Carbon, biodiversity, and livelihoods in forest commons: synergies, trade-offs, and implications for REDD+

We used proxies for three key sustainable forest management outcomes—biodiversity conservation, carbon storage, and livelihoods benefits (variable names in italics; table 1). As a proxy for carbon storage, we used a measure of forest biomass (the aggregate basal area of all trees >10 cm DBH per hectare, Chhatre and Agrawal 2009). As a proxy for biodiversity conservation, we used a measure of tree species richness (calculated using the Chao-1 estimator of tree species diversity, for trees >10 cm DBH, Persha et al 2011, using the R package {vegan}, Oksanen et al 2013). Finally, as a proxy for livelihoods benefits, we used a composite measure of the extent to which the forest provided subsistence and commercial livelihoods to the forest user (measured as the locally defined proportion of households in each user group that depends significantly on the forest for subsistence and commercial livelihoods, Chhatre and Agrawal 2009). To construct this measure we calculated the mean of the proportion of households depending on forest for subsistence and the proportion of those depending on forests for commercial livelihoods. Higher values of carbon and biodiversity are likely desirable from a conservation perspective. It is less clear whether higher values of livelihoods are desirable. On the one hand, greater dependency on the forest could indicate a forest in good condition that provides benefit flows to local communities. Alternatively, forest dependency could indicate higher levels of poverty. As such, in this paper we report the associations between the livelihoods outcome and our set of independent variables, but remain agnostic about the desirability of increasing or decreasing forest dependence.

We selected independent variables identified in several key earlier studies as being associated with variations in socio-ecological outcomes in community forests (e.g. Chhatre and Agrawal 2009, Persha et al 2011, table 1). These variables include features of forest and communities in terms of their (a) biophysical characteristics (forest size, in hectares); (b) institutional arrangements (legal designation as community forest (yes) or as government, private, or protected area forest (no), and the degree of forest-user rule-making autonomy, defined as yes if 'about right' and no otherwise); and (c) demographic and socioeconomic characteristics of forest user-groups (user-group size, measured as the number of individuals; ethnic diversity, calculated using the Simpson's Index; and collective action, calculated as an index of the extent of cooperation between forest users, and their investment in forest commons (table 1) (see supporting information).

We used an information theoretic and model averaging approach to statistical model building, to avoid the pitfalls of conventional stepwise regression analyses (which can inflate Type I error rates) and potential model selection uncertainty associated with the use of single statistical models (Burnham and Anderson 2004, Whittingham et al 2006). We constructed a series of multiple linear regression models in the R statistical environment using R version 3.1.3 (R core development team 2014), modeling carbon, biodiversity, and livelihoods as a function of our independent variables. Non-normally distributed residuals for the carbon and biodiversity models were corrected using a square root transformation of the dependent variables. We were not able to correct the heavy-tailed distribution of our residuals for the livelihoods model with a transformation, so we instead used a robust linear regression model, which accounts for the effect of outliers, using the 'rlm' function in the {MASS} package (Ripley et al 2015). For each of our dependent variables, we first ran all possible model combinations of independent variables using the 'dredge' function in the {MuMIn} package (Bartón 2015). We then used the 'model.avg' function in the {MuMIn} package to calculate model-averaged regression coefficients for all models with Δ AIC < 1. Finally, we calculated the model-averaged partial R² (for the linear regressions) and D² (for the robust linear regressions) to obtain relative estimates of the explanatory power of each variable (Guisan and Zimmermann 2000).

Ethnic diversity, legal designation, and collective action were all retained as predictors in our best fitting models (model averaged R² = 0.178, table 1). Of these, ethnic diversity had the strongest association with carbon (model averaged partial R² = 0.161). Forest user groups with lower ethnic diversity and collective action, and forests that were legally designated as community forests, were associated with higher carbon (figure 1(A), table 1).

Zoom In Zoom Out Reset image size

User group size, forest size, ethnic diversity, and legal designation were all retained as predictors in our best fitting models (model averaged partial R² = 0.122, table 1). Of these, user group size had the strongest association with biodiversity (model averaged partial R² = 0.074). Forest user groups that were smaller and that had lower ethnic diversity, and forests that were larger and not legally designated as community forests, were associated with higher biodiversity (figure 1(B), table 1).

User group size, collective action, legal designation, rule-making autonomy, and ethnic diversity were all retained as predictors in our best fitting models (model averaged partial D² = 0.315, table 1). Of these, user group size had the strongest association with livelihoods (model averaged partial D² = 0.208). Forest user groups that were smaller, and which had higher collective action, greater ethnic diversity, and greater rule-making autonomy, and forests that were not legally designated as community forests, were associated with higher livelihood outcomes.

Carbon was positively and significantly associated with biodiversity (r = 0.317, P = 0.026). Neither carbon (r = −0.15, P = 0.301) nor biodiversity (r = 0.06, P = 0.680) were significantly associated with livelihoods.

Each of the three sustainable forest management outcomes was associated with three to five of the independent variables used in our analyses (table 1; figure 1). Greater tree biomass (our proxy for carbon) was positively associated with legally-designated community forests, but was negatively associated with higher collective action and higher ethnic diversity. Greater tree species richness (our proxy for biodiversity) was positively associated with larger forests and higher ethnic diversity, and negatively associated with larger user-groups and legally-designated community forests. Higher subsistence and commercial benefit flows from forests to forest communities (our proxy for livelihoods) were positively associated with larger user-groups, higher collective action, higher ethnic diversity, and greater rule-making autonomy, and negatively associated with legally-designated community forests. Many of these variables have been shown to be important by previous studies of community forest management outcomes, including studies using IFRI data (e.g. Agrawal and Angelsen 2009, Persha et al 2011). A novel contribution of this study is to examine the relationships between a common set of independent variables and multiple separate forest management outcomes. Many previous studies either examine only a single outcome (e.g. Nagendra 2002) or combine multiple outcomes into a single categorical variable (e.g. Chhatre and Agrawal 2009).

Each outcome was associated with a different set of independent variables. If this is also the case elsewhere, this suggests that there is a need for policy-makers to clearly define their desired outcomes, and to design and target interventions accordingly. Forest size was significantly associated with biodiversity, but not with carbon or livelihoods. Similarly, rule-making autonomy was only associated with the livelihoods outcome. Such variation has evident implications for REDD+ management and policy. For example, community forest management might be used as a mechanism to achieve positive REDD+ objectives, either by building on the natural, human and institutional capital associated with existing community forest sites, or by expanding the network of community forests into sites where REDD+ objectives are a priority, such as in high carbon-stock biomes (Balooni and Lund 2014, Newton et al 2015). In Nepal, such high-carbon areas include the Terai, though there has historically been political resistance to the establishment of community forest sites in this region (Bampton et al 2007). Site selection for such initiatives might be guided in part by the associations between the selected variables and forest management outcomes. However, variation in these associations implies that, depending on the priority accorded by different decision makers to particular outcomes, the criteria applied in different sites are likely to vary. Factors that are consistently associated with improved outcomes in more than one dimension may be more attractive as a screening or selection tool: for example, smaller forest user-groups in our analysis were associated with both higher biodiversity and higher livelihood outcomes and thus could, in these sites, serve as a basis for seeking improved outcomes in both dimensions (if higher livelihood outcomes are deemed to be desirable) (table 1).

Our analyses describe the associations between multiple independent variables and three forest management outcomes of interest. The strength of these relationships was pronounced in some cases—for example, between user-group size and livelihoods (model averaged R² = 0.208). In other cases, the relationships were relatively weak—for example, between ethnic diversity and livelihoods (model averaged R² = 0.009). Here, we discuss ways in which the importance of these relationships can be understood and could be further explored, both in this context and elsewhere.

First, our analysis points to potential causal pathways that connect the independent variables and forest management outcomes. We selected the variables used in our analyses because they have proved to be relevant to community forest management in other contexts (Agrawal and Angelson 2009, Chhatre and Agrawal 2009, Persha et al 2011), and it is possible to conceive of logical mechanistic pathways through which these variables and outcomes are connected. For example, smaller forest patches are typically associated with lower biodiversity because of greater human disturbance, reduced population sizes, greater edge effects, and changes in community structure (Turner 1996). Similarly, higher biodiversity may be linked to smaller groups of forest-users because such groups are likely to exert a lower harvesting pressure and are less likely to overexploit species to the point of local extinction. As a second example, forests that are legally designated for community management often have rules that are locally created, contextually relevant, and which are respected and observed by forest-users (Larson and Soto 2008). Legally designated community forests may generate higher carbon benefits as a consequence of greater adherence to such locally generated rules. The ways in which these relationships variously play out in Nepal are additionally reported in many empirical case-studies and multi-site analyses (e.g. Nagendra 2002, Adhikari et al 2004), as well as comprehensive reviews that draw on more than multiple decades of community forestry experiences in the country (e.g. Acharya 2002, MFSC 2013).

Some of our results also appear contrary to previous evidence and assumptions. For example, forests that were not legally designated as community forests were associated with higher livelihood outcomes (defined as greater dependence on forest resources for livelihoods), whereas one might expect legally designated community forests to facilitate greater access to forest resources. Similarly, the association between lower levels of collective action and higher carbon outcomes might not have been predicted from the existing literature. More broadly, it is difficult to disentangle causality from association using observational studies, and while these data and our analyses enable us to describe associative links between our independent variables and these outcomes, they do not allow us to attribute a causal link. There remains a need to explain the associations. A quasi-experimental (e.g. matching-based) research design or longitudinal analysis using repeat-visit data would help to control for potential biases and facilitate more assured statements about causality (Ho et al 2007).

Second, the reported variations also need careful interpretation because the same independent variable may have contrasting directions of association with two different forest management outcomes. In our analysis, ethnic diversity, collective action, and the legal designation of forests as community- or state-owned all had negative associations with one outcome but a positive association with another. Although a deeper understanding of the causal linkages between these variables and the three outcomes is necessary to enable policy action, the results suggest complexity in the interactions between these variables and forest management outcomes and do not support a policy emphasis on win–win outcomes.

One means to further explore these relationships would be to collect data in a more targeted manner to test a specific hypothesis about the relationship between a specific independent variable and one or more forest management outcomes. Data could be collected on a larger number of communities with high levels of a variable of interest that are matched with other communities that are similar in terms of their basic demographic and socio-economic characteristics as well as in terms of their reliance on forests, but that have lower levels of the variable of interest. Such a priori research designs for data collection, coupled with matching based estimations post data collection, can permit more robust causal inference about the role of particular variables in enhancing multiple forest management outcomes.