Data from: Multiple facets of biodiversity are threatened by mining-induced land-use change in the Brazilian Amazon
Data files
Jul 06, 2023 version files 4.98 MB
-
Amazon_biome_treatment_zones_final.zip
-
Angio_sp_xy.shp.zip
-
Arthro_sp_xy.shp.zip
-
arvore_final10052016.zip
-
Indices_master_hex_n20_final.shp.zip
-
README.md
-
Vert_sp_xy.shp.zip
Abstract
Aim
Mining is increasingly pressuring areas of critical importance for biodiversity conservation, such as the Brazilian Amazon. Biodiversity data are limited in the tropics, restricting the scope for risks to be appropriately estimated before mineral licencing decisions are made. As the distributions and range sizes of other taxa differ markedly from those of vertebrates – the common proxy for analysis of risk to biodiversity from mining – whether mining threatens lesser-studied taxonomic groups differentially at a regional scale is unclear.
Location
Brazilian Amazon
Methods
We assess risks to several facets of biodiversity from industrial mining by comparing mining areas (within 70km of an active mining lease) and areas unaffected by mining, employing species richness, species endemism, phylogenetic diversity, and phylogenetic endemism metrics calculated for angiosperms, arthropods, and vertebrates.
Results
Mining areas contained higher densities of species occurrence records than the unaffected landscape, and we accounted for this sampling bias in our analyses. None of the four biodiversity metrics differed between mining and non-mining areas for vertebrates. For arthropods, species endemism was greater in mined areas. Mined areas also had greater angiosperm species richness, phylogenetic diversity, and phylogenetic endemism, although lower species endemism than unmined areas.
Main Conclusions
Unlike for vertebrates, facets of angiosperm and arthropod diversity are relatively higher in areas of mining activity, underscoring the need to consider multiple taxonomic groups and biodiversity facets when assessing risk and evaluating management options for mining threats. Particularly concerning is the proximity of mining to areas supporting deep evolutionary history, which may be impossible to recover or replace. As pressures to expand mining in the Amazon grow, impact assessments with broader taxonomic reach and metric focus will be vital to conserving biodiversity in mining regions.
Methods
Database Assembly
Mapping Mining Areas
We obtained spatial information on mineral prospecting and mineral mining leases within the Brazilian Amazon from SIGMINE (Sistema de Informações Geográficas da Mineração; DNPM, 2012). This database catalogues all registered legal mining activities within Brazil, detailing the extent of each activity, dates of operation, and mined commodities. To map ‘mining leases’ of industrial-scale mineral mines, we selected records greater than 100 hectares in area and classified as mining concessions (Concessão de Lavra) and omitted leases extracting water or those classified as small-scale artisanal operations (Lavra Garimpeira). This resulted in 411 polygons (including active leases and adjacent extensions of such leases) of 15,750 km2 in total area, with mining start dates ranging from 1944 to 2017 (mean = 1978, sd = 11.9; Fig. 1). To map ‘mining areas,’ which include the direct (i.e., immediate land-use change resulting from mineral extraction) and indirect (i.e., extensive land-use change associated with mineral extraction, processing, and transportation) impacts of mining on forests (Sonter et al., 2017), we created a 70 km buffer surrounding each mining lease. ‘Non-mining areas’ (i.e., areas unaffected by industrial mining) were mapped by extracting our mapped ‘mining areas’ and an additional layer representing all other legal mining leases excluded from our analyses (i.e., inactive leases, those targeting water, or operations smaller than 100 hectares in area; shown in white in Fig. 1) from the Brazilian Amazon (Fig. 1).
For interpolation analyses, hexagons are the most logical sampling unit shape as their centroids are equidistant, the distance of points from the edges to the centroid is the closest, and sampling biases are reduced due to their lower perimeter-area ratio compared to squares or triangles (Birch et al., 2007). Hexagons of approximately 0.5° with equal area were assigned to one of two study areas – mining areas or non-mining areas – based on where their centroid was located (Fig. 1). Hexagons were omitted from our analyses if they contained fewer than 20 occurrence records per taxonomic group or their centroid was located outside the Brazilian Amazon. We used 0.5° hexagon sampling units as sensitivity analyses conducted in previous studies utilising the same dataset indicated reduced variation in results for hexagon areas of 0.5° and above (Oliveira et al., 2017a; Strand et al., 2018) and so any fine-scale georeferencing inaccuracies remaining in the dataset after filtering are minimised (Oliveira et al., 2017b). This sampling unit area also ensured sufficient sample sizes would be assigned within and among mining-induced deforestation-affected areas to enable robust comparisons across the study area for all taxonomic groups, particularly arthropods, while reducing the amount of area hexagon interpolations may sample from outside their respective study area polygons.
Assembling Biodiversity Data
Data on species occurrences were obtained from (Oliveira et al., 2017a) and (Oliveira et al., 2019a) and represent the most comprehensive dataset of species occurrences in Brazil to date. These data were assembled from online databases spanning GBIF (gbif.org); CRIA (specieslink.net); Birdlife International (birdlife.org), Herpnet (herpnet.org), Nature Serve (natureserve.org); and Orthoptera Species File (orthoptera.speciesfile.org). These data were also supplemented with occurrence records obtained from taxonomic literature and biodiversity inventories (Oliveira et al., 2017a; Oliveira et al., 2019a). All species occurrence records were filtered to determine if they lacked geographic coordinates or exhibited location errors using a map of Brazilian municipalities (mapas.ibge.gov.br; Oliveira et al., 2017a; Oliveira et al., 2019a). Taxonomic validity for all occurrence records was confirmed using taxon-specific catalogues and expert reviews for each taxonomic group (Oliveira et al., 2017a; Oliveira et al., 2019a). After filtering for geographic and taxonomic accuracy, the final dataset comprised 113,790 occurrence records for all the Brazilian Amazon. The dataset contained 44,660 records of angiosperms (6899 species of families Asteraceae, Bromeliaceae, Fabaceae, Melastomataceae, Myrtaceae, Orchidaceae, Poaceae, and Rubiaceae), 24,374 records of arthropods (4630 species of bees, spiders, millipedes, Orthoptera, dragonflies, moths and Diptera), and 44,756 records of vertebrates (1584 species of birds, mammals, and anurans). Spatial distributions of occurrence record densities for each taxonomic group are provided in the supporting information (Fig. S1).
Phylogenetic trees were constructed from published figures into Newick code with TreeSnatcherPlus (Laubach & Von Haeseler, 2007) and supplemented with data from empirical phylogenetic studies synthesised by The Open Tree of Life project (Hinchliff et al., 2015). As branch lengths, when available, are not directly comparable between trees, all branch lengths were considered equal to one (Oliveira et al., 2017a; Oliveira et al., 2019a). Phylogenetic trees were compiled into a supertree using matrix representation with parsimony (Baum, 1992) and pruned to represent species restricted to Brazil. Our dataset represents the most extensive collection of species occurrence records and phylogenetic trees compiled in Brazil for this purpose to date (Oliveira et al., 2017a). However, data collected for environmental impact assessments that are not published online will inevitably be missing from our database, and rare, threatened, or range-restricted organisms may also not be included due to limited sampling.
Calculation of Biodiversity Facets
Sampling Effort
We first intersected mining lease and mining area polygons with species occurrence records to provide a coarse estimate of the proportion of occurrence records within mining leases and their more expansive impact areas from the total contained in our database. An equal area measure was calculated through the ‘Sampling Effort’ functor of the BioDinamica plug-in (Oliveira et al., 2019b) of Dinamica EGO (Ferreira et al., 2019), which was set with a 10 km search radius due to limited and sporadic biodiversity sampling in the Brazilian Amazon (Oliveira et al., 2016; Oliveira et al., 2017a). We then converted the output raster to points and summed the mean sample effort index values across 0.5° radius hexagon sampling units (Fig. 1). The ‘Sampling Effort’ functor in BioDinamica employs a Gaussian kernel density index function. For all analyses using BioDinamica, 0.5° hexagon sampling units were only created where ≥ 20 species occurrence records existed.
Biodiversity Metrics
We calculated four sampling-effort-corrected biodiversity metrics for each of the three taxonomic groups: species richness, species endemism, phylogenetic diversity, and phylogenetic endemism, since measuring biodiversity with species richness alone does not capture values pertinent to conservation at the landscape scale, such as endemism or evolutionary history (Faith, 1992; Faith et al., 2004). Indeed, the loss of species is not equivalent to the loss of evolutionary history (Vane-Wright et al., 1991), and conservation priority areas can differ when using species richness and phylogenetic diversity (Rodrigues et al., 2005; Forest et al., 2007). Furthermore, phylogenetic measures may capture the quantity and distribution of diversity better than species-based measures, especially when data are limited, but both are representative of different diversity components (Rosauer & Mooers, 2013; Tucker et al., 2017). Thus, here we employ a variety of biodiversity metrics for comparison between mining and non-mining areas in the Brazilian Amazon.
Species-based Metrics
Species richness (per unit area) is the most sensitive biodiversity measure to variation in sampling effort (Oliveira et al., 2016). To quantify species richness, we used a resampled species richness index to account for variation in sampling effort. The ‘Resample Species Richness’ functor (Oliveira et al., 2019b) operates by spatially resampling species occurrences. We set this functor to a minimum of 20 species occurrences per hexagon sampling unit, taking a random 25% subsample with 1000 iterations. This tolerance level retained the most variation in species richness while maintaining an adequate sample size to compare mining and non-mining areas. The output represents the mean resampled species richness per hexagon (Oliveira et al., 2017a; Oliveira et al., 2019a). This method provides a relative measure of species richness, simulating uniform sampling throughout the study area, thus addressing variation in sampling effort (Oliveira et al., 2017a; Oliveira et al., 2019a).
For comparisons of endemism, the level of geographic restriction among species, we used the weighted endemism index, a relative measure of endemism as opposed to an absolute measure (Williams & Humphries, 1994). We computed this using the ‘Weighted Endemism’ functor (Ferreira et al., 2019), which calculates the inverse of a species distribution area and sums it across hexagon sampling units (Oliveira et al., 2017a; Oliveira et al., 2019a). To control for variation in sampling effort and uncertainty in species distribution estimations, the functor generates a sampling effort-corrected and area-weighted endemism index using the equation: A * B / ((A * B) + ((1 - A) * (1 - C))) where ‘A’ is the weighted endemism as determined by the inverse of a species distribution area, ‘B’ is the product of weighted endemism and sampling effort (expressed as the mean kernel density index of species occurrence records within a 50 km radius of each occurrence point for each species analysed), and C is the total number of species sampled, with 150 species occurrences set as the maximum (0.999) as the frequency distribution of species records reaches an asymptote at this value (Oliveira et al., 2017a). Species with fewer records are assigned values beginning at 0.00001 for a single record upwards linearly (Oliveira et al., 2017a).
Phylogenetic Metrics
We considered two measures of spatial phylogenetic variation important for the maintenance and persistence of biodiversity, phylogenetic diversity (PD), a measure of divergence in phylogenetic relationships between species in an area (Faith, 1994), and phylogenetic endemism (PE), a measure of the restriction of phylogenetic lineages between given areas (Rosauer et al., 2009). Phylogenetic diversity is estimated by comparing summed distances between phylogenetic branches among species in a given area (Faith, 1992). Phylogenetic endemism is estimated by comparing the relative rarity of evolutionary lines among taxa between areas, with fewer branches at higher taxonomic classifications being afforded greater weighting in contributing to the endemism of a species’ evolutionary lineage (Rosauer et al., 2009).
We compared phylogenetic diversity using branch lengths as a surrogate for the uniqueness or similarity in features of species within a phylogenetic tree (Faith, 1992). The ‘Phylogenetic Diversity’ functor sums branch lengths through the root of phylogenetic trees using the shortest path between species connected within a sampling unit (Faith, 1992, 1994; Oliveira et al., 2017a). Due to the paucity of phylogenetic information for Brazilian species and hence within our database, despite it being the most comprehensive dataset compiled in Brazil to date (Oliveira et al., 2017a), branches were assigned equal lengths under an assumption of equivalent rates of feature descent across all phylogenetic pathways (Faith, 1992). In comparing phylogenetic measures of biodiversity, we used phylogenetic trees for taxa geographically restricted to Brazil (Oliveira et al., 2017a).
To compare phylogenetic endemism, a measure of geographic restriction of phylogenetic diversity and hence evolutionary history, we used the phylogenetic weighted endemism index (Rosauer et al., 2009). This index employs a relative measure of endemism rather than an absolute measure to compare geographic concentrations of evolutionary history, to address the sensitivity to spatial scale apparent in absolute endemism measures (Rosauer et al., 2009). The ‘Phylogenetic Endemism’ functor (Ferreira et al., 2019) interpolates phylogenetic weighted endemism indices across sampling hexagons by summing the phylogenetic branch lengths between species (Oliveira et al., 2017a; Oliveira et al., 2019b).
Data Processing and Analysis
Sampling effort and biodiversity indices were generated in the BioDinamica plug-in extension (Oliveira et al., 2019b) for the freely available Dinamica EGO software (Ferreira et al., 2019). The outputs from Dinamica were processed and assigned to study areas in ArcMap 10.7 (ESRI, 2018). Due to the amount of species occurrence records present in a sampling unit for each taxonomic group and the availability of phylogenetic tree data for each occurrence, the number of groups compared between mining and non-mining areas varied, as hexagon sampling units were removed from analyses if they contained < 20 occurrence records per taxonomic group (see N column in Table S1). Statistical comparisons between mining and non-mining areas for all metrics and all taxonomic groups were made using two-tailed Wilcoxon rank-sum tests, with all graphical representations created in R (R Core Team, 2018). For improved visualisation, the four biodiversity metrics were rescaled between 0 and 1 to facilitate interpretation using the scales package in R (Wickham & Seidel, 2020), which maintains identical data spread. Sensitivity analyses showing the small and largely inconsequential variation in results when using alternative potential impact buffers of 20 km and 50 km are included in the supporting information, noting that the largest difference in findings is observed for arthropod phylogenetic endemism where the sample size is substantially reduced (mining area hexagon n = 17) when using a 20 km buffer versus a 70 km buffer (mining area hexagon n = 53; Fig. S2, Fig. S3). Effect sizes are reported (Table S1) and compared in sensitivity analyses of mining risk buffers (Fig. S3) to show the magnitude of difference in metrics between study areas and to support comparisons with future analyses. However, they are not to be interpreted as measures of mining impact on biodiversity due to the abstract complexity of their biological interpretation as they relate to differences in interpolated biodiversity metrics.
Usage notes
GIS Software, e.g. ArcGIS, R, GRASS, QGIS.
Dinamica Environmental Modelling Software, with BioDinamica plug-in. Available at <https://csr.ufmg.br/dinamica/>