New Zealand Journal of Agricultural Research
ISSN: 0028-8233 (Print) 1175-8775 (Online) Journal homepage: https://www.tandfonline.com/loi/tnza20
Cadmium in New Zealand agricultural soils
Edward Abraham
To cite this article: Edward Abraham (2020) Cadmium in New Zealand agricultural soils, New
Zealand Journal of Agricultural Research, 63:2, 202-219, DOI: 10.1080/00288233.2018.1547320
To link to this article: https://doi.org/10.1080/00288233.2018.1547320
View supplementary material
Published online: 20 Dec 2018.
Submit your article to this journal
Article views: 127
View related articles
View Crossmark data
Citing articles: 4 View citing articles
Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=tnza20

link to page 19 link to page 18 link to page 19 link to page 19 link to page 19 link to page 19 link to page 18 NEW ZEALAND JOURNAL OF AGRICULTURAL RESEARCH
2020, VOL. 63, NO. 2, 202–219
https://doi.org/10.1080/00288233.2018.1547320
RESEARCH ARTICLE
Cadmium in New Zealand agricultural soils
Edward Abraham
Dragonﬂy Data Science, Wellington, New Zealand
ABSTRACT
ARTICLE HISTORY
I summarised cadmium concentrations from soil samples collected
Received 13 February 2018
by the fertiliser industry between 2006 and 2015, and from soil
Accepted 9 November 2018
samples collected by regional councils. Using these data, a
statistical
model
was
used
to
estimate
soil
cadmium
KEYWORDS
Cadmium; agriculture;
concentrations
on
agricultural
land
within
each
territorial
fertiliser; soil; statistical
authority. Across New Zealand, most territorial authorities (56 of
model; New Zealand
66), including all the South Island, had mean soil cadmium
concentrations below 0.6 mg Cd/kg. The highest mean soil
cadmium values on agricultural land were in the Otorohanga and
Matamata-Piako districts (0.90 and 0.89 mg Cd/kg, respectively). In
these districts, around 1% of properties were estimated to have
mean soil cadmium in excess of 1.8 mg Cd/kg. The fertiliser
industry requires that cadmium concentrations in phosphate
fertiliser are below a voluntary limit of 280 mg/kg P. The mean
cadmium concentration in fertiliser samples taken between 2003
and 2015 was 184 mg Cd/kg P, 66% of the voluntary limit.
Introduction
Cadmium (Cd) is a heavy-metal contaminant of phosphate (P) fertiliser. Where there is
sustained application of fertiliser, cadmium may accumulate in topsoil (e.g. Taylor
1997; Loganathan et al. 2003). From there, cadmium may be taken up into agricultural
produce (e.g. Roberts et al. 1994), and enter the food chain. Cadmium concentrations
in typical New Zealand diets are below thresholds established by the World Health Organ-
isation (Vannoort and Thomson 2009). In New Zealand, management of cadmium con-
centration in agricultural soils is carried out through the Tiered Fertiliser Management
System (TFMS; Warne 2011; Sneath 2015; Fertiliser Association of New Zealand 2016).
The TFMS is a voluntary system managed by the fertiliser industry. It is part of the
Cadmium Management Strategy, approved by the Cadmium Management Group (a
nationally representative group of stakeholders with representation from the fertiliser
industry, primary sector organisations, regional councils, and central government
agencies). The intent of the TFMS is to ensure that cadmium concentrations in agricul-
tural soils remain within an acceptable range that allows for protection of human
health and the environment, as well as long-term sustainable agricultural production.
The TFMS establishes concentrations of soil cadmium at which management actions
CONTACT Edward Abraham
edward@dragonﬂy.co.nz
Supplemental data for this article can be accessed http://dx.doi.org/10.1080/00288233.2018.1547320.
© 2018 The Royal Society of New Zealand

link to page 3 link to page 18 link to page 19 link to page 18 link to page 18 link to page 18 link to page 18 link to page 19 link to page 18 NEW ZEALAND JOURNAL OF AGRICULTURAL RESEARCH
203
Table 1. Soil cadmium concentrations speciﬁed by the voluntary Tiered Fertiliser Management System
(Fertiliser Association of New Zealand 2016).
Soil cadmium (mg
Tier
Cd/kg)
Management required
Tier 0
<0.6
Soil cadmium is within the range of natural background concentrations. No restriction
on phosphate fertiliser type or application.
Tier 1
0.6 to <1.0
Some restrictions on phosphate fertiliser application rates, and implementation of
appropriate management practices.
Tier 2
1.0 to <1.4
Increased restrictions on phosphate fertiliser type and application rates, and
implementation of appropriate management practices.
Tier 3
1.4 to <1.8
Further restrictions on phosphate fertiliser type and application rates, and
implementation of appropriate management practices.
Tier 4
≥1.8
No further cadmium accumulation allowed unless a detailed site-speciﬁc investigation
is undertaken to identify risks and pathways for potential harm.
should occur (Table 1). No accumulation of soil cadmium is allowed by the TFMS beyond
1.8 mg Cd/kg (Tier 4), without a detailed site-speciﬁc investigation to identify risks and
pathways for potential harm.
As part of the application of the TFMS, the New Zealand fertiliser industry has carried
out sampling for soil cadmium concentrations (Cavanagh 2014; Sneath 2015). The
sampling is focused on farms that are currently applying an average of phosphorus at
30 kg/ha/yr or more. The system is voluntary, and farmers applying fertiliser at a high
rate are encouraged to have soil samples analysed for cadmium concentrations, as part
of routine soil fertility tests. The cadmium tests are expected to be undertaken at least
once every ﬁve years. These data provide a basis for analysing cadmium concentrations
in New Zealand’s agricultural soils, providing information about soil cadmium concen-
trations in agricultural land by territorial authority; and investigating the association
between available factors and soil cadmium concentration. The analysis of earlier data
led to the recommendation to collect about 30 samples from each land use that makes
up a signiﬁcant proportion of land area within a region to increase the sampling represen-
tation for each land use in each region (Cavanagh 2014). This recommendation was fol-
lowed by the fertiliser industry, providing a broad spread of samples across New Zealand.
Soil cadmium sampling is also carried out by some regional councils as part of their
State of the Environment reporting (e.g. Land Monitoring Forum 2009; Drewry 2017).
This monitoring includes soil quality to provide information of the functioning of soil
in the context of biological productivity, environmental quality, and plant and animal
health (Hill and Sparling 2009). Within this monitoring framework, phosphate fertilisers
have been identiﬁed as a signiﬁcant anthropogenic source of additional cadmium to New
Zealand agricultural soils.
In order to reduce the rate of accumulation of soil cadmium, the fertiliser industry
manages the concentration of cadmium in phosphate fertiliser. Until the 1990s, most of
New Zealand’s phosphate fertiliser was derived from Nauru rock phosphate, which had
high cadmium concentrations of around 600 mg Cd/kg P (Syers et al. 1986). Since the
1990s, rock phosphate with lower cadmium concentrations has been sourced from
elsewhere. In 1995, the fertiliser industry established voluntary limits for cadmium
concentrations in fertiliser. The initial proposal was for a limit of 420 mg Cd/kg P with
a planned phased reduction in the limit. From July 1995, the cadmium concentration
in fertiliser was restricted to be below 340 mg Cd/kg P. Fertiliser companies were
rapidly able to manufacture fertiliser within this range, further reducing the upper limit

link to page 18 link to page 18 link to page 18 link to page 18 link to page 18 link to page 18 link to page 18 link to page 18 204
E. ABRAHAM
to 280 mg Cd/kg P from January 1997. Since 2001, this limit on cadmium concentrations
in fertiliser has been monitored and audited through the Fertmark programme, a New
Zealand fertiliser quality assurance scheme. The fertiliser companies have carried out
routine sampling of cadmium concentration in fertiliser, with the sampling programme
being audited by Quality Consultants of New Zealand (QCONZ).
This study used New Zealand soil cadmium data to develop a statistical model that
allowed the estimation of cadmium soil concentrations across a broad spatial scale, com-
bining diﬀerent data sources. The concentration of soil cadmium varies widely between
regions, and this analysis allowed concentrations to be estimated at the scale of Territorial
Authorities, potentially allowing management at a ﬁner spatial scale. Data included here
were derived from earlier summaries of soil cadmium data (Cavanagh 2014; Abraham
et al. 2016). These data were used in the statistical modelling to assess the inﬂuence of cov-
ariates on soil cadmium concentrations. The covariates included in the analysis were
sample source and protocol, soil type, land use, sample depth and year. Outcomes from
this study will support the Cadmium Management Group’s continuing risk management
of cadmium in New Zealand’s agriculture and food systems.
Methods
Cadmium in soil
Cadmium concentrations in soil were measured by the fertiliser industry, as part of on-
farm nutrient sampling. The samples were typically from composite samples collected
along a transect or grid that is representative of the size of monitoring paddocks, and typi-
cally includes around 15 cores taken to a depth of either 7.5 cm or 15 cm (from pastoral
and arable land, respectively) (Fertiliser Association of New Zealand 2016). For each soil
sample, the total cadmium concentration (mg Cd/kg) was analysed by accredited (IANZ,
International Accreditation New Zealand) laboratories following standardised methods
benchmarked against those speciﬁed by the Fertiliser Quality Control Council (Cavanagh
2014; Fertiliser Quality Council 2015). Variability was measured using the coeﬃcient of
variation (CV, deﬁned as the ratio of the standard deviation to the mean value of the
samples). Metadata associated with each sample included a customer code, a soil type, a
land use, the date of the sample, and a locality. Due to customer conﬁdentiality require-
ments, the precise location of the samples was not made available, and the localities
were standardised by identifying the associated territorial authority. The fertiliser compa-
nies Ballance and Ravensdown provided data from 6839 soil samples that were taken in
2007 and over the period 2012–2015 (in the presentation of results, the two companies
are anonymised by referring to them as Company A and Company B). Of the industry
samples, 3492 samples (51%) were from 2014.
In addition to the industry data, soil cadmium data from regional councils were pro-
vided by Landcare Research (with permission of the councils; as described by Cavanagh
2014). The regional council data included 1020 samples collected following State of the
Environment methods (e.g. samples taken to 10 cm depth; Hill and Sparling 2009), as
well as 443 other soil cadmium samples collected by Environment Canterbury as part
of an arable and pastoral regional monitoring programme (Lawrence-Smith and
Tregurtha 2013).

link to page 18 link to page 5 NEW ZEALAND JOURNAL OF AGRICULTURAL RESEARCH
205
The land use and soil type categories were standardised, with soil types of sedimentary,
ash, pumice, or peat/organic, and land use categories of dairy, sheep and beef, cropping,
orchard, background, or other. Due to variable reporting, the data did not consistently dis-
tinguish between arable and horticultural cropping, so these categories were merged. A
land use of ‘background’ was used for conservation land (conservation areas and indigen-
ous forest), while a land use of ‘other’ included forestry (exotic plantations), turf, and
urban land (see also Cavanagh 2014).
In preparation for the analysis, samples with incomplete metadata were removed from
the dataset. Of all the samples, 2026 had no location information (the territorial authority
was unknown); of the remaining samples, 623 samples had no sample depth information,
192 samples had no soil type indicated, and one sample had no sampling protocol infor-
mation. Excluding these samples left a ﬁnal dataset of 5459 samples with complete meta-
data (4210 samples from Ballance and Ravensdown, 928 samples from regional council
State of the Environment monitoring, and 321 samples from other Environment Canter-
bury sampling). In the model dataset, 1404 (33.3%) of the industry samples were from
sedimentary soils on dairy land (Table 2), with no data from background land, while
the regional council data was spread more evenly across land use types. In the data
since 2012, 89.8% of the fertiliser industry samples in the ﬁnal dataset had a customer
code (this code gave no address or other location information, but was a code that
allowed samples from the same farm to be identiﬁed). No farm information is available
for regional council samples, or for industry samples before 2012. Across the ﬁnal
dataset, 3708 samples had a customer code, with the samples being taken from 1905
unique farms.
Statistical modelling of soil cadmium
Statistical modelling was undertaken to analyse the inﬂuence of the covariates on the soil
cadmium concentration, including investigating whether there was any linear temporal
trend evident in the data, and whether the concentrations reported by the fertiliser
Table 2. Number of samples in the model dataset, summarised by soil type and by land use, for data (a)
from the fertiliser industry and (b) from regional councils.
(a) Number of industry samples
Land use
Soil type
Background
Cropping
Dairy
Orchard/Vineyard
Sheep & Beef
Other
Ash
121
678
17
231
10
Peat/Organic
1
62
1
4
Pumice
4
139
88
2
Sedimentary
383
1404
269
773
23
(b) Number of regional council samples
Land use
Soil type
Background
Cropping
Dairy
Orchard/Vineyard
Sheep & Beef
Other
Ash
12
14
29
15
26
15
Peat/Organic
3
5
15
2
12
3
Pumice
2
2
23
7
12
Sedimentary
161
205
267
45
287
87
Samples with incomplete metadata have been excluded. Fertiliser industry data are from the period between 2012 and
2015, regional council data are from all years between 2006 and 2014.

link to page 18 206
E. ABRAHAM
companies and the regional councils were consistent. The analysis also allowed for stan-
dardisation of the sample depth. The eﬀect of covariates on the concentration of soil
cadmium was estimated using a hierarchical Generalised Linear Model (GLM). The par-
ameters of the GLM were inferred from the data by using Bayesian Markov chain Monte
Carlo (MCMC) methods. These methods allow the uncertainty in the soil cadmium within
Territorial Authorities to be estimated, while accounting for the variability associated with
covariates such as land use and soil type.
Cadmium concentration is a positive quantity, and the cadmium concentration in each
sample was represented as a draw from a gamma distribution, G(a, b), where a is the
shape and b is the rate. The shape and rate were parameterised in terms of the mean,
m, and the coeﬃcient of variation, s (the coeﬃcient of variation is the standard deviation
divided by the mean). In terms of these parameters, the shape is a = 1/s2, the rate is
b = a/m, and the cadmium concentration, cs, of a sample, s, was estimated as:
cs G(1/s2, 1/(m s2
s
).
(1)
The coeﬃcient of variation was assumed to be the same for all samples (so the distribution
had the same shape, irrespective of the covariates). The mean concentration was rep-
resented as a linear function of the covariates:

N
log(m
b
s) =
ixi [s] + bd[d[s]] + bf [f [s]],
(2)
i=1
where the model had N covariates, including an intercept, with covariate i having the value
xi[s] for sample s. District and farm eﬀects were included as random eﬀects, bd and bf ,
respectively. The farm eﬀect was drawn from a normal distribution, independently for
each unique customer code. Two approaches were used for handling samples without a
customer code–in the ﬁrst approach, samples with missing codes were each given a
unique code, and in the second approach samples with missing codes were given a code
that was a composite of the district, year, and land use values. The ﬁrst approach would
overestimate the number of farms and second approach would underestimate it. The dis-
trict random eﬀect was drawn from a conditional auto-regression (CAR) model (Gelfand
and Vounatsou 2003), to account for any spatial correlation in the district data. Districts
with any common boundaries were considered to be neighbours. The CAR model
included a parameter a(0 , a , 1), specifying the degree of correlation between neigh-
bouring districts. A low value of α would imply that there is only a weak association
between the cadmium concentration in neighbouring districts, while a high value of α
would allow for a high association.
This model structure was chosen as there is a wide range in the number of samples
between districts. The spatial model for the district eﬀect allows an estimate to be made
of the mean cadmium concentration within each district, even for districts where there
are few samples. The farm random eﬀect is included as there are often multiple
samples per property, and this inclusion allowed within-farm variation in soil cadmium
concentrations to be represented. The Bayesian modelling is ﬂexible enough for these
sources of variation to be represented.
Diﬀerent combinations of the ﬁxed-level covariates were tested, to determine the eﬀects
that best explained variation in the cadmium concentration data. Potential ﬁxed eﬀects

link to page 19 link to page 19 link to page 18 link to page 7 link to page 18 NEW ZEALAND JOURNAL OF AGRICULTURAL RESEARCH
207
included a combination of source and protocol, protocol, soil type, land use, sample depth
and year. Sample depth and year were included in the models as scaled continuous vari-
ables, while the other covariates were included as factors (variables with discrete categori-
cal values). The model selection was based on minimising the leave-one-out information
criterion (LOOIC; Vehtari et al. 2016, 2017), with the lowest LOOIC value indicating the
closest model ﬁt. The model selection process allows for a balance to be found between
improving the model ﬁt, and increasing model complexity, guarding against overﬁtting
the model.
The Bayesian statistical model was ﬁtted using the software Stan (Carpenter et al. 2017),
with the priors given in Table 3. The priors were only weakly informative. Cauchy priors
were chosen for the scale parameters, following recommendations by Gelman and Hill
(2006). For the positive scale parameters, the a positive half-Cauchy prior was used.
The model was run for a warm up period of 1000 iterations, and then run for a further
20,000 iterations. The model was run with 3 chains, with a thinning interval of 10 iter-
ations, keeping a total of 6000 samples.
Using the posterior samples from the ﬁtted model, a standardised cadmium concen-
tration was estimated for agricultural land from each district. This value was calculated
6000 times, randomly sampling rows of original data from the same district to provide
soil type and land use values, drawing from the distribution of farm-level random
eﬀects, and randomly assigning to each of the fertiliser industry companies. For districts
that had less than 10 original cadmium samples, the random sampling of land use and soil
type was taken across the region. The depth of the standardised estimate was set to 15 cm,
reﬂecting the sample depth of the TFMS, and the year was set to 2015. From these esti-
mates, a mean concentration could be calculated and the number of estimates within
each TFMS level. The standardised values represented estimates of the mean soil
cadmium concentrations on farms within each district, to 15 cm, and in 2015. This stan-
dardisation assumed that the farms sampled by the fertiliser industry were representative
of the farms in each district (or region, if fewer than 10 samples were taken).
Cadmium in fertiliser
The cadmium concentration in superphosphate fertiliser (measured in mg Cd/kg P) is
monitored using samples from the main manufacturing sites of the fertiliser companies
Ballance and Ravensdown (currently at Mount Maunganui, Awarua, Ravensbourne,
Table 3. Priors used for ﬁtting the statistical model of soil cadmium.
Parameter
Symbol
Prior
Variation in gamma distribution
s
Cauchy (0, 10)
Variation in district random-eﬀect
sd
Cauchy (0, 10)
Variation in farm random-eﬀect
sf
Cauchy (0, 10)
Farm random-eﬀect
bf
Normal (0, sf )
District random-eﬀect
bd
CAR (a, sd, W)
CAR correlation parameter
a
Uniform (0, 1)
Fixed eﬀects
b
Cauchy (0, 10)
The priors are either broad Cauchy distributions (speciﬁed by the mean and half the interquartile range); a uniform distri-
bution between zero and one; a normal distribution with zero mean, and standard deviation estimated within the model;
or a conditional autoregression (CAR) submodel. The neighbour relations between districts is provided to the model as a
matrix (W).

link to page 18 link to page 9 link to page 9 link to page 18 208
E. ABRAHAM
Awatoto, and Hornby). The two fertiliser companies carry out weekly composite sampling
of fertiliser that is ready for dispatch, at each manufacturing site, with data available since
January 2003 (3188 samples). Samples are taken of each of the main superphosphate fer-
tiliser products, and an average cadmium concentration for each manufacturing site is cal-
culated, weighted by the production of each sampled product. The sampling is carried out
as part of the Fertmark fertiliser quality assurance programme, which speciﬁes the
sampling and analytical methods (Fertiliser Quality Council 2015). The industry sampling
programme is audited by Quality Consultants of New Zealand (QCONZ), who also collate
the data in a spreadsheet. In the present study, the weekly data were averaged to determine
the monthly mean concentration of cadmium between January 2003 and July 2015. The
average was an unweighted average across the manufacturing sites, as data on relative pro-
duction volumes were not available. The 5th and 95th percentiles of the data were calcu-
lated to indicate the range of variation in the spot samples.
Results
Soil cadmium sample data
Across all samples in the ﬁnal dataset from agricultural land, the mean cadmium concen-
tration was 0.45 mg kg−1 (median value of 0.32 mg kg−1), compared with the mean con-
centration of the 178 background samples of 0.09 mg kg−1 (median value of
0.06 mg kg−1). The maximum soil cadmium concentration in the ﬁnal dataset was
2.93 mg kg−1 (a sample with a soil cadmium value of 3.05 mg kg−1 was removed due to
incomplete metadata). There were strong regional patterns in the concentration of
cadmium (Figure 1A). The highest mean cadmium concentrations in agricultural land
were in the Waikato and Taranaki regions–in these regions the mean soil cadmium con-
centration was in the TFMS Tier 1 range. For all other regions, the mean soil cadmium
concentration was within the Tier 0 range. Over all the data, there were 16 samples
from Waikato, one sample from Auckland and four samples from Tasman that were in
the TFMS Tier 4 range (with a soil cadmium concentration above 1.8 mg Cd/kg).
It is worth noting that the Tier 4 Tasman samples all came from the same farm. There
were 19 samples taken at this property, with four of these samples recording soil cadmium
concentrations exceeding 1.8 mg Cd/kg. The mean soil cadmium value across all the
samples from this property was 0.93 mg Cd/kg. In general, there was considerable varia-
bility in the cadmium concentrations of repeated samples from individual farms. The
mean CV of the cadmium concentration of soil samples from the same farm was 0.30
(across the 95 farms that had ﬁve or more samples). When calculated for the 61 territorial
authorities with ﬁve or more soil samples, the mean CV was 0.56. By this measure, vari-
ation in cadmium concentration within farms was over half the variation within territorial
authorities.
Dairy was the land use with the highest cadmium concentration (Figure 1B), followed
by sheep and beef, and cropping. The agricultural land use with the lowest mean soil
cadmium concentration was orchard/vineyard. There was high variability in the data
however, and in a previous analysis of a subset of the industry data the mean soil
cadmium on 65 samples from orchard/vineyard samples was similar to dairy (Cavanagh
2014). All agricultural land uses had some samples that were within the Tier 4 range

link to page 18

NEW ZEALAND JOURNAL OF AGRICULTURAL RESEARCH
209
Figure 1. Summary of raw cadmium concentration in soil samples from (A) agricultural land in each
region, (B) all samples by land use and (C) agricultural land by soil type. The lines indicate the 5%
and 95% quantiles of the sample soil cadmium concentration, with the thinner whiskers extending
to the full range. The dot marks the mean cadmium value of the samples in each group. The
number of samples in each region is indicated at the bottom of each graph. The dashed orange
lines indicate the thresholds of the Tiered Fertiliser Management System (Fertiliser Association of
New Zealand 2016).

link to page 10

210
E. ABRAHAM
(16 samples from dairy, four samples from cropping, one sample from sheep and beef, and
one sample from orchard/vineyard). The dataset includes 178 samples collected by
regional councils on background land (land that has never been used for agriculture).
The mean soil cadmium on these samples was 0.087 mg Cd/kg (with 95% of the
samples having a concentration less than 0.24 mg Cd/kg). Mean cadmium concentrations
in samples from agricultural land were highest in volcanic ash or peat/organic soils, and
lowest in sedimentary soils.
The mean soil cadmium values, averaged by territorial authority and land use, were
typically lower when based on measurements made by regional councils than when
based on the industry data (Figure 2). Across the 41 territorial authorities and agricultural
land use groups that had ﬁve samples or more taken by both regional councils and indus-
try, the mean soil cadmium concentration from the regional council data was less than the
mean soil cadmium concentration from the industry data in 73% of the groups. There was
high variability in these data (a Pearson r2 of 0.83), reﬂecting both the low number of
samples in many of the territorial authority and land use groups, and the variability in
soil cadmium concentrations between samples. The raw data were collected from
diﬀerent sampling depths by the industry and the regional councils, with lower
cadmium concentrations for deeper samples. Some diﬀerences in the raw data may be
attributed to this diﬀerence in sampling depth.
Figure 2. Comparison of mean soil cadmium concentrations from data collected by regional councils
and by the fertiliser industry. Points are shown for each territorial authority and land use that had ﬁve or
more samples. The colour indicates the land use, and the diagonal line is the one-to-one line.

link to page 11 link to page 12 link to page 9 NEW ZEALAND JOURNAL OF AGRICULTURAL RESEARCH
211
Statistical modelling of soil cadmium
Selection of the best model of soil cadmium concentrations involved testing whether
source, protocol, or year should be included as eﬀects. The structure of the farm
random eﬀect was also investigated. The model with the smallest LOOIC value was
chosen (Table 4), and all models that included the farm random eﬀect had markedly
smaller LOOIC values than the model without this eﬀect; models that represented
missing property codes through the interaction of the main eﬀects also had comparatively
small LOOIC values. The latter meant that samples without a customer code, from the
same territorial authority, the same land use and soil type and taken in the same year
were treated as being from the same property. The model selection also indicated a
closer ﬁt for the model that included source and year eﬀects, but the diﬀerence to a
model that only included depth, soil and land use eﬀects was not signiﬁcant.
The model eﬀects showed increased soil cadmium concentrations in volcanic ash, peat,
and pumice soils (relative to sedimentary soils) with no signiﬁcant diﬀerence between the
eﬀects associated with these soil types (Table 5). Soil cadmium concentrations were lower
on land used for cropping, sheep and beef, and orchards than on dairy. The soil cadmium
concentration on background land use was estimated at 27.5% (95% c.i.: 24.3%–31.1%) of
the concentration on land used for dairy. These eﬀects followed the patterns seen in the
raw data (see Figure 1). The strength of these eﬀects ranged from a mean of 0.745 (for
orchard/vineyard land relative to dairy land), to a mean of 1.473 for peat/organic (relative
to sedimentary soil). In contrast, the mean territorial authority eﬀects ranged from 0.42 to
2.29, even after diﬀerences in soil, current land use, and the other model eﬀects were
accounted for. There was more variation associated with the territorial authority of the
sample than with the other predictors included in the model.
There was no signiﬁcant diﬀerence between the soil cadmium values measured by the
two diﬀerent companies, or by the Environment Canterbury regional monitoring samples.
Soil cadmium values measured by the regional councils following the State of the Environ-
ment (SOE) protocol was 83.6% (95% c.i.: 76.8%–90.8%) of the soil cadmium values
reported by the industry (Company A). Diﬀerences between the industry and regional
council data may be due to diﬀerences in the choice of properties that were included in
the sampling. Samples collected to a deeper sampling depth had lower soil cadmium con-
centrations: the concentration in 15-cm samples was 87.9% (95% c.i.: 82.8%–93.2%) of the
Table 4. Summary of model selection, listing the models in order of decreasing leave-one-out
information criterion (LOOIC).
Farm random eﬀects
Model factors
LOOIC
Diﬀerence
S.E.
Interactions
Depth + Soil + Land use + Source + Year
−6133.5
0.0
0.0
Interactions
Depth + Soil + Land use + Protocol
−6132.9
0.3
2.0
Interactions
Depth + Soil + Land use + Source
−6132.9
0.3
1.6
Interactions
Depth + Soil + Land use + Protocol + Year
−6131.3
1.1
1.9
Interactions
Depth + Soil + Land use
−6115.3
9.1
4.9
Interactions
Depth + Soil + Land use + Year
−6114.5
9.5
4.7
Samples
Depth + Soil + Land use + Source + Year
−6015.6
58.9
56.5
No farm random eﬀect
Depth + Soil + Land use + Source + Year
−4669.7
731.9
48.3
Samples without an identiﬁed farm identiﬁcation are either given a farm code based on unique combinations of the main
eﬀects (Interactions), or a unique farm code for each sample (Samples), or a farm random eﬀect was not included (No
farm random eﬀect). The models are then speciﬁed by including diﬀerent covariates. In addition to the LOOIC the table
gives the diﬀerence, relative to the ﬁrst model, and the standard error (S.E.) of the estimated diﬀerence.

link to page 13 link to page 12 212
E. ABRAHAM
Table 5. Posterior distributions of the model parameters, showing the mean and 95% credible interval
of the estimated eﬀects.
Parameter
Level
Mean
95% c.i.
Mean soil cadmium (Sedimentary, Dairy; Company A)
0.356
0.311–0.405
Soil type (relative to Sedimentary)
Ash
1.424
1.349–1.501
Peat
1.473
1.341–1.617
Pumice
1.372
1.245–1.509
Land use (relative to Dairy)
Cropping
0.838
0.786–0.894
Sheep & beef
0.829
0.795–0.865
Orchard
0.745
0.645–0.856
Background
0.275
0.243–0.311
Other
0.43
0.382–0.481
Source (relative to Company A)
SOE
0.836
0.768–0.908
Environment Canterbury
1.034
0.894–1.189
Company B
0.973
0.931–1.018
Sample depth
0.879
0.828–0.932
Year
0.995
0.978–1.012
Coeﬃcient of variation
District
0.713
0.543–0.951
Farm
0.287
0.271–0.305
Sample
0.383
0.374–0.392
Spatial correlation
0.836
0.577–0.967
The soil type, land use, and source parameters are shown exponentiated, and so can be interpreted as multiplicative eﬀects.
The sample depth parameter has been transformed to give the multiplicative eﬀect of a 15-cm sample depth, relative to a
7.5-cm sample depth. The year eﬀect estimates the annual multiplicative change in soil cadmium. The district and farm
standard deviations are converted to a coeﬃcient of variation of the exponentiated parameters. The spatial correlation
parameter is untransformed.
soil cadmium concentration in samples collected to 7.5 cm depth. There was no signiﬁcant
change in soil cadmium over the period of the study, with the year eﬀect implying an
annual change of −0.5% (95 c.i.: −2.2%– 1.2%).
The model allowed a standardised soil cadmium concentration to be estimated (as
described in the methods), primarily adjusting for the variation in sampling depth. The
standardisation also allowed the mean concentration to be estimated in territorial auth-
orities that only had a small number of samples. There are broad scale variations in soil
cadmium concentration across New Zealand. The territorial authorities with the highest
mean soil cadmium were in the Waikato and Taranaki regions (Figure 3, Table 6), with
Table 6. The ten Territorial Authorities (TAs) with the highest mean standardised soil cadmium
concentration.
Mean soil cadmium
Percentage of farms estimated to be in each TFMS
(mg/kg)
tier
Territorial authority
N
Raw
Standardised
Tier 0
Tier 1
Tier 2
Tier 3
Tier 4
Otorohanga District
267
1.03
0.90
15.1
52.2
26.7
5.0
1.0
Matamata-Piako District
154
0.91
0.89
16.8
52.6
23.9
5.7
1.1
Waipa District
207
0.94
0.86
17.4
55.9
22.1
4.1
0.5
Hamilton City
61
0.76
0.72
35.8
51.1
11.4
1.4
0.3
South Waikato District
52
0.85
0.72
32.5
57.5
9.2
0.8
0.1
Stratford District
72
0.84
0.72
33.3
54.6
11.3
0.8
0.1
South Taranaki District
180
0.77
0.69
37.6
53.1
8.7
0.5
0.0
Waitomo District
104
0.73
0.67
43.1
49.5
6.8
0.5
0.1
New Plymouth District
106
0.74
0.65
44.8
49.0
5.5
0.6
0.0
Waikato District
214
0.67
0.63
52.8
40.3
6.2
0.7
0.1
Shown are the number of samples (N); the mean soil cadmium concentration on agricultural land, from the raw data and
standardised from the statistical model; and the percentage of farms within each Territorial Authority estimated to be
within each Tiered Fertiliser Management System (TFMS) tier, based on standardised soil cadmium (data for all TAs
are included in Appendix A, Table A1).

NEW ZEALAND JOURNAL OF AGRICULTURAL RESEARCH
213
Figure 3. Mean standardised soil cadmium concentrations in agricultural land by territorial authority,
and the estimated percentage of farms within each tier of the Tiered Fertiliser Management System
(see detailed data in Appendix A, Table A1).
the highest mean standardised soil cadmium of 0.90 mg Cd/kg being in Otorohanga Dis-
trict. In Otorohanga District, 32.7% of farms were estimated to be in Tier 1 or above (a
mean soil concentration of 1.0 mg Cd/kg or over). It was estimated that 1.0% or more
of the farms in the Otorohanga and Matamata-Piako districts had a mean standardised
soil cadmium concentration over 1.8 mg Cd/kg, and so were in Tier 4 of the TFMS. In
the South Island, in contrast, all territorial authorities were estimated to have most
farms (at least 98%) in Tier 0 of the TFMS (a soil cadmium concentration less than
0.6 mg Cd/kg) (see Appendix A, Table A1). Soil cadmium concentrations were also
lower on the East Coast of the North Island, with all territorial authorities (other than
Napier City) estimated to have at least 97% of farms in Tier 0 (of the 29 samples taken
on farms in the Napier City territorial authority, 33% had soil cadmium in the Tier 1
range, between 0.6 and 1.0 mg Cd/kg).
The spread of standardised values is inﬂuenced by uncertainty in the territorial auth-
ority eﬀect, and this uncertainty is larger in territorial authorities that had low numbers
of samples. While the uncertainty will only have a minor eﬀect on the mean value
across farms in a territorial authority, the uncertainty will inﬂuence the proportion of
farms that are estimated to have soil cadmium in the higher tiers. The proportion will

link to page 14 link to page 19 link to page 18

214
E. ABRAHAM
increase as the spread of the farm-level estimates increases. This eﬀect is seen in Kawerau
District–Kawerau is a small area that is wholly within Whakatane District, and there were
no soil samples taken within Kawerau. Despite this, it was estimated that 0.3% of the farms
might be in Tier 4 (see Appendix A, Table A1).
Cadmium in fertiliser
Across all weekly samples from the main manufacturing sites, the average cadmium con-
centration between January 2003 and July 2015 was 184 mg Cd/kg P (Figure 4). Ninety-
ﬁve percent of the samples had cadmium concentrations less than 246 mg Cd/kg P, while
ﬁve percent of the samples had concentrations less than 108 mg Cd/kg P. The monthly
mean values ﬂuctuated within this range, falling below 108 mg Cd/kg P in the last two
months of the series (June and July 2015). The maximum monthly mean concentration
over the period from 2003 to 2015 was 243.5 mg Cd/kg P.
Since 2003, there have been two individual samples declared as exceeding the voluntary
limit of 280 mg Cd/kg P. These were slight exceedances at 285 and 299 mg Cd/kg P. Two
additional exceedances were recorded (with concentrations of 282 and 293 mg Cd/kg P),
but they were from samples mistakenly taken from rock phosphate prior to blending
(Peter Wood, QCONZ, pers. comm.).
The monthly-averaged concentration of cadmium in phosphate fertiliser was consist-
ently below the voluntary limits established by the fertiliser industry. There were some
exceedances reported in individual samples. The long-term sample average of cadmium
in phosphate fertiliser was 184 mg Cd/kg P. This average is 66% of the voluntary limit
of 280 mg Cd/kg P assumed for sulphuric acid- derived products, such as superphosphate
in developing the TFMS (Sneath 2015). Changes in the source of rock drive the variability
of the cadmium concentration in the fertiliser (Cadmium Working Group 2008). In the
last two available months of data, the monthly average sample concentration fell below
100 mg Cd/kg P for the ﬁrst time.
Figure 4. Cadmium concentration in fertiliser samples from the main manufacturing sites (based on
data from Quality Consultants of New Zealand Limited). The line shows the monthly mean concen-
tration, the straight line shows the mean value over the period, and the shading marks the 90th per-
centile interval of the sample data.

link to page 18 link to page 19 link to page 18 link to page 19 link to page 19 link to page 19 link to page 19 link to page 19 link to page 19 link to page 19 link to page 18 link to page 18 link to page 19 link to page 19 NEW ZEALAND JOURNAL OF AGRICULTURAL RESEARCH
215
Discussion
In New Zealand, large areas, including most of the South Island, have soil cadmium con-
centrations below levels requiring any immediate active management. Elevated soil
cadmium concentrations were primarily found in the Waikato and Taranaki regions,
which have previously been identiﬁed as regions with higher soil cadmium concentrations
(Kim 2005; Taylor et al. 2007; Cavanagh 2014; Staﬀord et al. 2014). Within those regions,
the analysis identiﬁed the territorial authorities with higher soil cadmium concentrations.
In the Otorohanga and Matamata-Piako districts, the mean standardised soil cadmium
concentration on agricultural land is within Tier 1 of the Tiered Fertiliser Management
System, with around one percent of properties estimated to have soil cadmium concen-
trations in Tier 4 (over 1.8 mg Cd/kg).
The statistical modelling identiﬁed factors that were related to soil cadmium concen-
trations, and these factors aligned with previous analyses of soil cadmium data (e.g.
Staﬀord et al. 2014): dairy farming was associated with higher cadmium concentrations,
and sedimentary soils were associated with lower soil cadmium concentrations. There
was, however, more variability associated with the territorial authority of a sample than
with either soil type or land use factors. It is expected that the variation in mean
cadmium concentrations between territorial authorities reﬂects diﬀerences in fertiliser
history, with high historical fertiliser use on dairy land in the Waikato and Taranaki
regions. Soil cadmium concentration has been found to be associated with total phos-
phorus (Roberts et al. 1994; Schipper et al. 2011; McDowell et al. 2013; Staﬀord et al.
2015; Salmanzadeh et al. 2016), suggesting that soil cadmium concentrations primarily
reﬂect phosphate fertiliser history. A similar relationship between phosphorus and
cadmium has been found in freshwater sediments (Kim 2005). No information on ferti-
liser history, or on associated elements such as total phosphorus, was available that
could have been used to test the relationship between fertiliser history and the current
soil cadmium concentrations.
The statistical model used a conditional auto-regression (CAR) approach to identify
correlation between neighbouring territorial authorities, which helped estimate soil
cadmium concentrations in areas of territorial authorities that a low number of
samples. The model only considered the main eﬀects, without considering the interactions
between the variables. This analysis was determined by the nature of the data. A key limit-
ation was that the spatial location of the samples was not made available, but data were
reported to the territorial authority level. This limitation prevented linking the dataset
of external sources of information, such as geological maps, slope, land use, or climatic
data that would have allowed estimation of cadmium concentrations across the landscape
(Marchant et al. 2010). It was also not possible to link the datasets from the two companies
and the regional councils, as diﬀerent identiﬁers prevented the matching of properties. If it
was possible to overcome the anonymity concerns, by developing an appropriate protocol
to manage the data, this improvement would greatly increase the long-term value of the
data. Soil cadmium may vary with paddock-scale variation in soil type, fertiliser history
or topography (Schipper et al. 2011; Staﬀord et al. 2015), and location data would allow
large-scale analysis of the inﬂuence of these factors. Precise locations would also allow
the model to identify areas within districts that have higher soil cadmium concentrations.
While Otorohanga and Matamata-Piako were identiﬁed as the districts with the highest

link to page 19 link to page 19 link to page 19 link to page 18 link to page 18 link to page 19 link to page 19 link to page 18 link to page 18 216
E. ABRAHAM
mean estimated soil cadmium concentrations, it is likely that there will be areas within
these districts that have generally higher soil cadmium concentrations, compared with
other areas with lower soil cadmium concentrations. If precise location information
cannot be provided to assess soil cadmium concentrations at a high spatial resolution,
there may be value in dividing the territorial authorities with high cadmium concen-
trations into smaller reporting areas.
A recent study has shown that soil cadmium concentrations may vary within a single
property (Staﬀord et al. 2015), with values ranging between 0.48 and 1.64 mg Cd/kg on a
Waikato property that had been used for dairy farming since the 1950s. This variation was
associated with variation in the total phosphorus concentration, in land slope, and in the
soil type. Strong variation (of around 0.6–1.3 mg Cd/kg) was encountered within a single
paddock. This variation was consistent with the statistical modelling that found, for
example, that if a farm had a mean soil cadmium value of 1.0 mg Cd/kg, the values of
soil cadmium in a single sample would have a 90% probability of being in the range
0.51–1.69 mg Cd/kg. Single samples may be over the TFMS tier 4 soil cadmium concen-
tration threshold of 1.8 mg Cd/kg, but further sampling may show that the average values
across the whole property are in a lower tier.
The data weakly constrained the overall rate of change of cadmium concentrations to
be between the 95% c.i. of −2.2% to +1.2% per year (for the period between 2007 and 2015
included in the analysis). Analysis of long-term soil cadmium data from the Whatawhata
Research Station, from between 1980 and 2010, found that cadmium has been either
decreasing or increasing, depending on the rates of fertiliser application and hill slope
(Schipper et al. 2011). Annual rates of change were between around −1% and +3%,
with a decrease in cadmium concentration on the steeper slopes and at low fertiliser appli-
cation rates. Schipper et al. (2011) ﬁtted a broken-stick model to their data, with higher
slopes before the late 1980s, when the switch to lower-cadmium fertiliser occurred.
Long-term data from the irrigation trials at Winchmore Research Station suggest that
cadmium accumulation at this Canterbury site has continued since the early 1950s,
with a close relationship between the total phosphorus fertiliser applied and the soil
cadmium (McDowell 2012; Kelliher et al. 2017). The current analysis did not attempt
to estimate variation in accumulation rate with region or land use, but estimated a
single rate of increase across all the data. Out of the 1905 identiﬁable properties in the
data, only 27 could be identiﬁed as having been sampled at diﬀerent times. The TFMS rec-
ommends sampling every ﬁve years. When the industry data have been collected over a
longer period, it would be appropriate to investigate the rate of change in soil cadmium
in more detail.
The company data showed that fertiliser cadmium concentration has remained below
the 280 mg Cd/kg P voluntary limit (apart from four slight exceedances, as discussed).
Between 2003 and 2015, the mean concentration was 184 mg Cd/kg P. There will con-
tinue to be changes in the source rock used for manufacturing fertiliser, which will
aﬀect this ratio. Data from the Whatawhata Research Station (Schipper et al. 2011),
shows that over the period when fertiliser cadmium concentrations have been lower,
accumulation rates of soil cadmium concentrations at this site have decreased. Depend-
ing on fertiliser application rates and factors such as irrigation or local topography, soil
cadmium concentrations may still be increasing (Schipper et al. 2011; McDowell 2012;
Kelliher et al. 2017).

link to page 18 link to page 18 link to page 18 link to page 18 NEW ZEALAND JOURNAL OF AGRICULTURAL RESEARCH
217
Based on an analysis of the industry soil data between 2006 and 2013, and by Cavanagh
(2014), several recommendations were made to improve the quality of the data. Some of
these recommendations, such as collecting a customer code and improving the coverage of
the sampling across New Zealand regions and main land use were followed. Cavanagh
(2014) recommended the collection of precise location information; however, to date
this information has not been made available by the fertiliser companies, due to concerns
for the anonymity of the participating farmers. In lieu of precise location, a more detailed
land use and soil classiﬁcation was proposed, but this information has not been collected.
Collection of precise location would reduce the need for this metadata as it could then be
consistently derived from external sources. Cavanagh (2014) also recommended recording
of longer-term farming practice. This recommendation could be addressed by collecting
total phosphorus data, which is directly related to long-term fertiliser use. Collection of
total phosphorus data is not necessary for the TFMS monitoring; however, these data
would be valuable for any study that aims to understand variation in soil cadmium
across the New Zealand landscape. I would also recommend that details of the sampling
(such as number of cores and number of paddocks included) were recorded in the industry
dataset. These records would allow clear identiﬁcation of which samples were from single
paddocks and which were combined composite samples from multiple paddocks, follow-
ing TFMS guidelines.
The industry data, which is collected to support the application of the TFMS, is a
unique large-scale dataset. It is the only dataset that is available for evaluating soil
cadmium concentrations across the New Zealand agricultural landscape. As time
passes, the value of these data will increase. With some improvements to the data collec-
tion, particularly collection of precise location information, collection of metadata describ-
ing the sampling protocol followed for each sample, and collection of ancillary
information on total phosphorus to allow fertiliser history to be quantiﬁed, the long-
term value of these data could be improved. I recommend that the industry revisit the rec-
ommendations raised by Cavanagh (2014), in light of the information presented here, to
keep developing this unique data collection programme.
Acknowledgements
Thanks to Greg Sneath of the Fertiliser Association of New Zealand (FANZ) for providing the
industry soil cadmium data; to Jo Cavanagh (Landcare Research) for providing the soil
cadmium data from the regional councils; and to Peter Wood (Quality Consultants of New
Zealand) for the fertiliser cadmium data. I am also grateful to Philipp Neubauer (Dragonﬂy Data
Science) for an introduction to using STAN for CAR modelling. An earlier analysis of this
dataset was presented to the 2016 Fertilizer and Lime Research Centre workshop, and I am grateful
to the many people who gave feedback following and during the preparation and presentation of the
workshop paper. Thanks to Greg Sneath (Fertiliser Association) and Ants Roberts (Ravensdown)
for comments on the draft manuscript.
Disclosure statement
No potential conﬂict of interest was reported by the author.
Funding
This work was funded by the Fertiliser Association of New Zealand.

link to page 4 link to page 14 link to page 7 link to page 3 link to page 3 link to page 2 link to page 4 link to page 6 link to page 7 link to page 3 link to page 16 link to page 15 link to page 3 link to page 4 link to page 2 link to page 15 link to page 16 218
E. ABRAHAM
References
Abraham E, Cavanagh J, Wood P, Pearson A, Mladenov P. 2016. Cadmium in New Zealand’s agri-
culture and food systems. In: Currie LD, Singh R, editors. Integrated nutrient and water manage-
ment for sustainable farming. Occasional Report No. 29. Palmerston North, New Zealand:
Massey University, Fertilizer and Lime Research Centre; p. 13.
Cadmium Working Group. 2008. Report one: cadmium in New Zealand agriculture. Wellington:
Ministry for Primary Industries. [accessed 2016 Mar 14]. https://www.mpi.govt.nz/document-
vault/10079.
Carpenter B, Gelman A, Hoﬀman MD, Lee D, Goodrich B, Betancourt M, Brubaker M, Guo J, Li P,
Riddell A. 2017. Stan: a probabilistic programming language. Journal of Statistical Software. 76
(1). Web. doi:10.18637/jss.v076.i01.
Cavanagh J. 2014. Status of cadmium in New Zealand soils: 2014. Wellington: Fertiliser Association
of New Zealand. [accessed 2016 Apr 1]. http://www.fertiliser.org.nz/includes/download.aspx?
ID=134299.
Drewry J. 2017. Soil quality state of the environment monitoring programme: annual data report
2016/17. Wellington: Greater Wellington Regional Council. Publication No GW/ESCI-T-17/101.
Fertiliser Association of New Zealand. 2016. Tiered Fertiliser Management System – for the man-
agement of soil cadmium accumulation from phosphate fertiliser applications. Wellington:
Fertiliser Association of New Zealand. [accessed 2018 Oct 24]. http://www.fertiliser.org.nz/
tools/search.aspx?sect=Site.
Fertiliser Quality Council. 2015. Code of practice for the sale of fertiliser in New Zealand.
Wellington: Fertiliser Quality Council. [accessed 2016 Mar 14]. http://fertqual.co.nz/?
ddownload=81
Gelfand AE, Vounatsou P. 2003. Proper multivariate conditional autoregressive models for spatial
data analysis. Biostatistics. 4(1):11–15.
Gelman A, Hill J. 2006. Data analysis using regression and multilevel/hierarchical models.
Cambridge: Cambridge University Press.
Hill RB, Sparling GP. 2009. Land and soil monitoring: a guide for SoE and regional council report-
ing. In: New Zealand: land monitoring forum chapter 3, soil quality monitoring. New Zealand:
Land Monitoring Forum. p. 28–89. https://www.mfe.govt.nz/sites/default/ﬁles/Land%20and%
20soil%20monitoring__A_guide_for_SoE%20and%20regional%20council%20reporting.PDF
Kelliher FM, Gray CW, Noble ADL. 2017. Superphosphate fertiliser application and cadmium
accumulation in a pastoral soil. New Zealand Journal of Agricultural Research. 60:404–422.
Kim N. 2005. Cadmium accumulation in Waikato soils. Environment Waikato Technical Report
2005/51. Hamilton: Environment Waikato. 129 p. https://www.waikatoregion.govt.nz/assets/
PageFiles/11184/TR2005-51.pdf.
Land Monitoring Forum. 2009. Land and soil monitoring: a guide for SoE and regional council
reporting. New Zealand: Land Monitoring Forum. [accessed 2018 October 24]. http://www.
mfe.govt.nz/publications/land/land-and-soil-monitoring-guide-soe-and-regional-council%C2%
A0reporting.
Lawrence-Smith EJ, Tregurtha CS. 2013. Regional environmental monitoring programme for soil
quality 2012–13: Arable & pastoral project – Final report. A report prepared for Environment
Canterbury. Plant & Food Milestone No. 51731. Plant & Food Research Contract No. 29222.
SPTS No. 9007.
Loganathan P, Hedley MJ, Grace ND, Lee J, Cronin SJ, Bolan NS, Zanders JM. 2003. Fertiliser con-
taminants in New Zealand grazed pasture with special reference to cadmium and ﬂuorine—a
review. Soil Research. 41(3):501–532.
Marchant BP, Saby NPA, Lark RM, Bellamy PH, Jolivet CC, Arrouays D. 2010. Robust analysis of
soil properties at the national scale: cadmium content of French soils. European Journal of Soil
Science. 61(1):144–152.
McDowell RW. 2012. The rate of accumulation of cadmium and uranium in a long-term grazed
pasture: implications for soil quality. New Zealand Journal of Agricultural Research. 55:133–
146. doi:10.1080/00288233.2012.662901.

link to page 15 link to page 2 link to page 15 link to page 15 link to page 2 link to page 15 link to page 15 link to page 3 link to page 15 link to page 2 link to page 2 link to page 7 link to page 7 link to page 2 NEW ZEALAND JOURNAL OF AGRICULTURAL RESEARCH
219
McDowell RW, Taylor MD, Stevenson BA. 2013. Natural background and anthropogenic contri-
butions of cadmium to New Zealand soils. Agriculture, Ecosystems & Environment. 165:80–87.
Roberts AHC, Longhurst RD, Brown MW. 1994. Cadmium status of soils, plants, and grazing
animals in New Zealand. New Zealand Journal of Agricultural Research. 37:119–129. doi:10.
1080/00288233.1994.9513048.
Salmanzadeh M, Balks MR, Hartland A, Schipper LA. 2016. Cadmium accumulation in three con-
trasting New Zealand soils with the same phosphate fertilizer history. Geoderma Regional. 7
(3):271–278.
Schipper LA, Sparling GP, Fisk LM, Dodd MB, Power IL, Littler RA. 2011. Rates of accumulation of
cadmium and uranium in a New Zealand hill farm soil as a result of long-term use of phosphate
fertilizer. Agriculture, Ecosystems and Environment. 144(1):95–101.
Sneath G. 2015. The Tiered Fertiliser Management System for managing soil cadmium in agricul-
tural soils in New Zealand. In: Currie LD, Burkitt LL, editors. Moving farm systems to improved
attenuation. Occasional Report No. 28. Palmerston North, New Zealand: Fertilizer and Lime
Research Centre, Massey University; p. 13. http://ﬂrc.massey.ac.nz/publications.html.
Staﬀord A, Anderson CW, Hedley MJ. 2015. Spatial distribution of soil cadmium in a long-term
waikato dairy farm. In: Currie LD, Burkitt LL, editors. Moving farm systems to improved attenu-
ation. Occasional Report No. 28. Palmerston North, New Zealand: Fertilizer and Lime Research
Centre, Massey University; p. 9. http://ﬂrc.massey.ac.nz/publications.html.
Staﬀord A, Cavanagh J, Roberts A. 2014. Soil cadmium – review of recent data in relation to the
Tiered Fertiliser Management System. In: Currie LD, Christensen CLC, editors. Nutrient man-
agement for the farm, catchment and community. Occasional Report No. 27. Palmerston North,
New Zealand: Fertiliser and Lime Research Center. Fertilizer and Lime Research Centre, Massey
University; p. 8. http://ﬂrc.massey.ac.nz/publications.html.
Syers JK, Mackay AD, Brown MW, Currie LD. 1986. Chemical and physical characteristics of phos-
phate rock materials of varying reactivity. Journal of the Science of Food and Agriculture. 37
(11):1057–1064.
Taylor M, Gibb R, Willoughby J, Hewitt A, Arnold G. 2007. Soil maps of cadmium in New Zealand.
Wellington: Ministry for Primary Industries. [accessed 2016 Mar 14]. https://www.mpi.govt.nz/
document-vault/10082.
Taylor MD. 1997. Accumulation of cadmium derived from fertilisers in New Zealand soils. Science
of the Total Environment. 208(1-2):123–126.
Vannoort RW, Thomson BM. 2009. New Zealand Total Diet Study: agricultural compound resi-
dues, selected contaminant and nutrient elements. Wellington: Ministry for Primary
Industries. [accessed 2016 Mar 14]. http://www.foodsafety.govt.nz/elibrary/industry/total-diet-
study.pdf.
Vehtari A, Gelman A, Gabry J. 2016. LOO: eﬃcient leave-one-out cross-validation and WAIC for
bayesian models. R package version 0.1.6. https://github.com/jgabry/loo.
Vehtari A, Gelman A, Gabry J. 2017. Practical Bayesian model evaluation using leave-one-out
cross-validation and WAIC. Statistics and Computing. 27(5):1413–1432.
Warne MJ. 2011. Review of, and recommendations for, the proposed New Zealand Cadmium
Management Strategy and Tiered Fertiliser Management System, MAF Technical Paper No:
2011/03, Prepared for the Cadmium Working Group. Wellington: Ministry for Primary
Industries. [accessed 2016 Mar 14]. https://www.mpi.govt.nz/document-vault/3659.

Document Outline