Introduction

Paying attention to contamination causes and sources and also to monitoring pathways is the main objectives in rational use of soil resources and environment. Increasing human activities during the last decades have induced widespread release of heavy metals in the environment. Industrial products, mines, transport, and even uncontrolled application of pesticides are the known sources to establish heavy metal contamination (Hutton and Meeus 2001). These metals may spread to the soil by sewage application, wastewater irrigation, and atmospheric deposition (Salomons 1992).

Late in the 1970s, pedologists realized the possible employment of geostatistics in soil survey. They also developed methods to make maps of individual soil properties without having to classify the soil and getting embroiled in all the doubts and controversy (Webster 2008). Due to the impossibility of making the samples of all data points, geostatistical approaches were used to predict the spatial and temporal distribution of a variable and create maps as well. These provide descriptive tools such as semivariograms to characterize the spatial pattern of continuous and categorical soil attributes (Goovaerts 1999). A lead risk map for Wolverhampton, England, was provided using a spherical model and kriging interpolation method, in which near surface soil lead concentrations showed a degree of spatial correlation on a city scale (Hooker and Nathanial 2006). A comparative study of interpolation methods for mapping soil properties revealed that the correlation coefficients between experimental data and estimated results of kriging were higher and its mean absolute errors were lower than those of Inverse Distance Weighting (IDW) method (Kravchenko and Bullock 1999). Nevertheless, the high accuracy of that method for studying of nitrogen distribution pattern has previously been reported (Gotway et al. 1996). This approach was also confirmed by others to predict the spatial distribution of heavy metals in floodplain soils around the Gule River in the Netherlands (Leenaers et al. 1990).

Climate change in Iran is likely to cause the conversion of agricultural lands into the marginal ones. As well known, the bioclimatic changes are out of our monitoring, whereas mining and manufacturing development should be perfectly controlled. Disturbance of mines, whose owners do not obey the laws of rehabilitation, is potentially detrimental to the environment (Shahbazi and De la Rosa 2010). Tailings dam are structures built to impound wastes from mining activities. Currently, thousands of tailings dam worldwide contain billions of tons of waste material at mining sites. Tailings dam should be constructed to achieve a stable and safe state. Unfortunately, since 1970, more than 35 tailings dam failures reported around the world. The Aznalcollar (Spain) and Sasa (Macedonia) accidents occurred in 1998 and 2003 caused an intensive flow of tailing materials through the rivers of Guadiamar and Kamenica, respectively (Vrhovnik et al. 2013). Copper is the main contaminant in the soils of copper mining areas of the world. Moreover, copper is an essential nutrient to all organisms but is toxic at high concentrations (Adriano 2001). Some of researchers have reported the spatial distribution of soil copper in relation to relevant soil properties (Grzebisz et al. 2001; Jafarnejadi et al. 2013; Sun et al. 2014).

The study site of this research work was around the Mazraeh copper mine which has been established since 1960 in the Northwest of Iran. The refining capacity of this refinery is expected to be 250 t copper per day. Copper ore concentration processes create wastes (tailings) that are directed to the ponds for storage. Therefore, the marginal agricultural areas are thought to be at risk of exposure to heavy metals, particularly copper. On May 6, 2010, after a heavy rain, the poor construction of the tailings dam caused to collapse and then to discharge the effluent into a seasonal stream, namely Mazraeh-Chai, that empties into Ahar-Chai River. Because of low capacity of the stream channel, massive flooding and triggered mudslide occurred in the marginal lands. The tailings flow material comprised high concentrations of some heavy metals, particularly copper which may adversely affect the environment. Consequently, we investigated the patterns of copper spatial distribution in the vicinity of the failed dam. This study deals with the characteristics of the soils and their surficial sediments after the accident.

Materials and methods

Study area

This study was performed in an area of about 1500 ha located in the Mazraeh region of Ahar County in East Azarbaijan province, Iran (Fig. 1).

Fig. 1
figure 1

Study area and sampling points setting

Full size image

The area lies within the coordinates of longitude 47° 02′ 47″ to 47° 03′ 46″E and latitude 38° 31′ 56″ to 38° 37′ 12″ N. There are cultivations, orchards, and voids along the pathway of seasonal stream. The geology of the study area belongs to quaternary period and characterized by high level piedmont fan and valley terraces deposits. The dominant soil orders in this region were classified as Inceptisols.

Sampling and preliminary analysis

Composite sampling was conducted at 30 sites close to the pathway of Mazra-Chai seasonal stream (Fig. 1) in May 2011 for determining the total concentration of copper not only at both upper (0–10 cm) and lower (10–30 cm) soil depths (indicated as surface and subsurface soils) but also at surface accumulated (surficial) sediments. The samples were collected on the line, spaced 300 m apart, from the tailings dam to about10 km away where the effluent was being discharged and expanded through the marginal lands. Moreover, six control soil samples were collected some meters away from the contaminated area but not covered by sediment. Global Positioning System (GPS) was used to provide the geographical coordination of sample points.

Soil samples were air-dried and then sieved through a 2-mm mesh. Some common physical and chemical properties including texture (Gee and Bauder 1986), electrical conductivity (EC) (Rhoades 1996), pH (Thomas 1996), cation exchange capacity (CEC) (Chapman 1965), organic carbon (OC) (Nelson and Sommers 1982), and calcium carbonate equivalent (CCE) (USDA-SCS 2004) were determined. The soil copper concentration (SCC) at two depths and copper concentration of surficial sediments (CCS) were measured using aqua regia method (Chen and Ma 2001). Half a gram (0.5 g) of the soil or sediment sample was digested with 12 mL of aqua regia solution (HNO3/HCl, 1:3 ratio) and allowed to stand overnight. Then, the sample was digested at 110 °C for 3 h. The digest was allowed to cool and filtered through Whatman no. 42 paper into a 100-mL standard volumetric flask. The filtrate was analysed for Cu using flame atomic absorption spectrometry, “Shimadzu, 6300.” To evaluate the pollution status of soils before and after the accident, a contamination factor (CF) was calculated using the concentration of copper in each sample compared to its concentration in the control samples. It was also divided into four categories according to previous investigations (Hakanson 1980).

Statistical analysis

Statistical analysis was performed using the Statistical Package for Social Science (version 16.0; SPSS Inc., Chicago, IL, USA). Descriptive statistics of the selected soil and sediment properties, i.e. mean, minimum, maximum, standard deviation, coefficient of variability, skewness, and kurtosis, were calculated. For normality, the original dataset were tested by one-sample Kolmogorov-Smirnov approach (Webster 2001). The relationships between SCC and soil physical-chemical properties were found via Pearson correlations using Eq. (1). The correlation coefficient, r, measures the strength and direction of linear correlation. A value close to +1 or −1 signifies a strong relation; one close to zero signifies a weak one.

$$ r=\frac{\frac{1}{N-1}{\sum}_{i=1}^N\left({Z}_i-\overline{Z}\right)\left({y}_i-\overline{y}\right)}{\sqrt{{s^2}_y\times {s^2}_z}} $$
(1)

where S 2 y and S 2 z are the variances of y and z, as two measured variables, respectively.

Modelling the variogram

The geostatistical analysis was carried out by the software GS+ for the environmental sciences, version 5.1. The datasets were normalized to improve the results of the modelling. Three theoretical models including spherical, Gaussian, and exponential models were adjusted to the semivariograms. The degree of spatial dependence of a random variable Z (x i ) over a certain distance was computed by the following semivariogram (Eq. 2).

$$ \gamma (h)=\frac{1}{2 N(h)}\sum {\left[ z\left({x}_i\right)- z\left({x}_{i+ h}\right)\right]}^2 $$
(2)

where, γ(h) is the semivariance for the interval distance class h, N(h) is the number of pairs of the lag interval, Z (x i ) is the measured sample value at point i, and Z (x i+h ) is the measured sample value at position (i+h). Uniform intervals were defined as a lag class distance intervals. The parameters of the appropriate data model including nugget (C 0), sill (C 0+C), and range of parameter (A 0) were derived with higher R 2 and lower residual sum of squares (RSS). The nugget effect represents the undetectable experimental error and field variation within the minimum sampling space and the sill represents total spatial variations.

Spherical model

The spherical function provides the main features of bounded variogram models (Eq. 3).

$$ \gamma (h)= c\left\{\frac{3 h}{2 a}-\frac{1}{2}\underset{a}{\overset{h}{{\left(-\right)}^3}}\right\}\ \mathrm{for}\ h\le a\kern0.5em \mathrm{and}\kern0.5em \gamma (h)= c\mathrm{for}\ h> a $$
(3)

where c is the sill variance and a is the range.

Exponential model

The exponential function is defined by Eq. 4, where c is the sill as before and r is a distance parameter.

$$ \gamma (h)= c\left\{1- \exp \left(-\underset{r}{\overset{h}{-}}\right)\right\} $$
(4)

Gaussian model

Unlike the two above models which increase from their origins with decreasing gradient, some others (like Gaussian model) appear to increase in gradient from the origin and then curve with decreasing gradient (Eq. 5).

$$ \gamma (h)= c\left\{1- \exp \left(-\frac{h^{\alpha}}{r^{\alpha}}\right)\right\} $$
(5)

where c and r have the same meaning as in the exponential model and α is an additional parameter, with 1 < α ≤ 2. Clearly, if α=1, then the model is exponential. However, α value greater than 1 creates reverse curvature near the origin. The limiting α = 2 defines the so-called Gaussian model.

Fitting the model

There are practical considerations determining the best approach to use in any given circumstance (Henderson et al. 2008). The method of IDW was used to spatial interpolation of all variables in the study area (Lark and Ferguson 2004). The models were fitted based on fundamental hypothesis of the IDW method which assumes that each measured point has a local influence that diminishes with distance. The Geostatistical Analyst Wizard tool was used to digitize the map of the study area. Both IDW and kriging methods are most widely used interpolation tools, but the IDW method was chosen due to limited data (i.e. <50 data points), poor distribution of these across the study area, and simplicity (Elliot et al. 2000). The IDW approach is based on the assumption that the attribute value of an unsampled point is the weighted average of known values within the neighbourhood, and the weights are inversely related to the distances between the prediction location and the sampled locations (Whelan et al. 1996). There are different kinds of IDW according to its power as well as our objective does not meet to apply all of them. Either isotropy hypothesis or IDW^2 was also conducted in this research work.

Spatial interpolation and its uncertainty will be used by automated anisotropic Inverse Distance Weighting Cross-Validation/Jackknife approach (Tipper 2008). In cross-validation, each measured point is removed and compared to the predicted value for that location (Lu and Wong 2008). It is suggested to have field observations to verify the model outputs by adding the coordinate of tested samples on digital maps which is missing in this article.

Geographic information system

Geographic information system (GIS) is a computer-based system which provides data capturing and preparation, data management, manipulation, and presentation. Spatial or georeferenced data can be produced by GIS thematic maps. Therefore, production, dissemination, and use of maps play an important role in cartography. The maps created represent certain levels of detail, which is dependent on the scale. One cannot only compare the results but also offer recommendations for point to point of the study area.

Finally, the data file was transferred to a congenial program, i.e. ArcGIS 10.1, a popular package with excellent graphics. ArcMap as a main application in ArcGIS was used for all mapping and editing tasks in the present study. The area of each mapping unit was then calculated to find the spatial variability of the copper contamination in the field scale. In a digital soil mapping (DSM) project, for example, the soil biological indices have been mapped using a regression-kriging method (Shahbazi et al. 2013).

Results and discussion

Data descriptions and statistical analysis

The statistical descriptive analysis of data is summarized in (Table 1). The high and low values of coefficient of variation (CV) belong to OC content and pH, respectively. All parameters except OC (both depths), SCC, and CEC (0–10 cm depth) had normal distributions. Logarithmic conversion was used to normalize their distributions. The normalized data of OC, SCC, and CEC were back transformed to the original scale in DSM. The analysis of geochemical mobility for the tailings of Sarcheshmeh mine showed that copper had the highest mobility among the heavy metals present (Khorasanipour and Eslami 2014). A point by point observation of total copper concentrations for soils and surficial sediments of the study area are fully presented in a Universal Transfer Mercator (UTM) coordinate system (Table 2). The values of CF in each sampling point at two depths are also presented. The mean total concentration of copper at upper 10 and 10–30 cm depth of the control soils were 49.7 and 48.8 mg kg−1, respectively. These values were used to calculate the contamination factors of copper for the contaminated soils. The mean total concentration of copper at upper 10 cm, 10–30 cm, and surficial sediments was 105.3, 86.4,and 1116 mg kg−1, respectively. The guideline values for copper were established for soils at different textures as follow (in mg Cu kg−1): clay, 100; loam, 60; and sand, 30 (Kabata-Pendias 2011).

Table 1 Descriptive statistics for selected properties of the surficial sediments and the investigated soils at two depths (no 30)
Full size table
Table 2 Total copper concentrations and contamination factors at the sampling points (no 30)
Full size table

Considering the dominance of sandy loam and loam soil textures in the study area, one can use the guideline value of 60 mg kg−1. Therefore, the mean total concentration of copper in upper and lower soil depths was approximately 1.75 and 1.44 times higher than the guideline value. However, the copper content of the control soils is lower than the guideline value. Much higher concentrations of copper (1900 and 925 mg kg−1) were respectively reported in paddy soils of China, around the Dexing Copper Mine, in Jiangxi Province, and the Daye Smelter, in Hubei province (Wu et al. 2011). A significant decrease in the heavy metal contents from surface to depth of contaminated sites has been previously reported (Rodríguez-Tovar and Martín-Peinado 2014). Furthermore, there was a significant decrease in standard deviation values with increasing depth. This indicates that the contamination originated from an influx of surface-transported materials rather than a geological event.

Fairly good direct correlations were found between SCC and OC. Copper can form stable complexes with organic matter (Pandey et al. 2000). Positive correlations between copper concentration and dissolved organic matter were similarly found in the soil percolates of contaminated soils (Kalbitz and Wennrich 1998) and in the soil solutions of some agricultural lands (Romkens and Salomons 1998). The mobilization of Cu and its uptake increased by rapeseed cultivation. Therefore, remediation via application of activated charcoal and phytoremediation using rapeseed may be recommended in the future investigations (Rinkelbe and Shaheen 2015). The results revealed a significant negative correlation between SCC and soil pH at two depths. However, the strength of relationship was higher in the subsurface soils than in the surface ones. The decrease in pH at high contaminated soils may be attributed to the copper hydrolysis. No significant correlation was found between the clay content and SCC in this research work. A positive correlation was found between SCC and EC of the soils at depth of 10–30 cm. This may be attributed to receiving and then leaching of soluble salts from transported materials. The SCC was also negatively correlated with sand content at depth of 0–10 cm. This may be due to low retention capacity of the coarse-textured upper horizons. The finding was in agreement with others (Usman 2008). The high correlation coefficient between SCC and CEC was a consequence of strong correlation between CEC and OC (Table 3).

Table 3 Correlation coefficients between soil properties and SCC
Full size table

Moreover, results showed a significant positive correlation (p < 0.01) between SCC and the thickness of surficial materials in both depths. This is another reason confirming again that the source of copper contamination in the study soils is the tailings dam failure. The results also revealed that the mean value of CFin the surface soil samples near the tailings dam was about 14 which were very high. This is in consistent with findings of the previous reports (Veinott et al. 2003). There are several copper mines in the entire world (Chile, the USA, China, Morocco, etc.). The copper concentration in the tailings of the Dabaoshan (Shaoguan region, China), Jiuhuashan (Anhui province, China), and Anti-Atlas (Southern Morocco) mines were reported 1486, 1899, and 2527 mg kg−1, respectively (Zhou et al. 2007). This concentration for the tailings dam materials in Mazraeh mine was 5870 mg kg-1 which is approximately three times greater than the above-mentioned concentrations. This could be the consequence of low recovery of copper from copper bearing ores. The mean value of CF for surface (0–10 cm) and subsurface soils (10–30 cm) was 2.1 and 1.77, respectively. This indicates the moderate contamination in the study area.

Geostatistical modelling and spatial variability

Spatial variability was investigated using semivariograms. The ratio of nugget to total semivariance, expressed as a percentage, was used to classify spatial dependence. A ratio of <25% indicated strong spatial dependence, between 25 and 75% indicated moderate spatial dependence, and >75% indicated weak spatial dependence (Cambardella et al. 1994). According to the best fitted models (considering higher R 2 and lower RSS) resulted from the geostatistical analyses, the ratio of nugget variance to the sill variance in surficial sediments was lower than soils. Additionally, this ratio for the upper 10 cm of the soils was lower than the next 20 cm. These results illustrated the more spatial dependence of copper distribution in sediments and upper 10-cm soils than in the next 20-cm soils. Either despite the high potential of soils to adsorb contaminants like Cu or these contaminants will all likely enter the surface and groundwater eventually (Jahanshahi et al. 2014). Therefore, the probable transmission of copper into the lower depths or into the runoff must be considered in the design of the proposed dam. Derived parameters from geostatistical analysis are summarized for both soil depths and sediments in Table 4.

Table 4 The best fitted models for semivariograms of the investigated soils at two depths and even surficial sediments using IDW^2 method
Full size table

When data are limited, to predict successfully at unsampled points strongly depends on the relationships between accuracy, sample size, and sample spacing and to what extent these factors are related to the property under investigation (Schloeder et al. 2001). In order to interpolate some of observed soil surface (0–10 cm) properties, the variogram models fitted to the experimental variograms are presented in (Fig. 2). The spherical model was appropriately fitted to interpolate pH variability (Fig. 2a). The similar results were previously reported (Silva et al. 2003). The exponential model was best fitted for interpolating soil clay content in the study area as well as it had been also found by others (Joshua et al. 2014). Moreover, the Gaussian model was also identified as a work horse modelling approach for interpolation the CEC, OC, SCC, and CCS (Fig. 2b–e).

Fig. 2
figure 2

The isotropic variogram models fitted to the selected soil surface properties. a Spherical model for pH. b–d Gaussian model for CEC, OC, and SCC. e CCS

Full size image

Integrating modelling results with GIS

Digital mapping after selecting the most suitable modelling approach is the subsequent process to observe the spatiality distribution. As it appears from the results, the tailings dam has occurred in the north part of study area. The raster graphics of five surface properties were reclassified to four classes using the Geometrical Interval method (Fig. 3a–e). Spatial distribution and overlay analysis showed that the values of all parameters decreased from the north to the south part of area in the direction of flow. Major parts of the released materials with high copper concentration have been percolated in soils near the source and rest of them have been diluted by storm water along the direction flow. Obtained results from geostatistical analysis graphically represented in Fig. 3 are in agreement with the results of correlation analysis (Table 3) that were discussed above.

Fig. 3
figure 3

Created maps of selected soil surface properties according to high accurate modelling. a pH. b CEC/ cmolc kg−1. c OC/ %. d SCC/ mg kg−1. e CCS/ mg kg−1

Full size image

Unfortunately, lacks of precise terrain data and application experience are the most important limitations of applying terrain analysis in low relief areas (Gallant and Dowling 2003). Geostatistical methods cannot substitute conventional soil mapping; they can be an excellent aid for a better understanding of soil distribution. They allow us to consider the soil as a continuum; in this issue, they overcome conventional soil mapping (Wälder et al. 2008). As we know, the topographic profiles of each digital elevation model (DEM) represents the relationships between elevation and soil formation factors as well as processes. In fact, it shows the change in elevation of a surface along a line. Even a very gentle surface relief may exert deep influences on soil properties (McKenzie and Ryan 1999). Here, attention has been focused on the relationships between DEM and selected soil surface properties in the present study (Fig. 4). According to the obtained graphs, the general scheme of pH variations was vastly different compared to other parameters. When the pH was in peak at 2 km away from the source, CEC, OC, and SCC values were at the lowest level. On the contrary, reverse results were observed at 5 km away from the source. The distance profile graphs revealed that there are similarities on distribution patterns of the SCC and CCS in stream pathway from upper to lower part of the study area. Also, general scheme of graphs (Fig. 4e–g) shows that the thickness of surficial sediments in the direction of flow is corresponding to SCC and CCS.

Fig. 4
figure 4

Distance profile graphs of the selected soil surface properties. a DEM/ m. b pH. c CEC/ cmolc kg−1. d OC/ %. e SCC/ mg kg−1. f CCS/ mg kg−1. g thickness of surficial sediments/ cm

Full size image