Resource-Use Patterns in Swidden Farming Communities: Implications for the Resilience of Cassava Diversity

Abstract

Resource-use patterns, especially through exchanges among farmers, may ultimately confer resilience to the local agrobiodiversity. We investigated the use of cassava ethnovarieties by swidden farming communities in Brazil, exploring the structure of networks depicting farmers and the varieties they cultivate. The emergent nested resource-use pattern indicated that all farmers shared a core of top-ranked ethnovarieties (most common/abundant) while some farmers also cultivate rarer varieties. This pattern may result of individual preferences. Due to the current loss of interest and cultivation area for traditional agriculture, we simulated the extinction of crop fields to evaluate whether nestedness conferred robustness to cassava diversity. The diversity of ethnovarieties of cassava tended to be conserved when farmers were randomly removed from the network than when we preferentially removed farmers with more diverse crop fields. Stem cuttings of ethnovarieties are commonly exchanged among farmers, thus the extinction of ethnovarieties within crop fields could be restored. Therefore, we suggest that the interplay between the farmer’s resource-use patterns and exchange system strengthens the resilience of cassava diversity, which is an important staple resource for such communities.

Appendix 1. Network Analysis

Individual-resource interactions were described as two-mode networks (e.g., Pires et al. 2011) in which the nodes representing local farmers were linked to other nodes representing the cassava ethnovarieties mentioned in the interviews. These individual-resource networks were plotted in an incidence matrix A to depict the interactions between the cassava ethnovarieties (rows) and local farmers (columns), where element aij of the matrix is 1 if ethnovariety j was cultivated by farmer i and zero if otherwise. We measured the degree of nestedness and modularity to describe the structure of the individual-resource networks.

Nestedness (N) is a particular network property in which the interactions are asymmetric, with a core of densely connected nodes and other less connected nodes that generally interact only with the core (e.g., Bascompte et al. 2003; Guimarães et al. 2006). In our case, nestedness may reveal heterogeneity among resource use: not all farmers use the entire regional pool of cassava ethnovarieties, i.e., some use mostly of the ethnovarieties available, whereas others use only the most cultivated ones. We calculated the degree of nestedness using the nestedness metric based on overlap and decrease fill (NODF, Almeida-Neto et al. 2008) using ANINHADO 3.0 software (Guimarães and Guimarães 2006). In highly nested matrices, NODF has a tendency to reach 1; otherwise it tends to be zero.

The farmers of Imbituba and Paraty regions differ in their cultivation strategies (Imbituba, communal; Paraty, household fields) and these differences may have an impact on the ethnovariety exchanges. Therefore, we evaluated whether the local farmers were clustered due to the varieties that they use. To do so, we evaluated the modularity of the networks. Modularity (M) quantifies the tendency of the nodes to cluster into cohesive groups: M measures the difference between the number of interactions among nodes that are within modules and the number that are between modules (e.g. Guimerà and Amaral 2005a, b). A modular network in our case would consist of weakly interlinked groups of farmers that are internally strongly connected due to the use of the same resource (e.g., Olesen et al. 2007). In other words, we would see high homogeneity of resource-use inside groups of farmers and high heterogeneity between them. We calculated M using a simulated annealing algorithm to identify the partition of a network into modules that yields the largest degree of modularity with the NETCARTO program (Guimerà et al. 2004).

Since the regional networks are different in size, i.e., the number of nodes and links, we used the relative nestedness (N*; see Bascompte et al. 2003) and relative modularity (M*; see Pires et al. 2011) to allow cross-network comparisons. The relative nestedness is defined as N* $=\left(N-{\overline{N}}_R\right)/{\overline{N}}_R$, where N is the observed nestedness and is the average nestedness of random matrices generated from the null model analysis (see below). Similarly, the relative modularity is defined as M* $=\left(M-{\overline{M}}_R\right)/{\overline{M}}_R$ (see Pires et al. 2011).

The significance of the network metrics was evaluated using the null model approach. Random networks were generated by a null model that shuffled the original total of 1 s among new matrix cells according to the frequency of citations of each ethnovariety (marginal totals of rows) and the number of ethnovarieties grown by each local farmer (sum of columns). Thus, each cell has different probabilities of being filled according to the two features of the observed dataset (see null model 2 in Bascompte et al. 2003): ${c}_{ij}=\frac{1}{2}\left(\frac{ Pi}{C}+\frac{ Pj}{R}\right)$, where Pi = the number of farmers that cultivate ethnovariety i (row sums); Pj = the number of ethnovarieties cultivated by farmer j (column sums); C = the number of local farmers (columns); and the R = number of cassava ethnovarieties (rows). The empirical values of nestedness and modularity were evaluated by checking whether the observed values were within the 95 % confidence intervals generated by the 1,000 randomized networks. The relative nestedness and modularity were evaluated by the z-score; the p-value was calculated as the proportion of randomized values that were higher than the observed values.

Appendix 2. Removal Simulation

To infer on the resilience of the cassava agricultural system, we tested the robustness of the regional pool of cassava ethnovarieties to the extinction of farmers from the communities. To accomplish that, we cumulatively removed farmers from one side of the network, and evaluated the magnitude of the secondary extinctions of ethnovarieties in the other side (cf. Mello et al. 2011). The removal of a local farmer mimics the loss of cultivation area or interest in the traditional swidden agriculture for to more profitable activities (a current trend in the studied regions; Cavechia 2011).

We simulated two scenarios: the removal of farmers according to their degree (i.e. number of interactions representing the diversity of cultivated ethnovarieties), and random removal. In the former, we removed the farmers sequentially from those with more interactions (i.e. cultivating more ethnovarieties) to those with fewer interactions. The latter was our benchmark, when all the farmers had the same probability of being removed from the network (i.e. abandon the agriculture practice). To incorporate uncertainty, we repeated the simulations 100 times for each region and used the average proportion of secondary “extinctions”. Simulations were interpreted through extinction curves, generated by plotting the average proportion of cumulative secondary extinction of ethnovarieties against the cumulative number of farmers removed. A positive relationship between primary and secondary extinction is expected in all cases: a extinction curve increasing slightly would represent a very robust network, because more farmers would have to be removed to extinct few ethnovarieties; accordingly, a curve with a very sharp increase would represent a very fragile network, in which the removal of few farmers would trigger the extinction of several ethnovarieties. The simulations were performed using package bipartite for R environment (Dormann et al. 2008).