3.2. Effects of drill cuttings addition on heavy metal concentrations in soils
Studies showed that drill cuttings addition statistically significantly increased the concentration of heavy metals in soil, proportionally to an increased in the share of drill cuttings in mixtures. Cadmium, the concentration of which did not increase following waste addition, and even slightly reduced instead, constituted an exception (Fig. 2).
The highest content of each heavy metal was found in the mixture containing 15% drill cuttings. However, even in this mixture, the content of one of the analyzed heavy metals did not exceed the limit concentrations stated in the Polish law for class II soils, which include agricultural lands (2016), which amount to: for Cd – 2 mg/kg; Cr – 150–500 mg/kg; Cu and Ni 100–300 mg/kg; Pb 100–500 mg/kg; Zn 300–1000 mg/kg.
The concentrations of Cu (0.28–6.82 mg/kg), Ni (7.65–10.26 mg/kg), Pb (10.88 − 13.27 mg/kg), Cd (0.82–1.03 mg/kg), Cr (13.9–17.19 mg/kg) and Zn (62.86–151.75 mg/kg) in the prepared substrates were within the ranges of values measured in Poland in non-contaminated soils, provided by Kabata-Pendias et al. (2010), amounting to: 10–25 mg/kg for copper; from 10 mg/kg (light soils) to 50 mg/kg (heavy soils) for nickel; from 20 mg/kg (light soils) to 60 mg/kg (heavy soils) for lead; from 0.3 mg/kg (light soils) to 2 mg/kg (heavy soils) for cadmium; from 15 mg/kg (medium soils) to 24 mg/kg (heavy soils) for chromium, and 50–100 mg/kg for zinc (Kabata-Pendias 2010).
The regulations in Canada (Directive 50 2016) permit introducing certain types of drilling wastes to soil, provided that the quality standards for these wastes and soils are met. While comparing the concentrations of metals measured in the prepared substrates to the requirements established in Canadian regulations for agricultural soils, to which drilling wastes (expressed in the ratio to dry mass) amounting to Cd – 1.4 mg/kg; Cr – 64 mg/kg; Cu – 3 mg/kg; Pb – 70 mg/kg; Ni – 50 mg/kg; Zn – 200 mg/kg, were introduced, it was noted that they are lower than determined in regulations. Similar results were obtained by Yao et al. (2014), in the studies conducted on the soils with the addition of spent potassium drill fluids (Yao and Naeth, 2014). The standards established in Canadian regulations were not exceeded also in the case of the soils to which spent oil-based frill fluids were added, although elevated mean multi-annual contents of Hg, Pb, Ni, Co, Cu, Cr, and Zn were observed in these soils, compared to the soils without additions (Kisic et al. 2009)
3.3. Effects of drill cuttings addition on plant growth and heavy metals accumulation and translocation in red clover biomass
Changes in the substrate conditions caused by the introduction of drilling wastes to acidic soil, affected the amount and chemical composition of the test plant – red clover (Trifolium pretense) (Fig. 3).
The highest amount of clover biomass was obtained in the case of a mixture with 5% drilling wastes addition; it was 2.5-fold higher in comparison with the control soil. The biomass of clover cultivated on the mixtures with 5% and 10% drill cuttings addition reached twice higher values, whereas on the mixtures with 15% drill cuttings addition, it was 1.5-fold greater than the biomass cultivated on the control soil. Tukey’s test indicated no statistically significant differences between the mass of clover roots cultivated on particular mixtures. However, differentiation occurred in the biomass of clover shoots (Kujawska and Pawłowska, 2020). The heavy metal concentration in the shoots of clover cultivated on the investigated mixtures was determined and the results are shown in Fig. 4.
The drilling wastes addition statistically significantly increased the content of cadmium and copper, whereas it reduced the content of Cr and Pb in clover shoots. No statistically significant changes in the concentrations of nickel and zinc in clover shoots cultivated on the non-modified soil and the soil with drilling wastes addition were observed. The increasing concentrations of these elements in biomass can be attributed to a change in the pH of the substrate, which affected the mobility and bioavailability of these elements (Kgopa et al. 2017). The mobility of these elements increases already at pH 6-6.5 values, slightly acidic and very slightly alkaline, which characterized the prepared mixtures. These values are within the range measured by Reeves and Baker (2000) for plants growing in metalliferous soils (5–25 mg kg− 1) (Reeves and Baker 2000). Cadmium is the most mobile and easily soluble heavy metal (Akhter et al. 2014). Although drill cuttings did not significantly increase the concentration of cadmium in substrates, the clover cultivated on these substrates took up cadmium easily.
The assessment pertaining to the usefulness of the investigated biomass as animal forage was based on maximum heavy metal contents in plants, determined by Kabata-Pendias et al. (1993). They are as follows: Cd – 0.5 mg/kg, Cr – 20 mg/kg; Cu – 30 mg/kg; Ni and Pb − 10 mg/kg; Zn – 100 mg/kg (Kabata-Pendias 1993). The plants cultivated on the prepared mixtures could be used as industrial plants, due to high content of zinc, which prevent them from being used as forage.
In order to evaluate the availability of heavy metals for plants and the capacity of plants for accumulating these metals, the bioconcentration factors (BCF) in shoots and roots of clover cultivated on the prepared mixtures and translocation factors (TF) were determined. Their values were presented in Figs. 5 and 6. Studies showed higher heavy metal accumulation capacity in the roots of plants, than in shoots.
Hyperaccumulation of cadmium was observed in the shoots of the plant cultivated on the soils without waste addition (BCF > 1). The values of BCF in the case of Cr, Pb, Zn, and Cu (for the Z–2.5 and Z–5 mixtures) in the aboveground parts reached the values < 0.06, whereas Ni and Cd were much higher; they were within the ranges of 0.18–0.37 and 0.38–0.79, respectively, which indicates moderate accumulation of these elements in plant shoots. The drill cuttings addition in the amount of 10% and 15% resulted in increased BFC for Cu in plant shoots to the level of 0.15–0.34, indicating its moderate accumulation.
In the case of roots, it was observed that the drill cuttings addition statistically significantly increased the clover capacity for Zn and Ni accumulation in all mixtures. Moreover, statistically significant increase of Cd accumulation in clover roots was observed, but only for the highest, 15% drill cuttings dose. However, the values of BCF for Cd were the highest, compared to other metals, and in all collected plants, the BCF value of this element was higher than 2. Such high bioconcentration factor was observed only in the case of Pb on the control sample. In turn, the BCF values higher than 1 were observed in root biomass in the case of Ni in the plants cultivated on the substrates containing drill cuttings (Fig. 6). It was observed that clover roots accumulated (0.48–0.52), Pb (0.20–0.45), Cu (0.16–0.19) i Zn (0.29–0.34) to a moderate degree.
It was observed that the drill cuttings addition to the substrate changed the metal accumulation capacity in particular plant parts. The accumulation of metals in the below-ground parts of plants cultivated on the control soil can be presented in the following order: Cd > Ni > Pb > Zn > Cr > Cu, whereas in roots, it is slightly different: Pb > Cd > Cr > Ni > Cu > Zn. After the highest drill cuttings addition the order was as follows: Cd > Cu > Ni > Cr > Pb > Zn in the above-ground parts and Cd > Ni > Cr > Zn > Pb > Cu in roots.
Accumulation of the investigated elements in clover roots was much higher than in the above-ground parts, which indicates the usefulness of this species in phytostabilization of polluted soils. In addition, introduction of 5% drill cuttings improved the growth conditions, which increased the accumulation of metals in roots and reduced their transport to the above-ground parts, which is especially evident in the case of cadmium, nickel, and zinc.
Although BCF can be a useful tool for assessing the influence of waste addition to soil on the accumulation of elements in biomass; however, interpretation of the obtained results is not easy, since bioaccumulation of metal by plants is dependent upon numerous factors, including variable soil conditions. As it was observed by McGrath and Zhao (2003), the values of BCF generally decrease with increasing metal concentration in soil (McGrath and Zhao 2003).
The obtained translocation factor (mobility) values of heavy metals in the clover cultivated on the mixtures with drilling wastes addition decreased under their influence (Table 3). All the determined TF values were lower than 1; hence, the mobility of metals in the root–above-ground part system of clover was very low. The reason for lower translocation of metals in the substrates containing drill cuttings might be alkalinization, which causes retention of metals in the root system. This phenomenon was also described by Kumpiene (2007) relating to the soils in which alkalinization occurred as a result of external organic matter addition (Kumpiene et al. 2007).
3.4. Artificial Neural Networks
On the basis of the experimental data: drilling waste doses, pH, organic matter of the mixtures with drill fluids, 100 networks were developed. The network quality was assessed using the following indicators: quality of training, quality of validation, training error and validation error from the least squares method, in order to select the most appropriate network type – MLP or RBF. The obtained network parameters were presented in Table 5.
Table 5
Characteristics of multi-layered perceptron (MLP) and radial basis function (RBF) networks for heavy metals
Network No.
|
Network Name
|
Quality (Training, %)
r
|
Quality (Validation, %)
|
Error (Training)
|
Error (Validation)
|
Activation (Hidden)
|
Activation (Output)
|
Cr
|
1
|
MLP3-10-1
|
99.26
|
97.11
|
0.008
|
0.002
|
Tanh
|
Sinus
|
2
|
MPL 3-6-1
|
99.74
|
91.57
|
0.004
|
0.008
|
Tanh
|
Tanh
|
3
|
RBF 3-7-1
|
98.37
|
94.04
|
0.001
|
0.008
|
Gaussian
|
Linear
|
Ni
|
4
|
MLP 3-3-1
|
99.99
|
99.90
|
< 0.001
|
0.004
|
Exponential
|
Logistic
|
5
|
MLP 3-6-1
|
98.13
|
99.98
|
< 0.001
|
< 0.001
|
Gaussian
|
Linear
|
6
|
RBF 3-8-1
|
98.97
|
99.99
|
< 0.001
|
0.001
|
Tanh
|
Sinus
|
Pb
|
7
|
RBF 3-9-1
|
99.71
|
99.72
|
0.002
|
0.001
|
Gaussian
|
Linear
|
8
|
MLP 3-7-1
|
99.97
|
99.84
|
0.001
|
< 0.001
|
Tanh
|
Ligostic
|
9
|
RBF 3-8-1
|
98.97
|
99.97
|
< 0.001
|
< 0.001
|
Gaussian
|
Linear
|
Cd
|
10
|
RBF 3-5-1
|
94.88
|
99.96
|
0.003
|
0.009
|
Gaussian
|
Linear
|
11
|
RBF 3-7-1
|
99.05
|
94.32
|
0.002
|
0.004
|
Gaussian
|
Linear
|
12
|
MPL 3-8-1
|
93.46
|
91.78
|
0.005
|
0.012
|
Sinus
|
Logistic
|
Cu
|
13
|
MLP 3-9-1
|
98.17
|
97.84
|
0.014
|
0.004
|
Sinus
|
Tanh
|
14
|
RBF 3-6-1
|
95.06
|
99.88
|
0.004
|
0.009
|
Gaussian
|
Linear
|
15
|
MPL3-10-1
|
99.97
|
99.94
|
< 0.001
|
0.003
|
Logistic
|
Logistic
|
Zn
|
16
|
RBF 3-2-1
|
99.05
|
99.99
|
< 0.001
|
0.003
|
Gaussian
|
Linear
|
17
|
MPL 3-7-1
|
99.96
|
99.65
|
< 0.001
|
0.012
|
Logistic
|
Logistic
|
18
|
MPL 3-3-1
|
99.87
|
99.88
|
< 0.001
|
< 0.001
|
Exponential
|
Tanh
|
Figure 7 shows a comparison of experimental data obtained via predicting the concentrations of selected metals in plants for selected networks.
Regression coefficients (R) for selected networks for training, validation, and test data assume the values over 90%. Such high regression coefficients indicate good fit of the network. As it was presented in Table 5, mean coefficients of correlation between the experimentally determined concentrations of heavy metals in plants and the values predicted by ANN reached were higher than 95%, which indicates that the ANN model was able to quickly and reliably predict the concentrations of heavy metals. Low values of errors (< 0.1) also prove high accuracy of neural networks.
In order to investigate the influence of drill cuttings addition, pH, and organic matter content of soil on the concentration of heavy metals, a sensitivity analysis (Table 6) was carried out. The network sensitivity analysis indicated the highest sensitivity to the impact of drill cuttings addition on the concentration of heavy metals in plants.
Table 6
Sensitivity analysis of the artificial neural network (mean values)
|
Share of wastes
|
pH
|
SO
|
Cd
|
17.26
|
13.54
|
11.60
|
Cr
|
214.26
|
179.77
|
51.70
|
Cu
|
82.34
|
27.29
|
5,84
|
Ni
|
24.45
|
17.11
|
16.68
|
Pb
|
13.39
|
8.54
|
0.68
|
Zn
|
72.97
|
23.41
|
1.23
|
On the basis of the obtained experimental results, an artificial neural network model for predicting the metal concentrations in plants. Such models can be created using various soil additives and soil quality parameters, which facilitates predicting the impact of wastes on the accumulation of metals in plants.
The results obtained by us and other researchers showed that ANN can be employed for predicting the concentrations of metals in plants,
Haatab et al. (2013) created an artificial neural network model for predicting the chromium concentration in the leaves of laboratory-cultivated dwarf French bean, in the soils with the addition of dolomite limestone, compost of poultry manure and pine bark (CPM), as well as mixture of dolomite limestone and compost of poultry manure and pine bark. The input data used in the model are: soil amendments, soil pH, electric conductivity and dissolved organic carbon in soil, and the obtained result is the concentration of Cr in dwarf French bean. Their ANN model indicated mean coefficient of correlation between the measured and predicted chromium values in dwarf French bean equal to 0.9998 (Hattab et al., 2013).
Jahantab et al. trained the neural network for Zn and Cr heavy metals in soil and plant, with the R2 value in most cases higher than 0.9 and close to 1, indicating the applicability of neural network for over-predicting data (Jahantab et al. 2020).
Gharaibeh and Ben-Hani (2003) created an artificial neural network model for predicting phytotoxicity, dry mass accumulation and reduction depending on the concentrations of metals used for irrigation. The input (selenium and nitrate levels) and the output data (growth reduction and selenium bio-tissues uptake) were simulated using artificial neural network program. Simulated data was then used to predict the interaction between selenium and nitrate in irrigation water at different levels of both nitrates and selenium (Gharaibeh and Bani-Hani 2003).
The versatility of artificial neural network tools is the feature, which enables to account for the selected quantity and quality of the investigated soil quality parameters. Moreover, artificial neural network models may be the basic tool for managers, engineers, and decision makers, aiding in designing, managing, and making decisions pertaining to the introduction of additives to soil.