Data collection

The Peruvian Forestry Law N°27308 obliged concession holders for the first time to present detailed inventories of the Brazil nut trees under their custody. Most inventories were carried out under the auspices of a number of institutions active in the region, most notably ACCA (Asociación para la Conservación de la Cuenca Amazónica), CAMDE (Conservación Ambiental y Desarrollo en el Perú), FONDEBOSQUE (Fondo de Promoción del Desarrollo Forestal), AIDER (Asociación para la Investigación y Desarrollo Integral), RNTAMB PRMRFFS (Programa Regional de Manejo de Recursos Forestales y Fauna Silvestre), Forestal Rio Piedras SAC and Conservation International. Field staff of these institutions georeferenced individual trees in the company of the respective Brazil nut harvesters who were systematically asked to provide their recollection of the average productivity of each individual tree. For the vast majority of trees these data were complemented with DBH and height measurements, and a description of each tree’s phytosanitary condition, by indicating whether a tree was infested by lianas, had broken branches or holes in its trunk, or showed evidence of termite nests, wound exudate or tumours. The seed production estimates, DBH and height measurements and phytosanitary characterization data of individual trees used in this study were collected during one single field mission. Brazil nut seeds are harvested by cracking open the lignified capsular fruits (pyxidia) with a machete after these have fallen on the ground. Individual seeds are also protected by wooden testa, but these are not opened in the field and harvesters expressed seed production weight as multiples of ´latas´ (tin cans) which contain approximately 11.66 kg of fresh in-shell seeds. We compiled a dataset containing 135,528 georeferenced Brazil nut trees (S1 File) which had diameters at breast height (DBH) ≥10 cm from 418 of the approximately 1,200 Brazil nut concessions in Madre de Dios, Peru (Fig 1), which were largely collected by the above institutions in the period between 2003 and 2007.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Distribution of Brazil nut concessions in Madre de Dios, Peru, with approximate locations of the 135,528 individual Brazil nut trees (DBH≥10 cm) from 418 concessions considered in this paper.

https://doi.org/10.1371/journal.pone.0183743.g001

A limitation of this dataset is the low precision of georeferences of numerous individual trees, related to the use of low-precision GPS equipment, further complicated by difficulties to capture satellite transmissions under the forest canopy. Furthermore, some coordinates were obtained through the use of compass and measuring tape, using points with known coordinates as references. However, as most of the spatial layers (climate and soil) we used for extracting environmental data had maximum resolutions of 30 arc seconds (~1km at the equator) and are themselves based on interpolated data (for further details see S2 File), imprecision issues are not expected to excessively influence the analyses carried out here.

In May 2012, we participated in a group interview organized by CIFOR (Center for International Forestry Research) with 8 female and 21 male concession holders of the Brazil nut association ASCAL to identify (through free listing) the variables that, according to the participants, influence the productive potential of Brazil nut trees (data used with permission). We also organized individual interviews with 53 additional concessionaires (12 female and 41 male) and asked them to free list the variables they believed influence seed production in Brazil nut. We did not obtain the approval of an Institutional Review Board or ethical committee mainly because of the nature of the interviews conducted. The purpose of the study was not to collect personal information of participants but rather aggregated data about the variables that influence the seed production of Brazil nut trees for the evaluation of our research hypotheses. We obtained oral consent of all participants, after explaining them that the goal of the interviews was to verify whether local people´s perceptions match or not with other scientific findings. Participant names, dates and places of interviews and oral consent were recorded in field notes. While we consider the collected data to classify as local or traditional ecological knowledge, most of it is publically available as anecdotal knowledge (and, for example, reported in local students' dissertations) and in our view has little to no chance of being misused. Actually some of the variables reported by the people may have been mentioned to them by the plethora of institutions or professionals working in the area, and it is not really possible anymore to confidently distinguish endogenously developed knowledge from knowledge introduced by outsiders. Participants were informed that the results of the interviews could potentially be published in aggregated format, and we intend to make available the findings of this study to concessionaires and Brazil nut harvesters during ongoing and future activities of the authors in the region.

Comparison of Brazil nut harvesters´ estimates with field measurements

In a recent study Rockwell et al. [9] measured seed production over two years of 499 Brazil nut trees (DBH≥40 cm) from five different concessions in Madre de Dios which are also included in our dataset of 135,528 trees. To evaluate the degree of correspondence between these measurements and the estimates provided by Brazil nut harvesters, we matched individual trees in the two datasets based on the correspondence between their geographical locations and measured diameters. For this, we used only the 457 trees in the Rockwell et al [9] dataset with unique coordinates and considered only trees whose diameters and locations could be matched with a 10% error margin and precisions of 10m, 20m and 40m, respectively. As the mean distances between Brazil nut trees sampled by Rockwell et al [9] varied between 57 and 81m, we expected the probability of matching the same trees in the two datasets to be inversely correlated with the degree of correspondence between their geographical locations.

Statistical analysis

We validated the correspondence between seed production estimates of individual trees by Brazil nut harvesters and measurements carried out by Rockwell et al [9] by means of linear regression models. Normality of the model residuals was confirmed by a Shapiro-Wilk test (P>0.10) and residuals did not show evidence of spatial autocorrelation (confirmed by means of autocorrellogram constructed in ncf package for R [15,16]).

We assessed the nature of the relationships between Brazil nut seed production estimates and tree size variables (height, canopy position and size, DBH and aboveground biomass), distances to the three nearest conspecific neighbours, geophysical, edaphic and climatic variables with those reported in literature and/or by Brazil nut harvesters by means of basic and mixed generalized linear and additive models (GLMs and GAMs, respectively). We constructed the models using the Poisson distribution, given that the production estimates of Brazil nut trees were expressed as multiples of tin cans (~ 11.66 kg each) and hence show strong similarities with count data. To account for overdispersion (<3 in all cases), we corrected the standard errors using quasi GLM and GAM models, where the variance is given by the product of the mean and the dispersion parameter. For the additive models (implemented in mgcv package for R [17]) we used thin plate regression splines and cross-validation to estimate the optimal amount of smoothing. The presence of positive spatial autocorrelation in model residuals (assessed by means of case-specific autocorrellograms), was accounted for by including the concession where each tree was sampled as a random effect variable. GLMMs were implemented using Penalized Quasi-Likelihood in MASS package for R [18]. As this approach does not allow calculating (pseudo) R² values or the proportion of the null deviance explained by the model, we additionally carried out Spearman regressions to complement GLMM results. To enhance interpretation of figures, response variables and fitted values were transformed to seed production estimates expressed as kilograms.

Each of the individual GLMs and GAMs is expected to explain only a tiny portion of the variability in seed production estimates for three main reasons. First, seed production of individual Brazil nut trees is influenced by the interplay of multiple genetic, physiological, demographic, phytosanitary, climate, weather, edaphic and anthropogenic variables, implying that single variables are likely to explain only a small fraction of the variation in seed production. Second, several of the explanatory variables (climate, soil, terrain) may have low precision as they themselves are based on interpolations. And third, the seed production estimates are expressed as multiples of the content of tin cans (~ 11.66 kg), and were provided by a large number of people, whose individual estimates are expected to show substantial variation in terms of quality. However, in line with the findings of Steinke et al. [5], we expect the averages of the pooled estimates to approximate true patterns between explanatory variables and the seed production of individual Brazil nut trees. We consider a pattern as significant if the average increase or decrease in seed production estimates along the regression line and across the range of an explanatory variable captured by our data is at least 10% of the average seed production estimate of all trees in our dataset (30.6 ± 26.9 (SD) kg per tree), i.e. ~3kg.

For comparisons of seed production estimates of Brazil nut trees across different categorical variables (phytosanitary conditions, growing close or not to rivers and roads, in and outside indigenous territories and protected areas), we used non-parametric Wilcoxon tests due to violation of the homoscedasticity assumption for the vast majority of comparisons, ruling out the use of t-tests.

Phenotypic variables and light exposure

We found that average production estimates increased with stem and crown height variables. This corroborates previous experimental research which related these variables to increased light exposure which in turn is believed to positively influence seed production [7–12]. Accordingly, we did not find an effect of the direction of the growth site slope (aspect) on Brazil nut productivity estimates (S1C Fig). This was expected, as most productive trees are emergent and that fact that owing to the closeness of the study area to the equator, the sun stands almost perpendicularly above the forest canopies at noon. However, lateral light regimes might influence fecundity, as suggested by the higher average seed production estimates documented for trees growing in the vicinities of roads, compared to trees at further distances.

The unimodal relationships we found between seed production estimates and variables related with wood biomass are much less documented. Some studies found positive correlations [10,12] between a tree´s DBH and its average seed production, but to the best of our knowledge only Kainer et al [11,19] reported a maximum of seed production in the middle diameter classes (100–150 cm DBH). The peak in our production estimate data similarly started around 100 cm DBH, but reached its true maximum between 125 and 250 cm DBH. This discrepancy may be because of the much smaller sample size of the studies of Kainer et al [11,19], combined with the fact that maximum tree sizes were lower in their study area (their biggest tree was 210 cm DBH). Interestingly, the maxima of estimated seed production of trees occurred roughly in the 100–200 cm diameter class, thus preceding the peak in average seed production estimates (red line in Fig 4D). This is because the vast majority of trees with zero estimated productivity were less than 150 cm DBH (Fig 3B). Peaking seed production in middle diameter classes is paralleled by the unimodal relation we found between the aboveground biomass estimate of a tree and its estimated seed production. There is evidence that some tree species produce most seed in middle age [24], followed by decline and senescence. While the relation between tree diameter and age in Brazil nut trees individuals can be highly variable, overall it seems to abide to an approximately linearly increasing trend [25].

Phytosanitary variables

Liana load is one of the best studied phytosanitary variables that negatively affect seed production of Brazil nut trees [7,9,19,20]. Our results not only corroborate previous studies, but also support assertions of Brazil nut harvesters that trees with broken branches, holes in their trunks or showing the presence of wound exudate tend to produce less (Fig 5). The negative effect of heavy winds on seed production identified by Brazil nut harvesters may be related to the association between wind and the incidence of broken branches.

Genetics and reproductive biology

Our data suggest that genetics influence seed productivity, at both individual and population levels. Individual-level genetics might explain why some trees never or hardly ever produce seeds (e.g. because of incompatibility issues) while others produce very high quantities of seeds year after year (e.g. by genetic superiority). Indications of the potential impact of population-level genetics on seed production are apparent in the relationship between the recorded seed production of a tree and the distance to its nearest conspecific neighbors. Our findings support the observation of concessionaires that seed production tends to increase in Brazil nut trees that grow closer to conspecific neighbours. This is called the “Allee” effect [26] and is related with effects of pollen limitation and increased selfing rates in more isolated trees [27]. It has been described for numerous tree species [28,29], particularly in predominantly outcrossing ones such as Brazil nut [30]. However, this effect only applied to Brazil nut trees growing at nearest conspecific distances higher than 100 m, whereas trees growing at very short distances of conspecific neighbours tended to yield slightly lower production estimates than trees at intermediate distances (Fig 6). Considering the low population density of Brazil nut in the study area (approximately one tree per two hectares), this may be due to the fact that close conspecific neighbours are more likely to be genetically related (siblings or half siblings). Crosses between conspecific neighbours with high local kinship result in biparental inbreeding which may increase seed abortion rates [31], and hence lower seed set of trees [32]. Further research is needed to confirm or refute this hypothesis.

Terrain and soil variables

Our results suggest that Brazil nut trees tend to produce more when they grow on higher lying, flat soils. Higher elevations in Madre de Dios are likely to correlate with terra firme soils (i.e. Brazil nut’s preferred habitat [33]) and lower elevations with floodplains/varzeas and proximity to rivers. This was confirmed by our finding that trees growing close to rivers yielded lower average seed production estimates than trees at further distances. Clay and Clement [34] argued that Brazil nut typically grows on well-drained ultisols and oxisols which have low pH. Accordingly, most of the Brazil nut trees in our dataset grew on soils with expected pH below 5 and there was a tendency for trees in more acid soils to yield higher production estimates (S2F Fig). In line with expectations, higher estimates were found for trees in soils with higher predicted organic carbon content (S2D Fig), which could also explain why Brazil nut harvesters reported that Brazil nut trees produce more in darker soils. The finding of Kainer et al [11] that higher CEC values tended to be associated with higher fecundity was only partly supported by our data (results Spearman correlation, but not GLMM) (S2E Fig). This does not necessarily mean our data were unable to convincingly detect this pattern. CEC is often mistakenly considered a proxy of soil fertility because soils of any given CEC can differ greatly in nutrient availability in Amazonia [35]. Further research is needed to assess the extent to which the nature of the relation between soil CEC and Brazil nut seed production is region or site specific.

Climate and weather variables

Moderate temperature regimes seem to positively influence seed production of Brazil nut trees. Higher temperatures during the warmest periods of year were associated with lower production estimates (S3A–S3C Fig). Particularly the latter observation is troublesome in light of climate change which is expected to result in rising annual temperatures and increases in the frequency of climate extremes, and hence might lead to lower production levels. Precipitation correlated positively with production estimates, particularly so in the drier periods of the year (S3D–S3F Fig), thus confirming previous studies [7,11]. The fact that this trend was strongest for precipitation in the driest month to some extent contradicts the common -but previously challenged [36]- assumption that Brazil nut needs at least two dry months (<60mm) for development and growth [37]. Anecdotal evidence from Madre de Dios that a combination of high temperatures and droughts in 2005 and 2010 led to massive flower abortions (William Moreno, per. communication), suggests that temperature and precipitation interact in their impact on productivity, which requires further research.

Anthropogenic variables

There has been a longstanding debate about the sustainability of harvesting Brazil nut seeds from natural stands with opposing views defended by different scholars [23,38]. While our data do not provide a firm stand in this debate, the fact that average seed production estimates of Brazil nut trees growing in protected areas were significantly lower than of trees from Brazil nut concessions, may support the hypothesis of Scoles and Gribel [23] that more intervened stands could be more productive. Brazil nut trees from protected areas are expected to be exposed less to anthropogenic disturbance compared to Brazil nut concessions where modifications of the understory and hunting of the main natural seed disperser (thus freeing more abandoned scatter hoards for germination) may promote natural regeneration and possibly seed production at stand level [23]. Alternatively, lower productivity estimates may reflect the fact that protected areas are located on suboptimal land for Brazil nut, as claimed by one of the Brazil nut harvesters we interviewed. A third scenario is that Brazil nut harvesters may have more difficulties to estimate seed production of individual trees in protected areas which they are allowed to enter only during the harvest season. It is widely accepted in ethnobotanical literature that more intensive contact with plant resources enhances knowledge on their useful traits [39], and this may also apply to seed production estimates.

The reason why productivity estimates of Brazil nut trees from indigenous territories tended to be lower than from Brazil nut concessions outside these areas is less clear. Aside from the possibility that at least some indigenous territories may also be located on suboptimal land for Brazil nut, the more sporadic harvesting practices of indigenous groups, or mistrust of the state authority at the time of data collection, coinciding with the first effort to make detailed inventories of Brazil nut concessions, may have resulted in their underestimating of average seed production levels.