Scholarly article on topic 'A comparison of influences on the landscape of two social-ecological systems'

A comparison of influences on the landscape of two social-ecological systems Academic research paper on "Earth and related environmental sciences"

CC BY-NC-ND
0
0
Share paper
Academic journal
Land Use Policy
Keywords
{"Eastern Panama" / Governance / History / Indigenous / "Land cover" / "Ostrom’s social-ecological systems framework"}

Abstract of research paper on Earth and related environmental sciences, author of scientific article — Divya Sharma, Ignacia Holmes, Gerardo Vergara-Asenjo, William N. Miller, Mitzy Cunampio, et al.

Abstract Case studies of social-ecological landscapes that consider local, spatially explicit land cover changes are necessary for the development of generalised knowledge on deforestation. This study focussed on two indigenous territories of eastern Panama that share the same settlement history, size and location but are perceived by local dwellers to differ in terms of land cover. By considering the territories social-ecological systems made up of Resource Systems, Resource Units, Actors and Governance Structures, following Ostrom’s framework for analysing the sustainability of social-ecological systems (McGinnis and Ostrom, 2014), we sought to determine which social-ecological factors could have led to divergent land cover outcomes to address local leaders’ concerns and inform future land management strategies. We conducted quantitative, spatial analysis using ArcGIS and multivariate statistics from numerical ecology on land cover data from participatory maps, and household level socio-economic data from semi-structured interviews and surveys. Results illustrate that the Resource System’s topography and Actors’ socioeconomics, namely number of people at home and household land ownership, are constraining variables on land cover and help explain divergent forest cover. To reconstruct the influence of history and Governance Structure on the landscapes, we conducted qualitative data collection, namely participatory pebble scoring of historical land cover, interviews with key informants, an archival search, and creation of a participatory historical timeline. Historical governmental timber extraction in the region pre-settlement, guided by topography constraints, may have led to degraded Resource Units (forests) susceptible to clearing. The Governance Structure’s self-organizing, monitoring and networking activities with outside institutions in scientific projects, enabled by Actors’ leadership and social capital, likely encouraged forest conservation in the forest-rich territory. Future land management could therefore benefit from establishment of a local non-governmental organisation to coordinate a communal vision of management and harness external conservation resources. Our findings suggest that inputting both qualitative and quantitative data obtained by participatory methods into Ostrom’s framework can help diagnose territories with divergent landscapes, and thereby inform both forest conservation science and local land management.

Academic research paper on topic "A comparison of influences on the landscape of two social-ecological systems"

Contents lists available at ScienceDirect

Land Use Policy

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

Land Use Policy

A comparison of influences on the landscape of two social-ecological systems

Divya Sharma3'*, Ignacia Holmes3, Gerardo Vergara-Asenjoab, William N. Millerc, Mitzy Cunampiod, Raquel B. Cunampiod, Mara B. Cunampiod, Catherine Potvina e

a McGill University, Department of Biology, 1205 Dr. Penfield Avenue, Montreal, QCH3A 1B1, Canada b Forest Research Institute, Fundo Teja Norte s/n, Valdivia, Chile

c McGill University, School of Environment, 3534 University St., Montreal, QC H3A 2A7, Canada d Community of Piriati-Emberd, Panama

e Smithsonian Tropical Research Institute (STRI), Apartado 0843-03092, Balboa, Ancon, Panama City, Panama

ABSTRACT

Case studies of social-ecological landscapes that consider local, spatially explicit land cover changes are necessary for the development of generalised knowledge on deforestation. This study focussed on two indigenous territories of eastern Panama that share the same settlement history, size and location but are perceived by local dwellers to differ in terms of land cover. By considering the territories social-ecological systems made up of Resource Systems, Resource Units, Actors and Governance Structures, following Ostrom's framework for analysing the sustainability of social-ecological systems (McGinnis and Ostrom, 2014), we sought to determine which social-ecological factors could have led to divergent land cover outcomes to address local leaders' concerns and inform future land management strategies. We conducted quantitative, spatial analysis using ArcGIS and multivariate statistics from numerical ecology on land cover data from participatory maps, and household level socio-economic data from semi-structured interviews and surveys. Results illustrate that the Resource System's topography and Actors' socioeco-nomics, namely number of people at home and household land ownership, are constraining variables on land cover and help explain divergent forest cover. To reconstruct the influence of history and Governance Structure on the landscapes, we conducted qualitative data collection, namely participatory pebble scoring of historical land cover, interviews with key informants, an archival search, and creation of a participatory historical timeline. Historical governmental timber extraction in the region pre-settlement, guided by topography constraints, may have led to degraded Resource Units (forests) susceptible to clearing. The Governance Structure's self-organizing, monitoring and networking activities with outside institutions in scientific projects, enabled by Actors' leadership and social capital, likely encouraged forest conservation in the forest-rich territory. Future land management could therefore benefit from establishment of a local non-governmental organisation to coordinate a communal vision of management and harness external conservation resources. Our findings suggest that inputting both qualitative and quantitative data obtained by participatory methods into Ostrom's framework can help diagnose territories with divergent landscapes, and thereby inform both forest conservation science and local land management.

© 2016 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND

license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

CrossMark

ARTICLE INFO

Article history:

Received 5 May 2015

Received in revised form 4June 2016

Accepted 15 June 2016

Keywords: Eastern Panama Governance History Indigenous Land cover

Ostrom's social-ecological systems framework

1. Introduction

Global deforestation is a well-documented phenomenon, with 13 million hectares of forest estimated as lost yearly between 2000

and 2010 worldwide (Food and Agriculture Organization, 2010). In

Central America and the Caribbean, satellite imagery and literature analysis show that 1.4% of forest cover was lost between 2000 and 2005 (Asner et al., 2009). A 'conversion of land cover and its effects' model for the neotropics predicted that Central America would be a hotspot of deforestation in 2010 (Wassenaar et al., 2007). These forests are also home to local people; in Latin America and the Caribbean, about 40 million Indigenous peoples live in forests (World Bank, 2004), and Indigenous peoples in Latin America own

* Corresponding author. E-mail address: divya.sharma2@mail.mcgill.ca (D. Sharma).

http://dx.doi.org/10.1016lj.landusepol.2016.06.018

0264-8377/© 2016 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/ 4.0/).

Fig. 1. Location of the two territories studied, Piriati- and Ipeti-Embera, off of the Pan-American Highway in eastern Panama.

almost half of the 9.1% of global forests that are community-owned (van Dam, 2011).

Landscapes are formed from a composite of local changes in land cover (Lambin et al., 2003) and can be viewed as social-ecological systems in which particular landscapes emerge from an array of environmental and human interactions. Thus case studies that consider fine-grained nuances on land cover change are necessary for the development of generalised knowledge on deforested landscapes (Lambin et al., 2001; Chazdon et al., 2009; Ellis et al., 2010).

Furthermore, there is a need to relate ecological and social factors to the landscape in a spatially explicit manner (Field et al., 2003; Pijanowski et al., 2009; Shkaruba and Kireyeu, 2013).

This paper adopts a spatially explicit approach to understanding factors that influence forest cover in Panamanian social-ecological systems. According to national reports and remote sensing, Panama had 44% forest cover in 2010 and, between 2005 and 2010, lost 0.36% of its land cover annually to deforestation (Food and Agriculture Organization, 2010). Eastern Panama, home to the

Embera, Wounaan and Guna Indigenous peoples where this study took place, was characterised by Wassenaar et al. (2007) as mostly forest with scattered forest loss to pasture and cropland. The conventional view of indigenous communities as forest stewards was recently supported in Panama, as Vergara-Asenjo and Potvin (2014) showed that indigenous territories, including claimed lands without legal title, and protected areas were more effective at avoiding deforestation than other tenure regimes between 1992 and 2008, and 2000 and 2008. Despite sharing the same ecoregion, recent settlement history in a frontier zone, cultural background, bounded property system and distance to market, two neighbouring Embera territories, namely Piriati- and Ipeti-Embera in the Choco-Darien moist forest of eastern Panama, are perceived by their inhabitants as disparate landscapes. While Ipeti maintains forest, deforestation is a focal concern of local leaders in Piriati, leading them to commission DS and CP to verify and determine the reasons for this disparity. This request provided a unique opportunity to identify combinations of social-ecological factors that drive deforestation.

We believe that Ostrom (2009) provides a useful framework for understanding the factors that determine the sustainability, or lack thereof, of social-ecological systems, by dividing the larger system into the Resource System, Resource Unit, Governance, and Actor subsystems. These systems interact with the biophysical environment and broader political, economic and social setting to inform focal action situations (interactions between actors leading to particular outcomes) (McGinnis and Ostrom, 2014).

We therefore compared the two territories by combining both quantitative and qualitative data, using Ostrom's framework, to examine how incremental shifts in the balance of these social-ecological factors can lead to land use changes. To do so we used quantitative GIS analysis of participatory and remotely sensed open source data from Global Forest Watch (Hansen et al., 2013), as well as qualitative participatory data. First, we determined whether the two territories in eastern Panama do indeed have divergent landscapes in terms of current forest cover and recent deforestation rates. We then identified which social-ecological, historical and governance factors that shape the landscapes, namely the Resource System's topography, Actor household socioeconomics, Resource Unit's historical quality and the Governance Structure, taken from Ostrom's framework, were different between the communities and could explain divergent forest cover. We then discussed the subtlety of factors at play on the landscape and their importance on differential forest cover and loss.

From a series of papers building upon each other, the framework emerging from Ostrom's work is a tiered, diagnostic scheme to study the links and interactions in complex social-ecological systems (Anderies et al., 2004; Ostrom, 2007, 2009; Ostrom and Cox, 2010; McGinnis and Ostrom, 2014). It attempts to structurally organise case studies in mutually comprehensible ways that could allow for meta-analyses and cumulative knowledge building, while ensuring that important factors are not ignored (McGinnis and Ostrom, 2014). Furthermore, it is a framework that gives equal weight to social and ecological system components (Binder et al., 2013). For example, biophysical characteristics of the Resource System, such as topography, can be key constraining variables on local land use decision-making. Six per cent of the cases examined in a meta-analysis identified topography and slope as "predisposing" explanatory factors of deforestation in Latin America; land characteristics such as these corresponded to deforestation via enabling shifting cultivation, frontier colonisation and conversion to pasture (Geist and Lambin, 2002). In an analysis of forest cover change in the Chorotega region of Costa Rica, secondary forest tended to be in areas of high slope and low soil depth and fertility, where pasture establishment would not be viable (Arroyo-Mora et al., 2005). Elevation and slope were most significantly related to deforestation in a study of land use/cover change in an indigenous coffee-growing

5" % Pasture

■ % Cropland

I I % Tall Fallow %F6rest

4-j I / / , " . .

-.©.....

-1 -I-1-1-1-1-1-*-1-'-1-'-

0 1 2 3 4 5

Fig. 2. Linear discriminant analysis illustrating the coefficients of percent land cover at the individual plot level that maximise discrimination of the territories of Piriati-(1) and Ipeti-Embera (2). Each point represents each row of data. The circle is the 95% confidence interval forthe mean of each territory. The more similarthe territory, the greater the overlap between the circles. The perpendicular projection of the forest vector onto the x-axis illustrates that forest cover was the most significant predictor of a plot belonging to Ipeti (2).

region in Mexico; pasture and agricultural lands tended to be located in areas of lower elevation (Ellis et al., 2010).

Actors' household level socioeconomics, and not just the influence of total population, are also potential drivers of land use change (Entwisle and Stern, 2005; Sydenstricker-Neto, 2012). Socioeconomic factors that have influenced land use change in rural Latin America include: age of household head (Abizaid and Coomes, 2004; Potvin et al., 2006; Gray et al., 2008), time since settlement (Potvin et al., 2006; Gray et al., 2008), household/hired/male labour force (Walker et al., 2000; Abizaid and Coomes, 2004; Mena et al., 2006; Gray et al., 2008; Sloan, 2008; Sydenstricker-Neto, 2012), education (Carr, 2005; Mena et al., 2006; Gray et al., 2008), ethnicity, previous land owned and off-farm employment (Carr, 2005).

Landscapes can further be shaped by Governance Structures, for example the presence of institutions like non-governmental organisations (NGOs) that, in their ability to change resource use patterns, can alter human-environment relationships (Bebbington, 2004). Either by reinforcing community forest governance or by directly engaging in forest conservation activities, NGOs can be involved in discouraging deforestation (Wright and Andersson, 2012). For example, involvement in community organisations was associated with a greater area under cultivation in a study across five indigenous populations in the Ecuadorian Amazon (Gray et al., 2008). A comparison of two communities in the Los Tuxtlas Biosphere Reserve in Mexico showed that the community that had contact

Table 1

First- and second-tier variables of a social-ecological system (Source: McGinnis and Ostrom, 2014) compared between Piriati- and Ipeti-Embera. Plot- and household-level variables were compared to land cover at the level of the individual plot. Landscape-level variables were compared across the territories. Households for which both factors and plot-level land cover data were available were analysed, accounting for different sample sizes. The codes are those given by Ostrom to distinguish each variable and the category in which it falls, shown here to allow comparison across studies.

First-tier variable

Second-tier variable (Code) Spatial level analysed

Factor

Source

Sample size

Statistical Test

Resource System

Resource Units

Actors

Governance System

Storage characteristics (RS8)

Number of units (RU5)

Temporal distribution (RU7)

Spatial distribution

Landscape

Landscape

Landscape

Landscape and plot

Number of relevant actors (A1)

Socioeconomic attributes (A2)

Landscape

Household

History or past experiences Landscape (A3)

Government (GS1) and non-governmental organisations (NGOs; GS2)

Landscape

Elevation and slope

Percent land cover in 2012

Land cover changes since settlement per decade Deforestation from 2004-2012

Maximum, minimum and average elevation and slope

Maximum, minimum and average distance to highway Maximum, minimum and average distance to river

Maximum, minimum and average distance to village Population growth per decade from 1980-2010 Plot size

Number of people available to help Year of household establishment Number of people at home Number of children at home Number of elders at home Age of eldest

Whether livestock owned (Y or N) Whether land owned (Y or N) Education level of household head (none, primary, secondary or post-secondary)

Place of origin of household head (community, the Bayano watershed or the Darién province/Colombia) Historical timber extraction

Digital Elevation Model (SRTM 90m) downloaded from the CGIAR Consortium for Spatial Information website.

Participatory maps from Piriati (Sharma et al., 2015) and Ipeti (Vergara-Asenjo et al., 2015) Participatory pebble scoring activities

Global Forest Watch (Source: Hansen/UMD/Google/USGS/NASA) Digital Elevation Model (SRTM 90m) downloaded from the CGIAR Consortium for Spatial Information website. Average per land cover class extracted using zonal statistics in ArcGIS 10.1 (see Appendix A)

Nearest distance of each land cover polygon to highway from 2012 participatory maps in ArcGIS 10.1

Contraloría General de la República de Panamá (2014)

2012 participatory maps (Sharma et al., 2015; Vergara-Asenjo et al., 2015)

Piriati: 35 semi-structured interviews with household heads, wives of household heads, land inheritors and youth conducted in

2013 (Sharma et al., 2015); Ipeti: 36 household surveys in 2009 (Raynaud and Shinbrot, 2009) as a follow-up to surveys by (Tschakert et al., 2007).

Participatory historical timeline; interviews with key informants

(nPiriati = 5; nIpeti = 2;

^government = 12); archival analysis Participatory historical timeline

nPiriatí = 47; nIpetí = 73

nPiriatí = 47; nIpetí = 73

nPiriatí nIpetí =

=47; 73

nPiriatí = 47; nIpetí = 73

nPiriatí = 13; nIpetí = 17

nPiriatí = 8; nIpetí = 18

Linear discriminant analysis

T-test

Correlation

Correlation

T-test Correlation

Correlation

Canonical

correspondence

analysis

Fig. 3. Topographic map of Piriati- and Ipeti-Embera overlaid with deforestation data between 2004 and 2012 from the Global Forest Watch (Source: Hansen/UMD/Google/USGS/NASA).

with outside NGOs and government agencies took greater responsibility for deforestation and demonstrated more concern about conservation (Durand and Lazos, 2008). In an analysis of the effect of NGOs on deforestation in 200 rural communities in Bolivia, the presence of more NGOs that were viewed as "important" was associated with lower rates of deforestation, but not to the presence of community forest institutions. The result suggests that NGOs do prevent deforestation in the country but not via enhancing community's forest governance capacity through the creation of institutions (Wright and Andersson, 2012). These cases illustrate the potential for incremental shifts in the balance of diverse social-ecological factors to result in land use changes.

The comparison between two territories in Panama using Ostrom's framework identifies these nuances that have led to disparate landscapes but also ways forward for future land management. This participatory project stemmed from local leaders' concern that land cover did not correspond to the landscape they wanted. Ostrom's framework elucidates the control knobs of the system that can be adjusted to redirect future decision-making. In particular, we found that a potential pathway for reforestation exists via enhancement of the Governance Structure, namely the creation of a local NGO that creates a communal vision of and funnels resources for forest conservation. Local leaders in the forest-poor territory should therefore be reassured that it was not simply household-level decision-making that led to disparate landscapes, and that there is room to form alliances to move the landscape in the trajectory they desire.

2. Methods & results

2.1. Study sites

This study focussed on two indigenous territories of eastern Panama, Piriatí- and Ipetí-Emberá, that share the same settlement

history, size and location but differ in terms of land cover. They are located in the Bayano watershed of Panama province, ~100 and ~120km east of Panama City, respectively (Fig. 1). Piriati's lands comprise 3867 ha, and Ipeti's 3205 ha. Both territories were formed following displacement of inhabitants along the Bayano River due to construction of a hydroelectric dam in the early 1970s (Potvin et al., 2006). Both were granted legally recognised communal land rights in 2014 and 2015, respectively. The territories are mostly inhabited by indigenous Embera, originally from the Darien province that shares an eastern border with Colombia. Shifting cultivation and cattle ranching are practised on plots of land managed by individual households. Ethical certificates for this study were obtained from the McGill University Research Ethics Board and the Instituto Nacional de Cultura (INAC) in Panama.

2.2. Comparison of land cover and deforestation between the territories

The first step in our analysis was to establish differences between the territories in terms of current land cover. Land cover came from 2012 participatory maps of Piriati (Sharma et al., 2015) and Ipeti (Vergara-Asenjo et al., 2015) created by landowners at the plot level. Aggregating to the level of the territory showed an overall forest cover of 10.7% and 42.5%, respectively. As land cover generally results from land use decisions taken at the individual household level, we conducted a linear discriminant analysis (LDA) using per cent forest, tall fallow, cropland and pasture cover, at the level of the individual plot in both communities, as explanatory variables to further understand differences in land cover between the communities. All households for which land cover data were available were used. In Piriati a concentration of plots in lands recently allotted to new families were too small to be drawn on the map and therefore excluded. This LDA and all subsequent statistical analyses were carried out on RStudio (version 0.98.484),

Table 2

Continuous social-ecological factors correlated (r >0.400 in bold) to percent land cover at the level of the individual plot in territories ofboth Piriati- and Ipeti-Embera in 2012, using the Pearson product-moment correlation.

Factor % Forest % Tall Fallow % Cropland % Pasture

Maximum elevation 0.694 -0.062 -0.315 -0.419

Minimum elevation 0.619 -0.138 -0.326 -0.407

Average elevation 0.702 -0.072 -0.336 -0.436

Maximum slope 0.670 -0.053 -0.339 -0.381

Average slope 0.649 -0.017 -0.371 -0.415

Maximum distance to highway 0.574 -0.031 -0.387 -0.393

Minimum distance to highway 0.687 0.085 -0.418 -0.484

Average distance to highway 0.671 0.065 -0.434 -0.465

# People at home -0.321 0.007 0.511 -0.238

# Children at home 0.050 0.334 0.167 -0.458

unless otherwise stated. The LDAbiplot was produced inJMP 11.0.0. Land cover at the level of the individual plot allowed discrimination between the two territories with 74.0% correct classification with individual plots in Piriati having lower forest cover than in Ipeti (Fig. 2). Forest cover was the most significant predictor of community grouping followed by pasture, with correlations to the discriminant function of 0.911 and -0.720 respectively, compared to 0.0762 and 0.454 for tall fallow and cropland (Fig. 2).

We then examined whether deforestation rates differ between the two territories. For Piriati, we had access to the participatory map made in 2012. In Ipeti, however, we were able to estimate deforestation by comparing forest cover from participatory maps created in 2004 (Potvin et al., 2006) and 2012. We therefore decided to also estimate annual deforestation in both communities from data made available by Global Forest Watch (GFW, 2014; Hansen et al., 2013) between 2004 and 2012 using ArcGIS 10.1 (see Appendix A for more details on spatial methods). GFW uses Landsat imagery at 30 m resolution, where trees are vegetation of greater than 5 m in height, and deforestation is a stand-replacement disturbance or total loss of tree cover in a pixel (Hansen et al., 2013). For Ipeti, laying the GFW data over the participatory maps yielded similar estimates of amount of deforestation: total deforestation of forest and tall fallow between 2004 and 2012 was estimated to be around 299 ha by the participatory maps, while GFW data estimated 240 ha of deforestation (see Appendix A for comparison of deforestation estimates by GFW and participatory mapping). GFW estimated 468 ha of deforestation in Piriati between 2004 and 2012. According to GFW data, the two territories' deforestation rates between 2004 and 2012 differ significantly (t(8) = 2.60, p = 0.0316; Fig. 3), with 12.1% and 7.49% total deforestation, and 1.34% and 0.833% average annual deforestation, for Piriati and Ipeti respectively.

2.3. Identification of social-ecological factors influencing land cover at the household and plot level

Next, we identified social-ecological factors influencing land cover and categorised them according to Ostrom's four first-tier variables: Resource Systems; Resource Units; Actors; and Governance Structure (Table 1). In theory, divergent land cover outcomes could emerge from differences between any of these variables. We therefore used quantitative data to characterise two Resource System factors, fifteen Resource Unit factors, and 11 Actor factors in both territories, at the level of individual plot, household, or landscape (Table 1). Qualitative information was then used to characterise Actors' history and the Governance Structure, presented in Sections 2.5 and 2.6.

At the landscape level, we compared average elevation and slope as key Resource System characteristics, and distance of land cover category to highway, village and river as factors relevant to Resource Units' spatial distribution. In Ipeti forest tended to be

located in areas of high elevation and slope (Fig. 4). In both territories forest was also, on average, located furthest from the village and furthest from the Highway compared to other land cover (Fig. 5).

At the plot and household level, we used Pearson product-moment correlations generated on JMP 11.0.0 to identify the continuous socioeconomic Actor factors and spatial distribution of Resource Unit factors that were most related to land cover. Per cent forest, tall fallow, cropland and pasture cover at the level of the individual plot were used as dependent variables. In both territories, number of people and children at home, elevation, slope and distance to highway were the continuous factors most correlated to land cover (Table 2). Households that had more people tended to have more cropland in their plots, and those that had more children at home tended to have less pasture, when using a moderate correlation of r> 0.400. Those household plots that had a higher elevation, slope and distance to highway tended to have more forest and less pasture, when using a stronger correlation (r>0.600). Although at the landscape level forest tended to be located further from villages, at the level of the individual plot the distance from plot to village was not strongly related to its land cover - i.e. its amount of forest. It should be noted that the strong positive relationship of nearest distance to highway on forest cover in Ipeti (0.581) is likely confounded by topography, since remote areas of the territories are on the foothill of a mountain ridge. Indeed, in Ipeti, nearest distance of plot to highway was positively correlated to average elevation (0.774) unlike in Piriati (-0.017). In Piriati, where land is flat, land cover was not strongly correlated to nearest distance to highway (0.368, 0.167, -0.313 and -0.087 for percent forest, tall fallow, cropland and pasture, respectively).

We used a canonical correspondence analysis (CCA) to identify the categorical socioeconomic Actor factors that were most related to land cover.The CCA shows that land and livestock ownership and education level of household head were those factors most related to land cover in both territories (Fig. 6). Those who own land tended to have a greater percentage of tall fallow than those who have either sold or rented their plot of land. Those who own their own livestock and have a greater level of education tended to have more forest. The two canonical axes of the CCA between the categorical explanatory factors and the per cent land cover response variables explained a cumulative proportion of variance of 95.1%; the first axis explained 60.8%.

2.4. Identification of social-ecological factors at the household and plot level that differ between the territories

Explanatory factors related to land cover at the individual plot level were used in a Linear Discriminant Analysis (LDA) to identify the variables that differentiate individual household plots from the two territories (using a subset of households for which both land cover and household characteristics were available: nPiriati = 9; nIpeti = 19; Table 3). Those continuous factors with a correlation to

Fig. 4. Landscape level comparison of average elevation, average slope and total area per land cover class in the territories of Piriati- and Ipeti-Embera in 2012.

Fig. 5. Landscape level comparison of average distance of each land cover type to village, river and highway in Piriati- and Ipeti-Embera in 2012.

land cover of r >0.400, excluding distance to highway due to its correlation with elevation, and three categorical factors most related to land cover were included, for a total of 10 factors (Table 3). Variables were centred and standardised before running the LDA.

With 60.7% correct classification, the LDA showed that number of people per household was the most significant, though only a moderately strong, predictor of membership to Piriati. Elevation, slope and ownership of land were the most significant predictors of membership to Ipeti (Table 3). That is, Piriati had more people per household, and Ipeti had a greater elevation, slope and proportion of landowners who own their land.1 It should be noted that

the inability to include newly allotted lands in Piriati in the analysis entailed the exclusion of some relatively young households. While number of people at home differed between the territories, population rate change per decade according to Contraloria (2013) between 1980 and 2010 did not; t(2) = 0.667, p = 0.574. Thus, recalling Ostrom's framework, differences in Actor socioeconomics and Resource Units' spatial distribution apparently explain land cover differences in 2012 among the two territories.

2.5. Identification of historical factors influencing land cover and deforestation

1 Although in theory collective lands such as those of Ipeti and Piriati cannot be sold, some families have indeed "sold" their land, albeit without legal titles.

Given that differences between Actors explain divergent land cover, we conducted a participatory pebble scoring activity in 2013 to document the effect of history and past experiences, acknowl-

io _ %'Pasture i.

- O ; • ^

O I Bayano ■ *t\

/ Darién/Colombia

o Ifts*. *........................................

m o Community %'Cropland^^^ ' \ xLivestocî^ %'TaN'Fallow" i \ \ • : i

A"""" 0 ■ Education %'Forest

!2 !1 0 1 2 CCA1

Fig. 6. Biplot of canonical correspondence analysis (CCA) between categorical social-ecological factors and per cent land cover dependent variables at the level of the individual plot in Piriati- and Ipeti-Embera in 2012. The axes are linear combinations of the explanatory social-ecological variables. The vectors illustrate the degree to which the explanatory variables account for the variation in the response matrix, i.e. land cover (Borcard et al., 2011). A perpendicular line from the tip of the vector to each axis shows its relative importance on the canonical axes. Explanatory variables are in italics, while percent land cover is in bold. The dots represent the "site" (household) scores of each row of data (Piriati is represented by white circles; Ipeti by black circles).

Table 3

Correlations of social-ecological factors, related to land cover from 2012 participatory maps, to the discriminant function (those most correlated in bold). The discriminant function is a linear combination of -0.146 xPeople at Home-0.403 xChildren at Home + 0.190 x Livestock - 0.132 X Land - 0.195 x Education + 0.155 x Max. Elevation-0.396 x Min. Elevation + 2.10 x Avg. Elevation +1.26 x Max. Slope - 1.98 x Avg. Slope.

Category Factor Correlation

Resource Units Maximum elevation 0.860

Minimum elevation 0.680

Average elevation 0.826

Maximum slope 0.888

Average slope 0.812

Actors # People at home -0.438

# Children at home -0.186

Livestock 0.345

Land 0.401

Education -0.081

edging their legacy on social-ecological landscapes (Coomes et al., 2011). We modelled the activity after a similar process carried out in Ipeti by Potvin et al. (2006). The deputy chief selected 8-12 participants with knowledge of the territories, including middle-aged to elderly men and women. Participants were asked to divide 20 pebbles between each land cover category for each decade from the 1970s until the 2010s to represent the composition of the territories over time.

Participants in pebble scoring perceived that both Piriati and Ipeti had similar levels of forest at the time of settlement until the 1980s, after which Piriati experienced a greater decline in forest (Table 4). Participants alleged that Piriati's forests had always been secondary due to timber extraction by the Panamanian government pre-settlement. They claimed that the drop in forest cover in the 1990s was partly due to allotment of previously ownerless lands by local leaders to new, landless families. These new landowners immediately began deforesting to cultivate for subsistence and sale of surplus. The further decrease in forest cover in the 2000s in Piriati, and subsequent high proportion of pasture, were attributed to a regional politician renting multiple parcels of land and converting forest and fallow to pasture for cattle ranching.

Table 4

Perceived per cent forest cover in the territories according to participatory pebble scoring activities in Piriati and in Ipeti (Ipeti's data from). Forest includes agro-forestry in Ipeti, as agroforestry was not used as a separate category in Piriati.

Decade Piriati Ipeti

1970s 80 86.3

1980s 70 70

1990s 35 50

2000s 15 43.3

2010s 15 -

The former finding prompted us to interview key informants in both communities (nPiriati = 5; nIpeti = 2), to understand if historical timber extraction left a legacy on differential forest cover in the two territories. We also conducted an archival search of historical timber extraction in the region and unstructured interviews with 12 key informants employed in governmental logging in the 1970s and 1980s, using a snowball sampling approach.

Interviews and archival analysis confirm the claims of pebble scoring participants: Piriati's forests were selectively harvested before the community settled on the territory. Interviewees from the community reported the presence of a logging court in the early 1970s where the villagers first settled in the western territories of Piriati, known as Parti. The court belonged to what became known as the governmental Corporation for the Integrated Development of the Bayano (hereafter the Bayano Corporation). The camp allegedly employed 20-100 workers who logged downriver and upriver (southward) until the terrain became too hilly. Villagers decided to relocate to the current village site of Piriati, at the time primary forest, due to the presence of peasant (campesino) farmers on Parti lands. At this point the Bayano Corporation agreed to clear the new community lands for the villagers. Once they returned to the new village to settle, community members observed a cut path from Parti leading to a courtyard of abandoned espavé (Anac-ardium excelsum) logs, suggesting that the Corporation had since engaged in selective logging. Meanwhile, villagers stated that the Corporation did not extract timber south of the Pan-American Highway due to its hilly terrain. Instead the operation moved eastward towards Ipeti, where timber extraction was also limited as the Corporation arrived post-settlement and extraction was possible

only in flat lands. Interviewees stated that, in the early 1980s, one independent logging operation was given a 2000-ha concession for three years in Piriatí, but by this point the Corporation had already extracted the valuable timber. Two teams of two locals were contracted by the group to harvest 300 trees per year, with additional pay for every extra tree harvested. An estimated 4200 trees were cut over the three years in Piriatí in the post-Corporation period. These claims were supported by interviews with former employees involved in timber extraction, who stated that the Corporation had been harvesting cativo (Prioira copaifera) and espavé along the Bayano River between approximately 1967 and 1973. After clear-cutting the reservoir and selling valuable timber species to fund dam construction, the Corporation began to log selectively eastward. Selective logging in Piriatí was of fine wood like Spanish cedar (Cedrela odorata), cedro espino (Pachira quinata), mahogany (Swietenia macrophylla) and oak (Tabebuia pentaphylla) and, less preferentially, espavé. There was at least one other logging camp in the eastern lands of Piriatí and logging occurred both east and west of these territories. South of the Highway, extraction was limited due to lack of valuable species. These interviewees further alleged that timber extraction was boosted by construction of the Bayano Bridge in the mid-1970s.

Archival research supports interviewees' assertions of timber extraction in the territories. A documentary from the time explicitly states that extraction was occurring in Piriatí (GECU, 1974). In eastern Panama bulldozers clearing logging roads, trucks and felled trees were observed east of the territories in the early to mid-1970s (Webb, 2008). Selective logging in eastern Panama in the mid-1970s was reportedly vital to funding the Bayano Corporation (Corporación Bayano, 1982). A study by an international organisation contracted to inform a management plan for the watershed reported that the Corporation was still harvesting wood in the Bayano watershed between 1980 and 1989 (Louis Berger, 1999). By the mid-1980s, thirty-nine independent logging operations with concessions of over 100,000 ha were reportedly running in eastern Panama (Rojas, 1985).

Thus, given the evidence for selective logging in the Bayano and, particularly, in Piriatí, we hypothesise that the communities did not have perfectly equivalent landscapes at the time of settlement circa 1980. Historical timber extraction appears to have occurred preferentially in Piriatí, possibly due to being closer to the hydroelectric dam being constructed in the Bayano and due to topographical constraints in Ipetí.

2.6. Identification of governance factors influencing land cover and deforestation

In order to historicise the presence of NGOs and determine whether the Governance Structure could account for divergence in the social-ecological systems, we created a participatory timeline in 2013 of the major events that occurred in Piriati since re-settlement, following methods outlined in Geilfus (2002). The deputy chief in Piriati chose 12-16 male and female community members across a range of ages to participate in a workshop, also made open to anyone willing to participate. Participants met in the communal meeting area of the village and were separated into four working groups according to gender and age. Each group generated its own timeline of events they considered important to the community's development and history, and presented it to the others, followed by a group discussion. Individual events were then compiled to create a comprehensive timeline based on events found in multiple timelines and/or events consensually deemed important by participants. Equivalent events in Ipeti were added to the timeline by gathering information from documents of the ongoing relationship between the Smithsonian Tropical Research Institute

(STRI), McGill and Ipetí, and through discussions with a key informant.

Historical events recalled by Piriatí's participants fell into three categories: territorial disputes; ecological disasters like crop diseases and floods; and infrastructure, development and reforestation projects by government, NGOs, scientists and churches (Fig. 7). The timeline illustrates how Piriatí and Ipetí share similar histories of land use conflict and ecological incidents, but Ipetí has had a stronger history of local governance and collaboration with outsiders. In 1994, Ipetí established a local registered NGO, OUD-CIE, which has been involved in a series of projects with external donors and collaborators, including co-author CP, since 1996. OUD-CIE enabled the community to network with these organisations; the NGO obtained small grants from the Global Environmental Facility (Holmes, 2016) and another from the Cervecería Nacional under a corporate social responsibility programme. After a decade of scientific research in Ipetí, the relationship with CP led to sale of carbon under a voluntary carbon project and agroforestry reforestation with STRI in 2008. Meanwhile in Piriatí, where no such local NGO yet exists, reforestation by government and NGOs (specifically, Global Brigades) has been at a small scale, focused on people rather than landscape-level land use, and collaboration with scientists began in 2013.

3. Discussion

The present case study illustrates, using Ostrom's framework for analysing social-ecological systems, the number and subtlety of social and ecological factors that accumulate to result in disparate landscapes. The comparison therefore underscores the relevance of a combination of both quantitative and qualitative participatory data to fully comprehend the range of factors leading to perceived differences in land cover. In eastern Panama, a storage characteristic of the Resource System, namely topography, has dictated differential forest loss in part, as have the socioeconomics of Actor households, the historical quality of the Resource Unit, in terms of inheritance of degraded forest, and the Governance Structure that determines whether the community is able to harness external resources for forest conservation (Fig. 8).

Quantitative analysis showed that harvesting by actors leading to differential forest loss was influenced by storage characteristics of the Resource System (topography) and subsequent spatial distribution of Resource Units - landowners leaving forest in areas of high slope and elevation (Fig. 8). Deforestation rates according to GFW have decreased in both territories in recent years; according to community members in Piriati, this is because remaining forests are in inaccessible areas. Similarly, in indigenous Mexico parcels of land further from the road were not actively used, i.e. were in fallow (Ellis et al., 2010). This latter finding conforms to the expected outcome of tropical deforestation - that rates will decrease as remaining forest becomes less accessible (Myers, 1993 and Rudel and Roper, 1996, as cited by Geist and Lambin, 2002). In both territories forest tended to be located in areas more remote from the villages and the Highway, congruent with a previous report that forests were in hilly areas in Ipeti (Kirby and Potvin, 2007). The analysis therefore suggests that partial divergence in the landscapes can be attributed to differences in the biophysical context of the territories, rather than disparities in decision-making.

The comparison also suggests that Actors' socioeconomic attributes inform harvesting by actors leading to differential forest loss (Fig. 8). Quantitative analysis suggested that deforestation is influenced by households' use of agriculture as a means to pursue a livelihood strategy shaped by the presence of more dependents. Piriati had more people per household than Ipeti, which tended to be associated with a greater proportion of land devoted to crop-

Fig. 7. Participatory historical timeline of Piriati- and Ipeti-Embera.

Fig. 8. The social-ecological systems framework (adapted from McGinnis and Ostrom, 2014) that illustrates those social and ecological factors found in this study to account for disparate forest cover in two indigenous territories of eastern Panama. The codes in brackets are taken from Ostrom's framework, shown here to allow for easy comparison across studies. Arrows originating from the Systems go to specific Focal Action Situations.

land. In this community, cultivation is a livelihood strategy pursued largely to enable subsistence (Sharma et al., 2015). Economic needs have previously been shown to incentivise forest clearing in an analysis of deforestation and reforestation in the frontier context of eastern Panama, where deforestation increased with presence of additional household labourers and forest cover (Sloan, 2008). In the study of land use/cover change in Mexico, population pressure was associated with the presence of pasture and agriculture, possibly for household consumption (Ellis et al., 2010). In rural Latin America, increased household size has encouraged deforesta-

tion by necessitating conversion to cropland for consumption or sale of surplus (Carr, 2005; Mena et al., 2006). Population pressure and associated subsistence needs are therefore confirmed here as drivers of differential forest clearing.

Beyond biophysical and socioeconomic influences, historical human land use activity can leave its imprint on current landscapes, thus underlining the importance of qualitative analysis (Rhemtulla and Mladenoff, 2007; Gray et al., 2008; Sloan, 2008; Rhemtulla et al., 2009; Moran, 2010). In the territories, a combination of three historical events emerging from qualitative analysis can help explain

differences in forest cover and loss. Firstly, the inherited quality of the Resource Unit likely influenced the perceived value of maintaining Resource Units (i.e. forests) (Fig. 8). Inheriting already degraded forests in Piriati due to historical timber extraction by the government pre-settlement may have incentivised forest clearing compared to in Ipeti, where forests were reportedly primary at the time of establishment (Potvin et al., 2006). If the pattern documented in the 1990s also prevailed in the 70s and 80s, we would expect historical timber extraction to have been greater closer to the market, which, at that time, would have been in Piri-ati (Simmons, 1997). Forty per cent of the Latin American cases in a meta-analysis of tropical deforestation showed that commercial timber extraction, including selective logging, combined with proximate drivers like shifting cultivation resulted in deforestation (Geist and Lambin, 2001). In the Brazilian Amazon, areas that were within 5-25 km of a main road and that underwent selective logging were up to four times more likely to be deforested in the subsequent four years than unlogged areas (Asner et al., 2006). Logging can leave forests vulnerable to droughts and forest fires (Asner et al., 2006; Matricardi et al., 2010) but subsequent forest clearing may also be motivated by the loss of perceived value of forests through depletion of culturally valuable tree species, as purportedly occurred in Piriati (Sharma et al., 2015). Though not explicitly stated in the context of Ostrom's framework, it has been similarly shown that historical land endowments influence future forest cover in bounded, rural territories that practise shifting cultivation; land use decisions and therefore current landscapes were constrained by a household's initial land type and amount, and forest type and age (i.e. its inherited Resource Units) in northeastern Peru (Coomes et al., 2011). The relative loss of forest in this present study was in part attributable to differential land endowment - location on flat areas conducive to forest clearing, but also inheritance of land that was already degraded due to being located closer to a centre of timber extraction - which ultimately influenced subsequent decisions to deforest.

Secondly, qualitative analysis of historical events, illustrating harvesting by local landowners leading to forest loss, supports the aforementioned influence of subsistence needs derived from quantitative analysis. As shown in participatory pebble scoring, local leaders' decisions to distribute unused lands, taken based on population pressure and subsistence needs, have influenced the landscape we see today. A similar transition to that in Piriati has been anticipated in Ipeti (Potvin et al., 2006).

Thirdly, current forest cover has been shaped by the external political setting that enabled a local politician (as well as the national government) to harvest, clear forest and degrade the land in the past (Fig. 8); forest loss was driven by agents of deforestation at different spatial and power scales. The finding that there are fewer landowners in Piriati who still own their land, and the negative relationship of land ownership to proportion of pasture both support pebble scorers' claim that pasture conversion on rented lands by a local politician exerting political power partially accounts for the disparate landscapes. The result is a comparable situation to that observed among indigenous coffee-growers in Mexico: few landowners own the majority of pasture and local community members who have sold or rented lands look after these landowners' cattle (Ellis et al., 2010). Apparently, incremental shifts in context accumulated to account for differences in forest cover in the territories.

In eastern Panama, the Governance Structure's self-organizing, monitoring and networking activities, enabled by Actors' leadership and social capital, likely also influenced the forest cover we see today (Fig. 8). Specifically, past collaboration between outside and local institutions in carbon projects and later agroforestry may help explain differences in forest cover. Scientific collaboration with Ipeti was begun at a time of strong local leaders who

had an apparent motivation to self-organise, network with external agents and ameliorate community members' quality of life. The local NGO, OUDCIE, was the point of collaboration for these scientific projects. Prior relationships inform the places that NGOs seek to establish projects (Bebbington, 2004); this initial scientific relationship enabled subsequent collaboration in reforestation and agroforestry with external NGOs in Ipetí. Thus the presence of strong conservation-oriented institutions, i.e. norms/social capital and leadership, in Ipetí may have incited the relative decrease in deforestation in the decade in which forest cover began to diverge from Piriatí. Indeed, community members (particularly women) in Piriatí, where there is no such active NGO, previously alleged that weak internal laws and lack of social organisation in the community - and thus lack of a cohesive, communal vision of land management - were factors that have led to forest loss in their territory (Sharma et al., 2015). Participants in pebble scoring in Piriatí anticipated that the arrival of external NGOs in the next decade would entail shifts from land devoted to tall fallow to agroforestry. In Mexico it was shown that community institutions could effectively manage forests when social capital was developed over time and when those in power had the desire to do so; collective use rules were facilitated by a communal vision of land management (Merino Pérez, 2004). Here, the continued presence of scientific projects in Ipetí, facilitated by their leaders' vision, appears to have helped perpetuate its culture of pro-conservation attitudes and harness external resources for conservation purposes.

4. Conclusions

This case study was initiated due to local indigenous leaders' concern over forest loss in their territory relative to that of their neighbours. Inputting both qualitative and quantitative participatory data into Ostrom's framework to compare the two territories illustrates that the landscapes' divergence is an outcome of differences in the Resource Systems and spatial distribution of Resource Units, i.e. topography, and in Actors' socioeconomics, i.e. people per household and land ownership. Furthermore, land cover was influenced by a series of historical events that determined past experiences of Actors and past landscapes, like harvesting activities by powerful external agents - namely, a local politician who razed rented lands for cattle ranching and a government corporation that felled valuable timber pre-settlement - leading to loss and degradation of Resource Units against which locals were largely powerless. Finally, disparate Governance Structures that influence long-term collaboration with NGOs and scientists and their feedback with local conservation-oriented institutions can also account for differential ability to maintain forest.

Prior to this study, local leaders faulted themselves for forest loss. Our results suggest however that current land cover is largely the result of forces outside the leaders' control. The comparison suggests, in a context of predisposition to forest clearing due to location on hilly, degraded lands, population pressure and subsistence needs, and the presence of powerful external agents, local leaders can steer the landscape trajectory by promoting self-governance, via establishment of a community NGO that mobilises social capital, promotes communal conservation attitudes and programmes, and networks with external conservation organisations. Women may have a distinct role to play in the creation of such an NGO, given their acute awareness of the influence of social cohesion and a communal vision on land use (Sharma et al., 2015). We thus see this study as a way to empower the local leaders, who, rather than blaming themselves for forest loss, can recognise the influence of historical endowment on forest cover as well as future pathways for potential reforestation.

Acknowledgements

This research was funded by the Natural Sciences and Engineering Resource Council of Canada (NSERC), the Fonds de recherche du Québec - Nature et technologies (FQRNT), and the Biology Department of McGill University. The broader project within which this research is embedded is funded by the Margaret A. Cargill Foundation. We thank our manuscript's anonymous reviewer for pushing us to increase the clarity of our arguments.

Appendix A. Elevation, slope and nearest distance calculations

Supplemental information on spatial analysis methods, operated with ESRI ArcGIS 10.1.

Elevation, slope and nearest distance calculations

The Digital Elevation Model (Digital Elevation Database SRTM 90m v4.1) was downloaded from the CGIAR Consortium for Spatial Information website (http://www.cgiar-csi.org/data/srtm-90m-digital-elevation-database-v4-1), which was clipped to the polygons of the territories using the Clip Raster tool. The Project Raster tool was used to convert to WGS 1984 UTM Zone 17N. The slope (per cent rise) was generated using Spatial Analyst, and the Zonal Statistics as Table tool was used to calculate the average slope and elevation per land cover class. To obtain the elevation at the level of the pixel and compare maximum, minimum and average elevation at the level of each household parcel, the DEM was converted from raster to polygon. It was then spatially joined to the participatory map file and the Zonal Statistics as Table tool was used to obtain elevation values per household. To obtain the slope at the level of the pixel, the Raster Calculator tool was used to obtain integer values of slope. Then the Raster to Polygon tool was used to enable a Spatial Join to the participatory map, and slope values per household were obtained using the Zonal Statistics as Table tool. To calculate the nearest distance of each land cover patch to the river, village and highway, we used the Near (Proximity) tool. Village, highway and river shapefiles were obtained from participatory maps of 2012, with the exception of the rivers to the east and west of the Piriati territories, which we sketched using the World Imagery Basemap.

Global Forest Watch calculations

Global Forest Watch data for Panama were obtained from the website www.globalforestwatch.org (Hansen et al., 2013). This raster contains all pixels that were deforested between 2001 and 2012, with the pixel value being the year the forest loss occurred. We used the Identity tool to identify GFW deforestation features in the participatory maps, and convert the data to the WGS 1984 UTM Zone 17N coordinate system. We then added an area field and calculated the area of each land cover patch in the participatory maps, using Calculate Geometry, and summarised the total area of each patch according to landowner. We used the Dissolve tool to remove the divisions of land covers within a single landowner, added an area field and calculated the total area of the parcel per landowner using Calculate Geometry. We then spatially joined the participatory map's attribute table to the GFW deforestation table to obtain the total area deforested per landowner. The Multipart to Singlepart tool was used to re-separate land cover patches within each landowner's parcel, and then we added the field "Patch No.". We then calculated the area of each patch using Calculate Geometry, and exported the table as a database file. We then used the Identity tool to identify GFW deforestation features in the multipart file and added the field "Patch No.Code", which combined the

patch number with the GFW gridcode (i.e. year of deforestation). An area field was added and the geometry was calculated and summarised per patch number to obtain area deforested per land cover patch per owner. The table was exported as a database file and joined to "Patch No.Code". An area field was added and the field calculator was used to calculate the per cent deforestation per land cover patch. This was done for both territories.

Comparing GFW deforestation to participatory mapping deforestation

GFW deforestation data was clipped to the Ipeti participatory map shapefile, and the Select Attributes tool was used to select for deforestation on or after 2005 (gridcode >5). The participatory maps of 2004 and 2012 were both converted from polygon to raster, with land cover as the value field. To generate a column of land cover change, we reclassified the 2004 map, where we multiplied by a value of 10 to the gridcode representing the land cover class (1 = forest; 2 = tall fallow; 3 = short fallow; 4 = plantation; 5 = pasture; therefore 1 became 10, etc.). We then used the Raster Calculator to add these new values to the comparable land cover gridcode values for 2012. Thus, a value of 14 meant that the patch was forest (1) in 2004 and plantation (4) in 2012. Therefore, grid-codes of 12, 13, 14 and 15 represented deforestation, and so we converted from raster back to polygon, and selected for deforestation by attribute (12 or 13 or 14 or 15 or 23 or 24 or 25 to include loss of tall fallow as deforestation).

To remove patches of deforestation deemed unable to be perceived by landowners (corresponding to single pixels of deforestation data), GFW deforestation data were dissolved based on location, then re-separated using the Multipart to Singlepart tool. We then added an area field, selected by attribute for area <0.1 ha. Patches of deforestation <0.1 ha were excluded from the analysis, as this was the minimum scale at which change in land cover was perceived in participatory mapping. Thus single pixels of deforestation (0.09 ha) were excluded. We switched the selection and extracted the file as a new shapefile of GFW deforestation to generate the final map comparing deforestation according to different sources. To identify the smallest area of change perceived by participants in mapping, we dissolved the participatory deforestation in Ipeti between 2004 and 2012 according to geographic location, then used the Multipart to Singlepart tool to separate the isolated patches of deforestation. We then added an area field and the minimum change perceived by mappers was 0.1 ha. (The smallest patches drawn by owners in the 2012 participatory map were at least 0.0275 ha.)

In order to generate a map of areas considered deforested by GFW but not by the participatory maps, we used the Identity tool with the shapefile of GFW and participatory deforestation between 2004 and 2012 as the input and the identity feature as the 2012 participatory map. We then removed the patches deforested according to the participatory map by selecting from the GFW data by location the patches that contain data within participatory deforestation patches. We then switched the selection in the attribute table to select the GFW patches that are outside patches deforested according to participatory mapping and exported the shapefile. We added the area field and summarised to calculate the total area per land cover class for which the two data sources were incompatible.

While aggregate deforestation estimated by GFW and by participatory maps agreed, the two methods were incongruent in terms of precise location of deforestation in the territories (Fig. A1). Those areas that were considered deforested according to GFW but not participatory mapping tended to be perceived as forested areas by community members (~1/2; 71.8 ha), while one third corre-

Fig.A1. Visual comparison of deforestation in the communal lands of Ipeti-Embera according to Global Forest Watch satellite data (Source: Hansen/UMD/Google/USGS/NASA) and participatory mapping (2004—Potvin et al., 2006; 2012—Vergara-Asenjo et al., 2015).

sponded to tall fallow (52.7 ha), and one sixth to pastured lands (24.2 ha).

Selection of ground-truthing sites

To better understand the discrepancies between GFW and participatory deforestation, local GPS (Global Positioning System) technicians ground-truthed 17 sites in 2014. Sites for ground technicians to visit were chosen based on being (a) deforested according to GFW but not according to participatory mapping; (b) patches >1.5 ha and (c) either forest (n = 10) or tall fallow (n = 7) according to 2012 mapping. These sites were chosen because they were relatively homogeneous patches of deforestation according to GFW, and therefore deforestation (or lack thereof) would be evident on the ground. In order to select these sites, we dissolved the land cover field of the shapefile containing non-overlapping deforestation from both sources. We then selected by attribute for patches

>1.5 ha and where the land cover in 2012 was forest or tall fallow. Sites considered forested by participatory mapping were chosen since ground-truthers would not be able to confirm deforestation, but only reject it. We then used the Feature to Point tool and added the XY coordinates and joined the data to the original participatory map shapefile to obtain landowner data and facilitate ground-truthing for technicians.

Twelve of 17 sites that were considered deforested by GFW between 2004 and 2012 were either forest or tall fallow in 2014 and, therefore, are unlikely to have been non-forested in 2012. Of all the points visited, 14 were characterised by mixed land cover in the immediate vicinity or surrounding hectare, perhaps accounting for the relative inaccuracy of satellite data at the local scale. Eight of the 12 points that were verified as forest or tall fallow by ground-truthers were also forest or tall fallow in the recent past, according to local technicians, suggesting they are unlikely to have been deforested in 2012 and then reforested by 2014. Four of these

17 sites were deforested between 2004 and 2007 and so could theoretically have been deforested and then reforested by 2012, but not considered deforested by a comparison of the 2004 and 2012 participatory maps. Out of the five points that could have been correctly classified as deforested by GFW (i.e. that were not forest or tall fallow in 2014), four had either forest or tall fallow in the hectare surrounding the GPS point visited and three out of the five were forest or tall fallow in the recent past. Recognising the limited sample size, these 17 points were incorporated into a dataset of 173 points by Vergara & Potvin, submitted, to validate GFW in the Bayano watershed at the regional level and draw conclusions from the validation.

GFW's reforestation data were not included in the present study as they are not annual data, but rather represent overall reforestation between 2001 and 2012.

References

Abizaid, C., Coomes, O.T., 2004. Land cover and forest fallowing dynamics in

seasonally dry tropical forests of the southern Yucatán Peninsula, Mexico. Land Cover Policy 21 (1), 71-84, http://dx.doi.org/10.1016/j.landusepol.2003.06.001.

Anderies, J.M., Janssen, M.A., Ostrom, E., 2004. A framework to analyze the

robustness of social-ecological systems from an institutional perspective. Ecol. Soc. 9(1), 18.

Arroyo-Mora, J.P., Sánchez-Azofeifa, G.A., Rivard, B., Calvo, J.C., Janzen, D.H., 2005. Dynamics in landscape structure and composition for the Chorotega region, Costa Rica from 1960 to 2000. Agric. Ecosyst. Environ. 106, 27-39, http://dx. doi.org/10.1016/j.agee.2004.07.002.

Asner, G.P., Broadbent, E.N., Oliveira, P.J.C., Keller, M., Knapp, D.E., Silva, J.N.M., 2006. Condition and fate of logged forests in the Brazilian Amazon. Proc. Natl. Acad. Sci. 103 (34), 12947-12950, http://dx.doi.org/10.1073/pnas.0604093103.

Asner, G.P., Rudel, T.K., Aide, T.M., Defries, R., Emerson, R., 2009. A contemporary assessment of change in humid tropical forests. Conserv. Biol. 23,1386-1395, http://dx.doi.org/10.1111/j.1523-1739.2009.01333.x.

Bebbington, A., 2004. NGOs and uneven development: geographies of

development intervention. Prog. Hum. Geogr. 28, 725-745, http://dx.doi.org/ 10.1191/0309132504ph516oa.

Binder, C.R., Hinkel, J., Bots, P.W.G., Pahl-Wostl, C., 2013. Comparison of frameworks for analyzing social-ecological systems. Ecol. Soc. 18 (4), 26, http://dx.doi.org/10.5751/ES-05551-180426.

Borcard, D., Gillet, F., Legendre, P., 2011. Numerical Ecology with R. Springer, New York, New York, USA.

Carr, D.L., 2005. Forest clearing among farm households in the Maya Biosphere Reserve. Prof. Geogr. 57,157-168, http://dx.doi.org/10.1111/j.0033-0124.2005. 00469.x.

Chazdon, R.L., Harvey, C.A., Komar, O., Griffith, D.M., Ferguson, B.G.,

Martinez-Ramos, M., Morales, H., Nigh, R., Soto-Pinto, L., van Breugel, M., Philpott, S.M., 2009. Beyond reserves: a research agenda for conserving biodiversity in human-modified tropical landscapes. Biotropica 41,142-153, http://dx.doi.org/10.1111/j.1744-7429.2008.00471.x.

Contraloría General de la República de Panamá (Contraloría), 2014. [online] URL: http://www.contraloria.gob.pa/lNEC/.

Coomes, O., Takasaki, Y., Rhemtulla,J.M., 2011. Land-use povertytraps identified in shifting cultivation systems shape long-term tropical forest cover. Proc. Natl. Acad. Sci. U. S. A. 108 (34), 13925-13930, http://dx.doi.org/10.1073/pnas. 1012973108/.

Contraloría General de la República de Panamá (Contraloría). 2013. [online] URL: http://www.contraloria.gob.pa/lNEC/.

Corporación Bayano, 1982. Prospecto para el Desarollo. Direccion de Evaluacion y Fiscalizacion Tecnica. Chepo, Panama.

Durand, L., Lazos, E., 2008. The local perception of tropical deforestation and its relation to conservation policies in Los Tuxtlas Biosphere Reserve, Mexico. Hum. Ecol. 36, 383-394, http://dx.doi.org/10.1007/s10745-008-9172-7.

Ellis, E.A., Baerenklau, K.A., Marcos-Martínez, R., Chávez, E., 2010. Land cover/land cover change dynamics and drivers in a low-grade marginal coffee growing region of Veracruz, Mexico. Agrofor. Syst. 80,61-84, http://dx.doi.org/10.1007/ s10457-010-9339-2.

Entwisle, B., Stern, P.C., 2005. Population, Land Cover and Environment: Research Directions. National Research Council, Washington DC.

Field, D.R., Voss, P.R., Kuczenski, T.K., Hammer, R.B., Radeloff, V.C., 2003. Reaffirming social landscape analysis in landscape ecology: a conceptual framework. Soc. Nat. Resour. 16,349-361, http://dx.doi.org/10.1080/ 08941920390178900.

Food and Agriculture Organization (FAO), 2010. Global Forest Resources

Assessment. Food and Agricultural Organization of the United Nations, Rome, ltaly.

Geilfus, F., 2002. 80 herramientas para el desarrollo participativo: diagnóstico, planificación, monitoreo, evaluación. Instituto Interamericano de Cooperación para la Agricultura (IICA), San José, Costa Rica.

Geist, H.J., Lambin, E.F., 2001. What Drives Tropical Deforestation? A Meta-analysis of Proximate and Underlying Causes of Deforestation based on Subnational Case Study Evidence. LUCC Report Series No. 4. LUCC International Project Office, Louvain-la-Neuve, Belgium.

Geist, H.J., Lambin, E.F., 2002. Proximate causes and underlying driving forces of tropical deforestation. BioScience 52 (2), 143-150, http://dx.doi.org/10.1641/ 0006-3568(2002)052[0143:PCAUDF]2.0.C0;2.

Global Forest Watch (GFW), 2014. Global Forest Watch. World Resources Institute, Washington DC, USA http://www.globalforestwatch.org/about/video.

Gray, C.L., Bilsborrow, R.E., Bremner, J.L., Lu, F., 2008. Indigenous land cover in the Ecuadorian Amazon: a cross-cultural and multilevel analysis. Hum. Ecol. 36, 97-109, http://dx.doi.org/10.1007/s10745-007-9141-6.

Grupo experimental del cine universitario (GECU), 1974. Project for the Integrated Development of the Bayano. University of Panama, Panama.

Hansen, M.C., Potapov, P.V., Moore, R., Hancher, M., Turubanova, S.A., Tyukavina, A., Thau, D., Stehman, S.V., Goetz, S.J., Loveland, T.R., Kommareddy, A., Egorov, A., Chini, L., Justice, C.O., Townshend, J.R.G., 2013. High-resolution global maps of 21st-century forest cover change. Science 342, 850-853, http://dx.doi.org/ 10.1126/science.1244693, Data available on-line from: http:// earthenginepartners.appspot.com/science-2013-global-forest.

Holmes, I., 2016. Restoring Carbon Stocks While Addressing Local Livelihoods: Opportunities and Challenges of the Global Climate Change Regime. PhD Thesis. McGill University.

Kirby, K.R., Potvin, C., 2007. Variation in carbon storage among tree species: implications for the management of a small-scale carbon sink project. For. Ecol. Manage. 246, 208-221, http://dx.doi.org/10.1016/j.foreco.2007.03.072.

Lambin, E.F., Turner, B.L., Geist, H.J., Agbola, S.B., Angelsen, A., Bruce, J.W., Coomes, O.T., Dirzo, R., Fischer, G., Folke, C., George, P.S., Homewood, K., Imbernon, J., Leemans, R., Lin, X., Moran, E.F., Mortimore, M., Ramakrishnan, P.S., Richards, J.F., Skánes, H., Steffen, W., Stone, G.D., Svedin, U., Veldkamp, T.A., Vogel, C., Xuy, J., 2001. The causes of land-use and land-cover change: moving beyond the myths. Glob. Environ. Change 11, 261-269, http://dx.doi.org/10.1016/ S0959-3780(01)00007-3.

Lambin, E.F., Geist, H.J., Lepers, E., 2003. Dynamics of land-use and land-cover change in tropical regions. Annu. Rev. Environ. Resour. 28, 205-241, http://dx. doi.org/10.1146/annurev.energy.28.050302.105459.

Louis Berger, 1999. Manejo Integral de la Cuenca del Río Bayano, Subcuenca del Río Majé y Áreas Adyacentes al Embalse. Consorcio Louis Berger International, Inc., Delca Consultores, S.A. Panama City, Panama.

Matricardi, Skole, D.L., Pedlowski, M.A., Chomentowski, W., Claudio Fernandes, L., 2010. Assessment of tropical forest degradation by selective logging and fire using Landsat imagery. Remote Sens. Environ. 114,1117-1129, http://dx.doi. org/10.1016/j.rse.2010.01.001.

McGinnis, M.D., Ostrom, E., 2014. Social-ecological system framework: initial changes and continuing challenges. Ecol. Soc. 19 (2), 30, http://dx.doi.org/10. 5751/ES-06387-190230.

Mena, C.F., Bilsborrow, R.E., McClain, M.E., 2006. Socioeconomic drivers of deforestation in the Northern Ecuadorian Amazon. Environ. Manage. 37 (6), 802-815, http://dx.doi.org/10.1007/s00267-003-0230-z.

Merino Pérez, L., 2004. Conservación O Deterioro: El Impacto De Las Políticas Públicas En Las Instituciones Comunitarias Y En Las Prácticas De Uso De Los Recursos Forestales. Instituto Nacional de Ecología, Mexico.

Moran, E.F., 2010. Environmental Social Science: Human-environment Interactions and Sustainability. Wiley-Blackwell, Chichester, West Sussex, UK.

Myers, N., 1993. Tropical forests: the main deforestation fronts. Environ. Conserv. 20, 9-16, http://dx.doi.org/10.1017/S0376892900037176.

Ostrom, E., Cox, M., 2010. Moving beyond panaceas: a multi-tiered diagnostic approach for social-ecological analysis. Environ. Conserv. 37 (04), 451-463, http://dx.doi.org/10.1017/S0376892910000834.

Ostrom, E., 2007. A diagnostic approach for going beyond panaceas. Proc. Natl. Acad. Sci. U. S. A. 104 (39), 15181-15187, http://dx.doi.org/10.1073/pnas. 0702288104.

Ostrom, E., 2009. A general framework for analyzing sustainability of

social-ecological systems. Science 325 (5939), 419-422, http://dx.doi.org/10. 1126/science.1172133.

Pijanowski, B.C., Iverson, L.R., Drew, C.A., Bulley, H.N.N., Rhemtulla, J.M., Wimberly, M.C., Bartsch, A., Peng, J., 2009. Addressing the interplay of poverty and the ecology oflandscapes: a grand challenge topic for landscape ecologists? Landsc. Ecol. 25,5-16, http://dx.doi.org/10.1007/s10980-009-9415-z.

Potvin, C., Tschakert, P., Lebel, F., Kirby, K., Barrios, H., Bocariza, J., Caisamo, J., Caisamo, L., Cansari, C., Casamá, J., Casamá, M., Chamorra, L., Dumasa, N., Goldenberg, S., Guainora, V., Hayes, P., Moore, T., Ruíz, J., 2006. A participatory approach to the establishment of a baseline scenario for a reforestation Clean Development Mechanism project. Mitig. Adapt. Strateg. Glob. Change 12, 1341-1362, http://dx.doi.org/10.1007/s11027-006-9056-3.

Raynaud, J., Shinbrot, X., 2009. Socio-economic Windows of Opportunity in Ipetí-Emberá, Panama. ENVR 451 Internship. McGill University, Montreal.

Rhemtulla, J.M., Mladenoff, D.J., 2007. Why history matters in landscape ecology. Landsc. Ecol. 22,1-3, http://dx.doi.org/10.1007/s10980-007-9163-x.

Rhemtulla, J.M., Mladenoff, D.J., Clayton, M.K., 2009. Historical forest baselines reveal potential for continued carbon sequestration. Proc. Natl. Acad. Sci. U. S. A. 106, 6082-6087, http://dx.doi.org/10.1073/pnas.0810076106.

Rojas, J., 1985. La explotación forestal en la region oriental de Panamá. In: González, S.H.M.J.E. (Ed.), Agonia de la Naturaleza. IDIAP/STRI, Panama.

Rudel, T., Roper, J., 1996. Regional patterns and historical trends in tropical deforestation, 1976-1990: a qualitative comparative analysis. Ambio 25, 160-166.

Sharma, D., Vergara-Asenjo, G., Cunampio, M., Cunampio, R.B., Cunampio, M.B., Potvin, C., 2015. Genesis of an indigenous social-ecological landscape in eastern Panama. Ecol. Soc. 20 (4), 37, http://dx.doi.org/10.5751/ES-07897-200437.

Shkaruba, A., Kireyeu, V., 2013. Recognising ecological and institutional landscapes in adaptive governance of natural resources. For. Policy Econ. 36, 87-97, http:// dx.doi.org/10.1016/j.forpol.2012.10.004.

Simmons, C.S., 1997. Forest management practices in the Bayano region of Panama: cultural variations. World Dev. 25, 989-1000, http://dx.doi.org/10. 1016/S0305-750X(97)00002-8.

Sloan, S., 2008. Reforestation amidst deforestation: simultaneity and succession. Glob. Environ. Change 18, 425-441, http://dx.doi.org/10.1016/j.gloenvcha. 2008.04.009.

Sydenstricker-Neto, J., 2012. Population and deforestation in the Brazilian Amazon: a mediating perspective and a mixed-method analysis. Popul. Environ. 34 (1), 86-112, http://dx.doi.org/10.1007/s11111-012-0173-5.

Tschakert, P., Coomes, O.T., Potvin, C., 2007. Indigenous livelihoods, slash-and-burn agriculture, and carbon stocks in Eastern Panama. Ecol. Econ. 60, 807-820, http://dx.doi.org/10.1016/j.ecolecon.2006.02.001.

van Dam, C., 2011. Indigenous territories and REDD in latin america: opportunity orthreat? Forests 2,394-414, http://dx.doi.org/10.3390/f2010394.

Vergara-Asenjo, G., Potvin, C., 2014. Forest protection and tenure status: the key role of indigenous peoples and protected areas in Panama. Glob. Environ. Change 28, 205-215, http://dx.doi.org/10.1016/j.gloenvcha.2014.07.002.

Vergara-Asenjo, G., Sharma, D., Potvin, C., 2015. Engaging stakeholders: assessing accuracy of participatory mapping of land cover in Panama. Conserv. Lett. 8 (6), 432-439, http://dx.doi.org/10.1111/conl.12161.

Walker, R., Moran, E., Anselin, L., 2000. Deforestation and cattle ranching in the Brazilian Amazon: external capital and household processes. World Dev. 28 (4), 683-699, http://dx.doi.org/10.1016/S0305-750X(99)00149-7.

Wassenaar, T., Gerber, P., Verburg, P.H., Rosales, M., Ibrahim, M., Steinfeld, H., 2007. Projecting land cover changes in the Neotropics: the geography of pasture expansion into forest. Glob. Environ. Change 17, 86-104, http://dx.doi.org/10. 1016/j.gloenvcha.2006.03.007.

Webb, R.L., 2008. Men, Mud and Motorcycles: Conquering 200 Miles ofJungle by Motorcycle: Panama to Columbia through the Darien Gap. South Carolina.

World Bank, 2004. Sustaining Forests: A Development Strategy. Washington DC.

Wright, G., Andersson, K., 2012. Non-governmental organizations, rural communities and forests: a comparative analysis of community-NGO interactions. Small Scale For. 12,33-50, http://dx.doi.org/10.1007/s11842-012-9206-2.