See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/306032757

Unsustainable landscapes of deforested Amazonia: An analysis of the

relationships among landscapes and the social, economic and environmental
profiles of farms at different ages fo...

Article  in  Global Environmental Change · September 2016

DOI: 10.1016/j.gloenvcha.2016.04.009

CITATIONS READS

11 824

39 authors, including:

Patrick Lavelle Sylvain Dolédec

Sorbonne Université Claude Bernard University Lyon 1
433 PUBLICATIONS   18,407 CITATIONS    155 PUBLICATIONS   10,344 CITATIONS   

SEE PROFILE SEE PROFILE

Xavier Arnauld de Sartre Thibaud Decaëns

French National Centre for Scientific Research Université de Montpellier
116 PUBLICATIONS   388 CITATIONS    202 PUBLICATIONS   4,997 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Rethinking Ecology View project

Crise econômica e estratégias empresariais: o Polo Siderúrgico de Carajás e as transformações pós-2008 View project

All content following this page was uploaded by Patrick Lavelle on 26 September 2016.

The user has requested enhancement of the downloaded file.

 Global Environmental Change 40 (2016) 137–155

Contents lists available at ScienceDirect

Global Environmental Change

journal homepage: www.elsevier.com/locate/gloenvcha

Unsustainable landscapes of deforested Amazonia: An analysis of the

relationships among landscapes and the social, economic and
environmental profiles of farms at different ages following
deforestation
Patrick Lavellea,b,* , Sylvain Dolédecc , Xavier Arnauld de Sartred, Thibaud Decaënse,
Valery Gondf , Michel Grimaldia , Johan Oszwaldg , Bernard Huberth , Bertha Ramirezi,
Iran Veigaj , Simão de Souzaj, William Santos de Assisj , Fernando Michelottij,
Marlucia Martinsk , Alexander Feijool , Pierre Bommelm, Edna Castañedai ,
Patricia Chaconn , Thierry Desjardinsa , Florence Dubsa , Erika Gordilloi, Edward Guevarab ,
Steven Fonteb , Maria del Pilar Hurtadob , Philippe Lenao , Tamara Limap ,
Raphaël Marichala , Danielle Mitjaq , Izildinha Mirandap , Tupac Oteror,
Catarina Praxedesk , René Poccardm, Pascale de Roberto , Gamaliel Rodriguezi,
Catalina Sanabrian , Stéphanie Tselouikoa , Alexander Velasquezi , Elena Velasquezr,
Jaime Velasquezi
a
Université Paris 6 UPMC/Institut de Recherche pour le Développement, UMR Biogéochimie des Ecosystèmes Continentaux, Paris, France
b
Centro Internacional de Agricultural Tropical, Cali, Colombia
c
Université Lyon 1, UMR 5023 LEHNA, Villeurbanne, France
d
Université de Pau et des Pays de l’Adour, UMR CNRS 5603 SET, Pau, France
e
Université de Rouen, UPRES-EA1293 ECODIV, Mont Saint-Aignan, France
f
CIRAD, UPR Bsef, F-34398 Montpellier, France
g
Université de Rennes 2, UMR CNRS 5664, LETG–COSTEL, France
h
Institut National de la Recherche Agronomique, Avignon, France
i
Universidad de la Amazonia, Florencia, Caqueta, Colombia
j
Universidade Federal do Pará, Belem, PA, Brazil
k
Museu Paraense Emilio Goeldi, Belém, PA, Brazil
l
Universidad Tecnológica de Pereira, Facultad de Ciencias Ambientales, Pereira, Colombia
m
Centre International de Recherche en Agriculture et Développement, Montpellier, France
n
Universidad del Valle, Cali, Colombia
o
Institut de Recherche pour le Développement, Paris, France
p
Universidade Federal Rural da Amazônia, Belém, Brazil
q
Institut de Recherche pour le Développement, UMR ESPACE-DEV, Montpellier, France
r
Universidad Nacional de Colombia sede Palmira, Palmira, Colombia

A R T I C L E I N F O A B S T R A C T

Article history:
Received 28 August 2015 In Amazonia, our knowledge of the trade-offs and possible thresholds in the relationships among social,
Received in revised form 18 April 2016 economic and environmental parameters remains quite limited and hinders the design of sustainable
Accepted 25 April 2016 socio-environmental systems. To fill this gap, we analyzed relationships among landscape metrics, socio-
Available online 6 August 2016 economic patterns, biodiversity and soil-based ecosystem services within a total of 51 farms located at 6
sites of the Colombian and Brazilian Amazon. Farms were representative of an initial set of 274
Keywords: and they represented colonization ages from 10 to 80 years and a range of public policies found in the
Socioecological systems region.
Amazonia
Cluster analysis separated farms in 7 types of production systems according to 5 main criteria (size of
Landscapes
the farm, human capital, incomes, farm products and production intensity) selected from an initial set of

* Corresponding author at: Université Paris 6 UPMC/Institut de Recherche pour le Développement, UMR Biogéochimie des Ecosystèmes Continentaux, Paris, France.
E-mail addresses: Patrick.Lavelle@ird.fr, plavelle48@gmail.com (P. Lavelle).

http://dx.doi.org/10.1016/j.gloenvcha.2016.04.009
0959-3780/ã 2016 Elsevier Ltd. All rights reserved.
138 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155

Biodiversity 18 criteria. Biodiversity was summarized into a composite index Bd built with data from 8 different
Ecosystem services groups: trees, shrubs, grasses and forbs, birds, Saturnidae and Sphingidae moths, Drosophilidae,
Social wellbeing earthworms and ants. Provision of ecosystem services was quantified by a composite indicator of 6 sub-
Economic efficiency
indicators of soil hydrological functions, C storage and chemical fertility. Increasing intensity of
production systems was linked to a significant decrease of indicators of natural capital biodiversity (Bd)
and soil based ecosystem services (Es) with 20% and 37.3% variance explained, respectively. No
relationship was observed between production systems and an indicator of human wellbeing (Sb) based
on a set of 5 criteria identified with the farmers. Findings indicate that early colonizers migrate when a
certain level of development has been achieved (as a result of their activities) and are replaced by
wealthier populations.
An overall indicator of sustainability (Su) – that combines production efficiency (Ep), Sb,Bd and Es
indices (ranging from 0.1–1.0) – decreases significantly with the landscape intensification (Li) with 18.7%
variance directly explained by this relationship. Su was also significantly related to production systems
(36.4% variance explained): while this indicator remained relatively stable with a value of 0.5 across the
early and intermediate phases following deforestation, it dropped down considerably (0.2) for
production systems based on cattle ranching on highly degraded pastures with less than 2% forest cover
remaining. Restoration with sylvopastoral systems allowed some farms of this former group to increase
sustainability to a value of 0.35 after less than 5 years. Agroforestry systems on sites deforested at the
same time maintained values around 0.5, as they allowed the maximum production efficiencies and
maintained relatively high values of the Biodiversity (Bd) and Ecosystem services (Es) indexes. This is
evidence that beyond the general negative trade-off between human development and natural capital
observed in Amazonia, agro ecological options to revert the trend are quite promising. A general
methodological approach for the reconstruction of sustainable landscapes in farms of the deforested
Amazonian region is proposed as a conclusion.
ã 2016 Elsevier Ltd. All rights reserved.

1. Introduction as biodiversity and provision of ecosystem services (Gunderson

With nearly 70% of forest cover remaining intact, the Amazon
region houses unrivalled levels of species diversity for a number of
taxonomic groups of plants, vertebrates and invertebrates (Da Silva
et al., 2005). It is emblematic of the challenges faced by human
development in a context of global environmental degradation and
climate change (Millennium Ecosystem Assessment, 2005).
Experts state that forest conservation is necessary to minimize
regional climatic disturbances and limit global temperature
increase to 2  C at the end of the century (Lenton, 2011). They
express concern about possible effects of climate change on the
integrity of this region (Giles, 2006). Others still consider this vast
and little populated area as a possible new frontier for expanding
food production (Simon and Garagorry, 2005; Morton et al., 2006).
Much emphasis has been placed upon creating large conservation
areas to prevent irreversible biodiversity losses that this option
would trigger (Peres, 2005; Walker et al., 2009; Gibson et al., 2011).
However, the threat of deforestation leading to an irreversible
shift to savanna has been expressed (Nepstad et al., 2008;
Lawrence et al., 2007; Staver et al., 2011). Decision support
strategies for conservation planning are now increasingly
discussed (Game et al., 2013; Gardner et al., 2009), while
considerably less effort has been directed towards improving
land-use efficiency in already cleared lands by trying to
understand trade-offs among ecosystem services and landscape
organization (Godar et al., 2012; Kareiva et al., 2007; Steffan-
Dewenter et al., 2007). In Amazonia, deforested lands comprise a
gradient from early phases of settlement (with relatively high
levels of forest cover) to more mature stages, where extensive
livestock production and/or intensive agriculture may sustain
development for a few decades before starting to decline
(Rodrigues et al., 2009). Although natural forests may be present,
they have all suffered some degree of perturbation.
Theory predicts that the socio-economic conditions prevailing
in the region and the production systems that are selected
determine the land cover mosaic of deforested landscapes as well
and Holling, 2002; Mattison and Norris, 2005; Nelson et al., 2009).
However, very few data have been collected thus far to test and
quantify these relationships in Amazonia and facilitate the
design of policies that simultaneously address the economic,
environmental and social aspects of development (Carvalho et al.,
2001; Grieg-Gran et al., 2005; Boerner et al., 2007; Foley et al.,
2007). In addition, linkages between the decrease in social and
economic performance, often observed after 3–4 decades follow-
ing forest colonization, and an overall impairment of
natural capital have not been clearly demonstrated (Rodrigues
et al., 2009).
In order to fill these important gaps, we analyzed the
relationships and trade-offs among the different social, economic
and environmental components of farming landscapes, while
testing the following hypotheses (Fig. 1).
H1. The production systems selected by farmers following
deforestation determine the subsequent composition and
structure of landscape and landuse intensity. Depending on
resources available (labor, finances, knowledge, equipment)
farmers convert forest to different types of agroecosystems and
determine the nature, size, and shape of managed plots and
their distribution in the farm landscape.
H2. Production systems, defined by human and financial
resources available, commodities produced on the farm and
incomes generated, directly affect production efficiency.
H3. Production systems also affect social wellbeing, such that
well-equiped and well-trained farmers are expected to have a
better chance to implement production systems with a high
productivity and/or commercial value and fit in competitive
marketing chains.
H4. Well performing production systems are therefore expected
to result in satisfactory levels of well being.
 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155
Fig. 1. Relationships hypothesized among the different components of Amazonian socio-environmental systems.
H5. Landscape composition and structure and landuse intensity
determine the provision of ecosystem services by soils. These
include hydrological functions (infiltration, storage, cleaning of
water), climate regulation through C storage and greenhouse
gas emissions (Grimaldi et al., 2014), primary production
affected by the management of chemical fertility, pollination,
pests and diseases, as well as soil formation and conservation
through aggregation processes and management of plant
cover.
H6. The quality and abundance of habitats provided by these
transformed ecosystems and landscapes determine the number
and functions of species that will compose local biodiversity.
H7. Biodiversity is linked to ecosystem services via multiple
physical, chemical and biological effects on soils and agro-
ecosystems.
H8. Biodiversity and soil ecosystem services are also linked to
human wellbeing, especially in poor communities that depend
more on local natural resources than wealthier communities
(MEA, 2005).
139
We finally designed a composite indicator of sustainability
based on social, economic and environmental variables to evaluate
trade-offs between these components following deforestation and
identify the drivers and necessary conditions for a sustainable
development. Variations of this indicator, expected to be
significantly linked to landscape management intensity and
production systems, should provide insight on the trade-offs
and relationships among the six socio-environmental components
analyzed in our study (Fig. 1).
Since landscape change is generally too slow to permit
monitoring in the limited time frame of a research project, we
adopted the “space for time” approach to studying the evolution of
socio-environmental systems through time. We also had to select
our sites in different geographical regions since the spatial
pattern of the deforestation processes would not allow us to find
coexisting sites covering the whole range of deforestation ages. We
therefore selected six sites that form a chronosequence ranging
from recently deforested areas (10 years) to those that were
cleared long ago (60–80 years; Fig. 2). Recently deforested
sites were selected in Brazil (Pará state), while older sites were
located in Colombia (Caquetá Department). We were thus able to
cover the desired sequence of ages and have distinct contexts
of development policy, with different legal status of the
140 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155

Fig. 2. Location of the study sites: two regions (1: Pará, Brazil) and 2: Caquetá, Colombia with six sites ranked from LU 1 to LU 6 according to land use intensity. Sites are
separated in Brazil and contiguous in Colombia.

settlements in Brazil and implementation of restoration policies in
Colombia.
2. Site characteristics and general methodology
2.1. General approach
We used widely tested and simple methods to measure all
relevant variables for each disciplinary field in order to assess a set
of farms within each of six sites, representing various stages of
transition from forest. We then reduced the initial set of several
hundred variables into a representative and easier to manage, set
of synthetic indicators that would express the largest possible
proportion of total variance in each data set. Reducing the
information into a low number of indicators would allow for the
testing of significant relationships among all indicators
(e.g., landscape intensification vs. biodiversity or ecosystem
services).
The initial set of 306 farms (51 per site) was reduced to 274 after
32 farms had to be discarded when, for example, farmers did not
allow for the research to be conducted, or information was
incomplete and/or obviously erroneous. Another set of farms was

Table 1
Main characteristics of the surveyed sites. Total woody cover is the sum of forests (primary and secondary) and fallows of different ages, but not tree-based crops.

Municipality X (UTM, Y (UTM, Start of Average distance Average Area of Total woody Main characteristics
State/Dept. WG584 WG584 Colonization to paved road or area of remaining cover% in
Country Mini/Maxi Mini/Maxi regional capital (km) farms original sampled
(ha) forest% farms
LU 1 Pacaja Parà, 493585.7 9582189.4 1978 35 82 70 75 Spontaneous colonization on
BR 496152.3 9591279.2 early pioneer front
LU 2 Nova Ipixuna 679602.5 9473375.9 1982 50 81 40 52 Agro-extrractivist project with
Parà, BR 686282.9 9467677.1 collective tenure + pastures
LU 3 Eldorada 625263.4 9343195.9 1995 38 25 44 48 Occupied by the landless
Carajas, Parà, 603403.3 9352609.8 workers movement
BR
LU 4 Belen de los 2324742.8 10187231.1 1950 52 49 2 19 Perennial crops in a rubber
A./Morelia 2312112.4 10173341.7 plantation with pastures
Caqueta, CO
LU 5 Florencia, 2352560.7 10178497.1 1940 20 65 6 18 Mixed pastures and perennials
Caqueta, CO 2335720.1 10163170.7
LU 6 Florencia, 2311340.8 10168671.4 1950 20 184 2 2 Dominated by long established
Caqueta, CO 2320397.3 10184388.2 degraded pastures
 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155
removed from the analysis when they had been abandoned or were
no longer dedicated to representative farming activities. We first
assessed these 274 farms for social and economic conditions; a
subset of 54 (9 per site) farms representative of this sample was
then analyzed for landscape metrics, ecosystem services and
biodiversity.
2.2. Site characterization
We conducted our study in two Amazonian regions with
distinct histories of colonization and management. A set of 51
contiguous farms forming landscape “windows” of 50–150 km2
was selected at each of the six sites, 3 in Brazil (Pará state) and 3
in Colombia (Caquetá Department), where smallholder agricul-
ture has been the basis for rural settlement (Fig. 2 and Table 1).
Site selection, based on knowledge gained in earlier research
activities and satellite observations was guided by the need
to have a gradient of land use intensification across the six
sites.
The selected sites represented distinct stages of colonization,
from 13 to 30 years in the Brazilian sites LU 1 to LU 3, to 58–68
years in Colombian sites LU 4 to LU 6 (Table 1). The proportion of
original forest remaining and total woody cover (including
secondary forest and old fallows) decreases regularly from the
LU 1 site (75% woody cover), to 52 and 48% in LU 2 and 3
respectively, 19 and 18% in LU 4 and 5, and only 2% in LU 6 (Fig. 3).
The size of farms is highly variable, from 25 ha on average in LU 3 to
184 ha in LU 6, as a result of local farm management strategies and
public policies. Distance to a main road and/or town also varies
significantly (Table 1).
Although our sampling design did not include any large extents
of pristine forest (i.e. the starting point of the deforestation
process), it has some areas of relatively intact forest and is
representative of typical environmental conditions that currently
prevail in colonized areas of Amazonia.
2.3. Socio-economic characterization
Production systems were first characterized by a set of 18
variables (Table 2). Preliminary analyses of this data set (not shown
here) allowed us to distinguish five multi qualitative variables that
Fig. 3. Landscape composition at the 6 different study sites.
141
best described and discriminated farms: farm size, human capital,
incomes from different farm products, productivity per unit of
labor and off-farm income. We used these variables to make a
classification of production systems and analyze their covariations
with socio-economic and environmental indicators measured at
each farm.
A Multiple Correspondance analysis (MCA) of the five multi
qualitative variables listed in Table 2, size of the farm, human
capital (5 categories), farm products (10 categories), productivity
(4 categories) and non agricultural revenues (3 categories) allowed
for comparison of production systems for the 274 farms. A cluster
analysis of the respective coordinates of farms along the first four
axes of the MCA allowed us to separate different types of
production systems.
Farm Size had a median value of 50 ha and was assessed in 5
classes: small 1–25 ha (68 farms); medium , from 25 to 45 ha
(59); medium +, from 46 to 50 ha (44); large, from 51 to 100 ha (66)
and very large, >100 ha (37).
Human capital separated farms into 5 classes according to the
occurrence of hired labor: intensive > 1.2 UTH (unit of human
labor) per year (19 farms), constant 0.8–1.2 UTH per year (35
farms), regular 0.2–0.8 UTH (18 farms), occasional < 0.2 UTH (74
farms), none (128).
Sources of Income separated farmers into three classes: agri-
cultural, with >75% income from agriculture (182), non-agricultural,
with >75% incomes from external activities (60) and mixed with
25–75% incomes generated in agriculture (32).
Five types of Farm Products were evaluated and 10 types of
farms separated according to the dominance of annual crops (36
farms), livestock milk + meat (81), perennial crops (21), other
animal production fish, pork, chicken (14), diversified farming with
no specific dominant production type (122). Production intensity
was assessed through rates of annual deforestation and conversion
to pastures as % of the farm area and the number of livestock units
per ha of pasture.
Finally, a Production efficiency index, Pe = income per labor UTH
unit, designed to synthesize farm economic performance, sepa-
rated four classes: wealthy > 14,118 US$ eq. Purchase Power Parity
PPP per labor UTH unit per year (83 farms), medium + 6559 to
14,117 US$ eq. PPP; (84 farms), medium 1826 to 6558 USD eq.
PPP; (67 farms) and poor < 1825 US$ eq. PPP; (45)
142 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155

Table 2
Multi-quantitative variables of production systems used for farm classification: median and mean values for each variable are provided for the 274 farms examined in the first
valuation phase.

Categories Variables Median Mean Range

General information Time elapsed since arrival years 9 13 1–80
Size (ha) 50 67.5 1–5000
Human & financial capital Family labor (full-time equivalent) 1.8 2.43 0.6–8.8
Number of employees per property (full-time equivalent) 0.1 0.37 0–2.0
Repayable credit received in the last 4 years PPP 0 1619 0–49000
Non repayable credit received in the last 4 years PPP 0 386.8 0–3100
Income Total annual gross income PPP 2929 5954 0–100874
Non-agricultural incomes as% of total gross income. 5.0 23.5 0–32
Gross income per deforested ha PPP units 104.56 180 0–2917
Total income/n. people working in farm PPP units 1318 3259 0–45360
Farm products Gross income from bovine livestock/PBt 13.0 29.6 0–100
Gross income from annual crops 17.5 29.5 0–100
Gross income from perennial crops 0 11.2 0–100
Production Intensity Annual deforestation rate (% farm area) 1.0 2.5 1–5
Annual rate of pasture installation (% farm area) 1.0 2.0 0.20 + 0.36

Social wellbeing was evaluated with 5 different metrics
identified by farmers as significant components of their overall
wellbeing: household material wealth, level of education, social
networking (with isolation from family and other people consid-
ered as a negative attribute), professional stability and autonomy
provided by ownership of the farm, geographical stability
indicated by the number of migrations during the farmer’s lifetime
(Chayanov, 1966; Courgeau and Lelievre, 1992; Arnauld de Sartre,
2006 (Table 1 SI). The sum of scores in each category provides a
value that is adjusted to an interval of 0–1.0 for the first three
indicators, and to 0–0.5 for the last two in accordance with the
importance that farmers assigned to the items measured by these
metrics.
Social wellbeing Sb, represents the sum of the scores obtained
for these five metrics, adjusted to a range of 0.1–1.0 to allow
comparison with the other indicators (Table 2 SI).
Based on the classifications of social and economic conditions of
the initial set of 274 farms, we selected a representative subset of
54 farms (9 at each site) where landscape intensification,
biodiversity and ecosystem services were assessed. Three farms
from Colombia were further discarded due to unavailability of
some types of information.
2.4. Biophysical characterization
2.4.1. Landscape intensification
A remote sensing approach allowed for the characterization of
landscapes in 1990, 1994, 1998, 2002 and 2007, at each site, using
Landsat TM and ETM+ images, 30 m spatial resolution. Field
observations conducted in 2007 and 2008 and image analysis
determined six classes of landscape elements: forest, fallow,
burned forest, ligneous pastures, pastures, undergrowth wetland,
plus water and bare soil (refer to Oszwald et al., 2011 for details of
the remote sensing analysis) (Fig. 4).
Land use maps were then established for each farm and at the
five time points. Metrics were extracted from these maps to
quantify changes in the composition and spatial organization of
farm landscapes.
Composition metrics were based on the relative proportions of
the six landscape elements identified above. Corrections for crop
components that could not be detected by remote sensing were
obtained from farmer interviews.
Structure was assessed with six variables describing fragmen-
tation, edge density, mean patch density, patch diversity richness
and evenness, shape perimeter/area and mean shape respectively.

Fig. 4. Landscape element classification established in the field and linked with Landsat TM spectral signature year 2007 (Oszwald et al., 2011).
 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155
These variables were used to constitute 3D matrices with
farms characterized by their composition and structure at five
dates. Four sub-indicators of land use intensity and an integrative
indicator were calculated using this data set (Oszwald et al.,
2011).
PCA analysis was used to convert each set of variables into a
reduced number of factors that express underlying trends and
effects responsible for the variations observed. We generally used
Factors 1 and 2 that in most cases expressed more than half of the
total variance observed in our data sets. The same approach was
used to construct indicators of biodiversity and ecosystem services.
The first two indexes, composition (Li1) and structure (Li2) in
2007 (Table 3), evaluate the respective departures of landscape
composition and structure in 2007 from the original forest
landscape. Values are adjusted to vary from 0 to 10, with the
lowest value of the composition index corresponding to a
completely deforested landscape and the highest value to a
landscape with 100% forest cover. The structure index Li2
emphasizes differences in landscape organization among farms.
A score close to 0 characterizes landscapes with large-sized
patches of deforested areas while a score close to 10 indicates a
high level of shape complexity in natural forest patches that
reflects on going successional processes. The third and fourth
indexes measure temporal dynamics in landscape composition
and structure from 1990 to 2007 and employ ACT-STATIS, a method
designed for comparing several tables in one run (Robert and
Escoufier, 1976). A score of the composition dynamics index (Li3)
close to 0 indicates a homogeneous agricultural landscape since
1990, whereas a score close to 10 indicates a forested landscape
with little to no deforestation since 1990. The fourth index (Li4)
assesses the temporal evolution of landscape structure variables,
such that a score close to 0 indicates a landscape that has
undergone intense transformation from forest to agriculture in the
1990’s or earlier. This is, for example, the case for landscapes that
have been fragmented for a long time. A score close to 10 indicates
a landscape having experienced only recent and limited defores-
tation with a low level of natural habitat fragmentation.
The final, synthetic indicator of land use integrity averages
scores of the four sub-indicators (Oszwald et al., 2011). In this
study, we transformed this indicator into a landscape intensification
(Li) indicator that ranks sites from little transformed to greatly
affected, with variations adjusted to a 0.1–1.0 interval (Table 2 SI).
2.4.2. Biodiversity
From march to july 2008, during the rainy season at all sites, we
assessed species richness and abundance of three plant and six
faunal groups: vegetation in three strata, trees, shrubs and
groundcover, earthworms, birds, Saturniidae and Sphingidae
moths, Drosophilidae and ants. Animal groups were chosen
according to their demonstrated link to soil ecosystem services
(earthworms, ants), their sensitivity to local disturbance (Droso-
philidae), or specific response to landscape composition and
migration ability (moths and birds). All groups were identified at
the species level, except ants that were identified at the genus
level. In this group, we used the number of genera as a proxy for
Table 3
Sub-indicators of landscape composition, structure and dynamics (see detailed explanation in Oszwald et al., 2011).
Name Code Description
Composition 2007 Li1 Proportion of different types of land use in 2007
Structure 2007 Li2 Structure described by fragmentation, diversity and shape of the fragments
Composition Dynamics 1990–2007 Li3 Temporal dynamics of composition
Structure Dynamics 1990–2007 Li4 Temporal dynamics of structure
Landscape Intensification Li Synthetic indicator of landscape intensification
143
species richness (Sanabria, unp. data). Within each farm, five
square sampling areas of 50 m  50 m, of uniform plant cover were
regularly spaced along the longest diagonal of the farm. For each
taxonomic group, species composition and the number of
individuals observed for each species were recorded. We choose
standard and rapid sampling techniques to allow collecting data in
short periods at 5 points in each farm (270 in total) separated by
200 m along transects of 1 km that were often far away from any
road or walking path.
2.4.2.1. Plants. Three vegetation strata were assessed using distinct
sampling areas: 1, trees (diameter > 10 cm) within a 10  50 m area; 2,
shrubs and young trees (diameter < 10 cm and height > 2 m) on a
5  50 m area; 3, herbaceous species, trees seedlings and saplings and
shrubs on ten 1 m2 plots, located every 5 m along a 50 m transect.
2.4.2.2. Moths. Saturniidae and Sphingidae were collected with
light trapping for a single point closest to the largest forest
remnant present in each farm area from 6:00 p.m. to 6:00 a.m.
2.4.2.3. Fruit flies. Drosophilidae were collected with cylindrical
traps of 10 cm diameter  25 cm in height, closed at one end and
with a funnel at the other, baited with ca. 200 g of fermented
banana. Three traps were placed at each sampling point and were
set out 24 h in the field.
2.4.2.4. Earthworms and ants. Earthworms and ants were collected
by hand-sorting from three blocks of soil at each sampling point, i.e.
one central 25  25  30 cm and two 25  25  10 cm, distant 5 m
from each other on either side of the center point (method ISO 23611-
5, 2011).
2.4.2.5. Birds. Visual and auditory bird counts were conducted at
each sampling point by two people during a single observation
session of 20 min in the morning between 6:00 a.m. and 11:00 a.m.
to consider the period of maximal bird activity.
The first axis PCA coordinates of farms (55% of total variability
explained) performed on rarefied richness of the eight taxa was
scaled to 0.1–1.0 and used as a synthetic index of biodiversity Bd
(Table 2 SI).
2.4.3. Soil-based ecosystem services
Measurements were conducted in 2008 during three months of
the rainy season, at the same points used for biodiversity
assessments, using standard methods to characterize soil con-
ditions. Chemical fertility was assessed with a set of eight variables
(Grimaldi et al., 2014; Table 4), further scaled into a synthetic
general Soil Quality Index ranging from 0.1 to 1.0 (Velasquez et al.,
2007) Table 2 SI).
Hydrologic functions were assessed from an initial set of 11 soil
physical variables and summarized by two separate measurements
that best explained the largest part of variability: surface infiltration
and available water at 0–10 cm. Water retention
curves determined for 27 points, from 108 cores per farm using
multiple regression models allowed for calculation of simple soil
Quantification method
PCA on composition metrics
PCA on structure metrics
ACT-Statis method
ACT-Statis method
Sum of sub indicators Li1 to Li4
144 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155

variables measured at all points (texture, bulk density, vertical
resistance to penetration, pH, CEC and C contents) (Grimaldi et al.,
2014). Plant available water capacity was calculated as the water
volume drained between matrix potentials of 30 kPa and 16
MPa.
The climate regulation function was assessed by the sum of C
stored in soil (0–20 cm) and C in aboveground plant biomass using
classical methodologies detailed in Grimaldi et al. (2014). Values
thus obtained were adjusted to a 0.1–1.0 interval to allow for the
analysis of their co variation with other indicators (Table 2 SI).
As for biodiversity, coordinates of farms along axis 1 and 2 of
a PCA performed on ES values were used to calculate a synthetic
index of soil ecosystem services (Es) (Grimaldi et al., 2014; Table 2
SI).
statistics and graphical displays were performed using the
packages, lme4, pbkrtest, lmerTest, lattice and ggplot2 developed
in R freeware (R Development Core Team, 2014).
3. Results
We first analyzed variations of the different parameters of
socio-environmental systems across sites, and tested the respec-
tive effects of production system and landscape intensification on
the observed dynamics. We then tested hypothesis 1 of a
significant relationship between production systems and land-
scapes. The other 7 hypotheses were then tested by analyzing the
relationship among production systems and landscapes with
socioeconomic and environmental variables.

2.5. General index of sustainability 3.1. Conditions of the socioenvironmental systems in the different sites

We designed an index of sustainability (Su) that combines the
social Sb, economic Pe, biodiversity Bd and soil ecosystem services
Es indices. The Su value of a given farm was set equal to the
coordinate along the first PCA axis 1 of the four above indexes.
Values of Su were further adjusted to an interval between 0.1 and
1.0. This index is expected to be highest when production systems
are highly productive and profitable (Pe), ensure a high level of
social wellbeing (Sb) while conserving the natural capital, assessed
through biodiversity (Bd) and soil ecosystem services (Es) (Table 2
SI).
2.6. Data modeling
Relationships among the different indexes Sb, Pe, Li, Bd, Es and
the general sustainability Su were analysed with mixed linear
models allowing the analysis of nested data (Pinheiro and Bates,
2000). We used linear mixed model for accounting fixed effects
under interest and random effects due to specific site characteristics
(nested sites). Fixed effects included landscape intensification Li or
one of its component, landscape composition in 2007 (Li1),
landscape structure in 2007 (Li2), dynamics of landscape
composition (Li3), dynamics of landscape structure (Li4) and the
type of production systems. Since Li1 was highly correlated to Li3
(r = 0.86; P < 5e 16), we considered only three components
describing land use intensity i.e. Li1, Li2, Li4. We used the proposal
of Nakagawa and Schielzeth (2013) for computing the variance
explained by the fixed (R2m) and both fixed and random
conditional (R2c) effects respectively. This allowed us to differen-
tiate site (“random”) effects from “fixed” effects due to, for
example, landscape composition or dynamics. The statistical
significance of model was estimated by a visual inspection of
residuals and by an analysis of variance between a null model (no
effect) and a model that included the explanatory variables. All
3.1.1. Farm landscapes
There was a clear decrease in the different sub-indicators and
in the synthetic indicator of landscape integrity measured on the
complete set of 274 farms, as time elapsed from initial
deforestation (Fig. 5). Farms at Pacaja (LU 1), in the earliest
stages of transition from forest, exhibited the highest values for
all sub indicators; they still had high proportions of natural forest
and relatively small patches of deforested areas, with limited
change during the period 1990–2007. The Maçaranduba site (LU
2; 5 to 10 years older) showed degradation in all indicators, a clear
demonstration that the agroextractivist status of the region was
no longer respected. However, in one of the farms where
deforestation had been greatly avoided and people made a living
on Brazil nut production (Ze Claudio, LU 2, farm n 308 Table 2 SI),
the composition indicator was 9, the second highest value across
all farms. Sadly, this farmer and his wife, who had conserved a
large proportion of the forest, were assassinated in 2011 by illegal
loggers who pressed them to cut down the protected tree species
of their land. At Palmares site (LU 3), increased deforestation and
conversion to annual crops further decreased the composition
index. The lowest values of the composite index (Li) were
observed In Colombia, in the conventional (LU 6) and in the
sylvopastoral (LU 5) sites. At site LU 5, sylvopastoral management
established recently on rather small areas had not yet impacted
the landscape (Ramirez et al., 2012) although the structure (Li2
subindicator was increased on average). Improved values were
observed in the agroforestry site LU 4, which demonstrated
the highest values for the landscape integrity indicator for
Colombia
3.1.2. Social well being
The Sb index of social well being increased along the
succession of sites, with minimum values in the early phases

Table 4
Variables measured to evaluate soil ecosystem services (Grimaldi et al., 2014).

Soil attribute Variables

1
C stock in soil Mg ha Bulk density and total C content at 0–10, 10–20 and 20–30 cm depths

1
C in biomass Mg ha Aboveground dry plant biomass of trees (BT) and bushes (BB)a

Water Infiltration rate mm h 1 Infiltration of a determined volume of water (250 cm3) poured in 20 cm diameter ring inserted at the soil surface to a depth of ca. 1 cm.
Soil available water in 0–10 cm Clay, silt and sand contents, rb, vertical (Rv) and tangential (Rt) resistances, Infiltration rate Vi
Chemical Fertility pH, CEC, exchangeable Al3+, Ca2+, Mg2+, K+, base catión saturation ratio, extractable Mehlich phosphorus P
a
BT = trees must have a diameter at breast height dbh  10 cm and BB (bushes) dbh < 10 cm and height >2,0 m were evaluated by applying allometric equations provided for
forest S1, S2 and fallows S3 after species identification and measurement of the diameter and height of individuals on plots of 50  10 m2 and 50  5 m2, respectively, and the
water content at 70  C of corresponding biomass aliquots all individuals with dbh 5 cm, applying otherwise for dbh  5 cm the factor 0.603 recommended by Hastie (1991). A
factor of 0.5 was used to convert plant biomass into C mass.
 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155 145

Fig. 5. Boxplot representation of values of the different indexes (from 0 to 10) of landscape integrity (see Table 3) across a chronosequence of sites from incipient deforestation
(LU 1) to highly deforested and degraded landscapes (LU 6). Middle bar is median, box extends from 25% to 75 % quartiles and horizontal bar stands for minimum and maximum
values.

at LU 1 and LU 2 sites of Pacaja and Maçaranduba, an increase at and lower values at the agroforestry sites (LU 4) (Fig. 6).
intermediate stages (LU 3, Palmares), maximum values at the Changes in the index reflect greater migration rates and isolation
traditional (LU 6) and silvopastoral (LU 5) sites of Colombia, from urban life and the associated issues of commodities
146 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155

transport, energy, health care and education in incipient 3.1.5. Soil ecosystem services
agricultural sites LU 1 to LU 3 as compared with peri-urban The Es index also decreased regularly across the deforestation
agriculture at the LU 6 and LU 5 Colombian sites (PCA analysis not chronosequence (Figs. 6 and 7) figured in our sites. A minimum
shown). value was attained in the traditional extensive livestock production
site (LU 6) in Colombia.
3.1.3. Production efficiency This general decrease of the Es index, however, covers different
The Pe index of Production Efficiency increased continuously situations as some soil ecosystem services may increase while
and exponentially from farming systems at very early stages LU 1 others dramatically decrease (Grimaldi et al., 2014) (Fig. 7).
and LU 2 where average annual incomes per ha are 37 and 74 PPP,
respectively, to 246 and 321 PPP, respectively, in the traditional 3.2. Effect of production systems and landscape patterns on
(LU 6) site of Colombia and the legal settlement at Palmares (PR) production efficiency, social wellbeing, biodiversity and the
(Figs. 6 and 7). In Colombia, the silvopastoral LU 5 site had slightly production of soil ecosystem services
lower incomes 152 PPP ha 1 while the agroforestry site LU 4
exhibited a remarkable maximum value of 733 PPP. Using mixed linear models, we investigated the proportion of
the large changes in socioeconomic and environmental conditions
observed along the sequence of sites that was linked to the farm
3.1.4. Biodiversity
production systems (H2–H4) and/or landscape features created by
The Biodiversity (Bd) index exhibited a continuous decline from
the farmers (H5–H7).
sites representing the early stages to the most deforested and
intensified sites (Figs. 6 and 7). In Colombia, however, agroecolog-
3.2.1. Classification of farms according to production systems
ical practices implemented in the silvopastoral (LU 5) and
We observed great differences among farms for the five attributes
agroforestry (LU 6) sites appear to have noticeable positive effects.
selected to characterize production systems, the overall area of
Early deforestation immediately affects birds, Sphingidae and ants
farms, the number of people working at each farm and farm products
as endemic species linked to pristine habitats soon disappear (Fig.
and internal and external incomes (Fig. 8). Along the gradient of
7). While the decrease in Sphingidae species richness continues
landscape intensification represented by our 6 sites, areas managed
and culminates in the extensive highly degraded pasture land-
per unit of labor increased from rather low values in the early
scapes of Colombia (LU 6), birds and ants maintain rather elevated
deforestation phases in the Pacaja and Palmares sites (6.5 and 7.0 ha
richness along the succession. This is mainly due to the
per person), respectively, to intermediate values in the agroforestry
colonization by a new set of species, some exotic and/or invasive.
LU 4 and agroextractivist LU 2 systems (11.1 ha and 13.3 ha
Plants, earthworms and Drosophilidae progressively disappear
respectively) (Fig. 8 left). Much greater areas were used in the LU
with forest clearance and degradation of soils. Of a total of 29
5 site with silvopastoral systems (27 ha), and particularly in the LU 6
earthworm species sampled in this study, 28 were endemic and all
window dominated by extensive livestock production on degraded
of these were new species, with 2 new genera discovered in Brazil
pastures (61 ha). There was also a general trend for increased
(Feijoo and Celis, 2010). The invasive species Pontoscolex coreth-
incomes per ha across the sequence of sites. At older sites,
rurus tended to progressively replace native communities in
agroforestry systems of the LU 4 site allowed greater incomes per
deforested areas transformed into pastures and crop areas
ha than traditional livestock productions of the LU 6 site. (Fig. 8 right).
(Marichal et al., 2010).

Social Well-being Ecosystem Services

Sb Es
0.2 0.4 0.6 0.8 1.0
0.8

Sb
0.6
0.4
0.2

LU1 LU2 LU3 LU4 LU5 LU6 LU1 LU2 LU3 LU4 LU5 LU6

Producon efficiency
Biodiversity
0.2 0.4 0.6 0.8 1.0

Ef
1000 Bd

100

10

LU1 LU2 LU3 LU4 LU5 LU6 LU1 LU2 LU3 LU4 LU5 LU6

Fig. 6. Variation of social wellbeing (Sb), Biodiversity (Bd) and Soil ecosystem services (Es) indexes (ranging from 0.1 to 1.0) and production efficiency Ef (in equivalent
purchase power PPP units) along a chronosequence of sites. Middle bar is median, box comprises extends from 25% to 75 % quartiles and horizontal bar stands for minimum and
maximum values.
 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155 147

Fig. 7. Variations of the indexes of biodiversity and soil ecosystem services at the three most contrasted sites LU 1: Pacaja, Brazil; extensive livestock production at the LU 6
site in Colombia; and agroforestry, at Colombian LU 4. Values expressed as % of the highest average value recorded among sites.

Cluster analysis of the respective coordinates of farms along the
first four axes of the MCA allowed us to distinguish seven types of
production systems (Fig. 9).
A first group was comprised of farms from early stages,
deforested 10–20 y. ago with mixed production systems of annual
and perennial crops and livestock with low incomes (types A and
B). Type A comprises farmers fully involved in agriculture on
relatively large areas, while farmers of the type B farms generally
engage in a complementary non-agricultural activity on farms of a
smaller size on average. Types C and D also have diversified
productions, on relatively small farms with low (C) or intermediate
(D) incomes. Types F and G are large farms with extensive livestock
production and high incomes. Both types rely on external labor,
production in F type farms being a bit more diversified than in
those of G. Finally, farms of the E type are small, with relatively
high incomes on agroforestry systems established on sites where
deforestation began 50–80 y. ago.
Distribution of production systems across the sites showed a
progressive shift of farms from the least (A, B) to the mostintensi-
fied systems (F and G), as time since colonization and the
proportion of area deforested increased (Fig. 10). Types A, B and C
comprised all the farms of the Pacaja (LU 1) and Maçaranduba
(LU 2) sites in Brazil, and still 8 out of 9 at Palmares (LU 3). Farm
types A and B were also found in the Colombian LU 4 agroforestry
site (2 out of the 8 Colombian farms) while the C type was
also found in the sylvopastoral LU 5 (3 of 8 farms) and the

Ha used per
UTF labor unit
Log Income ha -1
Eq. Purchase units
150

3.5
100

2.5
50

1.5
0.5
0

LU1 LU2 LU3 LU4 LU5 LU6 LU1 LU2 LU3 LU4 LU5 LU6

Fig. 8. Boxplot representation of variation of area used per capita (left) and incomes per ha per labor unit across the sequence of sites from early deforestation phase, LU 1 to
LU 6 (advanced almost totally deforested stage. Middle bar is median, box comprises extends from 25% to 75 % quartiles and horizontal bar stands for minimum and maximum
values.
148 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155

Fig. 9. (a) Projection of the seven types of farms in the F1/F2 factorial plan of a Multiple Correspondence Analysis and general characterization. The 274 farms from the first
phase evaluation are characterized by their size, human capital, productions, productivity and non agricultural revenues. (b) Proportion of variance explained by extracted factors (F1
and F2 highlighted).

conventional LU 6 sites (1 of 6 farms). The most intensified
systems (G and F) and the E agroforestry system were only found
in the Colombian sites.
3.2.2. Production systems and landuse intensity
Land use intensity was significantly correlated with production
systems, (Hypothesis H1, Fig. 1) (ANOVA; P < 0.01; Fig. 11A), but
this relationship, only explained 3% of the total variance.
3.2.3. Production efficiency and social well being
Production efficiency was influenced by the type of production
system (H2) (P < 0.006; 27.7% variance explained (Fig. 11B) but not
social well being (H3) (P > 0.40).
Social well being was weakly related to production efficiency
(H4) with only 5.7% of variance explained by a direct effect;
Fig. 12D). Social wellbeing was also negatively and significantly
related to biodiversity (H8) with 21.3% of variability explained
directly (Fig. 12E).
3.2.4. Production systems, landscapes and the natural capital
Soil Ecosystem Services were related to production systems and
landuse intensity (H5). Production systems explained a slightly
lower proportion of variance (24.5%, Fig. 11C) than landuse
intensity Li (37.3%, Fig. 12B). The relationship between the
indicator of social wellbeing (Sb) and soil ecosystem services
(Es), however, was not significant. Interestingly, the indicator of

Fig. 10. Proportion of farm types in the different sites. A, B: early phases; C, D mixed productions; E: agroforestry: F, G: extensive livestock production. LU 1 Pacaja and LU 2
Maçaranduba represent early deforestation phase, LU 3: Palmares, intermediate, LU 6 (Colombia) advanced almost totally deforested stage, LU 4 with agroforestry systems
and LU 5 with incipient sylvopastoral systems derived from degraded land.
 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155 149

soil ecosystem services was maintained at a rather high level in
systems found in the younger phases of development, while the
lowest values were observed in system G that relies on extensive
cattle production on largely eroded soils. Farms of the F (with
recently established sylvopastoral systems) and, especially, E
(agroforestry) types had higher values of the soil ecosystem
services Es index than type G farms with extensive livestock on
degraded pastures. Biodiversity was also significantly related to
Production systems (29% variance explained) (Fig. 11D). Biodiver-
sity decreased continuously across production system types A to G.
Biodiversity was also negatively related to landscape composition
in 2007 (Li1 with 20.0% of variance explained (H6) (Fig. 12A), but
not with the general indicator of landuse intensity (Li)
We also found a positive relationship between soil services (Es)
and biodiversity (Bd) indicators (H7) (53.1% of variance explained)
(Fig. 12C).
3.3. Sustainability
The indicator of Sustainability (Su) was: Su = 0.51  Pe + 0.41 
Sb + 0.58  Bd + 0.48  Es

Fig. 11. Variation of Landscape (A), Production efficiency (B), Soil ecosystem services (C), Biodiversity (D), and sustainability (E) index values in the sub-sample of 51 farms, as
a function of the production system implemented in the farms, from A and B: incipient farming systems to C and D: mixed farming in intermediate stages, E: agroforestry systems, F
and G: extensive livestock production on degraded pastures. Metrics generally vary from 0.1 to 1.0, except for Landscape composition (o to 10) and Production Efficiency expressed in
equivalent US $ corrected as power of purchase units.
P: probability to reject a significant relationship between production systems and indicators and% of variance explained by a direct effect, independent from site characteristics (mixed
linear models). Middle bar is median, box comprises extends from 25% to 75% quartiles and horizontal bar stands for minimum and maximum values. Values with similar letters are not
significantly different (ANOVA and Fisher test).
150 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155

with the production efficiency, Pe, social wellbeing, Sb,
biodiversity, Bd, and soil ecosystem services, Es, subindicators.
Values of Su were further adjusted to the interval 0.1–1.0.
Sustainability was significantly related to the production
systems, with 36.4% of variance explained by a direct effect of
production systems on sustainability (Fig. 11E). However, this
index was relatively stable in production systems of the early phase
(A and B), intermediate in mixed farming systems (C and D) and
under agroforestry (E), but demonstrating a sharp decrease with
extensive cattle ranching systems (G). Farms with incipient
silvopastoral systems exhibited a wide range of values, from low
values similar to those observed in the G production systems, to
intermediate values associated with farms of the A to E production
systems.
Sustainability was also significantly and negatively related to
the land use intensity index with 14.7% of the variance explained
by this relationship (Fig. 12F).

Fig. 12. Relationships among the different indicators of socio-environmental systems across sites. Statistical adjustment made with Mixed Linear Models. The grey area
represents uncertainty attributed to random effects (=effect of the sites). Each point represents one of the 51 sampled farms. P: probability for the relationship not to be significant; %
variance explained by a direct effect between variables and not by site random effects.
 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155 151

Fig. 13. Relationships among the components of socio-environmental systems obtained from mixed linear models (significance set for > 10% variance explained). In red: %
variance directly explained by the relationship. Red arrows indicate significant negative relationships and green arrows indicate significant positive relationships. (For
interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

4. Discussion Amazonian municipalities. Wellbeing of people increases as

In general, this study supports the view that whatever human
development and improvement of the wellbeing that has been
achieved, was done so at expenses of the natural capital. We also
find surprising low correlation between production systems and
landscape characteristics, as well as between human wellbeing and
production systems and their economic efficiency. Finally, our
findings confirm that production systems and landscape character-
istics are major drivers of the social, economic and environmental
aspects of sustainability.
Most elements of the socio-environmental systems analyzed
and compared here were significantly correlated and our
hypotheses were largely verified. The use of Mixed Linear Models
allowed us to attribute a sizeable part of the variance explained to
independent non-site specific relationships among the different
indicators. Direct effects were greater than site effects in
determining relationships between sustainability (Su) and landuse
intensity (Li), as well as between sustainability and production
systems. The same observation applied to relationships among
landuse intensity and soil ecosystem services as well as between
soil ecosystem services and biodiversity (Fig. 13).
While these relationships have been predicted and partly
measured in a number of studies, they had never been, to our
knowledge, verified and generalized in such a comprehensive way,
thus far in Amazonia (Foley et al., 2007; Vieira et al., 2008; Oliveira
et al., 2013).
4.1. Social sustainability
Deforestation and further exploitation of natural capital
allowed for significant improvement in social wellbeing, as
previously indicated by Rodrigues et al. (2009) for 286
development occurs, with improved roads, commerce, healthcare
and educational opportunities being associated with growing
urban centers and energy becoming available in formerly isolated
areas. However, we noticed a great turnover of the population.
Surprisingly, the residence time of people was about the same in
all our sites, which suggests that they leave a place once their
participation in the development process has come to a given
point. Poor migrants who initiate deforestation rarely have
enough capital and adequate training to make an efficient use of
the areas that they have deforested. Often, they sell their land
and/or are displaced by farmers with better access to resources,
and thus move on to the next pioneer front (Arnauld de Sartre,
2006). The average increase in the indicator of well being
measured in our study may, therefore, cover heterogeneous
situations as observed by Weinhold et al. (2013) in Eastern
Amazonia where cultivation of soybean tends to reduce poverty
indicators and raise median rural incomes, but is also associated
with increased inequality between large soy farmers who benefit
from the development and local indigenous populations who
don’t.
At our research sites, social progress is mainly linked to local
conditions and not to the type of production system implemented,
nor to the landscape. This observation suggests that the link
observed between landscape features and social conditions once
observed at our Palmares (LU 3) site (Billard et al., 2014) cannot be
generalized at the scale of the whole Amazonian region. Creation of
urban centers and policies that sustain social benefits and the
creation of public goods and services are likely to be the main
drivers. In our Brazilian site, at Palmares (LU 3), the combination of
proximity to a fast growing town and the legal status of the
settlement, allowed this site to have a much higher index of social
wellbeing than the adjacent Brazilian site of Maçaranduba (LU 2),
152 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155
reaching values rather close to the ones measured in Colombia
where development had occurred over much longer time periods.
Social wellbeing (quantified by our Sb synthetic indicator) was
not significantly related to production efficiency. This counter
intuitive result may be due to the indirect negative effects of
economic growth on natural capital. Depletion of the natural
capital is expressed by an inverse linear relationship of the
wellbeing Sb indicator with biodiversity. This result supports our
hypothesis (H8) that social development expressed by the
indicator of wellbeing (Sb) is achieved at the expense of the
natural capital represented by our Biodiversity indicator (Bd).
Although the Es indicator of soil ecosystem services was not related
to Sb, the strong link of the Es indicators with Bd and with
production systems, suggests that the effects of human develop-
ment on biodiversity have an indirect negative effect on soil
ecosystem services. This inverse relationship between human
wellbeing and indicators of natural capital has been previously
emphasized in the Millennium Ecosystem Assessment (2005).
Raudsepp-Hearne et al. (2010) proposed four possible interpre-
tations for this trend: (1) an incorrect evaluation of wellbeing; (2) a
strong relationship between wellbeing and food availability, which
increases while other services are declining; (3) decoupling of
wellbeing from nature by technology; (4) the existence of an
environmental debt that will further induce declines in wellbeing.
In the conditions of our Amazonian sites, the fourth hypothesis
seems to be the most appropriate. While we think that our
indicator provides a rather accurate measurement of wellbeing,
sufficient food production is not a problem in the region, with a few
exceptions, at any stage of the development process. Technology is
still weakly used in the types of agricultural systems that we
studied. Therefore, the remaining possibility is that consumption
of the natural capital will further create an environmental debt and
decrease wellbeing later, as observed after 30 years in municipali-
ties of Amazonia (Rodrigues et al., 2009). In particular, a decrease
in soil physical quality and removal of the plant cover leaves soils in
a fragile state, susceptible to erosion and compaction. A decrease in
regulating services (such as pollination, water purification and
provision, maintenance of soil fertility) and cultural services
(recreation, tourism, medicinal plants) is expected with increasing
land use intensity and soils are are at risk of loosing the capacity to
support food production for local populations (Foley et al., 2007).
We also observed in this study that the average area used per
working person increases along our chronosequence of landscape
intensification. The number of cattle units (>400 kg) per ha is only
0.57 in the LU 6 site, where production system G is dominant and
with the highest landscape intensification index. In the sylvopas-
toral (LU 5, mainly in the F group) and agroforestry (LU 6 in the E
group), cattle densities are 0.84 and 0.89 respectively. In such
conditions of a growing pressure for land, a great part of the initial
colonizing population will emigrate, either to new pioneer fronts
where living conditions are poor or to newly created urban areas
where their living conditions will depend more on education and
employment opportunities (Walker et al., 2000; Arnauld de Sartre,
2006).
4.2. Economic sustainability
Production efficiency (Pe), our indicator of economic develop-
ment, was correlated with production systems, with a rather high
proportion of variance explained, independently from local and
regional site effects. Incomes per ha managed are initially very low
in the early deforestation stages (LU 1 Pacaja Brazilian site) when
farmers mainly practice slash-and-burn agriculture and some
livestock management with few agricultural inputs, limited access
to technology and, often, very distant markets. Conditions improve
in sites that have been deforested for longer periods (Brazilian sites
of Palmares LU 3 and Maçaranduba LU 2) and in a few farms from
the Colombian sites, where production tends to diversify as urban
centers develop and markets for perennial crops become better
established. After 40–60 years, with over 95% of the land
deforested, the main activites shift towards low productive
livestock production on very large farms (farm type F and G),
mainly found at the traditional LU 6 Colombian site. We suspect
that the apparent need for larger areas to sustain livelihoods
as time passes (Fig. 8) is likely due to a decrease in natural
capital as land degrades. Soil compaction, loss of plant biomass
and overall biodiversity, and poor water infiltration observed in
farms of the F and G types provide clear evidence of this loss of
natural capital.
Production systems exhibited a clear relationship with natural
capital indices of biodiversity (Bd) and soil ecosystem services (Es)
that tended to decrease with production intensification, from the A
to G production systems.
Landuse intensification directly explained 20% of the biodiver-
sity loss and even larger proportion of the losses in soil ecosystem
services (37.3%). Although the relationship between biodiversity
and land use intensification was significantly explained by a linear
decreasing model, careful examination of point distribution in Fig.
12A suggests that the non-linear response, predicted by a number
of authors (Radford et al., 2005; Swift and Hannon, 2010; Zavaleta
et al., 2010), might become significant with the consideration of a
larger data set or data collected in a more homogeneous region.
Interestingly, the region of the relationship where biodiversity
visually drops off, corresponding to around 50% tree cover,
suggests the presence of a point beyond which biodiversity may
be difficult to restore. For this tree cover component, forest, fallows
and tree perennial plantations may play similar roles, but this
remains unclear. Clearly more research is required to elucidate the
possible presence of such a tipping point and the potential role of
various types of tree cover in supporting biodiversity.
We also show that agroecological practices developed in the
agroforestry systems (production group E) mainly found at the LU
5 site in Colombia, and to a lesser extent in sylvopastoral systems
common at the LU 6 site, can profoundly affect the economic
performance, with the maximum values recorded for the
Production efficiency Pe and incomes ha 1 UTF 1 indices (Ramirez
et al., 2012). Agroforestry systems in this study comprise an
association of former rubber plantations with cattle production
(milk and meat) and some annual crops. Sylvopastoral systems
combine shrubs and pasture for forage or protein bank production.
This result, in accordance with an abundant literature, clearly
identifies agroforesty systems as desirable options to maintain the
natural capital of the original forest at a relatively high level (Alfaia
et al., 2004; Lopez-Hernandez et al., 2004; Perfecto and Vander-
meer, 2008; Grimaldi et al., 2014) while sustaining livelihoods.
Agroecological farming generally requires improved management
of plants and soils, creativity and knowledge to be entirely
successful, as compared with more simplified “conventional”
farming (Browder and Pedlowski, 2000; Miller and Nair, 2006). A
combination of local and general scientific knowledge is necessary
to achieve currently expected levels of productivity (Altieri, 2004;
McKey et al., 2010) in a way that sustains or increases the natural
capital. This knowledge is not currently provided by general
teaching curricula and technical assistance to farmers on
agroecological technologies is almost completely absent. In
addition, we note that many of the soils that we analyzed would
require a preliminary restoration to allow for productive
agroecological farming.
Social wellbeing in these farms dedicated to agroforestry, as
measured by our Sb indicator, was the lowest in Colombia (LU 4),
probably due to location far from urban centers and the relatively
small area available per person (11.1 ha) as compared to 6–7 in
 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155
the pioneer fronts, and 60.7 in the LU 6 “traditional” Colombian
site.
Interestingly, we did not observe the “bust” part in evolution of
social wellbeing index observed by Rodrigues et al. (2009) after 30
years of deforestation and land occupation. In our sequence of
sites, the maximum values for the wellbeing index were observed
in sites where deforestation had begun 60–80 years ago. This is
probably due to differences in scales considered in the two studies.
At the scale of farms used in our study, development occurs and the
overall condition of farmers reaches a maximum value in the
traditional and silvopastoral sites of Colombia, where less than 5%
of the forest remains intact. However, the areas used for
production are far larger than in the early phases of development.
As a result, the overall number of farms decreases, which in the end
probably reduces the overall incomes at the scale of municipalities,
the scale considered by Rodrigues et al. (2009) in their study. In the
longer term, these production systems that do not invest in the
conservation and restoration of the natural capital are not likely to
be sustainable.
4.3. Environmental sustainability
Deterioration of natural capital by human activities is quite severe
in Amazonia, and the drivers are well identified in our study, since
our indicators of biodiversity and soil ecosystem services decreased
as indicators of human development and landscape intensification
improved. Deforestation and the implementation of annual or
perennial crops and pastures had rapid and profound effects on
biodiversity and soil quality (Grimaldi et al., 2014; Marichal et al.,
2010; Foley et al., 2007). However, the link between landuse
intensification and farming practices that define production
systems, although highly significant, was mainly generated by
random local effects. This is likely a consequence of the effects of local
and regional policies on such elements of production systems as non-
agricultural incomes or the type of labor available. Differences in
inherent soil quality are also likely to be important in this respect,
although in our case, the degraded G type of farms were found in soils
with better inherent fertility (Grimaldi et al., 2014).
Fig. 14. A three-step procedure to increase sustainability of farms based on
farmscape reconstruction. Su: sustainability index calculated as a function of social,
economic, biodiversity and soil ecosystem services indices.
153
4.4. General sustainability
Overall, our sustainability index remained rather stable in the
early phases of development, mostly observed in Brazil, with
production systems fom A to D. A decrease in values of
environmental indices were compensated by increases in the
social and economic indicators. In comparison, the G production
system, with extensive cattle grazing on degraded pastures
presented a very low value for the sustainability (Su) index.
However, farms with some sylvopastoral plots (F) and agroforestry
systems (E) exhibited much higher values of sustainability. At both
sites, public policies and/or direct intervention of university
research programs have favored the adoption of improved
management systems that have begun to restore biodiversity
and ecosystem services and ultimately increase the sustainability
index.
Our general indicator of sustainability, which combines the four
above described sub-indicators, was significantly linked with both
production systems and land use intensity. Almost all the variance
in the sustainability index due to production systems is explained
by a general impact (36.4%) and not by local conditions (1%).
Sustainability was also significantly linked to land use intensifica-
tion with almost equivalent proportions of variance explained by
general 18.7% and local random effects 15.7%.
Beyond a confirmation of our hypotheses, these results indicate
a great potential for the proposed indicator of sustainability to
evaluate the trade-offs achieved in a given socio-environmental
system. In our study, we first evaluated the different elements of
sustainability and synthesized them into a single indicator that
was statistically correlated with indicators that summarize the
management options and strategies. The shape (linear or non
linear) of the relationships and their intensity provide clues on
how much and in which direction a system must be changed to
improve its overall sustainability. With a general aim of improving
the sustainability of a farm, different mechanisms may be
envisaged to achieve this goal according to the level of education,
experience and resources available to the producer, either for
agroecological options or conventional agriculture. The indicator
will reveal in the end if changes incurred to a particular production
system can contribute to improve the situation globally (Ikerd,
1993).
4.5. The next step forward: reconstructing sustainable landscapes
By demonstrating linkages between sustainability and land use
intensity and production systems, and better elucidating the
different components involved, this work offers clues for improv-
ing sustainability across Amazonia any beyond. Optimizing trade-
offs and synergies among the different social, economic and
environmental drivers represents an obvious path forward.
Our study demonstrated the importance, across a wide range of
agricultural landscapes, of the role of production systems and
landscape composition and structure in regulating biodiversity,
soil ecosystem services and overall human wellbeing. The
diagnostic thus developed suggests ways to revert the negative
trade-offs observed between human development and natural
capital. Production systems generally had stronger relationships
with the sustainability index and its components than the general
landscape indicators. However, the task of modifying production
systems to improve sustainability is likely to be difficult. Elements
of production systems like the area of the farm and its human
capital are important to consider, but difficult to change. The other
aspect of production systems consists of production types that
determine the allocation of different types of landuse on the farm
landscape. These different landuse systems – that determine the
landscape composition, structure and overall dynamics – are
154 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155
known to have locally different impacts on biodiversity and
soil ecosystem services and at the same time, influence the
economic component of sustainability through the value of crops
produced.
Focusing on landscape level management may therefore
represent an optimal approach for improving sustainability, due
to the very strong link between landscape parameters and
sustainability components (Nelson et al., 2009; Jackson et al.,
2007; Rey Benayas and Bullock, 2012).
Economic performance of farms depends on the presence of
plots dedicated to profitable and competitive production systems.
The type of agricultural plots, especially, the existence of tree or
shrub components and their connection with natural ecosystem
patches likely determines biodiversity and the provision of soil
ecosystem services by soils on a farm (Nelson et al., 2009). The
social component, largely determined by components of public
policies outside the farm also depends on benefits gained from a
high quality environment. A few studies have attempted to develop
this view, although general methodologies proposed are still very
simple (Nelson et al., 2009; Smukler et al., 2010).
We propose a three-step procedure for the restoration of
sustainable landscapes based on the diagnostic developed in this
study and relationships tested among the components of
sustainability (Fig. 14).
A complete diagnostic of social, economic and natural capital
components should first be done at a number of farms represent-
ing the variability observed within a particular region. This
diagnostic could be conducted according to the general method-
ologies outlined in this study, trying to avoid as much as possible
measuring redundant variables so as to minimize costs. Synthetic
indicators would then be computed for each category.
A simulation model will be designed to relate landscape
variables to the different components of sustainability. This will
allow for the evaluation of different sub-indicators under different
scenarios, and of the general indicator of sustainability, associated
with any farmscape from the region.
Design of the new farmscape will be done in a participatory
exercise where farmers and other local stakeholders allocate parts
of the farm to different, existing or innovative, technologies chosen
from a bank of options. The bank of options would consist of set a
specification sheets prepared by agronomists and agroecologists
for diverse farmers in a region. These sheets comprise in a
synthetic and simple form all the technical and economic
information necessary for decision-making. Public policies are
essential in the process, since the considerable interventions
needed to reform farmscapes will require incentives (e.g., loans or
subsidies) to cover expenses linked to the restoration of natural
capital.
5. Conclusion
This comprehensive analysis of the evolution of farm sustain-
ability following deforestation allowed us to validate and quantify
the trade-offs between economic development and the conserva-
tion of natural capital across current and historical pioneer fronts
of the Amazon rainforest. While this research provided clear
evidence that some restoration by agroecological practices is
possible, our findings also indicated that these systems may be
implemented in combination with other practices (e.g., intensive
cash cropping or livestock breeding), into a well-designed farm-
scape in order to optimize multiple societal goals within these
landscapes. By improving the use of already deforested and often
severely degraded areas, we expect that less pressure would be
exerted on still pristine forests or forest regrowths. Our study also
shows that adequate and enforced public policies are necessary to
attain this goal.
Acknowledgements
This paper is specially dedicated to José Claudio Ribeiro Da Silva
and Maria do Espiritu Santo Da Silva, farmers at the Maçaranduba
site deeply involved in the project, who were assassinated on 24
may 2011 by illegal tree loggers (see homage: “Amazonia: a happy
soil” Youtube.com, Canal Lavelle1948).
The work was funded by the French Agence Nationale de la
Recherche through two grants: AMAZ_ES (ANR-06-PADD-001-011;
Agriculture Durable et Developpement programme) and AMAZ_BD
(ANR 06 BIODIV 009-01; IFB_ANR) and by Brazilian CNPq
(Conselho Nacional de Desenvolvimento Científico e Tecnológico)
project “Serviços ecossistêmicos e sustentabilidade das paisagens
agrosilvipastoris da AmazôniaOriental” no: 490649/2006-8.
The Institut de Recherche pour le Développement (IRD) played
a key role in funding temporary positions for people involved in the
operations and administering the project. We thank all of the
Brazilian, Colombian and French students that came to help as part
of their training, local technical staff and field workers. We are also
grateful to the farmers who participated in this research and
allowed us to work on their land.
Appendix A. Supplementary data
Supplementary data associated with this article can be
found, in the online version, at http://dx.doi.org/10.1016/j.
gloenvcha.2016.04.009.
References
Alfaia, S.S., Ribeiro, G.A., Nobre, A.D., Luizao, R.C., Luizao, F.J., 2004. Evaluation of soil
fertility in smallholder agroforestry systems and pastures in western Amazonia.
Agric. Ecosyst. Environ. 102, 409–414.
Altieri, M.A., 2004. Linking ecologists and traditional farmers in the search for
sustainable agriculture. Front. Ecol. Environ. 2, 35–42.
Arnauld de Sartre, X., 2006. Fronts pionniers d’Amazonie: Les dynamiques
paysannes au Brésil. Paris, CNRS éditions, Paris, pp. 223.
Billard, C., Gond, V., Oszwald, J., de Sartre, X.A., Pokorny, B., 2014. Patterns defining
small holder agricultural practice in Amazonia. Bois Et Forets Des Tropiques 53–
64.
Boerner, J., Mendoza, A., Vosti, S.A., 2007. Ecosystem services, agriculture, and rural
poverty in the Eastern Brazilian Amazon: interrelationships and policy
prescriptions. Ecol. Econ. 64, 356–373.
Browder, J.O., Pedlowski, M.A., 2000. Agroforestry performance on small farms in
Amazonia: findings from the Rondonia agroforestry pilot project. Agrofor. Syst.
49, 63–83.
Carvalho, G., Barros, A.C., Moutinho, P., Nepstad, D., 2001. Sensitive development
could protect Amazonia instead of destroying it. Nature 409 131–131.
Chayanov, A.V., 1966. The Theory of Peasant Economy. Richard Irwin Inc., Illinois,
USA.
Courgeau, D., Lelievre, E., 1992. Event History Analysis in Demography 1, and 2.
Clarendon Press Oxford.
Da Silva, J.M.C., Rylands, A.B., Da Fonseca, G.A.B., 2005. The fate of the Amazonian
areas of endemism. Conserv. Biol. 19, 689–694.
Feijoo, M.A., Celis, G.L., 2010. Tres nuevas especies de Righiodrilus Zicsi 1995.
Annelida, Oligochaeta: Glossoscolecidae de la Amazonía colombiana. Acta
Amazonica 40 (1), 231–240.
Foley, J.A., Asner, G.P., Costa, M.H., Coe, M.T., DeFries, R., Gibbs, H.K., Howard, E.A.,
Olson, S., Patz, J., Ramankutty, N., Snyder, P., 2007. Amazonia revealed: forest
degradation and loss of ecosystem goods and services in the Amazon Basin.
Front. Ecol. Environ. 5, 25–32.
Game, E.T., Kareiva, P., Possingham, H.P., 2013. Six common mistakes in conservation
priority setting. Conserv. Biol. 27, 480–485.
Gardner, T.A., Barlow, J., Chazdon, R., Ewers, R.M., Harvey, C.A., Peres, C.A., Sodhi, N.
S., 2009. Prospects for tropical forest biodiversity in a human-modified world.
Ecol. Lett. 12, 561–582.
Gibson, L., Lee, T.M., Koh, L.P., Brook, B.W., Gardner, T.A., Barlow, J., Peres, C.A.,
Bradshaw, C.J.A., Laurance, W.F., Lovejoy, T.E., Sodhi, N.S., 2011. Primary forests
are irreplaceable for sustaining tropical biodiversity. Nature 478, 378.
Giles, J., 2006. The outlook for Amazonia is dry. Nature 442, 726–727.
Godar, J., Tizado, E.J., Pokorny, B., 2012. Who is responsible for deforestation in the
Amazon? A spatially explicit analysis along the Transamazon Highway in Brazil.
For. Ecol. Manag. 267, 58–73.
Grieg-Gran, M., Porras, I., Wunder, S., 2005. How can market mechanisms for forest
environmental services help the poor? Preliminary lessons from Latin America.
World Dev. 33, 1511–1527.
 P. Lavelle et al. / Global Environmental Change 40 (2016) 137–155
Grimaldi, M., Oszwald, J., Dolédec, S., Hurtado, M.P., Miranda de Sousa, I., Arnauld de
Sartre, X., Santos de Assis, W., Castañeda, E., Desjardins, T., Dubs, F., Guevara, E.,
Gond, V., Lima, T.T.S., Marichal, R., Michelotti, F., Mitja, D., Noronha, N.C.,
Delgado Oliveira, M., Ramirez, B., Rodriguez, G., Sarrazin, M., Lopes da Silva Jr.,
M., Silva Costa, L.G., Lindoso de Souza, S., Veiga Pereira, I., Velasquez, E., Lavelle,
P., 2014. Ecosystem services of regulation and support in Amazonian pioneer
fronts: searching for landscape drivers. Landsc. Ecol. 29, 311–328.
Gunderson, L.H., Holling, C.S., 2002. Panarchy. Understanding Transformations in
Human and Natural Systems. Island Press.
Hastie, T.J., 1991. Generalized additive models. In: Chambers, J.M., Hastie, T.J. (Eds.),
Chapter 7 of Statistical Models. Wadsworth & Brooks/Cole.
ISO 23611-5, 2011. Soil quality—Sampling of soil invertebrates—Part 5: Sampling
and extraction of soil macro-invertebrates. https://www.iso.org/obp/ui/#iso:
std:iso:23611:-5:ed-1:v1:en.
Ikerd, J.E., 1993. The need for a systems approach to sustainable agriculture. Agric.
Ecosyst. Environ. 46, 147–160.
Jackson, L.E., Pascual, U., Hodgkin, T., 2007. Utilizing and conserving
agrobiodiversity in agricultural landscapes. Agric. Ecosyst. Environ. 121, 196–
210.
Kareiva, P., Watts, S., McDonald, R., Boucher, T., 2007. Domesticated nature: shaping
landscapes and ecosystems for human welfare. Science 316, 1866–1869.
Lawrence, D., D’Odorico, P., Diekmann, L., DeLonge, M., Das, R., Eaton, J., 2007.
Ecological feedbacks following deforestation create the potential for a
catastrophic ecosystem shift in tropical dry forest. Proc. Natl. Acad. Sci. U. S. A.
104, 20696–20701.
Lenton, T., 2011. 2 degrees C or not 2 degrees C? That is the climate question. Nature
473, 7.
Lopez-Hernandez, D., Araujo, Y., Lopez, A., Hernandez-Valencia, I., Hernandez, C.,
2004. Changes in soil properties and earthworm populations induced by long-
term organic fertilization of a sandy soil in the Venezuelan Amazonia. Soil Sci.
169, 188–194.
Marichal, R., Martinez, A.F., Praxedes, C., Ruiz, D., Carvajal, A.F., Oszwald, J., Hurtado,
M.D., Brown, G.G., Grimaldi, M., Desjardins, T., Sarrazin, M., Decaens, T.,
Velasquez, E., Lavelle, P., 2010. Invasion of Pontoscolex corethrurus
Glossoscolecidae, Oligochaeta in landscapes of the Amazonian deforestation
arc. Appl. Soil Ecol. 46, 443–449.
Mattison, H., Norris, K., 2005. Bridging the gaps between agricultural policy: land-
use and biodiversity. Trends Ecol. Evol. 20, 610–616.
McKey, D., Rostain, S., Iriarte, J., Glaser, B., Birk, J.J., Holst, I., Renard, D., 2010. Pre-
Columbian agricultural landscapes, ecosystem engineers, and self-organized
patchiness in Amazonia. Proc. Natl. Acad. Sci. U. S. A. 107, 7823–7828.
Millennium Ecosystem Assessment, 2005. Maweb.org.
Miller, R.P., Nair, P.K.R., 2006. Indigenous agroforestry systems in Amazonia: from
prehistory to today. Agrofor. Syst. 66, 151–164.
Morton, D.C., DeFries, R.S., Shimabukuro, Y.E., Anderson, L.O., Arai, E., Espirito-Santo,
F.D.B., Freitas, R., Morisette, J., 2006. Cropland expansion changes deforestation
dynamics in the southern Brazilian Amazon. Proc. Natl. Acad. Sci. U. S. A. 103,
14637–14641.
Nakagawa, S., Schielzeth, H., 2013. A general and simple method for obtaining R2
from generalized linear mixed-effects models. Meth. Ecol. Evol. 4.2, 133–142.
Nelson, E., Mendoza, G., Regetz, J., Polasky, S., Tallis, H., Cameron, D.R., Chan, K.M.A.,
Daily, G.C., Goldstein, J., Kareiva, P.M., Lonsdorf, E., Naidoo, R., Ricketts, T.H.,
Shaw, M.R., 2009. Modeling multiple ecosystem services, biodiversity
conservation, commodity production, and tradeoffs at landscape scales. Front.
Ecol. Environ. 7, 4–11.
Nepstad, D.C., Stickler, C.M., Soares-Filho, B., Merry, F., 2008. Interactions among
Amazon land use, forests and climate: prospects for a near-term forest tipping
point. Philos. Trans. R. Soc. B Biol. Sci. 363, 1737–1746.
Oliveira, L.J.C., Costa, M.H., Soares-Filho, B.S., Coe, M.T., 2013. Large-scale expansion
of agriculture in Amazonia may be a no-win scenario. Environ. Res. Lett. 8.
View publication stats
155
Oszwald, J., Gond, V., Doledec, S., Lavelle, P., 2011. Identification of land-use
indicators to assess changes in landscape mosaics. Bois Et Forets Des Tropiques
7–21.
Peres, C.A., 2005. Why we need megareserves in Amazonia. Conserv. Biol. 19, 728–
733.
Perfecto, I., Vandermeer, J., 2008. Biodiversity conservation in tropical
agroecosystems—s new conservation paradigm, Year in Ecology and
Conservation Biology, 2008, pp. 173–200.
Pinheiro, J.C., Bates, D.M., 2000. Mixed Effects Models in S and S-Plus. Springer-
Verlag, New York.
Radford, J.Q., Bennett, A.F., Cheers, G.J., 2005. Landscape-level thresholds of habitat
cover for woodland-dependent birds. Biol. Conserv. 124, 317–337.
Ramirez, B.L., Lavelle, P., Orjuela, J.A., Villanueva, O., 2012. Characterization of cattle
farms and adoption of agroforestry systems as a proposal for soil management
in Caqueta, Colombia. Revista Colombiana De Ciencias Pecuarias 25, 391–401.
Raudsepp-Hearne, C., Peterson, G.D., Tengoe, M., Bennett, E.M., Holland, T.,
Benessaiah, K., MacDonald, G.K., Pfeifer, L., 2010. Untangling the
environmentalist’s paradox: why is human well-being increasing as ecosystem
services degrade? Bioscience 60, 576–589.
Rey Benayas, J.M., Bullock, J.M., 2012. Restoration of biodiversity and ecosystem
services on agricultural land. Ecosystems 15, 883–899.
Robert, P., Escoufier, Y., 1976. Unifying tool for linear multivariate statistical
methods. RV coefficient. Appl. Stat. 25, 257–265.
Rodrigues, A.S.L., Ewers, R.M., Parry, L., Souza, C., Verissimo, A., Balmford, A., 2009.
Boom-and-bust development patterns across the Amazon deforestation
frontier. Science 324, 1435–1437.
Simon, M.F., Garagorry, F.L., 2005. The expansion of agriculture in the Brazilian
Amazon. Environ. Conserv. 32, 203–212.
Smukler, S.M., Sanchez-Moreno, S., Fonte, S.J., Ferris, H., Klonsky, K., O’Geen, A.T.,
Scow, K.M., Steenwerth, K.L., Jackson, L.E., 2010. Biodiversity and multiple
ecosystem functions in an organic farmscape. Agric. Ecosyst. Environ. 139, 80–
97.
Staver, A., Archibald, S., Levin, S.A., 2011. The global extent and determinants of
Savanna and forest as alternative Biome States. Science 334, 230–232.
Steffan-Dewenter, I., Kessler, M., Barkmann, J., Bos, M.M., Buchori, D., Erasmi, S.,
Faust, H., Gerold, G., Glenk, K., Gradstein, S.R., Guhardja, E., Harteveld, M., Hertel,
D., Hohn, P., Kappas, M., Kohler, S., Leuschner, C., Maertens, M., Marggraf, R.,
Migge-Kleian, S., Mogea, J., Pitopang, R., Schaefer, M., Schwarze, S., Sporn, S.G.,
Steingrebe, A., Tjitrosoedirdjo, S.S., Tjitrosoemito, S., Twele, A., Weber, R.,
Woltmann, L., Zeller, M., Tscharntke, T., 2007. Tradeoffs between income,
biodiversity, and ecosystem functioning during tropical rainforest conversion
and agroforestry intensification. Proc. Natl. Acad. Sci. U. S. A. 104, 4973–4978.
Swift, T.L., Hannon, S.J., 2010. Critical thresholds associated with habitat loss: a
review of the concepts, evidence, and applications. Biol. Rev. 85, 35–53.
Velasquez, E., Lavelle, P., Andrade, M., 2007. GIQS: a multifunctional indicator of soil
quality. Soil Biol. Biochem. 39, 3066–3080.
Vieira, I.C.G., Toledo, P.M., Silva, J.M.C., Higuchi, H., 2008. Deforestation and threats
to the biodiversity of Amazonia. Braz. J. Biol. 68, 949–956.
Walker, R., Moran, E., Anselin, L., 2000. Deforestation and cattle ranching in the
Brazilian Amazon: external capital and household processes. World Dev. 28,
683–699.
Walker, R., Moore, N.J., Arima, E., Perz, S., Simmons, C., Caldas, M., Vergara, D.,
Bohrer, C., 2009. Protecting the Amazon with protected areas. Proc. Natl. Acad.
Sci. U. S. A. 106, 10582–10586.
Weinhold, D., Killick, E., Reis, E.J., 2013. Soybeans, poverty and inequality in the
Brazilian Amazon. World Dev. 52, 132–143.
Zavaleta, E.S., Pasari, J.R., Hulvey, K.B., Tilman, G.D., 2010. Sustaining multiple
ecosystem functions in grassland communities requires higher biodiversity.
Proc. Natl. Acad. Sci. U. S. A. 107, 1443–1446.