Applied Geography 43 (2013) 171e181

Contents lists available at SciVerse ScienceDirect

Applied Geography
journal homepage: www.elsevier.com/locate/apgeog

Rebuilding the Brazilian rainforest: Agroforestry strategies for

secondary forest succession
Christine E. Blinn a, *, John O. Browder b, Marcos A. Pedlowski c, Randolph H. Wynne a
a
Department of Forest Resources and Environmental Conservation, College of Natural Resources and Environment,
Virginia Polytechnic Institute and State University, Blacksburg, VA 24061, USA
b
College of Architecture and Urban Studies, Virginia Polytechnic Institute and State University, Blacksburg, VA 24061, USA
c
Laboratório de Estudos do Espaço Antrópico, Universidade Estadual do Norte Fluminense, Campos dos Goytacazes, Rio de Janeiro, Brazil

a b s t r a c t
Keywords: Does the adoption of agroforestry by small farmers in the Brazilian Amazon promote secondary forest
Deforestation succession on the degraded pastures and crop fields? New results from a small-scale farm agroforestry
Development
demonstration project, the Rondônia Agroforestry Pilot Project (RAPP) that began in 1992 are presented
Frontier
Land use
in this paper. In 1992, 242 farmers were surveyed by a stratified random sampling protocol, 50 of whom
Landsat were selected to participate in the RAPP, constituting its experimental group. The remaining 191 farmers
Multi-temporal served as a control group. Farmers from both groups were re-surveyed in 2002 (after 10 years) and again
in 2010 (after 18 years). Annual site visits to the experimental group farms were conducted from 1993
through 2003 to monitor agroforest plot development and management, and changes in pertinent socio-
economic and household demographic characteristics such as household capacity, production systems,
and social participation. Differences in property size, number of people permanently residing on the
property, and social participation were found between the experimental and control group, with the
experimental group having larger properties, more residents, and more participation in mutual aid as-
sociations. Control group farmers were also more reliant on cattle production (based on 2009 sales)
despite having similar amounts of pasture as farmers in the experimental group. Within the experi-
mental group, very few differences were found between farmers based on the type of agroforestry plot:
timber, mixed or non-timber. Remote sensing analyses reveal long-term (10 yearsþ) spectral differences
in terms of the similarity to primary forest of both the agroforestry plots and the entire properties of the
farms in this study. Experimental group farmers with mixed or timber-based agroforestry plots allowed
more secondary forest succession to occur in and around their plots than farmers with non-timber plots.
Although, on average, farm properties have become less spectrally similar to primary forest since 1992,
properties with agroforestry plots tend to have more secondary succession and/or primary forest on their
land in 2011. Several example properties are shown to illustrate the tendency of farmers with agrofor-
estry plots to allow more secondary forest succession to occur on their land.
Ó 2013 The Authors. Published by Elsevier Ltd. Open access under CC BY license.

Introduction
Agroforestry and tropical forest succession
Socio-economic research articles on the role of agroforestry in
promoting secondary forest succession were absent from Mercer
* Corresponding author. Tel.: þ1 5402315525.
E-mail addresses: cblinn@vt.edu (C.E. Blinn), browder@vt.edu (J.O. Browder),
pedlowma@uenf.br (M.A. Pedlowski), wynne@vt.edu (R.H. Wynne).
and Miller’s (1998) content review of the journal Agroforestry Sys-
tems, spanning 1982 to 1996. Since then this subject has entered
into a broader scientific discourse on tropical agroforestry (e.g.,
Chowdhury, 2007; Ehiagbonare, 2006; JIRCAS, 2007; Lieberei &
Gasparotto, 1998; Meza, Sabogal, & Jong, 2006; Raman, Mudappa,
& Kapoor, 2009; Shono, Cadaweng, & Durst, 2007; Vieira, Holl, &
Peneireiro, 2009). Agroforestry, defined as a “system of land use
in which harvestable trees or shrubs are grown among or around
crops or on pastureland” (Agroforestry, 2011) has evolved in
numerous social and cultural contexts as a managed successional
land cover to achieve fallow enrichment, secondary forest cover,
riparian forestland rehabilitation, degraded forest recuperation or
recovery, and agro-successional restoration. As a vehicle for

0143-6228 Ó 2013 The Authors. Published by Elsevier Ltd. Open access under CC BY license.
http://dx.doi.org/10.1016/j.apgeog.2013.06.013
172 C.E. Blinn et al. / Applied Geography 43 (2013) 171e181
promoting reforestation through managed secondary forest suc-
cession, research has also focused on the factors that influence
farmers to adopt agroforestry practices (Smith, Dubois, Current,
Lutz, & Clement, 1998; Warner, 1993; Yokota, Martin, & Siki,
2009). Related research has also explored the impacts of agrofor-
estry and secondary forest succession on nutrient cycles (Sirois,
Margolis, & Camiré, 1998), on biodiversity and wildlife pop-
ulations (Bobo, Waltert, Fermon, Njokagbor, & Mühlenberg, 2006;
Harvey & Haber, 1998; Letcher & Chazdon, 2009; Lozada, de
Koning, Marché, Klein, & Tscharntke, 2006; Schulze et al., 2004),
on atmospheric carbon sequestration (Castro, Sanchez-Azofeifa, &
Rivard, 2003; Delaney, 1999; Fearnside & Guimarães, 1996; Lasco,
Guillermo, Cruz, Bantayan, & Pulhin, 2004; Roshetko, Delaney,
Hairiah, & Purnomosidhi, 2002; Schroth, D’Angelo, Teixeira, Haag, &
Lieberei, 2002; Takimoto, Nair, & Nair, 2009; Wise & Cacho, 2011),
and on agroforestry’s contribution to rural household livelihood,
not being limited to just income generation (Alavalapati & Nair,
2001; Barton, 1994; Budowski, 1980; Pattanayak & Mercer, 1998).
As the research literature on agroforestry and its potential contri-
bution to natural reforestation grows, questions remain regarding
what types of agroforestry systems produce greater positive results
in promoting secondary forest succession in the tropics.
Rondônia Agroforestry Pilot Project
This paper updates selected findings of the Rondônia Agrofor-
estry Pilot Project (RAPP), an on-farm experimental agroforestry
demonstration project involving small-scale farmers in the south-
western Brazilian Amazon state of Rondônia over an 18 year (1992e
2010) period (Browder & Pedlowski, 2000; Browder, Wynne, &
Pedlowski, 2005; Summers, Browder, & Pedlowski, 2004). One of
the long-term research questions that the RAPP sought to address
was: Is successful agroforestry a catalyst to secondary forest suc-
cession that might encourage reforestation of degraded lands on
small farms in the Amazon? And, more specifically, does the type of
agroforestry system (non-timber, mixed, or timber-based) adopted
influence the likelihood that a farmer will manage degraded land
for secondary forest regrowth as previously hypothesized (Browder
et al., 2005)? Furthermore, are there socio-economic household
characteristics that might predict secondary forest succession
based on agroforestry?
Considerable differentiation in the spectral signatures from
satellite images of the RAPP planting sites and their immediate
surroundings were noticed over time. In some cases, farmers had
allowed secondary forest vegetation to subsume their agroforest
plots and in others, farmers had more carefully managed their plots
sites to minimize secondary vegetation. The spectral differentiation
within the experimental group led the principal investigators to
speculate about socio-economic factors that might influence these
spatial patterns. They hypothesized that three variables, for which
household level survey data were available, might contribute to a
better understanding of these patterns, as follows: (1) Household
capacity of the control and experimental groups (number of
working age adults living and working on the farm and the size of
the farm area). The greater the household capacity the more likely a
household might be to pursue a more labor intensive non-timber or
mixed agroforestry experiment. (2) Dominant farming strategy
pursued by the experimental farmers (area devoted to perennial
cropping, annual cropping, and cattle). The more area in perennial
cropping the more likely a farmer would exhibit a tendency to
manage crops for annual harvest and income leading to a non-
timber agroforestry preference. (3) Associational activities of con-
trol and experimental groups (farmer participation in mutual aid
associations and rural workers unions). Farmers more active in
these associations would be more likely to innovate and adopt
agroforestry experimentally because of the greater network of
technical information and mutual support such participation
provides.
From these questions and observations the research objectives
for this paper, enumerated below, emerged e to assess the poten-
tial impact of agroforestry adoption on secondary forest succession
and primary forest conservation.
Objectives
In this paper, both 2010 land owner survey results and
contemporaneous remote sensing analyses were used to address
the following research questions:
1) Are there any significant differences in socio-economic char-
acteristics between:
a) The experimental group of agroforestry adopters (n ¼ 31)
and the control group of non-adopters (n ¼ 39) included in
the 2010 survey?
b) The three different sub-groups of agroforestry adopters;
timber, non-timber, and mixed, in the experimental group?
2) Are there any significant spectral differences in land cover in
and around the agroforestry plots between the sub-groups of
agroforestry adopters in the experimental group that might
indicate a positive synergy between agroforestry demonstra-
tion plots and subsequent secondary forest succession?
3) Are there any significant differences in the amount of remnant
primary forest cover on the properties in the project’s experi-
mental and control groups that might indicate a potential
natural forest conservation effect of agroforestry adoption?
Study sites
Both control and experimental groups were drawn from the
same rural population of family farmers who had migrated to
Rondônia between 1980 and 1985 and settled in the project’s two
study sites, the municipios (counties) of Nova União and Alto Par-
aiso. The largest proportion of these farmers originated in the South
and Southeast regions of Brazil, most having worked as share-
croppers and tenant farmers on coffee plantations. With the pro-
gressive government-sponsored conversion from coffee to soybean
production beginning in the mid-1960s this rural population
became progressively displaced from their traditional livelihoods
and the government actively encouraged their migration to Ron-
dônia as part of a larger regional development and resettlement
program called The Northwest Region Development Plan (POLO-
NOROESTE) (Browder & Godfrey, 1997; pp. 164e175). The farming
strategies pursued by these new homesteaders in Rondônia typi-
cally followed a similar pattern: Small-scale forest clearing and
planting of annual crops (corn, rice, beans) and a small area of
perennial crops (usually coffee and cacao). Small livestock and
eventually milk cattle, then beef cattle were added over time. For
various reasons, several patterns of socio-economic and land use
differentiation began to emerge leading some unsuccessful farmers
to sell all or part of their properties, whilst more successful
neighbors enlarged their holdings (for a more detailed examina-
tion, see Browder, 1994). The causes of these parallel processes of
property subdivision and enlargement and how they correlated to
changes in land cover, land use, household income, and numerous
other household level socio-economic and demographic charac-
teristics within the context of leading theories of frontier expansion
are presented elsewhere (Browder et al., 2008). Suffice it to say that
the study sites and their rural populations surveyed were
 C.E. Blinn et al. / Applied Geography 43 (2013) 171e181 173

representative of the general rural population of Rondônia during
the time period of the project.
Methods
The Rondônia Agroforestry Pilot Project (RAPP) e ground-level
analyses
Detailed descriptions of the RAPP study sites and project
methodology were previously published in Browder and Pedlowski
(2000) and Browder et al. (2005). In summary, during 1992, 50
farms were selected from a stratified random sample of 242 farms
surveyed in two colonization areas of the frontier state of Rondônia,
in the western Brazilian Amazon. Farmers self-selected their
participation in the project based on their responses to the initial
standardized baseline questionnaire, which was administered by
personal interview by trained Brazilian agronomy and forestry
students. Invited participants were those indicating a predisposi-
tion toward planting trees (for example, did they already plant trees
on their farm, could they recognize native species in the wild, and
the like) and experimenting with innovative agro-silvicultural
practices in the survey and subsequent open-ended interviews.
Participating farmers were invited to choose from a menu of 20
plant species (all native except for teak e Tectona grandis e an Asian
exotic introduced into the Amazon some 20 years before the RAPP)
each producing one or more commodities of local market demand.
Seedlings were produced on local project nurseries using regionally
available seed stock. Plots were designed to fit on a 1-ha area
containing no primary forest vegetation. Each plot consisted of
between 800 and 1000 individuals from between 2 and 18 species.
Plot composition typically included a mix of fruit, palm, oil and nut-
producing species (with potential to generate revenues in the mid-
term (4e5 years), along with various citrus and industrial timbers
(hard and soft-woods). To enhance farm income in the short-term,
participating farmers were offered a bee-keeping component, an
option which 36% of the original experimental farmers elected.
Seedlings were distributed to participants during the first half of
the rainy season. The farmers used household labor to plant the
seedlings according to each farmer’s plot plan. Within six months
of the initial planting each farm was visited, seedling growth was
measured, and problems with plot maintenance were described.
The project was monitored annually from 1993 to 2001 and then, in
2002, a 10-year follow-up survey of all 242 farms (by then sub-
divided into 283 farms) was undertaken (Browder et al., 2005;
Summers et al., 2004). A second follow-up study was conducted
in 2003 of a smaller subset of the original sample, to further
investigate the tropical forest management and silvicultural prac-
tices of this farming population (Summers, 2008). Then in 2010, a
third follow-up farm-level survey was completed that revisited all
31 (out of 50) surviving RAPP experimental farms, along with
another 39 farms from the original control group.
By 2010, three types of experimental farms had emerged: those
that had chosen initial agroforest designs that emphasized long-
term commercial timber components (timber-based agroforest
systems), and had subsequently managed those plots to maximize
long-term timber yield; those that opted for faster-growing ground
and bush cropping agroforestry systems, without a significant long-
term timber component (non-timber-based agroforest systems);
and those that combined timber and non-timber elements into
mixed agroforestry systems (Fig. 1). Mean values for each variable
of interest from the 2010 survey were calculated for the entire
sample, for both the experimental and control groups of farms, and
for the three types (timber, non-timber, and mixed) of farms in the
experimental group, described above. The socio-economic vari-
ables of interest fell into three types of characteristics: household
capacity (property size and number of adults residing on the

Fig. 1. Illustrative agroforest plots of each experimental farm type. Note: The figure shows examples of actual plots developed by RAPP farmers based on their interests (see Browder
& Pedlowski, 2000). Considerable variation exists within each farm type in terms of species composition. All participants were required to include one tree species in their
agroforest plots. Classification Criteria: “Non-timber” based systems incorporated just one tree species but predominantly consisted of perennial ground-covering bush species
producing fruits, seeds for commercial or medicinal purposes; “Timber based” systems usually contained one perennial ground cover crop but predominantly included 2 or more
timber producing tree species; “Mixed” systems consisted of more balanced proportions of elements of both non-timber and timber species.
174 C.E. Blinn et al. / Applied Geography 43 (2013) 171e181
property); production system type (based on annual cropping,
permanent cropping, and pasture/cattle); and social participation
(membership in farmers organizations).
The AndersoneDarling normality test was used to determine if
the continuous variables of interest were normally distributed and
could thus be evaluated with parametric statistical tests.
Nonparametric statistical tests were selected based on the
normality test results. Medians between the experimental and
control groups were tested for all continuous variables using the
ManneWhitney test. While medians between the intra-
experimental groups were tested using the KruskaleWallis test,
which can accommodate more than two groups. Binomial variables
were tested with Fisher’s exact test and the chi-square test of as-
sociation for comparisons of two or three groups, respectively. All
tests were conducted with MinitabÔ (Minitab Inc., 2012), a statis-
tical software program.
Remote sensing analyses
Two Landsat path/rows, 231/68 and 232/67, contained the study
area (Fig. 2). Five image dates (Table 1) at approximately five year
intervals were acquired from the United States Geological Survey
(USGS) for each path/row between 1991 or 1992 and 2011. All im-
ages were obtained in the standard level-one terrain-corrected
format and thus did not require image to image co-registration.
Each Landsat image was processed through the Landsat
Ecosystem Disturbance Adaptive Processing System (LEDAPS)
(Masek et al., 2006) to surface reflectance to minimize differences
caused by atmospheric conditions at the time of image acquisition.
Similar to Browder et al.’s (2005) methods, calculations were
standardized across image dates and path/row boundaries by using
a common area between Landsat scenes. The common area was ten
thousand pixels in size and contained primary forest in all images.
A mean surface reflectance vector for primary forest was calculated
Fig. 2. Study area location within Brazil and Landsat path/rows used.
for each image, path/row and date, using the common area. Images
with the spectral distance from primary forest were then created by
calculating the Euclidian distance (ED) between each pixel in an
image and that image’s primary forest mean surface reflectance
vector using five of the six non-thermal Landsat spectral bands, two
through five and seven.
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u7ðs6Þ
uX
ED ¼ t ðmi  xi Þ2 (1)
i¼2
where: i ¼ Landsat band number; mi ¼ mean value for band i; and
xi ¼ pixel value for band i. As an example, the ED calculation for a
pixel in the 2011 image from 232/67 is illustrated in Table 2.
Summary statistics were calculated for three buffer zones, 50,
85, and 117.5 m (Fig. 3), around the center location of each agro-
forestry plot using the spectral distance from primary forest im-
ages. These radii were selected to assess the synergistic “spillover”
effect of the agroforest plot on the land cover of its immediate
surrounding environment. The mean spectral distance of each
agroforestry plot from primary forest was averaged by agroforestry
plot type and buffer distance. The mixed and timber plot types
were combined in the remote sensing analyses because of similar
spectral distance trajectories through time (correlation of 0.79). A
forest similarity index, between 0 and 1, was created by subtracting
each plot’s mean spectral distance from the maximum mean
spectral distance of all properties and then dividing by the
maximum spectral distance. The forest similarity index was plotted
vs. time by agroforestry plot type and for an example plot where
secondary forest succession occurred (Fig. 4). The KruskaleWallis
nonparametric test of medians with an alpha of 0.05 was used to
test for the equality of the forest similarity index medians between
the mixed/timber and non-timber plot types at each time interval.
In order to compare farmers that were part of the experimental
 C.E. Blinn et al. / Applied Geography 43 (2013) 171e181 175

Table 1 considered as outliers in the population, the difference in median

Landsat TM dates used in analysis for each path/row. number of cattle sold between the control and experimental groups
231/68 232/67 is no longer statistically significant. Even though farmers in the
7/25/1992 6/28/1991
experimental and control groups allocated roughly the same
7/20/1996 6/09/1996 amount of land to “productive pasture” (although non-adopters had
8/03/2001 8/10/2001 a larger proportion of their total property in pasture), the control
7/16/2006 6/21/2006 group sold a larger number of cattle in 2009 (median of 23 without
8/15/2011 8/06/2011
outliers removed) than the experimental group sold (median of 6.0)
despite the borderline statistical significance for this variable.
Although both experimental and control groups share similar pro-
group, who had agroforestry plots on their land, with the control
duction systems, the control group was more dependent on cattle
group of farmers, who were surveyed, but did not have agroforestry
production for household livelihood than the agroforestry adopters.
plots, the average forest similarity index of all pixels within each
property boundary was calculated by farmer group (experimental
Social participation
vs. control). This surrogate measure of primary and/or secondary
Differences in social participation rates observed between the
forest retention or regeneration is plotted vs. year and by farmer
experimental and control groups were not statistically significant,
group in Fig. 5 and also tested for equality of medians with the
although agroforestry adopters appear to be more engaged in both
KruskaleWallis nonparametric test.
rural workers syndicates (unions) (77.4% vs. 61.5%) and mutual aid
associations (61.3% vs. 46.2%) for sharing labor among different
Results
households (Table 3b). There is an effect in these findings, associ-
ated with the survey instrument or sample design, since social
The 2010 survey results revealed some statistically significant
participation was one of the original criteria used for experimental
differences in household characteristics between the experimental
group selection. Given that the sample frames were constituted in
group and the control group, but no statistically significant differ-
1992, it is noteworthy that these social participation rates still
ences between the three types of agroforest farm designs adopted
remain higher among agroforestry adopters than the control group
(Tables 3a and 3b). Since all but one of the sample distributions
of non-adopters.
from the continuous variables (Table 3a) were not normally
distributed, nonparametric tests were used to compare medians
Intra-experimental group differentiation: timber vs. non-timber vs.
between and within the groups.
mixed-based agroforest systems

Experimental vs. control group
Household capacity
Statistically significant differences in median property size and
household composition between the experimental and control groups
emerged between the two groups by 2010, although there were no
significant differences in either at the time of the 1992 baseline survey
(Browder & Pedlowski, 2000), both having been drawn from the same
initial population (Table 3a). Both median property size and number of
permanent property residents were larger for the experimental
group, suggesting that farmers participating in the RAPP until 2010
enjoyed greater capacity for innovation (more land to work with and
more workers per farm to work the land).
Production systems
While there were no significant differences between the exper-
imental and control groups in terms of percentage producing an-
nuals and perennial crops (Table 3b), and number of hectares in
pasture, there was a statistically significant difference in medians at
an alpha of 0.055 in the number of cattle marketed in the previous
year (2009) (Table 3a). Three farmers in the control group sold more
than 200 cattle in 2009, with one selling 1200. If these farmers are
Table 2
Euclidian distance calculation for a pixel in the 2011 image from 232/67 with an
ED ¼ 363.398.
Unlike the differences observed between the experimental
group, as a whole, and the control group, no statistically significant
differences were observed between types of agroforest system
adopters. The lack of statistical significance is likely due to small
sample sizes and thus differences in four of the indicators seem
noteworthy. First, a greater proportion of non-timber system
households managed perennial crops (84.6%) than either of the
other two types of agroforest system adopters (Perennial Crops,
Table 3b). The second difference pertains to the number of cattle
sold (Cattle, Table 3a). Timber-based system adopters sold more
animal units in 2009 (30.4) than either of the other two types of
agroforest system adopters. A third difference pertains to the de-
gree of social participation among the 3 groups of agroforest system
adopters (Associations, Table 3b). A larger proportion of timber-
based system adopters reported participating in mutual aid asso-
ciations (71.4%) than either of the other two types. Finally, a higher
percentage of timber-based system adopters (71.4%) reported
obtaining benefits from their agroforest plots than either of the
other two groups (Utilization, Table 3b). Although only one exper-
imental group household sold any produce, participants typically
reported non-timber benefits of occasional game, shade, fresh
water, cleaner air, and cooler temperatures associated with the
agroforest plots.
Remote sensing results

Landsat Pixel reflectance Primary forest Difference Twenty eight of the original 50 agroforestry plot locations were
band Vectora (xi) mean reflectance squared (mi  xi)2
analyzed, six timber-based, thirteen non-timber, and nine from the
vectora (mi)
mixed sub-group. A total of 9, 26, and 50 pixels, respectively, were
2 392 313.702 6130.577
contained within the three buffer distances, 50, 85, and 117.5 m,
3 295 244.603 2539.858
4 2852 2525.466 106,624.5 around each agroforestry plot center. Mixed/timber agroforestry
5 1315 1198.933 13,471.55 plots had a decrease in their average forest similarity index be-
7 531 473.630 3291.317 tween 1991/1992 and 1996 (Fig. 4), presumably as a result of land
Sum 132,2057.8 clearing for plot establishment. During this same time period, non-
a
Units are 10,000 times the percent reflectance. timber agroforestry plots had very little change in their average
176 C.E. Blinn et al. / Applied Geography 43 (2013) 171e181

Fig. 3. Example plot with three buffer zones over a Landsat image displayed with a band 4,3,2 (RGB) band combination. (For interpretation of the references to color in this figure
legend, the reader is referred to the web version of this article.)

forest similarity index despite plot establishment activities. Be-
tween 1996 and 2001 mixed/timber agroforestry plots had a large
increase in their average forest similarity index while non-timber
plots showed little change or a slight decrease in their average
similarity index. From 2001 to 2006, non-timber plots had a
decrease in their average forest similarity index while mixed/tim-
ber plots had little change in their average forest similarity index.
All plot types had decreasing average forest similarity indices be-
tween 2006 and 2011 with non-timber plots having a very large
drop relative to the mixed/timber plot type. The mixed/timber plots
had a mean forest similarity index trajectory through time that was
more similar, albeit not identical, to that of a plot where secondary
forest succession occurred (Fig. 4).
The observed changes in the forest similarity index within a
50 m buffer around each agroforestry plot were very similar to
the changes within the 85 and 117.5 m plot buffers. At all buffer
distances, the areas surrounding each agroforestry plot were least
similar to primary forest in the non-timber plot type in 2001,
2006, and especially in 2011. Within each plot type, the mean
forest similarity index was almost identical for all three buffer
distances in 1991/1992 and 1996. In 2001, 2006, and 2011, the
mean forest similarity index decreased slightly with buffer size
across all plot types. In 2006 and 2011 the mixed/timber plots
had significantly higher median values of forest similarity index
than the non-timber plot types based on the KruskaleWallis test
of equality of medians at the alpha 0.05 level. The KruskaleWallis
 C.E. Blinn et al. / Applied Geography 43 (2013) 171e181
Fig. 4. Mean forest similarity index of pixels within a 50 m buffer of an agroforestry
plot’s center by agroforestry plot type. A plot that was allowed to return to “Secondary
Forest” is also included for comparison. The first image date was collected in different
years, 1991 and 1992, for the two path/rows and is thus labeled 1991/1992.
test results were not significant for the previous four time
periods.
Farm boundaries were available for 34 of the farmers in the
control group and 28 of the farmers in the experimental group. At
the farm scale, the average forest similarity index of all pixels
within the properties of farmers in the experimental group did not
start to diverge from the average forest similarity index of pixels
within properties of farmers in the control group until 2006 (Fig. 5).
The greatest difference between forest similarity indices of the
experimental and control groups occurred in 2011 (Fig. 5) and was
statistically significant at the alpha 0.05 level for this year only. This
suggests that farmers in the experimental group are retaining more
primary and/or secondary forest on their properties than farmers in
the control group. Both groups have moved farther from primary
forest since 1991/1992 with the control group moving at a faster
rate between 2001 and 2011.
Fig. 5. Mean forest similarity index of all pixels within all properties containing
agroforestry plots vs. control properties without agroforestry plots. A plot that was
allowed to return to “Secondary Forest” is also included for comparison. The first image
date was collected in different years, 1991 and 1992, for the two path/rows and is thus
labeled 1991/1992.
177
Discussion
The RAPP began with utilitarian assumptions: (1) farmers would
allocate effort in direct proportion to benefit; (2) such benefits may
or may not be pecuniary in nature, as even in fully commercialized
farms, some land devoted to subsistence production almost always
occurs; and (3) utility maximization always occurs in a contempo-
rary policy context, which frames the ways costs and benefits are
apportioned in society. No personal welfare maximization model of
behavior can ignore the political economy in which household [land
use] decisions are made. Hypothetically, then, successful experi-
mental group farmers are either: (1) deriving a tangible benefit from
their agroforest plots, monetary or subsistence; or (2) some exoge-
nous factor has shaped the context that either artificially encourages
or dissuades the farmer from continuing in the RAPP.
Socio-economic characteristics: experimental vs. control group
Household capacity
Perz (2005) found that agricultural diversity on small farms in
the Brazilian Amazon tended to be greater for households that were
more “asset-rich” with labor being the most important factor. We
similarly expected farmers with larger farms and more adult resi-
dents to have a greater capacity to adopt a potentially risky inno-
vation like an agroforestry experiment. The larger the farm, the
more likely a farmer could take a small area out of normal pro-
duction to dedicate it to an experiment without jeopardizing
household food security. Similarly, the more adult residents present
on the farm, the more likely a farm household could adopt an
additional productive activity minimizing the labor reallocation
costs of additional workload within the household. In this study,
the experimental group had larger farms on average and more
people per farm than the control group (Table 3a), which supports
our expectations. This finding corresponds with Sydenstricker-
Neto’s (2012) results showing that diversified agricultural sys-
tems in the Brazilian Amazon are associated with households
having more labor force and resource access.
Production system
Since the RAPP was intended to supplement, not substitute,
existing farming practices, no differences in farming systems (mix of
cropping regimes) between adopters and non-adopters were pre-
dicted. Very similar mixes of cropping regimes were found between
the experimental and control groups (Table 3b). The control group
was more dependent on cattle, but this could have been associated
with the fact that they had smaller farms on average (sensu
D’Antona, VanWey, & Hayashi, 2006; Vosti, Witcover, & Carpentier,
2002; Sydenstricker-Neto, 2012). However, the RAPP project design
did hypothesize that farmers adopting agroforestry would either
manage larger secondary forests or retain a larger proportion of
primary forests than non-adopters. We address this in Section 6.3.
Social participation
Rural households can diffuse the risks of adopting innovative
farming practices by pooling their energies in various forms of labor
sharing and social participation (e.g., Pannell & Vanclay, 2011). In
the Brazilian project areas, two forms of farmer organization are
prevalent: rural workers unions (sindicatos de trabalhadores rurais)
and associations of mutual (labor sharing) support (associações de
ajuda mútua). Rural workers unions, although often politicized with
specific partisan affiliations, frequently provide members with
some collective clout of political representation that individual
farm households usually lack. These can be important for defending
contested property claims, voicing protests to abusive third party
marketing practices, and confronting corruption by local officials,
178 C.E. Blinn et al. / Applied Geography 43 (2013) 171e181

Table 3a
Mean and median statistics of continuous socio-economic variables of interest in 2010 by farm and agroforestry plot type groupings.

Size (ha) Size (ha) People People Pasture area Pasture area Cattle (au) mean Cattle (au) median
mean median mean median (ha) mean (ha) median

Control group (n ¼ 39) 57.8 50.6c 3.3 3.0c 34.6 24.1 76.8a (30.1b) 23.0d (20.0b)
Experimental group (n ¼ 31) 75.7 77.1c 6.0 5.0c 30.3 21.7 18.2 6.0d
Timber (n ¼ 7) 74.7 72.3 5.4 5.0 34.4 28.9 30.4 10.0
Mixed (n ¼ 11) 76.0 77.1 6.3 5.0 30.5 19.3 16.7 6.0
Non-timber (n ¼ 13) 75.9 79.5 6.2 5.0 27.9 21.7 12.8 10.0
Overall sample (n ¼ 70) 65.7 68.7 4.5 4.0 32.7 21.7 50.5a 12.0

where:
Size e primary property size in hectares.
People e number of people permanently residing on property.
Pasture Area esize of pasture area in hectares.
Cattle e number of cattle (animal units) sold in 2009.
a
One farmer did not respond to this question and is thus not included in this statistic.
b
Statistic calculated with three outliers (value greater than 200) removed.
c
Statistically significant difference based on the ManneWhitney test at alpha of 0.050.
d
Statistically significant difference based on the ManneWhitney test at alpha of 0.055.

all of which are widely viewed to empower farmers to accept
greater risk. Mutual support associations are usually basic labor
sharing activities, where a group of workers spend a few days per
month working on one member’s farm in return for reciprocal
support from the association. Hypothetically, such participation
would reduce the risk of innovation and agroforestry adopters, it
was predicted, would be more involved in such organizations,
which is supported by the findings in Table 3b. Perz (2005) also
found a positive relationship between production diversity and
participation in local organizations.
Socio-economic characteristics: intra-experimental groups (timber
vs. non-timber vs. mixed-based agroforest systems)
Household capacity
There is no particular reason to presume that farm size would be
a factor influencing what type of agroforestry system a farmer
would adopt: timber, non-timber, or mixed. However, among the
experimental group of farmers adopting an agroforestry system, it
may be argued that those households adopting timber-based sys-
tems could afford to be smaller in number of residents than those
adopting either mixed or non-timber systems where regular labor
requirements are greater (e.g., in harvesting fruits, nuts, latex, etc.).
Although the differences were small, the timber based agroforestry
adopters did have slightly fewer people residing on the farm than
others in the experimental group (Table 3a).
Production system
It is not intuitive that there would be any differentiation in
farming systems resulting from the adoption of different types of
agroforestry systems, so none were predicted and again no sig-
nificant differences were observed (Table 3b). The slightly greater
proportion of non-timber agroforest adopters growing perennial
crops than other adopters (Table 3b) would be expected since
perennial crops constitute a substantial part of non-timber agro-
forest systems. The same, however, could be said for mixed system
adopters given that the proportion managing perennial crops is
the same as those pursuing timber-based systems. A logical reason
for the larger number of cattle sold by timber-based system
adopters (Table 3a) might be that timber-based system adopters
derive less income from the other components of their farming
system and are more reliant on cattle. There is no evidence from
the survey to confirm this possible reason. This difference may be
idiosyncratic to the particular year of the survey, but is notable
nonetheless.
Social participation
The greater social participation of timber-based system adopters
in mutual aid associations might have an ambiguous meaning. On
the one hand it seems counter-intuitive that households planting
trees would require additional non-household labor typically used
during the harvest periods for perennial crops. On the other hand,
timber-based system adopters may have more labor time available
to share with neighbors in mutual aid associations since they do not
seasonally harvest any crops from their agroforest plots; their plots
are long-term investments, with negligible regular labor re-
quirements. Again, there is no evidence from the survey to support
these speculations. Lastly, the higher percent of timber-based sys-
tem adopters reporting utilization of their agroforestry plots may
be a result of their plots potentially attracting a greater diversity of

Table 3b
Binomial socio-economic variable results from 2010 survey by farm and agroforestry plot type groupings.

Annual crops (%) Perennial crops (%) Syndicates (%) Associations (%) Utilization (%)

Control group (n ¼ 39) 46.2 74.4 61.5 46.2

Experimental group (n ¼ 31) 54.8 77.4 77.4 61.3 58.1
Timber (n ¼ 7) 57.1 71.4 85.7 71.4 71.4
Mixed (n ¼ 11) 54.5 72.7 63.6 54.5 54.5
Non-timber (n ¼ 13) 53.8 84.6 84.6 61.5 53.8
Overall sample (n ¼ 70) 50.0 75.7 68.6 52.9

where:
Annual Crops e percentage of sample producing annual/temporary crops.
Perennial Crops e percentage of sample producing perennial/permanent crops.
Syndicates e percentage of sample participating in rural workers syndicates.
Associations e percentage of sample participating in mutual aid associations.
Utilization e percentage of RAPP experimental group reporting utilization of agroforest plots.
 C.E. Blinn et al. / Applied Geography 43 (2013) 171e181 179

Fig. 6. Boxplot of the mean forest similarity index in 2011 of pixels within a 50 m
buffer of an agroforestry plot’s center by agroforestry plot type.

plant and animal populations than either of the other two more
intensively managed agroforest systems.

Impacts of agroforestry plots on forest cover

Fig. 7. Landsat time series of a farmer’s property with the agroforestry plot, which was
installed between 1992 and 1994, location circled. Only the area in and around the
Buffer zones around each agroforest plot were examined to agroforestry plot has not been converted to pasture or other nonforest uses. Images are
ascertain differences in secondary succession associated with plot displayed with Landsat bands 4, 3, and 2 displayed as red, green and blue, respectively.
maintenance. Circular buffers were used for ease of creation, but (For interpretation of the references to colour in this figure legend, the reader is
referred to the web version of this article.)
the buffers corresponded closely to a 3  3, 5  5, or 7  7 pixel
window around the center of each plot. We hypothesized that the cleared all of his/her property at one point in time or another, has
area outside, but in closest proximity to each plot would be more allowed more than one third of his/her property to return to sec-
influenced by the plot type and possibly managed as secondary ondary forest since the agroforestry plot was installed (Fig. 8).
forest. Secondary forest succession did occur in and around the Ethnographic research activities associated with the RAPP (i.e. in-
mixed/timber agroforestry plot types from 1996 to 2001 based on depth interviews with experimental group participants)
the forest similarity index statistics (Fig. 4). In contrast, the non- confirmed some consciousness of the benefits associated with
timber plots stayed fairly constant through 2001, but became less secondary succession and agroforestry: “We should have started
similar to primary forest between 2001 and 2006 and especially planting trees a long time ago. If we had, we would have money
between 2006 and 2011 (Fig. 4). The significant difference in forest
similarity index medians between mixed/timber and non-timber
plot types in 2006 and 2011, as illustrated in the boxplot in Fig. 6
for 2011, supports the hypothesis that agroforestry plot type in-
fluences the likelihood of secondary forest succession in and
around plots, the plot impact areas. Non-timber agroforestry plots
are less likely to result in secondary forest succession in plot impact
areas. Mixed and timber agroforestry plots followed a spectral
trajectory more similar to secondary forest succession between
1996 and 2001 with a leveling off and slight decrease in forest
similarity index between 2001 and 2011 (Fig. 4) presumably as a
result of increased management and/or utilization of the timber
products produced on the plots.

Property level impacts

Fig. 5 shows that properties in both the control and experi-

mental group are on average becoming less spectrally similar to
primary forest through time. The rate of this change has increased
since 2001, especially on properties in the control group, which
were significantly different from the experimental group at the
alpha ¼ 0.05 level in 2011. To visually illustrate the observed
changes, a Landsat time series of images for one farmer in the
experimental group is shown in Fig. 7. This particular farmer
cleared a majority of his/her entire property at one point in time or Fig. 8. Landsat time series of a farmer’s property with the agroforestry plot, which was
another, but has allowed both the agroforestry plot and the sur- installed between 1992 and 1994, location circled. More than one third of the property
has been allowed to return to a forested condition. Images are displayed with Landsat
rounding area to regenerate into secondary forest, which is in stark bands 4, 3, and 2 displayed as red, green and blue, respectively. (For interpretation of
contrast to the rest of the property which was all pasture in 2011 the references to colour in this figure legend, the reader is referred to the web version
(Fig. 7). Another farmer in the experimental group, who also has of this article.)
180 C.E. Blinn et al. / Applied Geography 43 (2013) 171e181
from that right now” (personal communication, author’s trans-
lation) (johnobrowder.wordpress.com).
Conclusions
Despite an overall land cover trend on small farms in the Bra-
zilian state of Rondônia towards less forest cover, the practice of
timber-based agroforestry appears to encourage landowners to
allow secondary forest succession to occur on their lands at least for
the short-term (11e15 years). The type of agroforestry imple-
mented by farmers matters, with more forest succession occurring
on properties with timber or mixed agroforestry plots. Landowners
with larger properties and more people residing on their farms
were more likely in this study to try innovative systems that
included agroforestry. Participants in this agroforestry research
project were also less dependent on cattle (based on 2009 sales),
although timber-based agroforesters appear to rely more on cattle
sales for household income than do non-timber agroforesters. The
experimental group was also more involved in mutual aid associ-
ations than the control group, probably because such participation
enables access to more extensive information networks and labor
sharing support than non-participation.
These findings would support the policy recommendation that
timber-based agroforestry on small family farms holds some
promise as an intermediate degraded tropical forest land-use
rehabilitation strategy. The secondary forest succession that
timber-based agroforestry appears to promote has potentially
synergistic implications for carbon sequestration, wildlife habitat
restoration, and long-term economic benefits for property owners,
who will eventually harvest the mature timber elements of the
agroforest plots, as some apparently have after 10 years. Important
questions regarding the long-term impacts of timber-based agro-
forestry on land cover cannot be answered in a longitudinal study
that extends less than 20 years. The evidence gained at this junc-
ture provides some encouragement for those seeking to rebuild
tropical forests through agroforestry.
Acknowledgements
The authors acknowledge the support of the John and Teresa
Heinz Charitable Trust and the Landsat Science Team (USGS con-
tract number G12PC00073).
References
Alavalapati, J., & Nair, P. K. R. (2001). Socioeconomic and institutional perspectives
of agroforestry. In M. Palo, J. Uusivuori, & G. Mery (Eds.), World forests, markets
and policies 2001 (pp. 71e81). Dordrecht, Netherlands.
Agroforestry. (2011). In The American heritage dictionary of the English language (5th
ed.). Retrieved from http://www.ahdictionary.com/word/search.html?
q¼agroforestry.
Barton, D. (1994). Indigenous agroforestry in Latin America: A blueprint for sustainable
agricultures?. In NRI socio-economic series 6: Chatham, UK: Natural Resources
Institute (NRI).
Bobo, K. S., Waltert, M., Fermon, H., Njokagbor, J., & Mühlenberg, M. (2006). From
forest to farmland: butterfly diversity and habitat associations along a gradient
of forest conversion in southwestern Cameroon. Journal of Insect Conservation,
10(1), 29e42.
Browder, J. O. (1994). Surviving in Rondônia. Studies in Comparative International
Development, 29(3), 45e69.
Browder, J. O., & Godfrey, B. J. (1997). Rainforest cities: Urbanization, development,
and globalization of the Brazilian Amazon. New York: Columbia University Press.
Browder, J. O., & Pedlowski, M. A. (2000). Agroforestry performance on small farms
in Amazonia: findings from the Rondônia agroforestry pilot project. Agroforestry
Systems, 49, 63e83.
Browder, J. O., Pedlowski, M. A., Walker, R., Wynne, R. H., Summers, P. M., Abad, A.,
et al. (2008). Revisiting theories of frontier expansion in the Brazilian Amazon:
a survey of the colonist farming population in Rondônia’s post-frontier, 1992e
2002. World Development, 36(8), 1469e1492.
Browder, J. O., Wynne, R. H., & Pedlowski, M. A. (2005). Agroforestry diffusion and
secondary forest regeneration in the Brazilian Amazon: further findings from
the Rondônia agroforestry pilot project (1992e2002). Agroforestry Systems,
65(2), 99e111.
Budowski, G. (1980). The place of agroforestry in managing tropical forests. In
Paper presented at the international symposium on tropical forests: Utilization and
conservation, New Haven, CT.
Castro, K. L., Sanchez-Azofeifa, G. A., & Rivard, B. (2003). Monitoring secondary
tropical forests using space-borne data: implications for Central America. In-
ternational Journal of Remote Sensing, 24(9), 1853e1894.
Chowdhury, R. R. (2007). Household land management and biodiversity: secondary
succession in a forest-agriculture mosaic in southern Mexico. Ecology and So-
ciety, 12(2). art. 31.
Delaney, M. (1999). Field test of carbon monitoring methods for agroforestry in the
Philippines. In Field tests of carbon monitoring methods in forestry projects (pp.
28e32). Arlington, VA: Winrock International.
D’Antona, Á. O., VanWey, L. K., & Hayashi, C. M. (2006). Property size and land cover
change in the Brazilian Amazon. Population and Environment, 27(5e6), 373e396.
Ehiagbonare, J. E. (2006). Effect of Taungya on regeneration of endemic forest tree
species in Nigeria: Edo State Nigeria as a case study. African Journal of
Biotechnology, 5(18), 1608e1611.
Fearnside, P. M., & Guimarães, W. M. (1996). Carbon uptake by secondary forests in
Brazilian Amazonia. Forest Ecology and Management, 80(1e3), 35e46.
Harvey, C. A., & Haber, W. A. (1998). Remnant trees and the conservation of
biodiversity in Costa Rican pastures. Agroforestry Systems, 44(1), 37e68.
JIRCAS. (2007). Agroforestry approach to the rehabilitation of tropical lands by using
nurse trees. Tsukuba, Japan: Japan International Research Center for Agricultural
Sciences. vii-70.
Lasco, R. D., Guillermo, I. Q., Cruz, R. V. O., Bantayan, N. C., & Pulhin, F. B. (2004).
Carbon stocks assessment of a secondary forest in Mount Makiling Forest
Reserve, Philippines. Journal of Tropical Forest Science, 16(1), 35e45.
Lieberei, R., & Gasparotto, L. (1998). Agroecological profile of plants used as pro-
duction factors and as management components in tropical polyculture sys-
tems. In Proceedings of the third SHIFT-workshop, manaus, March 15e19 (pp.
307e312). Bonn, Germany: BMBF.
Letcher, S. G., & Chazdon, R. L. (2009). Rapid recovery of biomass, species richness,
and species composition in a forest chronosequence in northeastern Costa Rica.
Biotropica, 41(5), 608e617.
Lozada, T., de Koning, G. H. J., Marché, R., Klein, A. M., & Tscharntke, T. (2006). Tree
recovery and seed dispersal by birds: comparing forest, agroforestry and
abandoned agroforestry in coastal Ecuador. Perspectives in Plant Ecology, Evo-
lution and Systematics, 8(3), 131e140.
Masek, J. G., Vermote, E. F., Saleous, N. E., Wolfe, R., Hall, F. G., Huemmrich, K. F., et al.
(2006). A Landsat surface reflectance dataset for North America, 1990e2000.
IEEE Geoscience and Remote Sensing Letters, 3(1), 68e72.
Mercer, D. E., & Miller, R. P. (1998). Socioeconomic research in agroforestry: prog-
ress, prospects, and priorities. Agroforestry Systems, 38(1e3), 177e193.
Meza, A., Sabogal, C., & Jong, W. (2006). Rehabilitation of degraded forest lands in the
Peruvian Amazon: Review of experiences and lessons learned. Bogor, Indonesia:
Center for International Forestry Research.
Minitab Inc. (2012). Minitab (Version 16.2.3) (Software).
Pannell, D., & Vanclay, F. (2011). Changing land management: Adoption of new
practices by rural landholders. Collingwood: CSIRO Publishing.
Pattanayak, S., & Mercer, D. E. (1998). Valuing soil conservation benefits of agro-
forestry: contour hedgerows in the Easter Visayas, Philippines. Agricultural
Economics, 18(1), 31e46.
Perz, S. G. (2005). The effects of household asset endowments on agricultural di-
versity among frontier colonists in the Amazon. Agroforestry Forum, 63(3), 263e
279.
Raman, T. R. S., Mudappa, D., & Kapoor, V. (2009). Restoring rainforest fragments:
survival of mixed-native species seedlings under contrasting site conditions in
the Western Ghats, India. Restoration Ecology, 17(1), 137e147.
Roshetko, J. M., Delaney, M., Hairiah, K., & Purnomosidhi, P. (2002). Carbon stocks in
Indonesian homegarden systems: can smallholder systems be targeted for
increased carbon storage? American Journal of Alternative Agriculture, 17(3),
138e148.
Sirois, M. C., Margolis, H. A., & Camiré, C. (1998). Influence of remnant trees on
nutrients and fallow biomass in slash and burn agroecosystems in Guinea.
Agroforestry Systems, 40(3), 227e246.
Schulze, C. H., Waltert, M., Kessler, P. J. A., Pitopang, R., Shahabuddin, Veddeler, D.,
et al. (2004). Biodiversity indicator groups of tropical land-use systems:
comparing plants, birds, and insects. Ecological Applications, 14(5), 1321e1333.
Schroth, G., D’Angelo, S. A., Teixeira, W. G., Haag, D., & Lieberei, R. (2002). Conver-
sion of secondary forest into agroforestry and monoculture plantations in
Amazonia: consequences for biomass, litter and soil carbon stocks after 7 Years.
Forest Ecology and Management, 163(1e3), 131e150.
Shono, K., Cadaweng, E. A., & Durst, P. B. (2007). Application of assisted natural
regeneration to restore degraded tropical forestlands. Restoration Ecology, 15(4),
620e626.
Smith, N., Dubois, J., Current, D., Lutz, E., & Clement, C. (1998). Agroforestry experi-
ences in the Brazilian Amazon: Constraints and opportunities. Brasilia, Brazil:
World Bank.
Summers, P. M. (2008). The post-frontier land use and social change in the Brazilian
Amazon (1992e2002). Blacksburg, VA: University Libraries, Virginia Polytechnic
Institute and State University.
Summers, P. M., Browder, J. O., & Pedlowski, M. A. (2004). Tropical forest man-
agement and silvicultural practices by small farmers in the Brazilian Amazon:
 C.E. Blinn et al. / Applied Geography 43 (2013) 171e181
recent farm-level evidence from Rondônia. Forest Ecology and Management,
192(2e3), 161e177.
Sydenstricker-Neto, J. (2012). Population and deforestation in the Brazilian
Amazon: a mediating perspective and a mixed-method analysis. Population and
Environment, 34(1), 86e112.
Takimoto, A., Nair, V. D., & Nair, P. K. R. (2009). Contribution of trees to soil carbon
sequestration under agroforestry systems in the West African Sahel. Agrofor-
estry Systems, 76(1), 11e25.
Vieira, D. L. M., Holl, K. D., & Peneireiro, F. M. (2009). Agro-successional restoration as a
strategy to facilitate tropical forest recovery. Restoration Ecology, 17(4), 451e459.
181
Vosti, S. A., Witcover, J., & Carpentier, C. L. (2002). Agricultural intensification by
smallholders: from deforestation to sustainable land use. Research Report 130.
International Food Policy Research Institute.
Warner, K. (1993). Patterns of farmer tree growing in Eastern Africa: A socioeconomic
analysis. Oxford, UK,: Oxford Forestry Institute. Tropical Forestry Papers, 27.
Yokota, Y., Martin, R. A., & Siki, R. (2009). Potentiality of local communities
participation in agroforestry program. JIRCAS Working Report, 60, 94e97.
Wise, R. M., & Cacho, O. J. (2011). A bioeconomic analysis of the potential of
Indonesian agroforests as carbon sinks. Environmental Science & Policy, 14(4),
451e461.