You are on page 1of 12

R E S E A R C H A N D A N A LY S I S

Spatio-Temporal Differentiation of Life


Cycle Assessment Results for Average
Perennial Crop Farm
A Case Study of Peruvian Cocoa Progression and
Deforestation Issues
Giancarlo Raschio,1 Sergiy Smetana ,1 Christian Contreras,2 Volker Heinz,1
and Alexander Mathys3
1
German Institute of Food Technologies, Quakenbrück, Germany
2
The Nature Conservancy, Lima, Peru
3
Laboratory of Sustainable Food Processing, ETH Zurich, Zurich, Switzerland

Keywords:
Summary
geo-spatial differentiation in LCA
industrial ecology The application of spatially and temporally explicit information to increase result precision is
land-use change gaining momentum in Life Cycle Assessment (LCA) studies. It is vital for the assessment of
life cycle assessment environmental impact of perennial crops with non-productive years, grown in combination
perennial crops impact assessment with shade crops. Available studies rely on differentiated life cycle inventory data for the
relative risk of deforestation
inputs in LCA or application of adapted impact assessment methodologies. This study
uses the identification of greenhouse gas emissions (GHG) hotspots (statistically significant
Supporting information is linked clusters of farms with either high or low GHG emission values) estimated from average
to this article on the JIE website LCA results and assesses a relative deforestation risk in such hotspots. A total of 1892 farms
in the Tocache province of San Martin region of Peru were evaluated between the year
2008 and 2010. Combination of average LCA results with farm size, age and deforestation
progression allowed for the identification of areas and farms with a high relative risk of
environmental impacts and potential deforestation. It was estimated that farms belonging
to high-GHG emission hotspots were twice more likely to expand their agricultural frontier
and cause deforestation than farms in low-GHG emission hotspots. Combining LCA with
geo-information systems and geostatistics is a viable path to explore the differentiation of
assessment results, which might lead to faster, more accurate, and resource-efficient ways
to tackle environmental impacts while also accounting for important environmental impacts
such as deforestation. Further research on the application of suggested approaches with
other perennial crops and other geographical areas is needed.

Introduction common principles of agricultural activities, we can observe a


wide range of real impacts on the environment caused by the di-
Farming is a major environmentally impacting stage of food
versity of the spatial, cultural, and temporal conditions affecting
production (Bessou et al. 2013; Foley et al. 2011). Despite
the agricultural practices (Cerutti et al. 2011, 2013; Bessou et al.

Conflict of interest statement: The authors have no conflict to declare.

Address correspondence to: Sergiy Smetana, German Institute of Food Technologies (DIL e.V.), Prof.-von-Klitzing-Straße 7, D-49610, Quakenbrück, Germany.
Email: S.Smetana@dil-ev.de

© 2017 by Yale University


DOI: 10.1111/jiec.12692 Editor managing review: Annie Levasseur

Volume 22, Number 6

1378 Journal of Industrial Ecology www.wileyonlinelibrary.com/journal/jie


R E S E A R C H A N D A N A LY S I S

2013). Consideration of such variations is becoming mandatory studies and available data. The authors suggested chronological
to increase the precision and avoid uncertainties and mistakes approach for the assessment of a specific stage production,
in environmental assessment. It is especially important for the spatial approach for a variety of farm stages presented in a given
results for public disclosure when false conclusions might lead area, and modeling modular assessment of the environmental
to increased impact actions of a large scale (Gavankar et al. impact based on different quality data. We argue that a modular
2015; Herrmann et al. 2014). Spatially and temporally variable approach for LCA combined with geo-spatial analysis via
results of environmental impact during cocoa farming are of a geographical information systems (GIS) can be an effective
special interest for policy makers, traders, and farm associations framework for the spatio-temporal assessment of perennial
as cocoa production is experiencing a shortage crisis to supply crops farming and the associated risk assessment of land use
global demand. They can use the data for the differentiated sup- and land-use change (LULUC) at the regional (subnational)
port of more sustainable farming activities on the one hand and scale.
identification of environmental “hotspot” areas, which require The idea of combination of LCA and GIS is not new. Such
additional attention for the reduction of negative impact, on tools complemented each other in the selection of the most
the other. appropriate locations for energy crops (Gasol et al. 2011), for
Perennial crops farming is especially sensitive to the spatio- more precise assessment of biodiversity and land-use impacts
temporal differentiation due to the different stages of perennial (Geyer et al. 2010), for regionalization of LCIA (Liu et al.
crops development, farm management, farm size, and, in many 2014; Heijungs 2012; Mastrucci et al. 2017; Mutel et al. 2012).
cases, clumped and uneven distribution (Mouron et al. 2006; At the same time, no evidence of the use of GIS tools for
Milà i Canals et al. 2006; Coltro et al. 2006; Cerutti et al. the spatial differentiation of LCA results has been found. For
2011; Bessou et al. 2013). Such spatio-temporal categorization example, carbon emissions linked to LULUC are practically
creates a complex system of perennial crops production, which not included in LCA studies on perennial crops (Bessou et al.
requires the adaptation of usual methods of life cycle assess- 2013).
ment (LCA) for the differentiation of impact assessment results Risk assessment on opposite is common as a complementary
(Basset-Mens et al. 2016; Cerutti et al. 2013). Research litera- applied method to LCA (Gemechu et al. 2016; Matthews et al.
ture considers a few potential solutions. The first solution aims 2002; Sonnemann et al. 2004; Mouron et al. 2006). Despite
toward the differentiation of input parameters in LCA through such a wide presentation of the approach, there is limited
introduction of localized spatio-temporal life cycle inventory literature on the combination of LCA, GIS, and risk assessment
(Maier et al. 2017; Ciroth et al. 2002; Finnveden and Nilsson tools.
2005). Another way to deal with the spatial differences is the The selection of the case study on cocoa farming in the
introduction of special characterization factors for the assess- province of Tocache in the San Martin region of Peru related
ment methodologies of LCA (Yang 2016; Mutel and Hellweg to the need to identify the environmental impact of new and
2009; Mutel et al. 2012; Nitschelm et al. 2016). established cocoa farms in the regions to establish proper rec-
The third solution should be the differentiation of the av- ommendations for a more sustainable farm management and
eraged LCA results according to the spatio-temporal charac- relevant policies development. This problem is becoming vital
teristics of the research area. The spatial differentiations then due to the predicted global shortages in cocoa and increasing
could include the area (size) of the farms, their shape, and demand. In Peru alone, the cultivated area of cocoa increased
their position according to the other farms, roads, or sensitive in 76,300 ha in the past 10 years (MINAGRI 2015) and pre-
areas (e.g., protected forest). Temporal characteristics are es- dicted to take over additional 28,000 hectares (ha) by 2016
pecially important for perennial crop plantations. They relate (El Comercio 2014). Thus, the development of a corresponding
to the rate of productivity, consumption of raw resources, and environmental impact assessment framework for the farm level
associated emissions. Comparing to the first two solutions, the is needed.
third approach aims at the use of average cocoa production The main objective of the study is to develop a regional
data and average characterization factors (typical for most life framework for the cocoa ecological (environmental) footprint
cycle impact assessment [LCIA] methodologies) with further assessment with the background application of LCA of an av-
geo-spatial differentiation of LCA results according to the re- erage farm. The framework should include the LCA of different
gional specifics. Even though this third solution is logical, it is stages of cocoa farm production (modules) with further combi-
not very exploited in research literature. nation of results into a model farm (performed for province of
The challenges of perennial crops assessment are connected Tocache in the San Martin region of Peru). The average LCA
also with the issue of productive and nonproductive years. results then should be differentiated according to the important
Nonproductive years could be responsible for a considerable aspects of cocoa production and distribution via means of GIS
impact on the environment depending on the type of crop and geo-spatial analysis with application of greenhouse gases’
species and farm location (6.5% to 29% of plantation life span (GHG) emission levels. Relying on the geo-spatially ranged
environmental impact) (Alaphilippe et al. 2014; Bessou et al. and distributed impacts, the framework then should identify
2014). Bessou and colleagues (2013) reviewed the studies of life the statistically significant spatial clusters or hotspots of the en-
span perennial farming LCAs and proposed a framework for the vironmental impacts and the relative risk of LULUC resulting
selection of assessment approach based on the objectives of the from the expansion of the cocoa agricultural frontier.

Raschio et al., Risk Assessment in the Supply Chain: LCA/Geostatistics 1379


R E S E A R C H A N D A N A LY S I S

Figure 1 Allocation factors applied for different functional units of the study. FU1ha = functional unit for farming of 1 hectare per year of
land to produce cocoa; FU1t = functional unit for production of 1 tonne of raw cocoa; GHG = greenhouse gas.

Methods Such a situation clearly called for proper impact allocation


rules.
Life Cycle Assessment
The assessment was performed for two functional units: (1)
Goal and Scope farming of 1 hectare per year (1 ha/yr) of land to produce cocoa
The implementation of environmental spatio-temporal (FU1ha) and (2) production of 1 tonne of raw cocoa (FU1t).
analysis of perennial crops’ production required the LCA re- These functional units are the most common in perennial LCA
sults obtained from the production of an average 1-ha cocoa studies (Vinyes et al. 2015; Cerutti et al. 2014; Mouron et al.
farm in a region (based on averaged primary data from farms). 2006). Moreover, spatial conditioning of FU was more precisely
Therefore, the goal of LCA was the identification of GHG associated with farmers’ inputs per unit of area, rather than with
emission from an average 1-ha cocoa farm in the province of production amount. At the same time, due to association of
Tocache in the San Martin region of Peru. The framework of most of the literature data with cocoa production (Utomo et al.
spatio-temporal environmental assessment of perennial crops 2016; Ntiamoah and Afrane 2008), we also preserved a mass
was tested with an impact category of global warming poten- FU.
tial (GWP) (global warming potential over the 100-year time The data used for the study were obtained from the farms
horizon; GWP100). This indicator is commonly used for the with different stages of cocoa growing. Five-year census data
representation of impacts of land-use changes and emissions included farms producing only cocoa (fourth year and older),
from agro-forestry activities (Schmidt et al. 2015; Sandin et al. cocoa and bananas (third year), and only bananas (first 2 years).
2016; Richards et al. 2017). It required certain allocation of inputs, which was performed
The study focused on the farming stage of cocoa production. according to the area occupied by the plants—the strategy sim-
Therefore, the system boundaries included the combination ilar to the approaches used in other studies (Mouron et al. 2006;
of farming activities (land preparation, production of comple- Utomo et al. 2016). Such allocation rules were applied to both
mentary shade crop [banana], and cocoa bean production before FU1ha and FU1t (figure 1). In the case of GHG emissions from
fermentation and drying1 ) and LULUC, which was quantified the use of inputs, they were distributed annually according to
for each farm by hectares of deforestation based on primary data the production amount per hectare and then adjusted to the
(Fornasiero and Graziani 2011). size of each farm (area in hectares). Raw material impact al-
The farms under consideration started producing cocoa location between banana and cacao was performed in the first
in the third year after their establishment (first year of 3 years. Regarding land-use change, in the case of FU1ha, GHG
cocoa production). Therefore, the inputs (fertilizers and emissions were equally distributed over 20 years and similarly
pesticides) in the production during the first 2 years of allocated between banana and cacao in the first 3 years. For
the plantation could not be completely allocated to cocoa the case of FU1t, GHG emissions were equally distributed over
production. During the first 2 years, farms are oriented toward 18 production years of cocoa production and allocated between
banana production: a complementary shade crop of cocoa banana and cacao in the third year (first year of cacao produc-
seedlings. At the same time, a certain proportion of resources tion) (WBCSD and WRI 2011). This way deforestation will
(land, GHG, and water) was used by the cocoa seedlings. not be allocated only to the initial years of farm establishment,

1380 Journal of Industrial Ecology


R E S E A R C H A N D A N A LY S I S

Figure 2 (a) Location of the Tocache province (study area) in Peru, (b) with its four districts and the 1,892 farms analyzed in the study.

but it will be distributed over the whole productive lifetime of a each farm (average farm LCA results). Data on land-use change
cacao farm, thus avoiding negative incentives for cocoa buyers caused by cocoa farming (deforestation effect) for each year
(i.e., delaying reporting or buying only from farms in their late (2006 through 2010) were obtained from an official data set
production years). generated by Peru’s Ministry of Environment (MINAM 2015).
The study was framed by several assumptions which allowed The changes in forest cover were determined for each year via
for the development of a more holistic assessment of environ- overlapping layers of deforestation with farms’ expansion areas.
mental impact and reduced the uncertainties of limited data This study is based on the modular approach for the assess-
on it. Given the average lifetime of a cocoa plantation, it was ment of environmental impact (Bessou et al. 2013) using an
assumed that the overall life span of the farms in the region was average annual impact for different stages of cocoa production
20 years. The study utilized the data of the first 5 years of a co- in Peru (the province of Tocache in the San Martin region).
coa farm lifetime (2 years of cocoa growth and 3 years of cocoa For the LCA, we chose OpenLCA 1.4 software (GreenDelta
production); therefore, it was assumed that the production of GmbH, Berlin, Germany) to calculate GWP100 using National
cocoa farms remains stable at the level of the fifth year (Darvin Renewable Energy Laboratory U.S. inventory database and the
2014). IPCC2013 impact assessment method (IPCC 2014). As the aim
of the study was spatio-temporal assessment of current state of
Life Cycle Inventory cocoa production for the more precise regionalized assessment
The data analysis was based on the data set on 1892 farms and establishment of allocation rules at the farm level, it re-
in the Tocache province of San Martin region (figure 2). The lied on an attributional approach. The impact of production
data set included a geo-referenced point at the estimated center changes on the further functioning of the system was not a part
of each farm as well as the information about the implementa- of the study, which excluded the use of consequential approach.
tion year of the farm, size in hectares, and annual production
per hectare. The average production per hectare for the region
Geospatial Analysis and Geographical Information
was differentiated depending on the year of cocoa plantation
Systems
(0.46 tonne/ha for the third year to 0.70 tonne/ha for the fifth
year). Moreover, the data on inputs (fertilizers and pesticides) ArcGIS 10.1 was used to analyse shapefiles of calculated
per hectare depending on the farm age (table 1) were acquired GWP100. Its application together with spatial statistic tools
from the data set provided by DEVIDA (National Commis- allowed for mapping the clusters of intensive GHG emissions
sion for Development and Life Without Drugs) from a census and their spatio-temporal analysis. Geo-spatial tools included
the institution conducted on cocoa farmers in 2011 as part of Incremental Spatial Autocorrelation analysis (ESRI 2015) that
DEVIDA’s annual monitoring activities for the performance enabled measuring the autocorrelation of a data set by identify-
assessment of its Alternative Crops Program (DEVIDA 2011). ing the distance, at which clustering events were more intense
This data set encompassed farms established between 2006 and (peak distance). Conceptualization of Spatial Relationships was
2008; therefore, it represented farms at different cocoa produc- used to reflect the inherent relationships among the features
tion stages for the time of the census in 2011. under analysis through the spatial weight matrix (ESRI 2015).
DEVIDA’s data set did not include data on land-use change; Conceptualization of Spatial Relationships represented a mod-
thus, our study complemented it with the information on land- eling approach to determine the neighborhood of influence
use change and GHG emissions from annual production for for the assessment of GHG emissions clusters. It is a helpful

Raschio et al., Risk Assessment in the Supply Chain: LCA/Geostatistics 1381


R E S E A R C H A N D A N A LY S I S

Table 1 Amount of fertilizers and pesticides inputs according to the age of the farm

Year after farm installation

Inputs (product names) Units 1 2 3 4 5

Compomaster cacao 50 kg/ha 3 4 5 6 7

Nitrophoska special 50 kg/ha 1 1 1 2 2

Urea 50 kg/ha 1 1 1 2 2

Potassium chloride 50 kg/ha 1 1 1 1.5 1.5

Triple calcium phosphate 50 kg/ha 1 1 1 1.5 1.5

Benomyl kg/ha 0 0 0.2 0.4 0.4

Matalaxyl L/ha 0 0 0.5 1 1

Matalaxyl + mancozeb L/ha 0 0 0.5 1 1

Carbendazim L/ha 0 0 0.5 1 1

Cymoxanil + mancozeb L/ha 0 0 0.5 1 1

Chlorpyrifos L/ha 0 0 0.5 1 1

Methamidophos L/ha 0 0 0.25 0.5 0.5

Note: kg/ha = kilograms per hectare; L/ha = liters per hectare.

parameter to model the influence of the spatial distribution of and subsequent deforestation; Hv – high-value hot spots; and
features (cocoa farms) on the values (GHG emissions) of such Lv – low-value hot spots.
features.
The objectives of using geospatial tools were to: (1) identify
the presence of GHG emission clusters; (2) map clusters, if Results
any; and (3) identify and analyze spatial and temporal patterns Greenhouse Gas Emissions of Cocoa Farming
of clusters of high and low GHG emission intensity from
cocoa farming. Clustering was performed according to the Application of predefined allocation rules of impact factors
statistically significant relation (99% of confidence interval between banana (shade crop) and cocoa indicated that, on
[CI]) of being in the same group (high or low GHG emission average, banana trees require 20%, 16%, and 14% of the total
intensity). amount of inputs applied per hectare of a cocoa farm in the
first 3 years. It resulted in the application of relevant allocation
factors (0.8, 0.84, and 0.86).
Relative Risk Assessment Although total deforestation conducted by cocoa farmers is
Geospatial clustering of GHG emissions into different not as significant as that from a commercial cocoa plantation,
intensity groups allowed estimation of relative risk (RR) of which can cause deforestation at an annual average rate of 500
related variables appearance. RR was calculated as a ratio that ha/yr (Finer and Novoa 2015; Harris et al. 2015), the main
compares the probability of manifesting an event or variable issue is the expansion of the agricultural frontier for cocoa at
in two groups, and it is calculated by dividing the cumulative the expense of forests. Annual GHG emissions attributable to
incidence of one group over the cumulative incidence of the cocoa production from deforestation in the study area increased
other (Boston University School of Public Health 2016). from 2 tonnes carbon dioxide equivalent (CO2 -eq) in 2008 to
The approach (equation 1) was tested with characteristics 562 tonnes CO2 -eq in 2010 based on an FU1t (the calculation
of initial deforestation (due to the farm establishment) procedure is presented in the Supporting Information available
and subsequent deforestation (further expansion of farming on the Journal’s website). In the case of FU1ha, annual GHG
area). emissions from deforestation increased from 2 tonnes CO2 -eq
in 2008 to 515 tonnes CO2 -eq in 2010. These results indicated
that, in the study area, GHG emissions from deforestation
P (i ) Hv
R R(i ) = , (1) can represent between 4% and 49% the total GHG emissions
P (i ) L v from cocoa production based on FU1t (figure 3). In the case
of a FU1ha, GHG emissions from deforestation can represent
where: RR(i) – relative risk of variable appearance; P(i) – prob- between 4% and 46% the total GHG emissions from cocoa
ability of presenting variable “i”; i – variables of analysis: initial production (figure 3).

1382 Journal of Industrial Ecology


R E S E A R C H A N D A N A LY S I S

application of country average GWP of 0.32 kilograms (kg)


CO2 -eq per 1 kg of cocoa (Ntiamoah and Afrane 2008) to the
researched region in Peru would result in overestimation of
average GHG emission levels of 36% in 2008 and underestima-
tion of about 400% for 2009 and 2010 (figure 3). Therefore, the
adoption of country-level average values for GHG emissions for
perennial crops can negatively affect the prioritization process
of areas requiring technical and financial support to mitigate
GHG emissions from agricultural practices, thus becoming
a barrier for regional low-carbon development. Performing
an LCA analysis for hundreds of farms as a first approach
to identify GHG emissions clusters might not be a plausible
Figure 3 Relative temporal distribution of GHG emissions in
first approach, either for a national program or private sector
studied farms. FU 1 ha = functional unit for farming of 1 hectare per
stakeholder. However, methodologies like the one presented
year of land to produce cocoa; FU 1 ton = functional unit for
in this study could be further developed and adapted to local
production of 1 tonne of raw cocoa; GHG = greenhouse gas; ha =
circumstances to use available data on deforestation to identify
hectare.
those clusters of farms with higher GHG emissions at the local
level. Therefore, this method could be employed as a low-cost,
Spatial Statistics Analysis preliminary filter in selecting the farms that should be priori-
tized in low-carbon strategies with a higher spatial resolution
Statistically significant low- and high-impact clusters were
than the one offered by country-level LCA averages. Finally,
indicated for 2008, 2009, and 2010. Low-value emission clusters
country-wide average impact factors will affect “responsible”
were constantly localized northwest from the marketplace at the
farmers who conduct low-carbon agriculture by overlooking
boundary between Polvora and Tocache districts, while high-
their good practices and hindering their access to green buyers.
value impact clusters were highlighted in close vicinity in the
Also, our results showed that the selection of an FU, either
Tochache district and in the Nuevo Progreso district (figure 4).
mass or area, influenced the proportion of GHG emission from
deforestation regarding total annual emissions. This article also
Relative Risk highlights the effect of the FU selection on GHG emissions
accounting. On the one hand, both FUs generated the same
Results demonstrated that farms in high-value clusters were
amount of annual GHG emissions from cacao production (use
statistically more likely to conduct land-use change than farms
of inputs). On the other hand, a FU1t resulted in up to 5%
in low-value hotspots (table 2). For clusters in 2008, there is
more total annual GHG emissions and 11% more GHG emis-
no statistically significant evidence that farms in high-value
sions from deforestation than an FU1ha. Additionally, GHG
hotspots are more likely to incur in initial deforestation (caused
emissions from land-use change represent a lower share of total
by the establishment of the farm within a 20-year period) or
annual emissions when using an FU1ha rather than an FU1t.
subsequent deforestation (caused by the expansion of the farm
Therefore, the selection of FU can affect the results of LCA,
borders during the year of analysis) when compared to farms
the argument supported by the other literature (Smetana et al.
in low-value hotspots. On the other hand, results for clusters
2015; Nemecek et al. 2016). For this reason, it becomes impor-
in 2009 and 2010 indicate that there is statistically significant
tant to take into consideration the FU used in the calculations
evidence that new farms in high-value clusters were about twice
on which GHG emission reports rely on.
as likely to establish at least part of their cocoa farming land
Cocoa plantations often involve the co-production with
in previously forested areas and to expand their farming area at
other “shade crops” in the initial years of plantation genesis
the expense of forest cover loss than farms in low-value clusters
(Utomo et al. 2016). Other LCA studies on cocoa did not con-
(table 2).
duct an explicit allocation of environmental impacts between
In both 2009 and 2010, farms belonging to low-value clusters
the main production and a supplementary crop at the early
appear to have a larger average farm area than those in high-
stages of production (Utomo et al. 2016; Ntiamoah and Afrane
value clusters (table 3). Also, in both years, farms in high-
2008). Proper allocation of the environmental impact to shade
value clusters appear to have more primary and secondary forests
crops adds an additional point to the discussion of nonproduc-
within their boundaries in comparison to farms in low-value
tive years of perennial crops LCA. Moreover, the calculations
clusters (table 3).
could be complicated using different varieties of complementary
crops. This study introduces two approaches to allocate GHG
emissions, one for an FU1t and one for an FU1ha, for those
Discussion
emissions taking place in years when shade crops are present.
Our findings indicated that the use of average environmental Furthermore, the application of specified allocation rules and
impacts at the national scale present limitations for perennial averaged GHG emission spatial distribution approach should
crops characterization at the regional scale. For example, the indicate the impact of cocoa production more precisely.

Raschio et al., Risk Assessment in the Supply Chain: LCA/Geostatistics 1383


R E S E A R C H A N D A N A LY S I S

Figure 4 High- and low-value emission clusters (tonnes CO2 -eq/tonne cocoa) in (a) 2008, (b) 2009, and (c) 2010. High and low emission
clusters are not comparable in between different years. CO2 -eq = carbon dioxide equivalent.

Our results suggest that spatial statistic tools, in particular turning into illegal crops. Therefore, farms in high-value clus-
cluster analysis also known as hotspot analysis, can provide ters would take an extensive approach to improve their cocoa
relevant instruments to understand the relation between high production volumes.
GHG emission levels from agricultural production, as well as Another point to consider would be the fact that farms pro-
initial and subsequent deforestation. The possibility of farms ducing cocoa in high-value clusters for the first year in 2009 or
association into clusters of high and low levels of GHG emis- 2010 presented a higher relative risk of being implemented in
sions allowed the identification of environmental hotspots, previously forested areas than those farms in low-value clusters.
which appear to be related to the risk of land-use change. For It might be the case that the establishment of new farms in
example, it is provoking to highlight a trend in the years 2009 2007 and 2008 (so producing cocoa for the first time in 2009
and 2010 on how farms in high-emission-level clusters have a and 2010, respectively) might have been conducted by migrant
higher RR of expanding their farming areas into forested areas farmers motivated by the incentives of subsidized and/or free
(subsequent deforestation). This result makes sense in the light inputs for agricultural production. These migrant farmers, who
of the characteristics of the farms in high-value clusters, which own no property and have little options to generate incomes
have smaller average areas, but larger cover of both primary beside agriculture, usually are prone to being engaged in slash-
and secondary forest. It is usually the case that smaller farms and-burn agriculture (Tubiello et al. 2015; Raschio et al. 2017).
will find deforestation a viable option to expand their farming The identification and prediction of high environmental
area. Also, in the areas where the application of inputs is not impacting areas might be beneficial for carbon reduction
efficient due to the provision of those inputs via an interna- incentives of chocolate producers, but might also hold certain
tional development program to eradicate illegal cocoa leaf crop negative consequences for cocoa farmers. Even though our
(DEVIDA), it can be expected that farmers might find it attrac- methodology is aiming toward the promotion of more environ-
tive to expand their farming areas claiming benefits to avoid mentally sustainable production, it might result in association

1384 Journal of Industrial Ecology


R E S E A R C H A N D A N A LY S I S

Table 2 Relative risk (RR) analysis between high-value and low-value clusters

CI 95%

Relative Risk (RR)

Significance test
Standard error
Lower limit

Upper limit

Z-statistic
Variables

P-value
Year

2008 Initial deforestation 2.09 0.20 22.24 0.52 0.61 0.553 Discarded

Subsequent deforestation 6.27 0.79 50.02 0.46 1.73 0.083 Discarded

2009 Initial deforestation 1.95 1.07 3.55 0.13 2.18 0.029 Significant

Subsequent deforestation 2.15 1.37 3.40 0.10 3.30 0.001 Significant

2010 Initial deforestation 1.94 1.01 3.74 0.14 1.99 0.046 Significant

Subsequent deforestation 1.87 1.12 3.13 0.11 2.38 0.017 Significant

Note: CI = confidence interval.

Table 3 Main characteristics of identified high-value and low-value clusters

2009 2010
Farms in high-value Farms in low-value Farms in high-value Farms in low-value
Variables clusters (total = 220) clusters (total = 474) clusters (total = 244) clusters (total = 474)

Farms conducting organic 87 239 99 226


farming
Farms participation in a farmers’ 30 190 40 193
association
Average farm size (ha) 1.17 1.53 1.24 1.51

Average primary forest within 0.9 0.17 1.42 0.29


farms (ha)
Average secondary forest within 0.46 0.29 0.42 0.29
farms (ha)
Note: ha = hectares.

of low-carbon farms into regions of high-impact clusters or clusters distribution. While outlining certain trends in their
opposite low-impact ones. In both cases, we argue that our distribution, it was possible to significantly predict only the ap-
model showed evidence that it was possible to relate emission pearance, but not the magnitude, of subsequent deforestation.
intensity at farm level with the likelihood of manifestation of Therefore, this methodology could be used in two ways: (1)
environmental variables. use LCA results to help identify clusters of farms that might
Our approach can provide a preliminary indicator of areas incur in deforestation activities and (2) use free and available
location with high land-use change risk without the necessity satellite imagery to assess deforestation and identify clusters
of developing full land-use change models. However, the appli- of farms with high GHG emission from farming processes.
cation of the suggested approach is limited to those agri-food As an accepted practice, quantification of past deforestation
companies that can trace their supply chains to the regional should be conducted via an assessment of historical land-use
level. Our approach narrows down the location of clusters, thus and land-cover change whereas the quantification of subse-
facilitates the delimitation of an assessment area for a local as- quent deforestation can be estimated via spatial prediction
sessment. We recommend that clusters should be assessed to models.
identify outliers that might respond to the specific characteris- The study relied on the assumption that all the farmers use
tics of a study area. the same annual amount of resources per hectare of a farm,
Our study relied on spatially and temporally variable data, which led to the averaged LCA results for the region. They
which resulted in variations of high-emission and low-emission might be different for other crops and areas. Even though such

Raschio et al., Risk Assessment in the Supply Chain: LCA/Geostatistics 1385


R E S E A R C H A N D A N A LY S I S

a practice is not commonly used in LCA studies, the testing Acknowledgments


indicated the possibilities of such approach application on small
The authors express their gratitude to the generous financial
regional scales. At the same time, we agree that diversification
support granted by the Alexander von Humboldt Foundation
of impact assessment model according to the group of farms
under an International Climate Protection Fellowship.
with similar conditions might improve its resolution abilities.
Further improvement of risk and geo-spatial analyses could be
performed with the consequential LCA approach with a longer
Funding Information
temporal dimension. Such approach will not only identify more
precisely the potential future impact of cocoa harvesting in the The research leading to these results has received fund-
region, but also provide a feedback on the indirect impacts of ing from the Alexander von Humboldt Foundation under an
further expansion of cocoa production (Weidema 2000; Suh International Climate Protection Fellowship.
et al. 2010; Weidema and Schmidt 2010).
Based on the available information from our data set, we
focused on GHG emission analysis only on the first 5 years Note
of a cocoa farm lifetime (from implementation of the farm
1. The study did not include the data on drying and fermentation
to the third year of cocoa production). Ideally, the method as those processes took place at the processing facilities of cocoa
we presented could be applied to a larger data set covering buyers.
longer periods and/or to other perennial crops (i.e., coffee) that
are known to be drivers of deforestation in various locations.
The investigation on the change of impacts with the use of References
other than banana shade crops could be an additional aspect
for further studies. Alaphilippe, A., J. Boissy, S. Simon, and C. Godard. 2014. Using of the
LCA methodological framework in perennial crops: Comparison
of two contrasted European apple orchards. In Proceedings of the
9th International Conference on Life Cycle Assessment in the Agri-
Conclusions Food Sector(LCA Food 2014), San Francisco, CA, USA, 8–10
October, 2014, 11–15. Tacoma, WA, USA: American Center
Addressing land-use change issues in agri-food supply chains
for Life Cycle Assessment.
demands data at an adequate level that can be provided by mod-
Basset-Mens, C., H. Vannière, D. Grasselly, H. Heitz, A. Braun, S.
els of higher spatial and temporal resolution allowing for more Payen, P. Koch, and Y. Biard. 2016. Environmental impacts of
accurate calculations of product footprint (Godar et al. 2015). imported and locally grown fruits for the French market: A cradle-
Such spatially explicit models facilitate the understanding of lo- to-farm-gate LCA study. Fruits 71(2): 93–104.
cal drivers of environmental change that motivate subsequent Bessou, C., C. Basset-Mens, C. Latunussa, A. Vélu, H. Heitz, H.
deforestation in supply chains, thus avoiding the pitfall of as- Vannière, and J. P. Caliman. 2014. LCA of perennial crops:
suming that the impacts of one unit of product are homogeneous Implications of modeling choices through two contrasted case
at the country or regional levels (Meyfroidt et al. 2014). It is studies. In Proceedings of the 9th International Conference on Life
a viable path to explore the combination of LCA with GIS Cycle Assessment in the Agri-Food Sector (LCA Food 2014), San
and geostatistics, which might lead to faster, more accurate Francisco, CA, USA, 8–10 October, 2014, 125–133. Tacoma,
WA, USA: American Center for Life Cycle Assessment.
and resource-efficient ways to tackle environmental impacts
Bessou, C., C. Basset-Mens, T. Tran, and A. Benoist. 2013. LCA
while also accounting for important environmental impacts
applied to perennial cropping systems: A review focused on the
such as deforestation. It is only through better understanding of farm stage. The International Journal of Life Cycle Assessment 18(2):
the strengths and weaknesses of suppliers that organizations 340–361.
can identify opportunities for environmental improvement Boston University School of Public Health. 2016. Measures of asso-
(Hajmohammad et al. 2013). ciation. Boston, MA, USA: Boston University School of Public
Spatial variables, such as land-use change and the temporal Health.
dynamics of environmental impacts, should be incorporated Cerutti, A. K., G. L. Beccaro, S. Bruun, S. Bosco, D. Donno, B. Notar-
into regional LCA studies of perennial crops. From one side, nicola, and G. Bounous. 2014. Life cycle assessment application
there is the potential of diversification of regional average LCA in the fruit sector: State of the art and recommendations for
results and, from the other side, it has the potential to increase environmental declarations of fruit products. Journal of Cleaner
Production 73: 125–135.
spatial and temporal precision of regionalized impact assessment
Cerutti, A. K., S. Bruun, G. L. Beccaro, and G. Bounous. 2011. A review
results. This article discussed the potential of using the first
of studies applying environmental impact assessment methods on
approach to indicate high-value impact clusters, which have a fruit production systems. Journal of Environmental Management
higher probability of emissions and deforestation in the future. 92(10): 2277–2286.
At the same time, the study has not addressed a few issues on Cerutti, A. K., S. Bruun, D. Donno, G. L. Beccaro, and G. Bounous.
results variations caused by the different crops, locations, and 2013. Environmental sustainability of traditional foods: The case
management techniques, which would be a viable approach for of ancient apple cultivars in Northern Italy assessed by multifunc-
further research. tional LCA. Journal of Cleaner Production 52: 245–252.

1386 Journal of Industrial Ecology


R E S E A R C H A N D A N A LY S I S

Ciroth, A., M. Hagelüken, G. W. Sonnemann, F. Castells, and G. Environment and Energy (Editorial Series of Italian Commod-
Fleischer. 2002. Geographical and technological differences in life ity Science Academy and Engineering Association of Messina)
cycle inventories shown by the use of process models for waste Franco Angeli, 165–176.
incinerators. The International Journal of Life Cycle Assessment Herrmann, I. T., M. Z. Hauschild, M. D. Sohn, and T. E. McKone.
7(6): 363–368. 2014. Confronting uncertainty in life cycle assessment used for
Coltro, L., A. Mourad, P. Oliveira, J. Baddini, and R. Kletecke. 2006. decision support. Journal of Industrial Ecology 18(3): 366–379.
Environmental profile of Brazilian green coffee. The International IPCC (Intergovernmental Panel on Climate Change). 2014. Climate
Journal of Life Cycle Assessment 11(1): 16–21. change 2014: Synthesis report. Contribution of Working Groups I, II
Comercio, El. 2014. Perú Podrı́a Ser El Lı́der Mundial En Exportación and III to the Fifth Assessment Report of the Intergovernmental Panel
de Cacao. [Peru could become the world leader in cacao ex- on Climate Change, edited by R. K. Pachauri and L. A. Meyer.
ports]. El Comercio. Lima, Peru. http://elcomercio.pe/economia/ Geneva, Switzerland: IPCC.
peru/peru-podria-lider-mundial-exportacion-cacao-noticia- Liu, K. F.-R., M.-J. Hung, P.-C. Yeh, and J.-Y. Kuo. 2014. GIS-based
1704459. Accessed 22 January 2014. regionalization of LCA. Journal of Geoscience and Environment
Darvin, P. 2014. Personal communication with P. Darvin, Field Expert Protection 2(2): 1–8.
on Cacao Farms Survey. DEVIDA. Lima, Peru. May 28. Maier, M., M. Mueller, and X. Yan. 2017. Introducing a localised spatio-
DEVIDA. 2011. Alternative Crops Monitoring Data for 2011-2012. temporal LCI method with wheat production as exploratory case
Edited by DEVIDA. Lima. study. Journal of Cleaner Production 140(pt 2): 492–501.
ESRI. 2015. ArcGIS Help 10.1—Spatial Statistics Toolbox. Mastrucci, A., A. Marvuglia, E. Popovici, U. Leopold, and E. Benetto.
ESRI Press. USA. http://resources.arcgis.com/en/help/main/ 2017. Geospatial characterization of building material stocks for
10.1/index.html#//005p0000000n000000. Accessed 17 July 2014. the life cycle assessment of end-of-life scenarios at the urban scale.
Finer, M., and S. Novoa. 2015. Patrones y Drivers de la Deforestación Resources, Conservation and Recycling 123: 54–66.
en la Amazonı́a Peruana [Patterns and drivers of deforestation in Matthews, H. S., L. Lave, and Heather Maclean. 2002. Life cycle im-
peruvian Amazon]. MAAP Sı́ntesis #1. pact assessment: A challenge for risk analysts. Risk Analysis 22(5).
Finnveden, G. and M. Nilsson. 2005. Site-dependent life-cycle im- www.ce.cmu.edu/greendesign/docs/life-cycle-impact.pdf. Ac-
pact assessment in Sweden. The International Journal of Life Cycle cessed 26 September 2017.
Assessment 10(4): 235–239. Meyfroidt, P., K. M. Carlson, M. E. Fagan, V. H. Gutiérrez-
Foley, J. A., N. Ramankutty, K. A. Brauman, E. S. Cassidy, J. S. Gerber, Vélez, M. N. Macedo, L. M. Curran, R. S. DeFries, et al.
M. Johnston, N. D. Mueller, et al. 2011. Solutions for a cultivated 2014. Multiple pathways of commodity crop expansion in
planet. Nature 478(7369): 337–342. tropical forest landscapes. Environmental Research Letters 9(7).
Fornasiero, P. and M. Graziani, eds. 2011. Renewable resources and https://doi.org/10.1088/1748-9326/9/7/074012.
renewable energy: A global challenge. Boca Raton, FL, USA: CRC. Milà i Canals, L., G. M. Burnip, and S. J. Cowell. 2006. Evaluation
Gasol, C. M., X. Gabarrell, M. Rigola, S. González-Garcı́a, and J. of the environmental impacts of apple production using life cy-
Rieradevall. 2011. Environmental assessment: (LCA) and spatial cle assessment (LCA): Case study in New Zealand. Agriculture,
modelling (GIS) of energy crop implementation on local scale. Ecosystems & Environment 114(2–4): 226–238.
Biomass and Bioenergy 35(7): 2975–2985. MINAGRI (Ministry of Agriculture of Peru). 2015. Agriculture and
Gavankar, S., S. Anderson, and A. A. Keller. 2015. Critical compo- livestock Dynamics 2004–2013. Chapter III. Lima: Ministry of
nents of uncertainty communication in life cycle assessments of Agriculture of Peru.
emerging technologies. Journal of Industrial Ecology 19(3): 468– MINAM (Ministry of Environment of Peru). 2015. Estrategia Na-
479. cional Sobre Bosques Y Cambio Climático—Mapa de Perdida
Gemechu, E. D., C. Helbig, G. Sonnemann, A. Thorenz, and A. Tuma. de Bosque Amazonica 2001–2013. [National Strategy on Forests
2016. Import-based indicator for the geopolitical supply risk of and Climate Change—Amazonic Forest Loss Map 2001–2013].
raw materials in life cycle sustainability assessments. Journal of edited by MINAM. Lima: MINAM.
Industrial Ecology 20(1): 154–165. Mouron, P., T. Nemecek, R. W. Scholz, and O. Weber. 2006. Manage-
Geyer, R., D. M. Stoms, J. P. Lindner, F. W. Davis, and B. Wittstock. ment influence on environmental impacts in an apple production
2010. Coupling GIS and LCA for biodiversity assessments of land system on Swiss fruit farms: Combining life cycle assessment with
use. The International Journal of Life Cycle Assessment 15(5): 454– statistical risk assessment. Agriculture, Ecosystems & Environment
467. 114(2–4): 311–322.
Godar, J. U., M. Persson, E. J. Tizado, and P. Meyfroidt. 2015. Mutel, C. L. and S. Hellweg. 2009. Regionalized life cycle assess-
Towards more accurate and policy relevant footprint analy- ment: Computational methodology and application to inven-
ses: Tracing Fi Ne-Scale socio-environmental impacts of pro- tory databases. Environmental Science & Technology 43(15): 5797–
duction to consumption. Ecological Economics 112: 25–35. 5803.
https://doi.org/doi:10.1016/j.ecolecon.2015.02.003 Mutel, C. L., S. Pfister, and S. Hellweg. 2012. GIS-based regionalized
Hajmohammad, S., S. Vachon, R. D. Klassen, and I. Gavronski. 2013. life cycle assessment: How big is small enough? Methodology
Reprint of lean management and supply management: Their role and case study of electricity generation. Environmental Science &
in green practices and performance. Journal of Cleaner Production Technology 46(2): 1096–1103.
56: 86–93. Nemecek, T., N. Jungbluth, L. Milà i Canals, and R. Schenck. 2016. En-
Harris, N., O. A. Payne, and S. A. Mann. 2015. How much Rainforest vironmental impacts of food consumption and nutrition: Where
is in that chocolate bar? Washington DC, USA: World Resources are we and what is next? The International Journal of Life Cycle
Institute. Assessment 21(5): 607–620.
Heijungs, R. 2012. Spatial differentiation, GIS-based regionalization, Nitschelm, L., J. Aubin, M. S. Corson, V. Viaud, and C. Walter. 2016.
hyperregionalization, and the boundaries of LCA. Milano, Italy: Spatial differentiation in life cycle assessment LCA applied to an

Raschio et al., Risk Assessment in the Supply Chain: LCA/Geostatistics 1387


R E S E A R C H A N D A N A LY S I S

agricultural territory: Current practices and method development. Suh, S., B. Weidema, J. H. Schmidt, and R. Heijungs. 2010. Gen-
Journal of Cleaner Production 112: 2472–2484. eralized make and use framework for allocation in life cy-
Ntiamoah, A. and G. Afrane. 2008. Environmental impacts of co- cle assessment. Journal of Industrial Ecology 14(2): 335–353.
coa production and processing in Ghana: Life cycle assessment https://doi.org/10.1111/j.1530-9290.2010.00235.x.
approach. Journal of Cleaner Production 16(16): 1735–1740. Tubiello, F. N., M. Salvatore, A. F. Ferrara, J. House, S. Fed-
Raschio, G., S. Smetana, C. Contreras, A. Mathys, and V. Heinz. 2017. erici, S. Rossi, R. Biancalani, et al. 2015. The contribution
Spatiotemporal tools for regional low-carbon development: Link- of agriculture, forestry and other land use activities to global
ing LCA and GIS to assess clusters of GHG emissions from cocoa warming, 1990-2012. Global Change Biology 21(7): 2655–2660.
farming in Peru. In Sustainability Through Innovation in Product Life https://doi.org/10.1111/gcb.12865
Cycle Design. Singapore: Springer. Utomo, B., A. A. Prawoto, S. Bonnet, A. Bangviwat, and S. H. Ghee-
Richards, M., M. Pogson, M. Dondini, E. O. Jones, A. Hastings, D. wala. 2016. Environmental performance of cocoa production from
N. Henner, M. J. Tallis, et al. 2017. High-resolution spatial mod- monoculture and agroforestry systems in Indonesia. Journal of
elling of greenhouse gas emissions from land-use change to energy Cleaner Production 134: 583–591.
crops in the UK. GCB Bioenergy 9(3): 627–644. Vinyes, E., C. M. Gasol, L. Asin, S. Alegre, and P. Muñoz. 2015. Life
Sandin, G., G. M. Peters, and M. Svanström. 2016. LCA of forest cycle assessment of multiyear peach production. Journal of Cleaner
products—Challenges and solutions. In Life cycle assessment of Production 104: 68–79.
forest products, 25–67. Part of the SpringerBriefs in Molecular WBCSD (World Business Council on Sustainable Development) and
Science series. Cham, Switzerland: Springer. WRI (World Resources Institute). 2011. Product Life Cycle Ac-
Schmidt, J. H., B. P. Weidema, and M. Brandão. 2015. A framework counting and Reporting Standard. Washington, DC, USA: World
for modelling indirect land use changes in life cycle assessment. Business Council for Sustainable Development and World Re-
Journal of Cleaner Production 99: 230–238. source Institute.
Smetana, S., A. Mathys, A. Knoch, and V. Heinz. 2015. Meat alter- Weidema, B. 2000. Avoiding co-product allocation in life-cycle assess-
natives: Life cycle assessment of most known meat substitutes. ment. Journal of Industrial Ecology 4(3): 11–33.
The International Journal of Life Cycle Assessment 20(9): 1254– Weidema, B. P. and J. H. Schmidt. 2010. Avoiding allocation in life
1267. cycle assessment revisited. Journal of Industrial Ecology 14(2): 192–
Sonnemann, A. G., F. Castells, and M. Schuhmacher, eds. 2004. In- 195.
tegrated life-cycle and risk assessment for industrial processes. Boca Yang, Y. 2016. Toward a more accurate regionalized life cycle inven-
Raton, FL, USA: CRC. tory. Journal of Cleaner Production 112: 308–315.

Supporting Information
Supporting information is linked to this article on the JIE website:
Supporting Information S1: This supporting information includes details of GHG emissions calculations from deforestation
and GHG emissions calculations from cocoa production.

1388 Journal of Industrial Ecology


Copyright of Journal of Industrial Ecology is the property of Wiley-Blackwell and its content
may not be copied or emailed to multiple sites or posted to a listserv without the copyright
holder's express written permission. However, users may print, download, or email articles for
individual use.

You might also like