHO limit of 0.003 mg/L The UPL95 for copper ranged from 0.007 to 0.076 mg/L, and all monitoring stations were below the WHO guideline of 2 mg/L The UPL95 for lead varied between 0.007 and 0.263 mg/L, and only one monitoring station (BV162) presented a UPL95 value below the WHO limit of 0.01 mg/L For mercury, all monitoring stations were below the WHO limit of 0.006 mg/L The UPL95 calculated for nickel varied between 0.010 and 0.093 mg/L, and the UPL95 was above the WHO guideline of 0.07 mg/L for nickel in only one monitoring station (BV063) All UPL95 values for selenium were below the WHO limit of 0.01 mg/L For zinc, UPL95 values varied between 0.070 and 0.328 mg/L, below the WHO suggested limit of 3.0 mg/L at all monitoring stations

The dendrogram obtained by CA is shown in Figure 2 A division into four clusters, 1 to

4, is observed and UPL95 values for each group and metal are presented as box plots in Figure 3

Fig 2 Dendrogram with four groups using UPL95 for As, Cd, Cu, Hg, Ni, Pb, Se, and

Zn at the Velhas River basin monitoring stations

Fig.3 Variation of UPL95 metal concentrations at the Velhas River basin monitoring stations separated in four groups after hierarchical cluster analysis Groups with the same letter are not significantly different (Friedman test)

The number of monitoring stations in each group was: twelve for Group 1, four for

Group 2, ten for Group 3, and three for Group 4 The analysis in Fig 3 shows that, except for selenium (for which UPL95 values were equal to the detection limit in 93%

of the monitoring stations), all metals showed significantly different concentration values between groups, according to Kruskal-Wallis and Friedman tests Even though the study did not examine pollution sources from specific human activities taking place

in each area, a few of these sources are suggested here Nonetheless, further studies are needed to assess how adjacent cities handle their wastewater effluents and concentration

of pollutants in such effluents

Monitoring stations from Group 1 presented the lowest UPL95 values for all metals analyzed, indicating little anthropogenic influence In fact, all monitoring stations from Group 1 are located on the Velhas River tributaries with little human occupancy (Figures 1 and 4) Group 2 includes four monitoring stations having higher UPL95 values for Cu and Hg than those observed in the other groups Copper is widely used in electrical conductors, coinage alloys as well as in bronze and brass It is also an essential element; hence, the observed concentrations, which are within guideline limits, are safe for human consumption (WHO, 1998) Several human activities, not directly related to mercury, account for substantial releases into the environment These include the burning of fossil fuel, the production of steel, cement, and phosphate, and the smelting of metals from their sulfide ores (WHO, 1989a) All these activities are found

in the Velhas River basin, particularly in the higher and middle river segments For the other metals analyzed, except for cadmium and selenium, monitoring stations from Group 2 showed higher values than those found in Group 1

Fig 4 Spatial distribution of the four groups at the Velhas River basin, considering

incremental influence areas, as obtained by SRTM image processing

Groups 2 and 3 presented the highest UPL95 values for As (no statistical difference between them) Group 3 includes the areas immediately upstream and downstream from the heavily populated region and shows higher UPL95 values than Group 1 for all metals, except Cd, Hg and Se These results are in agreement with the intense mining and mineral-related activities in this area The southern region of the basin includes the area known as “Iron Quadrangle” (one of the world’s largest iron ore producers), with important mineral reserves of Fe, Mn, Cu, Sb, As, Au, Al, and U (Comig, 1974) The intense mining activities in this area release tons of waste in open air, water, sediment, and soil The significant accumulated concentration of heavy metals and toxic elements