Science of the Total Environment 881 (2023) 163372
Contents lists available at ScienceDirect
Science of the Total Environment
j ourna l homepage: www.e lsev ie r .com/ locate /sc i totenvThe distribution of cadmium in soil and cacao beans in PeruEvert Thomas a,⁎, Rachel Atkinson a, Diego Zavaleta a, Carlos Rodriguez b, Sphyros Lastra a,c, Fredy Yovera a,d,
Karina Arango a, Abel Pezo a, Javier Aguilar b, Miriam Tames b, Ana Ramos b, Wilbert Cruz c, Roberto Cosme c,
Eduardo Espinoza d, Carmen Rosa Chavez e, Brenton Ladd fa Bioversity International, Lima, Peru
b Servicio Nacional de Sanidad y Calidad Agroalimentaria (SENASA), Lima, Peru
c Instituto Nacional de Innovación Agraria (INIA), Lima, Peru
d Cooperativa Agraria Norandino, Piura, Peru
e Ministerio de Desarrollo Agrario y Riego del Perú (MIDAGRI), Lima, Peru
f Universidad Científica del Sur, Lima, PeruH I G H L I G H T S G R A P H I C A L A B S T R A C T⁎ Corresponding author at: Alliance of Bioversity Internati
E-mail address: e.thomas@cgiar.org (E. Thomas).
http://dx.doi.org/10.1016/j.scitotenv.2023.163372
Received 28 December 2022; Received in revised for
Available online 11 April 2023
0048-9697/© 2023 The Authors. Published by Elsevi• Cadmium content in cacao exceeds regu-
latory thresholds from some regions in
Peru and limits access to international
markets
• We developed nation-wide predictive
maps of soil and cacao bean cadmium
• The most important predictors of soil and
cacao bean cadmium are geology, precipi-
tation seasonality, rainfall, pH, and geo-
graphical location
• Elevated concentrations of soil and cacao
bean cadmium are largely restricted to
the northern parts of the country
• While in the north of Peru most cacao
farmers are impacted, at a national level,
<20%will be affected by the current regu-
lationsA B S T R A C TA R T I C L E I N F OEditor: Charlotte Poschenrieder
Keywords:
Geology
Seasonality
Precipitation
pH
Soil texture
Random forestPeru is the eighth largest producer of cacao beans globally, but high cadmium contents are constraining access to in-
ternationalmarkets which have set upper thresholds for permitted concentrations in chocolate andderivatives. Prelim-
inary data have suggested that high cadmium concentrations in cacao beans are restricted to specific regions in the
country, but to date no reliable maps exist of expected cadmium concentrations in soils and cacao beans. Drawing
on>2000 representative samples of cacao beans and soils we developed multiple national and regional random forest
models to develop predictive maps of cadmium in soil and cacao beans across the area suitable for cacao cultivation.
Our model projections show that elevated concentrations of cadmium in cacao soils and beans are largely restricted to
the northern parts of the country in the departments of Tumbes, Piura, Amazonas and Loreto, as well as some very lo-
calized pockets in the central departments of Huánuco and San Martin. Unsurprisingly, soil cadmium was the by far
most important predictor of bean cadmium. Aside from the south-eastern to north-western spatial trend of increasing
cadmium values in soils and beans, the most important predictors of both variables in nation-wide models were geol-
ogy, rainfall seasonality, soil pH and rainfall. At regional level, alluvial deposits and mining operations were alsoonal and CIAT, The Americas – Lima Office, Centro Internacional de la Papa, Avenida La Molina 1895, Lima 12, P.O. box 1558, Peru.
m 1 April 2023; Accepted 4 April 2023
er B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
E. Thomas et al. Science of the Total Environment 881 (2023) 163372associated with higher cadmium levels in cacao beans. Based on our predictive map of cadmium in cacao beans we
estimate that while at a national level <20 % of cacao farming households might be impacted by the cadmium
regulations, in the most affected department of Piura this could be as high as 89 %.1. Introduction
Cacao production in Latin America currently accounts for some 18 % of
global production, and has been on the rise in recent years, for reasons that
include its use in development programs to alleviate rural poverty, promote
peace in post-conflict regions and replace illicit crops (DEVIDA, 2017;
Abbott et al., 2018). However, access to markets is potentially limited due
to the presence of cadmium (Cd) in soils which may in turn lead to Cd accu-
mulation in cacao beans. To date this problem appears restricted to Latin
America and the Caribbean. The accumulation in beans can result in levels
above the thresholds for chocolate and cocoa products set bymarkets, most
importantly the European Union (European Commission regulation (EU)
488/2014), but also the USA state of California (Industrial Agreement Prop-
osition 65 (19/02/2018)). Limits on cadmium levels in chocolate have also
been put in place byAustralia andNewZealand (Food Standards Code Stan-
dard 1.4.1), Indonesia (IndonesianNational Standard) and the Russian Fed-
eration (SanPin 2.3.2-1078-01). Cd is a non-essential metal which can
cause health risks above certain intake levels, and the limits have been set
to protect the health of consumers (Meter et al., 2019).
The presence of cadmium in cacao farm soils of Latin-America has been
attributed to both geogenic sources such as the weathering of bedrock, allu-
vial deposition of sediments, and anthropogenic activities including the use
of cadmium-contaminated mineral fertilizers (Argüello et al., 2019; Bravo
et al., 2021; Meter et al., 2019; Tantalean Pedraza and Huauya Rojas,
2017; Zug et al., 2019). Among the most important edaphic and climatic
variables correlated with cadmium accumulation in cacao beans identified
to date are soil cadmium levels, soil characteristics such as pH, soil organic
matter and micronutrient levels, as well as seasonality of rainfall (Argüello
et al., 2019; Gramlich et al., 2018; Scaccabarozzi et al., 2020; Wade et al.,
2022).
However, while higher cadmium levels are more pronounced in Central
and Latin America than in Africa and Asia, cadmium levels above the regu-
latory thresholds are not ubiquitous across all cacao cultivation areas
(Arévalo-Gardini et al., 2019; Bravo et al., 2021). This is due to the varied
geological history of cacao producing regions, and upper water catchments
that provide sediments and water to farms downstream (Argüello et al.,
2019; Chavez et al., 2015). Because of this variation, mapping the spatial
distribution of cadmium in soils and beans is critical to guide decisions on
mitigation efforts, to estimate potential impact of the regulatory interven-
tions, and help actors in the value chain comply with regulations. This
has already been carried out for Ecuador (Argüello et al., 2019),
Colombia (Bravo et al., 2021) and Honduras (Gramlich et al., 2018). In
Peru, previous and localized studies indicate that there are areas with
high cadmium within the country (Arévalo-Gardini et al., 2019; Remigio,
2014) but to date a complete mapping has not been done.
Peru is an important cradle of cacao diversity (Thomas et al., 2012) and
has experienced continued growth of its cacao sector over the past years. It
is the eighth largest exporter of cacao globally, with about 70 % of its ex-
ports historically going to Europe. It is also the second largest exporter of
organic beans, after the Dominican Republic, the main market also being
Europe. High Cd concentrations have already led to reductions in cacao
bean exports from Peru to European markets, with farmers from the region
of Piura reporting income losses of 31 % on average (Villar et al., 2022).
Our objective here was to develop predictive maps of the expected cad-
mium concentration in soils and cacao beans across the environmental
niche suitable for cacao cultivation in Peru (Ceccarelli et al., 2021). Draw-
ing on extensive national soil and cacao bean cadmium measurements,
we constructed random forest models at ~250 m resolution using climate,
geology, terrain, soil and anthropogenic disturbance variables available as2
spatial data layers to develop national and regional spatial maps of pre-
dicted soil and bean cadmium. The random forest method can obtain
equally accurate and unbiased predictions as the krigging techniques,
used in previous Cd mapping studies (e.g. Argüello et al., 2019), but has
the advantage that it does not require any strict statistical assumptions to
bemet and is moreflexible when combining explanatory variables of differ-
ent types, among others (Hengl et al., 2018). The resulting maps allow for
estimation of the number of farmers potentially affected by cadmium limits,
can guide the identification of low cadmium areas for cacao cultivation and
help prioritize areas for mitigation action.
2. Methodology
2.1. Farm selection and sampling
Cacao is grown widely across Peru, by >100,000 farmers and covering
>180,000 ha, mostly located in the regions of Tumbes, Piura, Amazonas,
Cajamarca, San Martin, Ucayali, Huánuco, Pasco, Junín and Cusco
(MIDAGRI, 2022). To capture the variability of farms in terms of soils and
cacao genotypes, we aimed to sample 100–150 cacao trees in each of
these regions, with a lesser number in the regions of Madre de Dios and
Ayacucho where cacao production covers a much smaller area. The princi-
pal cacao cultivation areas in each region were identified based on consul-
tation with local experts. Available spatial information on cadmium in
cacao was also used to ensure that hotspots identified previously were in-
cluded (Arévalo-Gardini et al., 2019; HuamanI-Yupanqui et al., 2012;
Remigio, 2014). Samples were collected from at least 5 farms in each culti-
vation area, with farm choice based on prior coordination with key actors,
accessibility, and owner consent. Germplasm banks and clonal gardens
were also included in the sampling. Sampling was carried out between
April 2018 and September 2019 and a total of 563 farms were visited in
35 provinces of 13 departments of Peru, estimated to represent 90 % of
the producing areas. In each farm, once the consent of the owner was
given, a brief questionnaire was carried out to understand current and pre-
vious crop management (Supplementary material 1). The owner was asked
to select trees that represented the diversity of genotypes present and up to
a maximum of 6 trees were sampled per farm, resulting in a nationwide
total of 2194 trees. We collected leaf material of each tree for posterior ge-
netic characterization, but this information was not included in the current
modelling. Each selected tree was given a unique code and marked for fu-
ture reference and the GPS coordinatewas recorded. The sampling was car-
ried out using research permits N° 001 -2021-MIDAGRI-INIA/DGIA and N°
0003-2021-INIA-DGIA.
The method of sampling soil and beans for individual trees was adapted
from similar studies carried out in the region (Chavez et al., 2015; Gramlich
et al., 2018; Ramtahal et al., 2016; Zug et al., 2019). For soil samples, the
leaf litter was cleared under the crown and a soil sample was taken at
eight equidistant points 70 cm from the trunk using a soil auger to a
depth of 20 cm. The eight samples were combined to form a single compos-
ite sample. For bean samples, up to four cacao pods were harvested from
each tree. The pods were opened on site and beans stored in a plastic bag.
To avoid running lots of sampleswith cadmium levels below the limit of de-
tection, the Cd concentration from one randomly selected tree from each
farm was analysed initially. If the total soil cadmium > 0.05 mg kg−1, or
total bean Cd above the limit of quantification limit (0.047 mg kg−1; see
below), the remaining samples from the farm were also analysed.
Additionally, 334 composite soil and 484 bean samples taken at a farm
level were included in this analysis. Farm-level soil samples were collected
using a zig zag methodology with 8 samples per hectare that were
E. Thomas et al. Science of the Total Environment 881 (2023) 163372combined to provide one composite soil sample. For beans, 5 to 10 pods
were collected to give a total of approximately 1 kg of fresh beans. The rea-
son for using two sampling methodologies is because data were collected
under different projects (see funding details) each with their specific objec-
tives. The tree level sampling had the objective of understanding whichFig. 1. Sampling locations of soil and cacao bean samples. The area suitabl
3
variables (including genetics) influence variation in cadmium uptake in
beans (results will be discussed in a separate manuscript) while the second
was to assess the average cadmium concentration at farm gate. Both meth-
odologies took composite samples of soil and beans, the main difference
being that the tree-level sampling yielded multiple samples per farm.e for cacao cultivation is shown in green (sensu Ceccarelli et al., 2021).
E. Thomas et al. Science of the Total Environment 881 (2023) 163372In sum, we collected 1717 soil and 1534 bean samples from individual
trees from 563 farms, in addition to 334 composite soil and 484 bean sam-
ples from484 farms,making a total of 2051 data points for soil Cd and2018
for bean Cd (Fig. 1). The soil samples were air-dried for 1 to 2 weeks,
ground and sieved through a 2 mm mesh to give 400 g of sample. The
bean samples were drained and dried in an oven at 70 °C for 48 h on card-
board containers until a humidity of about 7 % was reached.
The total Cd concentration in soil was quantified using the analytical
services of SGS, a certified multinational laboratory with facilities in Peru.
A 300 g sample was requested by the laboratory, who digested a 2.5 g ali-
quot in agua regia (1:3 HCl:HNO3) for 1.5 h on an open block. Total soil
cadmiumwas quantified by inductively coupled plasmamass spectrometry
(ICP-MS) (Nexion 2000C, Perkin Elmer). The laboratory uses a certified ref-
erence standard (OREAS906)with 0.42±0.04mgkg−1 Cd. It carries out a
duplicate analysis on 5 % of the samples, and blanks are run every 50 sam-
ples. The limit of quantification is 0.01 mg kg−1.
The bean cadmium concentration collected at individual tree level was
quantified in the ISO certified laboratory of the Peruvian government
agency SENASA (Servicio Nacional de Sanidad Agraria del Peru), which is
the competent authority for food safety in Peru. Whole beans were frozen
overnight and finely ground in a food mill. 0.5 g of this powder was
digested in 6 ml of HNO3 and 2 ml H2O2 for at least 1.5 h at 200 °C prior
to analysis by ICP-MS. The laboratory runs blanks and duplicates every
10 samples. Internal controls (bean Cd samples) are also used. The limit
of quantification is 0.047 mg kg−1. The composite bean samples taken at
farm level were analysed in SGS, following the AOAC Official Method
2013.06, modified in 2019. The limit of detection is 0.01 mg kg−1 and
limit of quanti −1fication 0.03 mg kg . Average recuperation is 105 %
(80–110 %). Both laboratories took part in an inter laboratory ring test
that involved 24 labs from Peru, Ecuador and Colombia running internal
reference samples. The results from both laboratories yielded z scores of
<2, indicating acceptable levels of accuracy according to ISO 13528
(2015) and ISO 17043 (2010) (Dekeyrel, 2021).
To assess the importance of irrigation water as a source of Cd contami-
nation in cacao plantations, we used DGT passive membranes (Davison and
Zhang, 1994) to determine the Cd load of irrigation channels used to irri-
gate cacao plantations in LaQuemazón and Las Lomas, Piura,with contrast-
ing bean Cd concentrations. Cacao beans from LaQuemazón and Las Lomas
have among the lowest and highest Cd in the region, respectively. One
membrane was left in the irrigation channel for one week every month
over a full year at each site (a total of 12membranes per site), and themem-
branes were analysed using ICP MS following EPA Compendium Method
0.0010 IO-3.1; IO-3.5 (1999) by the certified laboratory TYPSA. The
time-averaged concentration of Cd in the DGT membrane (CDGT nmol
ml−1) was calculated using the following formula:
MΔg
CDGT ¼
DmdlApt
where M (nmol) is the mass of analyte accumulated in the binding layer,
calculated from the measured concentration; Δg is the total thickness of
the materials in the diffusion layer (0.094 cm); D mdl (cm2 s−1) is the diffu-
sion coefficient of Cd in the diffusion layer at the deployment temperature;
Ap is the physical area of the exposed lter membrane (3.14 cm2fi ) and; t
(s) is the deployment time (Davison and Zhang, 2016). A yearly average
of CDGT for each site was calculated from the data.
2.2. Variable selection
One limitation of building spatial predictions of expected cadmium con-
tent in soils and cacao beans is that predictor variables need to be available
as spatial layers covering the area of interest. For example, soil nutrients
such as zinc or manganese have been shown to be useful predictors of cad-
mium in soil or beans (Argüello et al., 2019; Bravo et al., 2021), but no
maps exist of their distribution across the cacao cultivation areas in Peru.
We therefore only used available spatial layers of geology, soil, terrain,4
geographical position, vegetation, climate, deforestation and the presence
of mining. To characterize geology, we used the map developed by
INGEMMET (Instituto Geológico Minero y Metalúrgico, Ministry of Energy
and Mines, Peru) at a scale of 1:100,000 (https://portal.ingemmet.gob.pe/
web/guest/mapa-geologico-nacional). During initial explorations of the
map in combination with our cadmium sampling data, we noticed that
many high cadmium soils tended to be located on top, or downstream of
one geological layer named “continental Quaternary Holocene” (Qh_c).
We therefore created a new categorical map identifying both the location
of the Qh_c layer and the areas downstream of it within the same water-
shed. We obtained soil type and seven major edaphic variables, from
ISRIC-World Soil Information (Hengl et al., 2017): organic carbon (SOC),
pH in H2O (pH), % sand (Sand%),% silt (Silt%),% clay (Clay%), Cation Ex-
change Capacity (CEC), Bulk density (BLD), Coarse fragments >2 mm
(CRF). For the edaphic variables we calculated a weighted mean across
0–5, 5–15, 15–30, 30–60, and 60–100 cm soil depth values to derive a sin-
gle data value for 0–100 cm. For terrain variables we used altitude, slope,
topographical position, terrain ruggedness, and direction of water flow,
constructed with the raster package for R (Hijmans, 2019). We additionally
used the vegetation map developed by the Ministry of Environment of Peru
(https://sinia.minam.gob.pe/mapas/mapa-nacional-ecosistemas-peru) to
distinguish terra firme areas from alluvial plains, based on vegetation
types. To characterize climate across the cacao growing areas we used 19
bioclimatic variables obtained from worldclim (Fick and Hijmans, 2017).
As anthropogenic disturbance variableswe considered on-site deforestation
and the presencemining activities in the upstream areas of thewatershed as
both these activities can lead to increased movement of cadmium from the
soil and bed rock and posterior deposition in cacao growing areas. For de-
forestation, we used the Global Forest Watch maps of deforestation in-
curred over the last 12 years (Global Forest Watch, 2022), while the map
of evidence of mining was constructed based on the 2019 inventory of min-
ing sites by INGEMMET. As there was a trend of increasing cadmium con-
centrations in both soil and cacao bean samples from south to north and
from east towest, we additionally included rasters of longitude and latitude
among the predictor values, correspodning to the longitude and latitude of
the grid cell centres in decimal degrees. The categorical maps available as
shapefiles (geology, vegetation) were rasterized to a spatial resolution of
7.5 arc sec which was the unit area of analysis and of model projection.
To reduce the redundancy among predictor variables, we removed co-
linear soil, climate and terrain variables based on stepwise calculations of
variance inflation factors (VIF), retaining only variables with VIFs <5.
The retained continuous variables for model construction were: Longitude
(in decimal degrees), mean diurnal temperature range (Temp Range Day;
in °C), the maximum temperature of the warmest month (Max Temp; in
°C), precipitation seasonality (Rain Season; no unit), precipitation of the
warmest quarter (Rain Warm Quart; in mm), precipitation of the coldest
quarter (Rain Cold Quart; in mm), the direction of water flow (Water
Flow Dir; no unit), the topographical position index (TPI; in m), terrain rug-
gedness index (TRI; in m), soil clay content (Clay%; in g/kg), soil sand con-
tent (Sand%; in g/kg), soil silt content (Silt%; in g/kg), soil cation exchange
capacity (CEC; inmmol(c)/kg), soil organic carbon (SOC; in dg/kg), soil pH
in water (pH). The categorical explanatory variables were soil type (Soil
Type), geological layer (Geology), the presence of continental quaternary
Holocene deposits on site or upriver in the watershed (Quatern Geology),
vegetation type (Vegetation), number of years since deforestation (Defor-
est), and presence of mining upriver in the watershed (Mining).
2.3. Statistical analyses
We developed multiple predictive random forest (RF) models of cad-
mium concentrations in soil and cacao beans using the cforest function in
the party package for R (Strobl et al., 2007). Considering that in nearly all
cases individual farms were located in different grid cells, to assess the in-
fluence on model performance of multiple observational (tree-level) data
per grid cell, versus just one measurement (composite farm-level samples),
we developed RF models both for cadmium measurements at tree/farm
E. Thomas et al. Science of the Total Environment 881 (2023) 163372level and averaged per grid cell. Furthermore, considering the dramatically
different growing conditions in cacao cultivation regions in Peru (eg pH
neutral soils and irrigation in the coastal valleys of the north vs acid soils
and high rainfall in the Amazon basin), we compared the performance of
models constructed at regional and national level. The regions considered
were (using spatial extent of departments): the coastal valleys of Tumbes
and Piura, the northern Amazon of Cajamarca and Amazonas, the central
Amazon of San Martin, Ucayali, Huánuco and Pasco, and the southern Am-
azon of Junín, Ayacucho, Cusco and Puno.
We cross-validated each of the models based on 15 iterations of model
calibrations and testing using spatial blocks with the blockCV package for
R (Valavi et al., 2019) which divides a study area (defined by the locations
of datapoints) in spatial blocks, while optimizing evenness of the number of
datapoints per block. For the national models we divided our study area in
100 km wide squares (blocks), and for the regional models 20 km sized
squares, based on approximate median spatial autocorrelation ranges of
continuous explanatory raster variables across the respective study regions,
as implemented in the blockCV package. Eachmodel calibration was based
on the data contained in a random selection of four fifths of the blocks,
while the remaining data were used for model testing. Cross-validation
with spatial blocks provides a better measure of the predictive power of a
model in areas for which no data points are available. To assess model per-
formance, we calculated the root mean squared error (RMSE), a measure of
model accuracy, the mean absolute error (MAE) and the R2 of simple linear
models between predicted and observed values.
To quantify uncertainty of model predictions, both in areas with data
and areas without data, we calculated the coefficient of variation for each
grid cell based on the spatially projected calibration models we developed
as part of the 15 cross-validation iterations.
Variable importance values were calculated using the varimp function in
the party package and their variabilitywas quantified using importance values
for each of the 15 cross-validation models mentioned above. Importance
values were based on the mean decrease in model accuracy and were stan-
dardized across runs by dividing by the value of the most important variable.
2.4. Assessing the impact on cacao cultivation areas and farming households
To estimate the proportion of cacao cultivation areas and the number of
cacao farming households potentially affected by the regulations limiting
Cd in chocolate and cocoa powder, we extracted predicted bean cadmium
values from our final projected map for the ~20,000 farms used by
Ceccarelli et al. (2021). Based on these values we calculated proportions
of farms per department according to five classes of predicted bean Cd
(0–0.5; 0.5–0.8; 0.8–1.2; 1.2–2; >2 mg kg−1). Recent publications have
proposed 0.6 mg kg−1 as an unofficial threshold used by importers
(Vanderschueren et al., 2021), but in Peru a threshold of 0.5mg kg−1 is typ-
ically used for bulk cacao and 0.8 to even 1.2 for single origin fine flavour
cacao. We then used these proportions to estimate the number of cacao
farmer households potentially affected based on numbers per department
according to the agrarian statistics of 2018 census which reported approx-
imately 84,000 cacao farmers accounting for a total production of
155,500 metric tonnes of cacao beans in Peru (MIDAGRI, 2022).
3. Results
The random forest models calibrated at tree/farm level performed consis-
tently better in predicting observed measurements than the models trained
based on average values per grid cell. In comparisons of predicted and ob-
served values RMSE increased 14 to 93 % for soil Cd and 11 to 62 % for
bean Cd, respectively. In what follows we therefore focus on the results ob-
tained for the random forest models calibrated at tree/farm level. Principal
components analysis identified two main clusters in the data and shows
that soil and bean cadmium tend to increase towards the north-western
part of the country, an areawith lower rainfall and higher rainfall seasonality,
and sandy rather than silty soils (Fig. 2). These same variables were among
the most important predictors in the nation-wide random forest models for5
soil Cd, in addition to the geology of the sampling location and soil pH
(Fig. 3). Soil cadmium content was by far the most important predictor vari-
able in the random forest model for bean Cd, in addition to the geographical
position (represented by longitude), rainfall seasonality and the climax vege-
tation type of the sampling location. Both random forestmodels showed good
performance in predicting observed soil and bean Cd (R2 of 0.77± 0.15 and
0.83± 0.03 for calibration data, respectively), but the soil Cd models scored
considerably lower in spatial block cross validations (R2=0.24±0.11) than
the bean Cd models (R2 = 0.48 ± 0.26).
The performance of random forest models developed on regional basis
were variable (R2 for soil and bean Cd models ranging from 0.36 to 0.87
and 0.22 to 0.85, respectively; Figs. 5 and 6). Interestingly, the most impor-
tant variables in the models differed between regions. Rainfall seasonality
and geology were most important in the north west departments of Piura
and Tumbeswhich have among the highest soil and bean Cd concentrations
in the country, and where cacao is exclusively grown under irrigation
owing to the very dry and highly seasonal climate. In the northern Amazo-
nian departments of Cajamarca and Amazonas by far the most important
predictors for soil Cdwere the location of the continental Quaternary Holo-
cene geological layer within each watershed, followed by rainfall and geol-
ogymore generally. Themost important predictors of bean Cd in the region
were the climatic variables relating to rainfall and temperature, after soil Cd
concentration. In the departments of San Martin, Ucayali, Huánuco and
Pasco, where most of the national cacao production is concentrated, geol-
ogy and location of the continental Quaternary Holocene geological layer
within each watershed were the most important predictors of soil Cd,
followed by soil type, while for bean Cd these were geology, climatic vari-
ables of rainfall and temperature, after soil Cd concentrations. Finally, in
the southernmost regions from Junin toMadre de Dioswhere Cd concentra-
tions in bean and soil are mostly low to very low, soil CEC and pHwere the
most important predictors of soil Cd. By contrast terrain ruggedness was the
by far most important predictor for bean Cd, even before soil Cd, though
this was the least performing random forest model of all (Fig. 6).
While themost important predictor variables tended to be similar in the
random forest models for both soil and bean cadmium, at a national and re-
gional level, the nature of the relationships between numeric predictors and
response variables was more variable. For example, while low rainfall and
high soil pH tended to be associatedwith higher soil Cd concentrations, the
opposite was true for bean Cd concentrations in the national-level random
forest models (Figs. 3 and 4). Similarly, the functional relations between
predictor and response variables often differed between partial dependence
plots obtained from national and regional random forest models, as well as
between regional models for soil and bean Cd.
As for the categorical predictors, the geological layers with highest and
lowest partial dependence scores in the nation-wide random forest model
for soil Cd corresponded with the most extreme ones of the regional
models. Soil samples collected on top of the continental Quaternary Holo-
cene geological layer (Qh_c) had consistently higher Cd concentrations
than samples from other areas.
One observable trend in bean Cd concentrations at a national level is
that they tended to be higher in areas close to lakes or rivers, or subjected
to alluvial flooding, and on the fluvial deposit geological layers. Further-
more, in the regions of Piura-Tumbes and San Martin-Huánuco-Ucayali-
Pasco, cacao beans collected in areas with active mining tended to have
higher Cd concentrations than in other areas.
The year average of CDGT Cd inwater in Las Lomas, where soil and bean
Cd concentrations of our samples were 2.80 ± 0.82 mg kg−1 and 7.12 ±
2.70 mg kg−1, respectively, was 0.315 ± 0.267 μg l−1, compared to
0.085 ± 0.108 μg l−1 for the La Quemazón area where average soil and
bean Cd were 0.54± 0.1 mg kg−1 and 0.72± 0.36 mg kg−1. Considering
that approximately 8000m3 water per hectare is used for irrigating cacao,
and mature plantations have approximately 800 cacao trees per hectare,
this means an average yearly influx of 3.15 ± 2.67 and 0.85 ± 1.08 mg
of bioavailable Cd to the soil from irrigation water per cacao tree in the
Las Lomas and Quemazón areas respectively. Cd concentrations in irriga-
tion water showed seasonal patterns, with highest values observed in Las
E. Thomas et al. Science of the Total Environment 881 (2023) 163372
Fig. 2. Biplot of a principal component analysis of soil and bean cadmium and continuous explanatory variables considered here. The colour gradient indicates regions of the
highest (dark red) to lowest (light green) densities of data points, through use of Kernel density estimations. Temp Range Day, mean diurnal temperature range; Max Temp,
max temperature of thewarmest month; Rain Season, precipitation seasonality; RainWarmQuart, precipitation of thewarmest quarter; Rain Cold Quart, precipitation of the
coldest quarter; TPI, topographical position index; TRI, terrain ruggedness index; Clay%, soil clay content; Sand%, soil sand content; Silt%, soil silt content; CEC, soil cation
exchange capacity; SOC, soil organic carbon; pH, soil pH in water.Lomas between June and October and in La Quemazón between May and
June (Fig. S27).
Spatially explicit predictions of soil and bean Cd by the random forest
models showed high consistency formost areas. The coefficient of variation
was lower than 0.26 and 0.36 for 95 % of the grid cells for soil and bean Cd
respectively (Fig. 7).
Our findings suggest that approximately 16 to 45 % of cacao farming
households in Peru might be impacted by the regulation, having concentra-
tion of Cd in beans above 0.8 and 0.5 mg kg−1, respectively (Fig. 8). These
levels are commonly used as upper threshold by buyers of fine flavour and
conventional cacao beans, respectively.Most of these farms are found in the
northern parts of the country in the departments of Tumbes, Piura, Amazo-
nas and Loreto (with 59–89 %, 89–100 %, 51–79 %, 27–95 % of farms in
each respective region potentially impacted), as well as some very localized
pockets in the central departments of Huánuco and San Martin. In terms of
the national cacao production in Peru, 9.2 to 41 %might be affected by the
Cd restrictions, for the 0.8 and 0.5 ppm thresholds, respectively (Fig. 8).
4. Discussion
Here we present the first maps of Cd in soils and cacao beans at a na-
tional scale for Peru. With 2051 data points from 1047 farms, it represents
the most comprehensive collection of data for any cadmiummapping exer-
cise to date (Ecuador 159 farms, 560 samples (Argüello et al., 2019),
Honduras 55 farms, 110 samples (Gramlich et al., 2018), Colombia 1827
data points, one per farm but only soil analysis (Bravo et al., 2021).6
Although our analysis is limited by fewer soil parameters per sample
than, for example Argüello et al. (2019), our modelling approach allows
for the integration of spatially explicit layers available at the national
level which leads tomore power in spatial predictions beyond the sampling
areas. Further, comparison of the functional relations between predictor
and response variables obtained for nation-wide and regional random for-
est models permits making inferences about their causal or circumstantial
nature. Climate, soil and terrain variables with consistent trends in partial
dependence curves at different scales and geographies are likely to play a
causal role in explaining soil and/or bean cadmium. For example, higher
rainfall tended to be associated with lower soil Cd concentrations both at
national and regional level, potentially due to higher likelihood of leaching.
Variables with contradicting or inconsistent trendsmight on the other hand
point to either regional differences, or the possibility that these variables
serve as proxies for other variables not considered in our analysis. For ex-
ample, inconsistent trends observed for soil texture variables at national
and regional levels might reflect the effect that differences in clay
minerology may have. This is masked when simply using percent clay in
soil texture analysis. Regardless of the underlying mechanisms, the results
clearly point to the complexity of predicting Cd uptake across a country
with enormous differences in geology, soils, climate, and topography.
4.1. Soil Cd
The most important variable explaining Cd in soil is geology indicating
that for cacao, high levels of cadmium in beans is not manmade but based
E. Thomas et al. Science of the Total Environment 881 (2023) 163372
Fig. 3. Predicted soil Cd (mg kg−1) in areas suitable for cacao cultivation (sensu Ceccarelli et al., 2021). The map (A) shows a projected random forest model calibrated with
all data points and using the predictors shown in the right-hand-side dot plot, ranked in terms of importance (B). Model performancemetrics and variable importances (B) are
based on test and train data obtained from 15 cross validations in spatial blocks. A comparison of predicted versus observed soil Cd based on the final random forest models
trained using all data for production ofmap in A yielded an R2 of 0.81 (C; plotted on log scale axes). Partial dependence curves of the sixmost important continuous variables
are displayed at the bottom (D). For partial dependences of categorical variables please refer to Figs. S1–2; partial dependences curves of all continuous variables are given in
Fig. S3.on natural soil levels. The low usage of fertilizers, even if they are contam-
inated with cadmium suggests they are not an important input of Cd
for farms in Peru (McLaughlin et al., 2021). The importance of the geolog-
ical formation has been noted previously both in Latin America and the Ca-
ribbean (Argüello et al., 2019; Gramlich et al., 2018), and further afield
(Birke et al., 2017; Marchant et al., 2010). Our results show that aside
from the localized importance of different geological layers, layers that
formed during the Quaternary Holocene in continental Peru seem to have
a higher likelihood of contributing higher Cd concentrations to soils.
These are young layers and may have experienced less leaching. More7
broadly, the geological formations associated with elevated soil Cd concen-
trations in our dataset show a degree of spatial clustering, such that there is
a clear gradient of increasing soil Cd concentrations from south to north and
east to west. The soils with highest Cd are found in the coastal valleys of
Piura and Tumbes in the northwest of the country, an areawith very limited
and highly seasonal rainfall which could explain why rainfall seasonality is
an important predictor of soil Cd models both at national level and for the
Tumbes-Piura region. Seasonality of rainfall was an important predictor in
other regions too, but in the opposite direction with higher seasonality as-
sociated with lower soil Cd. It is therefore likely that seasonality covaries
E. Thomas et al. Science of the Total Environment 881 (2023) 163372
Fig. 4. Predicted Cd in cacao beans in areas suitable for cacao cultivation (sensu Ceccarelli et al., 2021). The map (A) shows a projected random forest model calibrated with
all data points and using the predictors shown in the right-hand-side dot plot, ranked in terms of importance (B). Model performancemetrics and variable importances (B) are
based on test and train data obtained from 15 cross validations in spatial blocks. A comparison of predicted versus observed bean Cd based on the final random forest models
trained using all data for production of map in A shows yielded an R2 of 0.84 (C; plotted on log scale axes). Partial dependence curves of the six most important continuous
variables are displayed at the bottom (D). For partial dependences of categorical variables please refer to Figs. S4–5; partial dependences curves of all continuous variables are
given in Fig. S6.with spatial trends in soil type and geology among others. For example, in
the highly seasonal department of Cajamarca soil Cd is much lower than
in Amazonas department which has a more tropical rainforest climate, in
addition to different soil types and underlying geology.
Even though Cd is more available in acidic soils, alkaline soils in our
study region are associated with higher Cd. This would appear to be linked
to soil formation and weathering, as well as the importance of alluvial soils
whose composition reflects their geological origin. Accordingly, the inverse
association between rainfall and soil Cd content, both at national and re-
gional level is likely to reflect increased migration of naturally occurring
cadmium down the soil profile or through run-off leaching of Cd in soils
under stronger rainfall regimes (Kabata-Pendias and Szteke, 2015;
Rieuwerts, 2007). The typical neutral to alkaline soils from the Piura-
Tumbes region have suffered less leaching than acidic soils typical of the8
Amazon basin, have a different type of clay (2:1 clays as opposed to 1:1
clays in the Amazon basin, see below) and thus may retain a higher
heavy metal composition regardless of their higher soil pH.
The positive relation we found between CEC and total soil Cd both in
the nation-wide and regional models, apart from the Piura-Tumbes region
where the opposite is found, is expected as soils with a higher cation
binding capacity are likely to adsorb Cd more strongly. The opposite re-
lation in the Piura-Tumbes region might be because the lowland soils in
this region have a higher percentage of 2:1 clay minerals (e.g. montmo-
rillonite) than tropical soil clays in the other parts of the country. These
are rich in cations (especially Magnesium and Sodium), and have a
higher pH. The difference in this predictor is thus probably based on
soil origin and type rather than anything else. This is discussed in
more detail in the next section.
E. Thomas et al. Science of the Total Environment 881 (2023) 163372
Fig. 5. Variable importance scores (A), comparison of predicted versus observed soil Cd based on final random forest models trained using all data (B; plotted on log scale
axes), model performance metrics (C) and partial dependence curves of the six most important continuous variables (D) for each of the four regional random forest
models predicting soil Cd (mg kg−1). Variable importance scores and model performance metrics are based on 15 cross validations in spatial blocks. For partial
dependences of categorical variables and partial dependence curves of all continuous variables in regional soil Cd models please refer to Figs. S7–17.
9
E. Thomas et al. Science of the Total Environment 881 (2023) 163372
Fig. 6. Variable importance scores (A), comparison of predicted versus observed bean Cd based on final random forest model trained using all data (B; plotted on log scale
axes), model performance metrics (C) and partial dependence curves of the six most important continuous variables (D) for each of the four regional random forest models
predicting bean Cd (mg kg−1). Variable importance scores and model performance metrics are based on 15 cross validations in sptial blocks. For partial dependences of
categorical variables and partial dependence curves of all continuous variables in regional bean Cd models please refer to Figs. S18–26.
10
E. Thomas et al. Science of the Total Environment 881 (2023) 163372
Fig. 7. Uncertainty maps of predicted soil (A) and cacao bean Cd (B), expressed as coefficient of variation calculated based on 15 cross validations in spatial blocks.4.2. Cacao bean Cd
The availability of Cd to plants in general is complex and depends upon
factors such as pH, organicmatter content, soil texture andmineralogy, cat-
ion exchange capacity, electrical conductivity, macro- and micro-nutrient
content and the presence of microorganisms (Adriano, 1986; Correa
et al., 2021; McLaughlin et al., 2021; Shahid et al., 2016; Singh et al.,
1995). Previous studies tend to agree that key soil properties for cadmium
bioavailability and subsequent accumulation in cacao beans include total
soil cadmium, pH, organic matter, geological substrate and soil sand con-
tent. The effect of clay is inconsistent and factors such as micronutrients
and CEC have not been studied in detail yet (Arévalo-Gardini et al., 2019;
Argüello et al., 2019; Barraza et al., 2017; Gramlich et al., 2018, 2017).Fig. 8. Proportions of farms within 5 ranges of predicted average cacao bean Cd con
production levels expressed by department in Peru.
11Unsurprisingly, in this study the most important predictor of bean Cd at
a national level is soil Cd, followed by geology, suggesting that the nature of
the geological parent material influences both total soil Cd content and its
availability. However, bean Cd is not only influenced by the geological
layer upon which the tree is growing, but also layers located in upstream
areas. Both ancient and more recent alluvial and fluvial deposits, as well
as ancient lakes and riverbeds, were associated with higher bean Cd, de-
pending on the nature of the parent material. This suggests that rivers
and streams running through areas with high levels of cadmium can deliver
not only nutrients but also cadmium and other heavy metals released by
weathering of the bed rock or mining to agricultural areas downstream,
confirming previous research. Gramlich et al. (2018) suggested that the de-
position of sediments from river flooding may be a key source of topsoilcentrations (mg kg−1), and the corresponding estimated numbers of farmers and
E. Thomas et al. Science of the Total Environment 881 (2023) 163372cadmium in their study sites in Honduras. In Peru, Llatance et al. (2018) re-
corded differences in cadmium concentrations in soil samples taken from
non-inundated (<0.008 mg kg−1), inundated (0.043 mg kg−1) and
semi-inundated soils (0.11 mg kg−1) in which less water is retained but
for a longer period of time. A collaborative study in Ecuador led by the
French cooperative Ethiquable and the French Research Institute for Devel-
opment (IRD) similarly found that farms that were regularly flooded by the
river had the highest cadmium concentrations in cacao beans (with concen-
trations reaching 4.3 mg kg−1; Maurice L., pers. com.). In line with this,
within the watershed of the Santiago River in the department of Amazonas
we found a 3.5-fold difference in Cd translocation factor (Cd bean/Cd soil)
comparing farms found in the alluvial plain that are flooded naturally at
least every decade with hillside farms that are not.
Water does not have to carry high levels of cadmium to affect plant up-
take: several factors such as saline water conditions (high electrical conduc-
tivity), and flood-drought cycles can increase the availability of cadmium
present in the soil (Singh et al., 1995). While there are no studies looking
at the effect of flooding on cadmium accumulation in cacao, in rice this
has been studied in detail and there appears to be an interaction between
micronutrient, soil type, water content and Cd speciation (de Livera et al.,
2011; Rassaei et al., 2020). A flooded soil is anaerobic. Under these condi-
tions, Fe and Mn oxyhydroxides dissolve while metal sulphides precipitate.
This results in increased sorption sites for Cd, and precipitation as CdS, in
both cases reducing its availability. During drying conditions Fe and Mn
oxyhydroxides precipitate, taking them out of solution and resulting in an
increased ratio of Cd to Fe and Mn, which, in addition to a decrease in pH
(as the sulphides dissolve) results in increased bioavailability of Cd and
thus its uptake by the plant.
In the Piura-Tumbes region cacao is grown exclusively under irrigation.
Our data show that the Cd load of irrigation water can be substantial but
varies depending on the water origin. Leaching of Cd can be exacerbated
by mining and land degradation or other operations (Oporto et al., 2007;
Smolders et al., 2003; Sun et al., 2010; Yang et al., 2006; Zhai et al.,
2008), which could explain why water in the Las Lomas irrigation canal
which originates from an active mining area upstream contained four
times more Cd than water used for irrigation in La Quemazón which origi-
nates from an area with fewer mines and different bedrock. In a study
across Ecuador Argüello et al. (2019) similarly found that the bean samples
with the highest cadmium concentration (5.28–10.4mg kg−1) came from a
farm in a regionwith artisanalmining. However, asmentioned above, Cd in
irrigation water is only part of the story as Cd bean concentrations in Las
Lomas cacao trees were ten times higher than in La Quemazón, despite
only a four times higher Cd load in irrigationwater. This could be explained
by higher levels of soil Cd and by differences in soil characteristics that en-
hance Cd uptake by the plant.
In accordance with soil Cd, among the remaining most important vari-
ables explaining bean Cd are geographical location, rainfall, rainfall season-
ality and pH. However, there are some key differences in the direction of
the effects. Thus, while there is a positive relation between pH as a predic-
tor and soil Cd, in beans it is negative. This can be explained by the higher
solubility of Cd under acidic conditions leading to a higher translocation
factor by the cacao plant (McLaughlin et al., 2021). There is a positive rela-
tion between rainfall and bean Cd in all regions except for Piura. The under-
lying explanationmay be related to soil pH and CEC. Heavy rains can cause
nutrient leaching and result in soil acidification which increases Cd avail-
ability and thus uptake. In Piura, with much lower rainfall and generally
neutral to alkaline soils, this process is unlikely to happen. The type of
clay present is also different: Type 2:1 clays (e.g. montmorillonite) predom-
inate in Piura, and type 1:1 clays (e.g. kaolinite) in the Amazon basin. In the
former CEC does not change with pH as it is dependent on the permanent
charges of soil particles, while in the latter pH and organic matter content
play much more important roles (Solly et al., 2020; Haghiri, 1974)). This
means that that 2:1 clays are less prone to acidification under heavy rainfall
(Adriano, 1986). It is of note in Piura that there is a negative relation be-
tween the predictor soil CEC and Cd in bean. At higher CEC there is higher
capacity for soil particle surfaces to retain cations which can lead to a12decrease in cadmiumbioavailability. As CEC decreases there is an increased
competition between H+ and Cd2+ ions for binding sites which results in
cadmium desorption from soil particles into the soil solution.
4.3. Implications and way forward
While the regulatory Cd thresholds are for the final products (chocolate
and derivatives), buyers of cacao beans interpret these to inform their pur-
chases. Thus, although the limits in the existing regulations range from 0.3
to 0.8 mg kg−1 Cd depending on the product, buyers of beans for chocolate
bar production often request concentrations of Cd of <0.8 mg kg−1 (Meter
et al., 2019). Using this 0.8 mg kg−1 threshold, our results show that, al-
though Cd accumulation in cacao beans is of serious concern in some spe-
cific locations, affecting >89 % of farms in Piura for example, in general
cacao production and cacao farmers in Peru are not expected to suffer sig-
nificant economic losses due to Cd regulations imposed by the European
Union and other markets.
The production of cacao in Peru has two main markets: bulk cocoa and
fine flavour cocoa. Bulk cacao is mainly used for the manufacturing of
cocoa butter and powder, as well as high-volume mainstream chocolate
products and is usually mixed with cocoa from other countries prior to pro-
duction of thefinal product. This makes up themajority of Peru's cacao pro-
duction, focussed in the centre and south of the country (Huánuco, San
Martin, Ucayali, Pasco). Bulk cacao is further divided into organic and con-
ventional, with 37 % of Peruvian exports to the EU being organic
(9600 tonnes). While the organic bulk cacao market is not origin-specific,
it is likely to be highly sensitive to Cd levels. It is interesting to note that
the EU has dramatically increased imports of organic cacao from Sierra
Leone since 2020, a country with no known issues of Cd accumulation in
beans (Profound, 2022). Fine flavour cacao is a niche product where varie-
ties with interesting flavours are sold directly to chocolate makers. Here or-
igin plays a key role in marketing due to unique flavour profiles which are
linked to a specific cacao cultivar or varieties, the particular environmental
conditions where it is grown, or both. Not only are flavour profiles of inter-
est, but buyers usually require organic and fairtrade certification. Fine fla-
vour cacao in Peru can garner a higher price than conventional bulk
cacao, and is thus an important option for small-scale farmers
(Tschartntke et al., 2023). To date, the most famous fine flavour cacaos
come from Piura, Amazonas and Cusco, but there are many more entering
the market (Villar et al., 2022).
As themapping presented here indicates, the location of a farm is key to
predicting levels of cadmium in beans. For bulk cacao, a buyer sources
beans across a wide geographical area, so that cadmium thresholds can
be met through mixing of the beans from high and low Cd areas. For exam-
ple, Villar et al. (2022) show that in Huánuco, farmers to date have not re-
ported economic losses due to cadmium levels even though Cd bean
concentration is high in farms located in alluvial soils along the Huallaga
river.We assume this is due to themixing of this cacaowith beans produced
further away from the river. This is despite some buyers requiring maxi-
mum concentrations of as low as 0.3 mg kg−1, an interpretation of the reg-
ulatory limit for Cd in cocoa powder for final consumption. Hence the
0.5 mg kg−1 limit commonly used by bulk cacao buyers is unlikely to im-
pact this market segment significantly. However, it waits to be seen if
bulk cacao for the organic market has been affected. In contrast, Piura is
home to a highly prized cultivar (cacao Blanco de Piura)which has a unique
flavour profile that even varies by valleywithin the department. It is in high
demand byfine flavour cacao buyers and bean to bar chocolate makers and
typically fetches prices well above global stock market rates. However,
owing to the high bean Cd values inmost of Piura, many farmers are no lon-
ger able to sell their cacao to those high-profit markets, and are forced to
sell their cacao to national markets, resulting in moderate to severe income
loss to farmers with limited resources and disintegration of associations
(Villar et al., 2022).
The predictive maps we present here are available through the website
www.cacacodiversity.org. This allows users to obtain (ranges of) measured
values of Cd levels in beans and soil in grid cells where these are available,
E. Thomas et al. Science of the Total Environment 881 (2023) 163372and predicted values in other cells within the suitable areas of cultivated
cacao calculated using random forestmodels calibrated on all observational
data. For predicted values we provide 95 % confidence intervals calculated
from predicted values of each of the 15 cross-validation models in spatial
blocks. While the maps have reasonable predictive power, they are limited
to the layers available at a national level, which also restricts our under-
standing of the issue. First the explanatory variables have uncertainty issues
of their own (Fick and Hijmans, 2017; Hengl et al., 2017) with unclear im-
plications on soil and bean Cd predictions. Second, the spatial resolution of
the explanatory variable rasters (here ~5.4 ha) is generally lower than that
of the cacao plantations and hence grid cell values represent an average of
the different land uses covered by it. For example, a single grid cell may
contain old growth forest and crop land with marked differences in SOC
(Duarte-Guardia et al., 2020). Third, the explanatory variables could fur-
ther be improved through the inclusion of important predictors of bean
Cd such as key macro and micronutrients which are currently not available
as spatial data layers. It should also be acknowledged that the map and
analysis has been carried out at regional scale and we are aware of substan-
tial variation even within a farmer's field. This may be in part due to varia-
tion in farm management and genetic makeup of the cacao (Lewis et al.,
2018). However, we believe that the maps presented here are an important
step in understanding the origin of cadmium accumulation in cacao, and
add to the increasing volume of information available to enable better deci-
sionmaking to help actors in the cacao value chain adapt to the regulations
for Cd, and thus minimize its impact on small scale rural communities.
CRediT authorship contribution statement
Evert Thomas: Conceptualization, Formal analysis, Funding acquisi-
tion, Investigation, Methodology, Project administration, Resources, Super-
vision, Validation, Visualization, Writing – original draft, Writing – review
& editing. Rachel Atkinson: Conceptualization, Formal analysis, Funding
acquisition, Investigation, Methodology, Project administration, Resources,
Supervision, Validation, Visualization, Writing – original draft, Writing –
review & editing. Diego Zavaleta: Data curation, Investigation. Carlos
Rodriguez: Investigation. Sphyros Lastra: Data curation, Investigation.
FredyYovera: Investigation.Karina Arango:Data curation, Investigation.
Abel Pezo: Investigation. Javier Aguilar: Investigation, Project adminis-
tration, Resources, Validation. Miriam Tames: Investigation, Resources.
AnaRamos: Investigation, Resources.Wilbert Cruz: Investigation, Project
administration, Resources. Roberto Cosme: Project administration, Re-
sources. Eduardo Espinoza: Investigation. Carmen Rosa Chavez: Con-
ceptualization. Brenton Ladd: Investigation, Resources, Writing – review
& editing.
Data availability
The authors do not have permission to share data.
Declaration of competing interest
The authors declare that they have no known competing financial inter-
ests or personal relationships that could have appeared to influence the
work reported in this paper.
Acknowledgements
We are grateful to all participating cacao farmers, field collaborators
and other partners for collaborationwithfield data collection. This work re-
ceived financial support from the Peruvian Ministry of Agricultural Devel-
opment and Irrigation (MIDAGRI, Peru) through the STC-CGIAR fund, the
Government of Peru through the Public inversion program for Amazonian
watersheds (PIICA-1) implemented by the Plan Binacional de Desarrollo
de la Region Fronteriza Peru-Ecuador, Capitulo Peru, USAID through the
USDA-led program ‘Cacao seguro’ (Funding Opportunity number USDA-
FAS-10960-0700-10.-19-0007), the European Commission programme on13Development-Smart Innovation through Research in Agriculture (DESIRA)
through the Clima-LoCa project (contract number FOOD/2019/407-158),
and of the CGIAR Fund Donors.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi.
org/10.1016/j.scitotenv.2023.163372.References
Abbott, P., Benjamin, T., Burniske, G., Croft, M., Fenton, M., Colleen, R., Lundy, M., Rodriguez
Camayo, F., Wilcox, M., 2018. An Analysis of the Supply Chain of Cacao in Colombia.
United States Agency for International Development.
Adriano, D., 1986. Trace Elements in the Terrestrial Environment. 1st ed. Springer, New York.
Arévalo-Gardini, E., Arévalo-Hernández, C.O., Baligar, V.C., He, Z.L., 2019. Heavy metal accu-
mulation in leaves and beans of cacao ( Theobroma cacao L.) in major cacao growing re-
gions in Peru. Sci. Total Environ. 606, 792–800.
Argüello, David, Chavez, E., Lauryssen, F., Vanderschueren, R., Smolders, E., Montalvo, D.,
2019. Soil properties and agronomic factors affecting cadmium concentrations in cacao
beans: a nationwide survey in Ecuador. Sci. Total Environ. 649, 120–127. https://doi.
org/10.1016/j.scitotenv.2018.08.292.
Barraza, F., Schreck, E., Lévêque, T., Uzu, G., López, F., Ruales, J., Prunier, J., Marquet, A.,
Maurice, L., 2017. Cadmium bioaccumulation and gastric bioaccessibility in cacao: a
field study in areas impacted by oil activities in Ecuador. Environ. Pollut. 229,
950–963. https://doi.org/10.1016/j.envpol.2017.07.080.
Birke, M., Reimann, C., Rauch, U., Ladenberger, A., Demetriades, A., Jähne-Klingberg, F.,
Oorts, K., Gosar, M., Dinelli, E., Halamić, J., 2017. GEMAS: cadmium distribution and
its sources in agricultural and grazing land soil of Europe — original data versus clr-
transformed data. J. Geochem. Explor. 173, 13–30. https://doi.org/10.1016/j.gexplo.
2016.11.007.
Bravo, D., Leon-Moreno, C., Martínez, C.A., Varón-Ramírez, V.M., Araujo-Carrillo, G.A.,
Vargas, R., Quiroga-Mateus, R., Zamora, A., Rodríguez, E.A.G., 2021. The first national
survey of cadmium in cacao farm soil in Colombia. Agronomy 11, 761. https://doi.org/
10.3390/agronomy11040761.
Ceccarelli, V., Fremout, T., Zavaleta, D., Lastra, S., Imán Correa, S., Arévalo-Gardini, E.,
Rodriguez, C.A., Cruz Hilacondo, W., Thomas, E., 2021. Climate change impact on culti-
vated and wild cacao in Peru and the search of climate change-tolerant genotypes. Divers.
Distrib. 27, 1462–1476. https://doi.org/10.1111/ddi.13294.
Chavez, E., Superior, E., Esp, L., Li, Y., 2015. Concentration of cadmium in cacao beans and its
relationship with soil cadmium in southern Ecuador. Sci. Total Environ. 533, 204–214.
https://doi.org/10.1016/j.scitotenv.2015.06.106.
Correa, J., Ramírez, R., Ruíz, O., Leiva, E., 2021. Effect of soil characteristics on cadmium ab-
sorption and plant growth of Theobroma cacao L. seedlings. J. Sci. Food Agric. 101.
https://doi.org/10.1002/jsfa.11192.
Davison, W., Zhang, H., 1994. In situ speciation measurements of trace components in natural
waters using thin-film gels. Nature 367, 546–548. https://doi.org/10.1038/367546a0.
Davison, W., Zhang, H., 2016. Principles of measurements in simple solutions. In: Davison, W.
(Ed.), Diffusive Gradients in Thin-films for Environmental Measurements. 2021. Profi-
ciency Test on the Determination of Total Cd Concentration in Cacao Samples: Final Re-
port as Part of Project STDF/PG/681 and the Clima-LoCa Project. Cambridge University
Press, Cambridge Dekeyrel, J. Leuven, Belgium.
de Livera, J., McLaughlin, M.J., Hettiarachchi, G.M., Kirby, J.K., Beak, D.G., 2011. Cadmium
solubility in paddy soils: effects of soil oxidation, metal sulfides and competitive ions. Sci.
Total Environ. 409, 1489–1497. https://doi.org/10.1016/j.scitotenv.2010.12.028.
Dekeyrel, J., 2021. Proficiency Test on the Determination of Total Cd Concentration in Cacao
Samples: Final Report As part of project STDF/PG/681 and the Clima-LoCa project Leu-
ven, Belgium.
DEVIDA, 2017. Estrategia Nacional de Lucha contra las Drogas 2017–2021. Comisión
Nacional para el Desarrollo y Vida sin Drogas, Peru.
Duarte-Guardia, S., Peri, P., Amelung, W., Thomas, E., Borchard, N., Baldi, G., Cowie, A.,
Ladd, B., 2020. Biophysical and socioeconomic factors influencing soil carbon stocks: a
global assessment. Mitig. Adapt. Strateg. Glob. Chang. https://doi.org/10.1007/
s11027-020-09926-1.
Fick, S.E., Hijmans, R.J., 2017. WorldClim 2: new 1-km spatial resolution climate surfaces for
global land areas. Int. J. Climatol. https://doi.org/10.1002/joc.5086.
Global Forest Watch (2022) https://www.globalforestwatch.org. Consulted May 2022
Gramlich, A., Tandy, S., Andres, C., Chincheros Paniagua, J., Armengot, L., Schneider, M.,
Schulin, R., 2017. Cadmium uptake by cocoa trees in agroforestry and monoculture sys-
tems under conventional and organic management. Sci. Total Environ. 580, 677–686.
https://doi.org/10.1016/j.scitotenv.2016.12.014.
Gramlich, A., Tandy, S., Gauggel, C., López, M., Perla, D., Gonzalez, V., Schulin, R., 2018. Soil
cadmium uptake by cocoa in Honduras. Sci. Total Environ. 612, 370–378. https://doi.
org/10.1016/j.scitotenv.2017.08.145.
Haghiri, F., 1974. Plant uptake of cadmium as influenced by cation exchange capacity, or-
ganic matter, zinc, and soil temperature. J. Environ. Qual. 3, 180–183. https://doi.org/
10.2134/jeq1974.00472425000300020021x.
Hengl, T., Mendes de Jesus, J., Heuvelink, G.B.M., Ruiperez Gonzalez, M., Kilibarda, M.,
Blagotić, A., Shangguan, W., Wright, M.N., Geng, X., Bauer-Marschallinger, B., Guevara,
M.A., Vargas, R., MacMillan, R.A., Batjes, N.H., Leenaars, J.G.B., Ribeiro, E., Wheeler,
I., Mantel, S., Kempen, B., 2017. SoilGrids250m: global gridded soil information based
on machine learning. PLoS ONE 12. https://doi.org/10.1371/journal.pone.0169748.
E. Thomas et al. Science of the Total Environment 881 (2023) 163372Hengl, T., Nussbaum, M., Wright, M.N., Heuvelink, G.B.M., Gräler, B., 2018. Random forest as
a generic framework for predictive modeling of spatial and spatio-temporal variables.
PeerJ 6, e5518. https://doi.org/10.7717/peerj.5518.
Hijmans, R.J., 2019. Introduction to the ‘raster’ Package (Version 2.9-5). pp. 1–26.
HuamanI-Yupanqui, H.A., Huauya-Rojas, M.Á., Mansilla-Minaya, L.G., Florida-Rofner, N.,
Neira-Trujillo, G.M., 2012. Presencia de metales pesados en cultivo de cacao (Theobroma
cacao L.) orgánico. Acta Agron. 61, 339–344.
Kabata-Pendias, A., Szteke, B., 2015. Trace Elements in Abiotic and Biotic Environments. CRC
Press.
Lewis, C., Lennon, A.M., Eudoxie, G., Umaharan, P., 2018. Genetic variation in bioaccumula-
tion and partitioning of cadmium in Theobroma cacao L. Sci. Total Environ. 640–641,
696–703. https://doi.org/10.1016/j.scitotenv.2018.05.365.
Llatance, W.O., Gonza Saavedra, C.J., Guzmán Castillo, W., Pariente Mondragón, E., 2018.
Bioacumulación de cadmio en el cacao (Theobroma cacao) en la Comunidad Nativa de
Pakun, Perú. Rev. For. Perú 33, 63. https://doi.org/10.21704/rfp.v33i1.1156.
Marchant, B., Saby, N., Lark, R., Bellamy, P., Jolivet, C., Arrouays, D., 2010. Robust analysis of
soil properties at the national scale: cadmium content of French soils. Eur. J. Soil Sci. 61,
144–152.
McLaughlin, M.J., Smolders, E., Zhao, F.J., Grant, C., Montalvo, D., 2021. Managing cadmium
in agricultural systems. Advances in Agronomy. Elsevier, pp. 1–129 https://doi.org/10.
1016/bs.agron.2020.10.004.
Meter, A., Atkinson, R., Laliberte, 2019. Cadmium in Cacao From Latin America and the Ca-
ribbean. A Review of Research and Potential Mitigation Solutions. CAF, Caracas.
MIDAGRI, 2022. Perfil productivo y competitivo de los principales cultivos del sector.
Oporto, C., Vandecasteele, C., Smolders, E., 2007. Elevated cadmium concentrations in potato
tubers due to irrigation with river water contaminated by mining in Potosí, Bolivia.
J. Environ. Qual. 36, 1181–1186. https://doi.org/10.2134/jeq2006.0401.
Profound, 2022. Entering the EuropeanMarket for Organic Cocoa. CBI Ministry of Foreign Af-
fairs, The Netherlands.
Ramtahal, G., Yen, I.C., Bekele, I., Bekele, F., Wilson, L., Maharaj, K., Harrynanan, L., 2016.
Relationships between cadmium in tissues of cacao trees and soils in plantations of
Trinidad and Tobago. Food Nutr. Sci. 07, 37–43. https://doi.org/10.4236/fns.2016.
71005.
Rassaei, F., Hoodaji, M., Abtahi, S.A., 2020. Cadmium speciation as influenced by soil water
content and zinc and the studies of kinetic modeling in two soils textural classes. Int.
Soil Water Conserv. Res. 8, 286–294. https://doi.org/10.1016/j.iswcr.2020.05.003.
Remigio, J., 2014. Determinacion de procedimientos, interpretacion de resultados de analisis
y elaboracion de interrelaciones de los diferentes estudios para determinar la
concentracion de cadmio en los granos de cacao. Central Piurana de Cafetaleros
CEPICAFE, Piura.
Rieuwerts, J.S., 2007. The mobility and bioavailability of trace metals in tropical soils: a re-
view. Chem. Speciat. Bioavailab. 19, 75–85.
Scaccabarozzi, D., Castillo, L., Aromatisi, A., Milne, L., Castillo, A.B., Muñoz-Rojas, M., 2020.
Soil, site, and management factors affecting cadmium concentrations in cacao-growing
soils. Agronomy 10. https://doi.org/10.3390/agronomy10060806.
Shahid, M., Dumat, C., Khalid, S., Niazi, N.K., Antunes, P.M.C., 2016. Cadmium bioavailabil-
ity, uptake, toxicity and detoxification in soil-plant system. In: de Voogt, P. (Ed.), Reviews
of Environmental Contamination and Toxicology. Reviews of Environmental Contamina-
tion and Toxicologyvol. 241. Springer International Publishing, Cham, pp. 73–137.
https://doi.org/10.1007/398_2016_8.
Singh, B.R., Narwal, R.P., Jeng, A.S., Almas, Å., 1995. Crop uptake and extractability of cad-
mium in soils naturally high in metals at different pH levels. Commun. Soil Sci. Plant
Anal. 26, 2123–2142. https://doi.org/10.1080/00103629509369434.14Smolders, A.J.P., Lock, R.A.C., Van der Velde, G., Medina Hoyos, R.I., Roelofs, J.G.M., 2003.
Effects of mining activities on heavy metal concentrations in water, sediment, and macro-
invertebrates in different reaches of the Pilcomayo River, South America. Arch. Environ.
Contam. Toxicol. 44, 0314–0323. https://doi.org/10.1007/s00244-002-2042-1.
Solly, E.F., Weber, V., Zimmermann, S., Walthert, L., Hagedorn, F., Schmidt, M.W.I., 2020. A
critical evaluation of the relationship between the effective cation exchange capacity and
soil organic carbon content in Swiss forest soils. Front. For. Glob. Change 3.
Strobl, C., Boulesteix, A.-L., Zeileis, A., Hothorn, T., 2007. Bias in random forest variable im-
portance measures: illustrations, sources and a solution. BMC Bioinf. 8, 25.
Sun, L.-N., Zhang, Y.-F., He, L.-Y., Chen, Z.-J., Wang, Q.-Y., Qian, M., Sheng, X.-F., 2010. Ge-
netic diversity and characterization of heavy metal-resistant-endophytic bacteria from
two copper-tolerant plant species on copper mine wasteland. Bioresour. Technol. 101,
501–509. https://doi.org/10.1016/j.biortech.2009.08.011.
Tantalean Pedraza, E., Huauya Rojas, M.Á., 2017. Distribución del contenido de cadmio en los
diferentes órganos del cacao CCN-51 en suelo aluvial y residual en las localidades de
Jacintillo y Ramal de Aspuzana. Rev. Investig. Agroproducción Sustentable 1, 69.
https://doi.org/10.25127/aps.20172.365.
Thomas, E., van Zonneveld, M., Loo, J., Hodgkin, T., Galluzzi, G., van Etten, J., 2012. Present
spatial diversity patterns of Theobroma cacao L. in the neotropics reflect genetic differen-
tiation in pleistocene refugia followed by human-influenced dispersal. PLoS ONE 7,
e47676. https://doi.org/10.1371/journal.pone.0047676.
Tscharntke, T., Ocampo-Ariza, C., Vansynghel, J., Ivañez-Ballesteros, B., Aycart, P.,
Rodriguez, L., Ramirez, M., Steffan-Dewenter, I., Maas, B., Thomas, E., 2023. Socio-
ecological benefits of fine-flavor cacao in its center of origin. Conserv. Lett. 16, e12936.
https://doi.org/10.1111/conl.12936.
Valavi, R., Elith, J., Lahoz-Monfort, J.J., Guillera-Arroita, G., 2019. blockCV: an r package for
generating spatially or environmentally separated folds for k-fold cross-validation of spe-
cies distribution models. Methods Ecol. Evol. 10, 225–232. https://doi.org/10.1111/
2041-210X.13107.
Vanderschueren, R., Argüello, D., Blommaert, H., Montalvo, D., Barraza, F., Maurice, L.,
Schreck, E., Schulin, R., Lewis, C., Vasquez, J.L., Pathmanathan, U., Chavez, E., Sarret,
G., Smolders, E., 2021. Mitigating the level of cadmium in cacao products: reviewing
the transfer of cadmium from soil to chocolate bar. Sci. Total Environ. 781, 146779.
Villar, G., Yovera, F., Pezo, A., Thomas, E., Roscioli, F., Sandy da Cruz, R., Jimenez, E., Lopez,
A., Aguilar, F., Espinoza, E., Davila, C., Chavez Hurtado, C., Lastra, S., Zavaleta, D.,
Charry, A., Atkinson, R., 2022. Caracterización socioeconómica de las cadenas de valor
de cacao con énfasis en la problemática de cadmio en Piura y Huánuco, Perú. Alianza
Bioversity & CIAT, Lima Peru Available at: https://cgspace.cgiar.org/handle/10568/
125328.
Wade, J., Ac-Pangan, M., Favoretto, V.R., Taylor, A.J., Engeseth, N., Margenot, A.J., 2022.
Drivers of cadmium accumulation in Theobroma cacao L. beans: a quantitative synthesis
of soil-plant relationships across the Cacao Belt. PLoS ONE 17, e0261989. https://doi.
org/10.1371/journal.pone.0261989.
Yang, Q.W., Lan, C.Y., Wang, H.B., Zhuang, P., Shu, W.S., 2006. Cadmium in soil–rice system
and health risk associated with the use of untreated mining wastewater for irrigation in
Lechang, China. Agric. Water Manag. 84, 147–152.
Zhai, L., Liao, X., Chen, T., Yan, X., Xie, H., Wu, B., Wang, L., 2008. Regional assessment of
cadmium pollution in agricultural lands and the potential health risk related to intensive
mining activities: a case study in Chenzhou City, China. J. Environ. Sci. 20, 696–703.
Zug, K.L.M., Huamaní Yupanqui, H.A., Meyberg, F., Cierjacks, J.S., Cierjacks, A., 2019. Cad-
mium accumulation in Peruvian cacao (Theobroma cacao L.) and opportunities for miti-
gation. Water Air Soil Pollut. 230, 1–18. https://doi.org/10.1007/s11270-019-4109-x.