Environmental Pollution 236 (2018) 366e372

Contents lists available at ScienceDirect

Environmental Pollution
journal homepage: www.elsevier.com/locate/envpol

Quadratic discriminant analysis model for assessing the risk of

cadmium pollution for paddy fields in a county in China*
Xiumei Wang a, Xiujian Li b, Ruoyu Ma b, Yue Li b, Wei Wang b, Hanyu Huang a,
Chenzi Xu a, Yi An a, *
a
Agro-environmental Protection Institute, Ministry of Agriculture, Tianjin 300071, China
b
Department of Computer and Control Engineering, Nankai University, Tianjin 300071, China

a r t i c l ein fo abstract

Article history: In China, the cadmium (Cd) levels in paddy fields have increased, which has led to the excessive uptake of
Received 15 September 2017 Cd into rice grains. In this study, we determined the physicochemical properties of soil samples, including
Received in revised form the pH, soil organic matter (SOM) content, cation exchange capacity (CEC), and total Cd content (Cd soil) in
21 January 2018
order to establish a quadratic discriminant analysis (QDA) model for assessing the risk of Cd in rice and
Accepted 27 January 2018
to calculate its prior probability. Decision tree and logistic regression models were also estab - lished for
comparison. The results showed that the accuracy rate was 74% with QDA, which was significantly higher
than that obtained using the decision tree (67%) and logistic regression (68%) models. The correlation
Keywords:
coefficients between the soil pH and the other three factors (CEC, SOM, and Cd soil) were higher in the
Cadmium
Decision tree inaccurate set than the accurate set, whereas the correlation coefficients were smaller in the inaccurate
Logistic regression set than the accurate set.
Paddy field © 2018 Elsevier Ltd. All rights reserved.
Quadratic discriminant analysis model

1. Introduction complexes with humic substances (Halim et al., 2015; Xu et al.,

The pollution of arable soils with heavy metals is a great risk for
agricultural production (Bermudez et al., 2012). In China, pollution
by cadmium (Cd) is particularly important among the different types
of heavy metal pollution because a large area is contaminated (about
7.0% of arable soil) and there is great risk of it being trans- ferred
from the soil into rice (Liu et al., 2014), which is the major crop
produced in China.
Many studies have attempted to predict the risk of Cd trans-
ferring from soil into rice based on the physicochemical properties
of soil, such as the pH, soil organic matter (SOM) contents and cation
exchange capacity (CEC). According to lots of literature, these
properties can change the soil Cd species and then affect its phy-
toavailability. Among them, the accumulation of Cd in rice decreased
with CEC (Sebastian and Prasad, 2014; Honma et al., 2016) and pH
(Miller et al., 1976), while the effect of SOM to Cd phytoavailability is
a little complicated. The SOM can reduce the Cd phytoavailability in
soil through adsorption or forming stable
*
This paper has been recommended for acceptance by Dr. Chen Da.
* Corresponding author. Tel.:þ86 22 23610159; fax: þ86 22 23610159.
E-mail address: simon8601@126.com (Y. An).
2010), but also supply organic chemicals, acting as chelates, to the
soil solution and enhance Cd availability to rice (Halim et al., 2015;
Zhao et al., 2015). The models used previously for predict- ing risk
are based on decision trees, linear or logistic regression, and neural
networks. Regression algorithm can state the relationships between
soil inputs (soil properties) and soil output (heavy metal content)
(Ro€mkens et al., 2009; Elzinga et al., 1999), while these
formulated relationships are not necessary for decision tree (Vega et
al., 2009, 2010) and neural networks models (Hattab et al., 2013;
Liu et al., 2011). However, few of these models obtained ac- curate
quantitative predictions of the Cd concentration in rice. This may be
because other factors affect the transfer process, such as different
crop varieties, hydrological conditions, meteorological factors, and
farming methods (Song et al., 2015; Ye et al., 2012; Yu et al., 2014; Hu
et al., 2015).
Excessive predicting indexes can lower the accuracy and in-
crease the monitoring costs. Thus, Geisser (1964) suggested using
quadratic discriminant analysis (QDA) to obtain posterior distri-
butions for estimating the classes of predictions in order to simplify
the predicted index as a prior probability (Khasnobish et al., 2011).
The discriminant function is given by

https://doi.org/10.1016/j.envpol.2018.01.088
0269-7491/© 2018 Elsevier Ltd. All rights reserved.
 X. Wang et al. / Environmental Pollution 236 (2018) 366e372
1 1
dk ðxÞ ¼ log pðkÞ — log.Sc . — ðx — mc ÞT S—c 1 ðx — mc Þ;
2 2
where k is the predicted class, x is the set of measurements, mc is
the mean vector, p(k) is the prior probability, and Sc is the
covariance matrix.
In the present study, we developed a QDA model for assessing the
risk of Cd in rice based on the total Cd content of the soil (Cdsoil),
physicochemical properties (pH, SOM, and cation exchange ca-
pacity (CEC)), and the calculated prior probability. We also estab-
lished decision tree and logistic regression models for comparison.
2. Materials and methods
2.1. Survey of study area
The study area (Fig. 1) was located in Xiangtan county, Hunan
Province, which is a typical area with paddy fields polluted by Cd
according to previous studies (Wang et al., 2015; Lei et al., 2015).
Previous research (Williams et al., 2009; Wei et al., 2009; Du et al.,
Fig. 1. Map of Xiangtan County showing the locations of the sampling plots.
367
2013) has shown that the Cd in this region is mainly derived from
the metallurgical industry, chemical industry, and sewage
(1)
irrigation.
In addition, severe Cd pollution in the soil means that there is a
high risk of Cd absorption by paddy rice, which is the dominant
crop (7.09×105 t/y of rice from 1.35 ×105 ha of land) in this region
because of the suitable subtropical humid monsoon climate with an
average annual temperature of 17.4 ○ C and mean annual precipi-
tation of 1300 mm.
2.2. Locations of sampling plots
In total, 536 soilerice paired samples were collected during
2012e2014 in Xiangtan County (126 pairs in 2012, 360 pairs in 2013,
and 50 pairs in 2014). The layout of the plots was fairly random (as
shown in Fig. 1) but the sampling density was higher in heavily
polluted towns (historically), including Yangjiaqao Town, Hekou
Town, Meilinqiao Town, Tanjiashan Town, and Yisuhe Town. Soil
samples were taken from the topsoil (0e25 cm) using a stainless steel
auger at the harvest time when the paddy field was drained. Rice
samples were also collected from the sampling sites.
368 X. Wang et al. / Environmental Pollution 236 (2018) 366e372

2.3. Sample pretreatment and analysis

K
X
○C, GiniðpÞ ¼ pkð1 — pkÞ; (3)
The soil samples were air-dried in the laboratory at 25e30
k¼1
before removing gravel, plant roots, and other foreign matter, and
they were divided into two subsamples. One subsample was ground Where K is the number of categories and pk is the probability when
in an agate mortar until it passed through a 2-mm mesh nylon sieve, the category is K.
before it was used for determining the pH and CEC. The second In addition, the formula used for fitting the logistic regression is
subsample was ground until it passed through a 100-mesh nylon given by Eq. (4), where the gradient descent method is used to obtain
sieve, before it was used for determining the SOM contents and the maximum likelihood value for the parameter q:
Cdsoil.
1
A grain huller was used to remove the outer shell of the rice h ðx ; (4)
Þ¼ T

samples and obtain brown rice grains. The rice grains were then q
1 þ e—q x
cleaned three times with deionized water, oven dried to constant

weight at 60 ○C, and passed through a 0.25-mm sieve. where q is the mean vector comprising the soil pH, SOM, Cd soil, and
The pH of each soil solution was measured potentiometrically CEC.
(water:soil slurry ratio ¼1:2.5) using a pH meter (PB-10, Sartorius,
Go€ttingen, Germany). The SOM contents were determined using
2.4.3. Model verification
the WalkleyeBlack method (Nelson et al., 1982). The CEC was
Finally, in order to calculate the accuracy of the data fitting by
measured using 1 M NH4OAc buffered at pH 7.0 (USDA, 1996),
the QDA, decision tree model, and logistic regression model, 30
where a soil sample weighing 2 g was digested with an acid
groups of data were randomly selected from the testing set and
mixture containing concentrated HNO3, HF, and HClO4 at a ratio of
repeated 50 times to compare the effectiveness of the model fitting
1:2:1, and then analyzed with a flame atomic absorption spectro-
procedure.
photometer (AA240, Varian, Palo Alto, CA, USA). A brown rice grain
sample weighing 1 g was digested using an acid mixture containing
concentrated HNO3 and HClO4 at a ratio of 9:1, before analyzing 3. Results and discussion
with an atomic absorption spectrometer (AA800, PerkinElmer,
Norwalk, CT, USA) in a graphite furnace. 3.1. Regional monitoring results and classic assessment model

2.4. Model description In general, a higher value for Cd soil can lead to a greater value for
Cdrice. However, a clear relationship between Cd soil and Cdrice has not
In this study, all of the models were constructed using Matlab been reported previously (Li et al., 2003). To verify this rela-
2012a, and statistical analyses were performed with Origin 9.0 and tionship, Cdsoil is plotted versus Cdrice in our study region in Fig. 2.
IBM SPSS Statistics 20.0 software. For the data shown in Fig. 2, Pearson's correlation coefficient was
only 0.168, thereby indicating the lack of a linear correlation
between Cdsoil and Cdrice in this study. We also used the Cd con-
2.4.1. QDA model
centrations comprising the limitation II level standard of the Chi-
We used the data set obtained in 2012 to calculate the prior
nese Soil Environmental Quality Standard (SQS) and the Chinese
probability for the QDA model, and the data sets from 2013 to 2014
Food Safety Standard (FQS) as cut-off rules to obtain four quad-
were employed as the training set and testing set, respectively. It
rants: N1 where Cdsoil > SQS and Cdrice > FQS; N2 where Cdsoil ≤ SQS
should be noted that the number of data used in the training set was
and Cdrice > FQS; N3 where Cdsoil ≤ SQS and Cdrice ≤ FQS; and N4
actually 330 after eliminating noise and outliers.
where Cdsoil > SQS and Cdrice ≤ FQS.
The QDA model is given by Eq. (2):
1
h i
p j2pS j— 2 exp — 1ðx — m ÞT S—1ðx — m Þ 10
i i 2 i 0 i
Critical line y = 0.724+0.663x
i¼0 ij Sij 2
— 2ð — iÞ Si ð — iÞ R2 = 0.168

pðy ¼ ijx;qÞ¼ P n h io ;
—1 1x
Cd concentration in brown rice / FQS

i¼1
p 2p exp m T —1 x m
(2) 8

(2) 8
N2=12 N1=309

where the value of i is 0 when the Cd concentration in rice

(Cdrice) ≤0.2 mg kg—1 and 1 when Cdrice > 0.2 mg kg—1; p is the 6
prior probability; m is the mean vector comprising the soil pH, SOM,
Cdsoil, and CEC; and S is the covariance matrix for the soil pH, SOM,
Cdsoil, and CEC. 4
Critical line

The cycle was repeated until the complete decision tree was generated
N =11 N =204
2.4.2. Comparison of models and it was then clipped to obtain an optimal deci- sion tree:
3 4

Decision tree and logistic regression models were also estab-

lished to compare with the QDA. It was not necessary to consider the
prior probability, so data from 2012 were not used in these two types
of models.
For the decision tree model, the Gini coefficient (Eq. (3)) of the
probability was calculated for each variable and the variable with the
smallest Gini coefficient was selected as the classification condition.
 2

0
0 2 4 6 8 10
Cd concentration in soil / SQS

Fig. 2. Distributions of Cd in soil (Cdsoil) and Cd in rice grains (Cdrice). The vertical
line is a critical line showing whether Cdsoil exceeded the Chinese Soil
Environmental Quality Standard (SQS). The horizontal line is a critical line
indicating whether Cdrice exceeded the Chinese Food Safety Standard (FQS).
 X. Wang et al. / Environmental Pollution 236 (2018) 366e372
We found that NN1 þN3
¼ 1.48, which suggests that the concen-
2þN4
trations of Cd in both the soil and rice relative to the safety stan-
dards differed in nearly 40% of the paired samples.
Previous studies have suggested that Cd rice is correlated well
with Cdsoil when certain soil physicochemical properties (e.g., pH,
CEC, and SOM) are considered together (Ro€mkens et al., 2009; Brus
et al., 2005; Japenga et al., 2007; Kirkham, 2006). Thus, we also
determined the values of the soil pH, CEC and SOM to build models,
as shown in Table 1.
According to the results, the soil pH ranged from 4.6 to 7.8 with
an average of 5.6. The average values for CEC and SOM were
45.07 ± 5.13 and 12.70 ± 1.32, respectively. The results of the cor-
relation analysis showed that Cdrice was significantly correlated with
the pH, SOM, CEC, and Cdsoil at p < 0.01.
Two classic models were employed in this study to investigate the
relationship between Cdsoil (with pH, CEC, and SOM) and Cdrice. The
first was a logistic regression model where after repeated training,
the equation was obtained as follows.
1
h ðx
Þ¼
q
1 þ EXPð8:0529 — 1:5615pH þ 0:1404SOM þ 0:2804Cdsoil — 0:4127CECÞ
The second was a decision tree model, which is a classifier and
not a mathematical model. Thus, there was no equation for the
decision tree model and its classification process is shown in Fig. 3
instead.
According to tests of 50 groups, the accuracy rate for the deci-
sion tree model was 67% and that for the logistic regression model
was 68% (Fig. 4). Both of these percentages were larger than 60%,
which was the proportion represented by (N1þN3) in the whole
samples.
3.2. QDA model for Cd risk assessment
Many other factors could be introduced to further improve the
accuracy of the assessment models, where some region-specific
factors might affect the transfer of Cd from the soil to food crops,
including the crop species or breed, meteorological conditions, soil
physicochemical properties (excluding the pH, SOM, and CEC, such
as Fe, Mn, Si, Ca, and Se), agricultural management practices, and
geographical differences (Zhao et al., 2015; Song et al., 2015; Ye et
al., 2012; Kheir et al., 2010; Sarwar et al., 2010). Xu et al. (2016).
suggested that certain prior conditions can be used to enhance the
reliability of a model for assessing the risk of Cd accumulation in
rice.
In this study, we used the historical rate in this region of Cd rice
exceeding the specified standard as the prior probability. Therefore,
Table 1
pH, soil organic matter (SOM), cation exchange capacity (CEC), Cd in soil (Cd soil), and Cd in rice grains (Cd rice) in each year. Minimum, maximum, and average (in brackets)
values are presented.
Year pH SOM
(g kg—1)
2012 4.6e7.8 (5.5) 28.3e71.1 (44.3)
2013 4.8e7.7 (5.7) 36.6e55.1 (45.2)
2014 4.8e6.3 (5.4) 38.4e52.1 (45.9)
* ND: Indicates lower than the detection limit (0.0284 mg kg —1).
369
the value of p was calculated with the following equation using the
data from 2012:
Ni
p ¼ ; (6)
i
N
where the value of i is 0 when Cdrice 0.2≤mg kg—1 and 1 when Cdrice
> 0.2 mg kg—1, and N is the number of samples in 2012.
In order to assess the risk of Cd in rice, a QDA model was
established by analyzing whether the probability of Cdrice exceeding
FQS was larger than 50%. After repeated training, the parameters
obtained for the QDA model are shown in Table 2.
The different QDA parameters for the groups where Cdrice
0.2 mg≤kg—1 and Cdrice > 0.2 mg kg—1 can be explained by the different
pathways comprising the links among the influential factors. Thus,
path analysis was performed, as shown in Fig. 4. For the group where
Cdrice exceeded the limit, pH and Cdsoil directly
affected Cdrice, while pH, SOM, and CEC indirectly affected Cd rice via
Cdsoil. For the group where Cdrice did not exceed the limit, pH, SOM,
and CEC directly affected Cdrice, while pH and SOM indirectly
affected Cdrice via CEC. However, Cdsoil had no direct effect on Cdrice,
(5)
and it indirectly affected Cdrice via pH, SOM, and CEC.
According to the results obtained by the QDA model, the accu-
racy rate was 74% (Fig. 5), which was higher than that for the de-
cision tree or logistic regression models. Similar conclusion could be
drawn by comparing to the researches of Xu et al. (2016), Adams et
al. (2004) and Guo et al. (2007). Also, the accuracy rate of QDA
model was almost the same with some neural network models (Liu
et al., 2011).
In terms of the model's stability, the standard deviation of the
accuracy rate was 0.032 with QDA, and the results were 0.035 and
0.025 for the decision tree and logistic regression models, respec-
tively. Thus, there were no significant differences among the three
models according to the 95% confidence intervals.
We also analyzed the characteristics of the error data for the QDA
model. Fig. 6 shows the sites where the QDA model obtained the
opposite results compared with the actual results.
According to the results obtained by analysis of variance, there
were no significant differences (p < 0.05) in the pH, CEC, SOM, or
Cdsoil in the accurate set and inaccurate set for the QDA model. In
addition, the correlation analysis results (Table 3) showed that the
correlation coefficients between the soil pH and the other three
factors (CEC, SOM, and Cdsoil) were higher in the inaccurate set than
the accurate set, whereas the correlation coefficients were smaller in
the inaccurate set than the accurate set. This indicates that the QDA
model might make an incorrect prediction if the relationship
CEC Cdsoil Cdrice grains
(cmol[þ] kg— 1) (mg kg—1 ) (mg kg—1 )
7.49e20.29 (12.76) 0.13e7.70 (0.64) ND*e3.20 (0.38)
10.94e14.19 (12.69) 0.17e2.80 (0.60) NDe3.31 (0.40)
10.97e14.16 (12.62) 0.30e1.70 (0.71) 0.04e3.31 (0.61)
370 X. Wang et al. / Environmental Pollution 236 (2018) 366e372

Fig. 3. Decision-tree for predicting Cd in rice grains (Cdrice) based on soil properties (where Cds is Cd in soil and Cdr is Cdrice).

Fig. 4. Path analysis for soil properties with different limits on Cd in rice grains.

Table 2
Parameters in the quadratic discriminant analysis (QDA) model. Cd soil: Cd in soil and Cdrice: Cd in rice grains.

Parameter Cdrice ≤ 0.2 mg kg— 1 Cdrice > 0.2 mg kg— 1

p 0.365 0.635
2 3 2 3
S 0:384 0:386 0:0469 0:0127 0:126 —0:0120 0:00690 0:0293
0:386 10:9 0:0820 0:468 —0:0120 12:2 0:403 0:383 7
6 7 6
4 0:0469 0:0820 0:0172 —0:0125 5 4 0:00690 0:403 0:0424 —0:0408 5
0:0127 0:468 —0:0125 0:289 0:0293 0:383 —0:0408 0:263

m ð5:83; 47:2; 0:494; 12:9Þ ð5:49; 45:9; 0:591; 12:7Þ

 X. Wang et al. / Environmental Pollution 236 (2018) 366e372 371

Table 3
Analysis of correlations between soil properties in the accurate set and inaccurate
set. SOM: soil organic matter; Cd soil: total Cd content in soil; CEC: cation exchange
capacity.

Category Variable pH SOM Cdsoil CEC

Accurate set pH 1 0.170 0.234 0.094

SOM 1 0.324 0.179
Cdsoil 1 —0.042
CEC 1
Inaccurate set pH 1 0.337 0.498 0.335
SOM 1 0.258 0.021
Cdsoil 1 —0.223
CEC 1

Fig. 5. Accuracy rates for the quadratic discriminant analysis (QDA), decision tree
(DTA), and logistic regression (LRA) models after testing 50 times.

Fig. 6. Assessment of the results obtained by the quadratic discriminant analysis model using data from 2014.
372 X. Wang et al. / Environmental Pollution 236 (2018) 366e372
between the soil pH and other physicochemical properties is
stronger.
4. Conclusions
In this study, was established a QDA model for assessing the risk
of paddy fields being polluted by Cd. We can make the following
conclusions based on the results obtained.
(1) According to tests based on 50 groups, the accuracy rate was
74% using QDA, but only 67% and 68% with the decision tree
and logistic regression models, respectively. There were no
significant differences in the stability of the three models.
(2) The correlation coefficients between the soil pH and the other
three factors (CEC, SOM, and Cdsoil) were higher in the
inaccurate set than the accurate set, whereas the correlation
coefficients were smaller in the inaccurate set than the ac-
curate set.
Acknowledgement
This work was funded by the Special Fund for Agro-scientific
Research in the Public Interest (No. 201403014) and Agricultural
science and technology innovation project (No. CAAS-ASTIP-2017-
AEPI-lyj).
References
Adams, M.L., Zhao, F.J., McGrath, S.P., Nicholson, F.A., Chambers, B.J., 2004. Pre-
dicting cadmium concentrations in wheat and barley grain using soil proper- ties.
J. Environ. Qual. 33, 532e541.
Bermudez, G.M., Jasan, R., Plá, R., Pignata, M.L., 2012. Heavy metals and trace ele-
ments in atmospheric fall-out: their relationship with topsoil and wheat
element composition. J. Hazard Mater. 214, 447e456.
€ mkens, P.F., 2005. Probabilistic quality standards for
Brus, D.J., Gruijter, J.J.D., Ro
heavy metals in soil derived from quality standards in crops. Geoderma 128,
301e311.
Du, Y., Hu, X.F., Wu, X.H., Shu, Y., Jiang, Y., Yan, X.J., 2013. Affects of mining activities
on Cd pollution to the paddy soils and rice grain in Hunan province, Central South
China. Environ. Monit. Assess. 185, 9843e9856.
Elzinga, E.J., Van Grinsven, J.J.M., Swartjes, F.A., 1999. General purpose Freundlich
isotherms for cadmium, copper and zinc in soils. Eur. J. Soil Sci. 50, 139e149.
Geisser, S., 1964. Posterior odds for multivariate normal distributions. J. Royal Soc.
Series B Methodol. 26, 69e76.
Guo, H.Y., Liu, T.S., Chu, C.L., Chiang, C.F., Ro€ mkens, P.F., 2007. Prediction of Heavy
Metal Uptake by Different Rice Species in Paddy Soils Near Contaminated Sites of
Taiwan.
Halim, M.A., Majumder, R.K., Zaman, M.N., 2015. Paddy soil heavy metal contami-
nation and uptake in rice plants from the adjacent area of Barapukuria coal mine,
northwest Bangladesh. Arab. J. Geosci. 8, 3391e3401.
Hattab, N., Hambli, R., Motelica-Heino, M., Bourrat, X., Mench, M., 2013. Application
of neural network model for the prediction of chromium concentration in
phytoremediated contaminated soils. J. Geochem. Explor. 128, 25e34.
Honma, T., Ohba, H., Kanekokadokura, A., Makino, T., Nakamura, K., Katou, H., 2016.
Optimal soil eh, ph, and water management for simultaneously minimizing arsenic
and cadmium concentrations in rice grains. Environ. Sci. Technol. 50, 4178.
Hu, P., Ouyang, Y., Wu, L., Shen, L., Luo, Y., Christie, P., 2015. Effects of water man-
agement on arsenic and cadmium speciation and accumulation in an upland rice
cultivar. J. Environ. Sci. 27, 225e231.
€ mkens, P.F., 2007. A feasibility test to estimate
Japenga, J., Koopmans, G.F., Song, J., Ro
the duration of phytoextraction of heavy metals from polluted soils. Int. J.
Phytoremediat 9, 115e132.
Khasnobish, A., Bhattacharyya, S., Konar, A., Tibarewala, D.N., Nagar, A.K., 2011. Two-
fold Classification for Composite Decision about Localized Arm Movement from
EEG by SVM and QDA Techniques, International Joint Conference on Neural
Networks. IEEE, pp. 1344e1351.
Kheir, R.B., Greve, M.H., Abdallah, C., Dalgaard, T., 2010. Spatial soil zinc content
distribution from terrain parameters: a GIS-based decision-tree model in Lebanon.
Environ. Pollut. 158, 520e528.
Kirkham, M.B., 2006. Cadmium in plants on polluted soils: effects of soil factors,
hyperaccumulation, and amendments. Geoderma 137, 19e32.
Lei, M., Tie, B.Q., Song, Z.G., Liao, B.H., Lepo, J.E., Huang, Y.Z., 2015. Heavy metal
pollution and potential health risk assessment of white rice around mine areas in
Hunan Province, China. Food. Secur. 7, 45e54.
Li, F., Okazaki, M., Zhou, Q., 2003. Evaluation of Cd uptake by plants estimated from
total soil Cd, pH, and organic matter. Bull. Environ. Contam. Toxicol. Environ.
Contam. Tox 71, 0714e0721.
Liu, M., Liu, X., Wu, M., Li, L., Xiu, L., 2011. Integrating spectral indices with envi-
ronmental parameters for estimating heavy metal concentrations in rice using a
dynamic fuzzy neural-network model. Comput. Geosci. 37, 1642e1652.
Liu, J.G., Qu, P., Zhang, W., Dong, Y., Li, L., Wang, M.X., 2014. Variations among rice
cultivars in subcellular distribution of cd: the relationship between trans- location
and grain accumulation. Environ. Exp. Bot. 107, 25e31.
Miller, J.E., Hassett, J.J., Koeppe, D.E., 1976. Uptake of cadmium by soybeans as
influenced by soil cation exchange capacity, pH, and available phosphorus.
J. Environ. Qual. 52, 157e160.
Nelson, D.W., Sommers, L.E., Sparks, D.L., Page, A.L., Helmke, P.A., Loeppert, R.H.,
1982. Total Carbon, Organic Carbon, and Organic Matter. Methods of Soil Analysis
Partdchemical Methods, pp. 961e1010.
€ mkens, P.F., Guo, H.Y., Chu, C.L., Liu, T.S., Chiang, C.F., Koopmans, G.F., 2009.
Ro
Prediction of cadmium uptake by brown rice and derivation of soil-plant
transfer models to improve soil protection guidelines. Environ. Pollut. 157,
2435e2444.
Sarwar, N., Saifullah, M., Malhi, Z.S.S., Zia, M.H., Naeem, A., Bibi, S., 2010. Role of
mineral nutrition in minimizing cadmium accumulation by plants. J. Sci. Food Agr.
90, 925e937.
Sebastian, A., Prasad, M.N.V., 2014. Cadmium minimization in rice. A review. Agron.
Sustain. Dev. 34, 155e173.
Song, W.E., Chen, S.B., Liu, J.F., Li, C., Song, N.N., Ning, L.I., 2015. Variation of Cd
concentration in various rice cultivars and derivation of cadmium toxicity
thresholds for paddy soil by species-sensitivity distribution. Integr. Agric 14 (9),
1845e1854.
USDA, Soil, 1996. Survey Laboratory Methods Manual. Soil Survey Investigations,
Washington, DC.
Vega, F.A., Matías, J.M., Andrade, M.L., Reigosa, M.J., Covelo, E.F., 2009. Classification
and regression trees (carts) for modelling the sorption and retention of heavy metals
by soil. J. Hazard Mater. 167, 615e624.
Vega, F.A., Andrade, M.L., Covelo, E.F., 2010. Influence of soil properties on the
sorption and retention of cadmium, copper and lead, separately and together,
by 20 soil horizons: comparison of linear regression and tree regression ana-
lyses. J. Hazard Mater. 174, 522e533.
Wang, M., Chen, W., Peng, C., 2015. Risk assessment of Cd polluted paddy soils in the
industrial and township areas in Hunan, Southern China. Chemosphere 144,
346e351.
Wei, C., Cheng, W., Yang, L., 2009. Characterizing spatial distribution and sources of
heavy metals in the soils from mining-smelting activities in Shuikoushan, Hunan
Province, China. J. Environ. Sci. 21, 1230e1236.
Williams, P.N., Ming, L., Sun, G.X., Huang, Q., Lu, Y., Deacon, C., 2009. Occurrence and
partitioning of cadmium, arsenic and lead in mine impacted paddy rice: Hunan,
China. Environ. Sci. Technol. 43, 637e642.
Xu, W.H., Li, Y.R., He, J.P., Ma, Q.F., Zhang, X.J., Chen, G.Q., Wang, H.X., Zhang, H.B.,
2010. Cd uptake in rice cultivars treated with organic acids and EDTA. J. Environ.
Sci. 22, 441e447.
Xu, J.J., Wu, B., Zhang, L.Y., Guo, S.H., Li, G., Li, F.M., 2016. Risk assessment of cad-
mium in rice in Xiangtan, Hunan, China based on Bayesian method. Chin. J. Appl.
Ecol. 27, 3221e3227 (in Chinese).
Ye, X., Ma, Y., Sun, B., 2012. Influence of soil type and genotype on Cd bioavailability
and uptake by rice and implications for food safety. J. Environ Sci. 24, 1647e1654.
Yu, L., Zhu, J., Huang, Q., Su, D., Jiang, R., Li, H., 2014. Application of a rotation system
to oilseed rape and rice fields in Cd-contaminated agricultural land to ensure food
safety. Ecotox. Environ Safe 108, 287e293.
Zhao, K., Fu, W., Ye, Z., Zhang, C., 2015. Contamination and spatial variation of heavy
metals in the soil-rice system in nanxun county, southeastern China. Inter. J.
Environ. Res. Pub. Heal 12, 1577e1594.