|
|
Received: 3 November 2016 Revised: 29 May 2018 Accepted: 14 July 2018 DOI: 10.1002/ece3.4448
ORIGINAL RESEARCH
Functional Redundancy in bird community decreases with riparian forest width reduction Lucas A. Maure1
| Rodolpho C. Rodrigues2 | Ângelo V. Alcântara1 |
Bruno F. C. B. Adorno1 | Douglas L. Santos1 | Eduardo L. Abreu1 | Rafael M. Tanaka1 | Rute M. Gonçalves1 | Erica Hasui1 1 Instituto de Ciências da Natureza, Universidade Federal de Alfenas, Alfenas, Minas Gerais, Brazil 2
Departamento de Ecologia, Instituto de Biociências, Universidade de São Paulo, São Paulo, São Paulo, Brazil Correspondence Lucas A. Maure, Instituto de Ciências da Natureza, Universidade Federal de Alfenas, Alfenas, MG, Brazil. Email:
[email protected] Funding information CNPq
Abstract 1. Riparian ecosystems are suffering anthropogenic threats that reduce biodiversity and undermine ecosystem services. However, there is a great deal of uncertainty about the way species composition of assemblages is related to ecosystem function, especially in a landscape fragmentation context. 2. Here, we assess the impact of habitat loss and disturbance on Functional Diversity (FD) components Functional Redundancy (FRed), Functional Evenness (FEve), and Functional Richness (FRic) of riparian forest bird assemblages to evaluate (a) how FD components respond to riparian forest width reduction and vegetation disturbance; (b) the existence of thresholds within these relationships; (c) which of the main birds diet guild (frugivores, insectivores, and omnivores) respond to such thresholds. We predict that FD components will be affected negatively and nonlinearly by riparian changes. However, guilds could have different responses due to differences of species sensitivity to fragmentation and disturbance. We expect to find thresholds in FD responses, because fragmentation and disturbance drive loss of specific FD components. 3. Our results show that FRed and FEve were linearly affected by width and disturbance of riparian habitats, respectively. FRed was significantly lower in riparian forests assemblages below 400 m wide, and FEve was significantly higher above 60% disturbance. These responses of FD were also followed to the decline in insectivores and frugivores richness in riparian forests most affected by these changes. 4. Consequently, our study suggests communities do not tolerate reduction in riparian forest width or disturbance intensification without negative impact on FD, and this becomes more critical for riparian area 0.8).
ern Minas Gerais state, southeastern Brazil (Figure 1, Supporting Information Table S1). Regional climate is classified as Cwa (humid subtropical climate) by Köppen-Geiger system, with a tropical rain-
2.3 | Predictor variables
fall pattern in summer and dry winters. Average annual temperature
In addition to width, we used four disturbance indices to evaluate
is 20.2°C, and annual precipitation being around 1,516 mm (available
the disturbance levels at our riparian forest study sites:
in https://en.climate-data.org/info/sources/). The relief is formed by gentle hills that are primarily structured formed by old orogens and the altitudes are often >800 m a. s. l. The studied forest remnants lie in a transitional region between
Disturbance index (%): We created a disturbance index that expressed the relative local disturbance intensity at each riparian forest. Within each bird sampling plot (30 m radius), we measured the
the Atlantic Forest and Cerrado, two of the most biologically rich and
number of vertical forest layers, maximum vegetation height, per-
highly threatened biomes in Brazil (Myers, Mittermeier, Mittermeier,
centage of canopy cover and the frequencies of epiphytes, vines
da Fonseca, & Kent, 2000). The original forest type in this region
(i.e., herbaceous or sub-woody climbing plants, which commonly
is described as semideciduous seasonal Atlantic Rainforest (IBGE
grow in disturbed places or forest edges), and invasive plant spe-
2012), but in much of the region, this has been drastically reduced
cies (i.e., Brachiaria spp.) (Imaflora 2008). We counted the number
to small and sparse forest fragments. In the study area, 99% of re-
of vertical stratification layers (i.e., emergent, canopy, understory,
maining forest patches are smaller than 20 ha, and 78% of the land-
shrub, herbaceous ground cover), using the maximum height
scape in which these patches occur have forest cover below 20%.
of the different trees, ontogenetic stages, and plant life form.
Even with specific protection laws for riparian forests (New Forest
Usually, preserved forests had greater proportions of tall trees
Code (Law 12,651/2012: available at: http://saema.com.br/files/
(height 20–25 m), more vertical strata (>2 strata), a more closed
Novo%20Codigo%20Florestal.pdf), almost 75% of these habitats
canopy, presence of young individuals of tree species in the un-
are deforested and used for pasture, coffee, and sugarcane planta-
derstory forest, and higher epiphyte frequency (i.e., Cactaceae,
tions or other agricultural uses (E. Hasui, personal communication).
Bromeliaceae, and Orchidaceae). In contrast, disturbed forests were characterized by more open canopy, fewer strata (1–2
2.2 | Riparian forest selection We selected the riparian forest sites by combining information from
strata), higher densities of short trees (1–15 m), higher frequencies of vines, and presence of invasive species. We standardized these data to reduce the variance between them and used principal
land cover and drainage network maps. The land cover map was
components analysis (PCA) to summarize information for these
built by unsupervised classification (oriented object method) of six
vegetation variables into one orthogonal variable (first axis) (Zar,
RapidEye satellite images (5-m resolution, total area = 3,586 km²)
1996). The first PCA axis was positively affected by the number
from 2010 using ENVI-E X 4.8 software. To assess the land cover
of vertical strata, canopy cover, total height and was negatively
map accuracy, we performed a visual inspection using Google Earth
affected by the frequency of invasive plant species (Supporting
satellite images and validated this information with field visits.
Information Table S3). It accounted for 45.4% of the total variance
The drainage network map was digitalized from topographic maps
in vegetation measurements (Eigenvalue = 2.73) and carried in-
(1:50.000 scales) produced by IBGE (available at http://www.ibge.
formation about vegetation structure and disturbance conditions
gov.br/home/geociencias/download/arquivos/index1.shtm). We
of the sample points in riparian forests. Therefore, higher values
identified riparian forests by then using this map to produce images
can be interpreted as well-protected old-growth forest and lower
containing buffers 15-m wide beside each stream and 200 m around
values as young-regenerating forests or forest growing under reg-
the Furnas water reservoir.
ularly disturbed conditions. Prior to further using this variable, we
Next, we divided the region in 102 landscape hexagons (each
transformed it with Min-Max-Scaling means (Mf- min Mf)*100/
landscape area = 500 ha) and measured patch and landscape met-
(max Mf – min Mf) to standardize the range of values between
rics (patch size, shape, riparian forest width, and percentage of for-
0 and 100. After this, we subtracted 100 from each value. Thus,
est cover) for each fragment lying within these landscapes. We used
disturbance index = 100% is the most disturbed condition and
these metrics to select 24 riparian forests that varied widely in riparian forest width and local and landscape contexts. Those selected
disturbance index = 0% is the least impacted condition. NDVI mean: We used dry season (cloud-free) Landsat TM 5 2011
sites ranged from 30 to 1,420 m in width considering both sides of
images to calculate the normalized difference vegetation index
the stream, from 0.88 to 496.18 ha in patch size from 3% to 67% in
(NDVI) for each riparian forest cell (30 m resolution) using the for-
forest cover, presented a similar forest shape (we excluded complex
mula: NDVI = (NIR − red)/(red + NIR) (Marabel & Alvarez-Taboada,
forms or riparian corridors connected to other forest fragments),
2013). Then, we used zonal statistical analysis to calculate NDVI
|
5
MAURE et al.
values for each riparian study forest. NDVI is sensitive to photo-
indicate that riparian forest is composed by a mosaic of patches at
synthetically active biomass and is positively correlated with plant
different successional stages, whereas narrow range of NDVI point
productivity (Pettorelli et al., 2005). NDVI values range from −1.0
to a greater homogeneity in photosynthetic activity within areas.
(indicate nonvegetated surface features such as water, barren
NDVI coefficient β: We created this index to express the “historical
rock) to 1.0 (maximum green vegetation). Thus, the maximum val-
dynamic” or “the resilience” of the riparian forest as the slope of
ues are indicative of riparian forests possessing fully recovered
the curve (coefficient β) of NDVI per unit time (i.e., per year). Here,
above-ground biomass levels (old-growth forests), while lower
high positive values mean that riparian forests possess periods of
values indicate low values for, or absence of, photosynthetic ac-
higher growth rates, while low values indicate periods of lower
tivity due to the existence of canopy gaps, young regrowth, or
growth rates or higher stability in primary productivity (Viedma,
extensive edge effects.
Meliá, Segarra, & García-Haro, 1997). Negative values represent
NDVI range: Within a given riparian forest, spatial and biological dis-
forest degradation or loss over the time. To calculate this index,
turbance can create a mosaic of patches at different successional
we used 10 Landsat Tm 5 dry season images from 1985 to 2011.
stages, implying spatial variation in the timing or intensity of distur-
First, we used zonal statistical analysis to calculate mean normal-
bance, and consequently varying capacities for ecosystem services
ized difference vegetation index (NDVI) values for each riparian
delivery (Ferraz et al., 2014). To represent this spatial pattern, we
forest in each image. Then, we applied linear regression between
used NDVI range values for each riparian forest for 2011. This index
the mean NDVI values by year and estimated the slope of the
expresses the spatial heterogeneity of riparian forest in terms of
curve of this relationship for each riparian forest site (Supporting
plant productivity (i.e., above-ground biomass). Thus, higher values
Information Figure S1).
TA B L E 1 List and description of traits used to calculate predictive variables (Functional Redundancy, Functional Evenness, and Functional Richness) and define functional group richness
a
Ecological relevance to ecosystem process
Trait
Scale
Description
Source
Diet (seven items)
Continuous
Based on the percentage contribution of each food item to the total dietary records for the species: seeds, fruits, nectar, other plant material, scavenging, invertebrates, and vertebrates
Percentage of each food category defines an important niche dimension and can reflect the niche breadth in terms of specialist and generalist. From the percentage food category, it possible to infer to which ecological processes each species is most likely to be linked (i.e., seed dispersal, pollination, removal of carcasses, controlling invertebrates, regulation of vertebrates
Wilman et al. (2014)
Foraging strata (three items)
Continuous
Indicates whether foraging stratum estimates are based on species level data: ground, understory, and canopy
Foraging strata relate to the location of resource acquisition
Wilman et al. (2014)
Body mass
Continuous
Based on the average of adult body mass (g)
Ecologists think about diet niche breadth in terms of prey size range, and the general pattern observed was that prey size tends to be directly proportional to the size of the predator, both within and between species
Wilman et al. (2014)
Dependence on forested habitat a
Categorical
Response trait to habitat loss and fragmentation (categories: high, medium, and low forest dependence)
Forest dependence reflects environmental tolerances, habitat, or ecological preferences of bird species and, consequently, relates to the location of resource acquisition (Violle et al., 2007). In addition, the trait attribute varies in response to changes in habitat loss and fragmentation
http://datazone. birdlife.org/home
Dependence on forested habitat: we used habitat specialization as one of the traits to calculate overall species functional diversity. For habitat specialization groups, we used this trait a priori to define each group and not to calculate functional diversity.
|
MAURE et al.
6
2.4 | Bird sampling
Functional Richness (FRic): FRic values represent the niche space occupied by the species present in a community. This space is a
We used point counts to sample birds (Develey, 2003). We selected
convex hull volume defined by linked extreme values of species
three stream-side point count locations in each riparian forest, with
traits. From this, an algorithm calculates the volume inside the hull
each point separated by at least 200 m. Each sampling location
(Supporting Information Table S2). Thus, communities composed
was visited once between 2013 and 2017, and we defined as the
of species with similar functional traits have lower FRic values
sampling area a zone of 30-m radius around each point. Sampling
(Villéger et al., 2008).
observations occurred always in the three first hours after sunrise and consisted in recording all individuals seen and/or heard inside
We also calculated Functional Group Richness (FGR), which
the 30-m buffer during 10 min of observation at each point. We ex-
represents the number of functional groups per riparian forest
cluded Trochilidae (hummingbirds) from our samples because of the
based on a dendrogram of species traits. We used Gower distances
difficulty of identifying them using point counts. We calculated the
(15 height) to cluster species in this dendrogram (Supporting
relative abundance of each species by adding the number of all indi-
Information Figure S2, Laliberté & Legendre, 2010). Although the
viduals observed (avoiding double counting of the same individuals)
choice of cut-off point in the dendrogram depends on many ana-
in the three point locations of each riparian forest.
lytical decisions and may be arbitrary, we reduced this arbitrari-
For each sampled species, we compiled information of morpho-
ness by checking each functional group a posteriori. We observed
logical traits and foraging attributes (Table 1) that are related to bird
that species with traits values distance lower 15 height are asso-
species function in the ecosystem (Flynn et al., 2009). We chose
ciated with similar ecosystem functions (Supporting Information
traits reflecting habitat preferences and resource use requirements,
Table S2).
including body size, diet, foraging habitat, and location (Wilman et al., 2014; Birdlife database available at http://www.birdlife.org/ datazone/species/search). We define a functional group as a set of species that respond similarly to a particular habitat condition.
2.6 | Data analyses and modeling To analyze the relationship of FD metrics with riparian habitat and vegetation variables, we tested the relationships of depend-
2.5 | Response variables We used two matrices, containing the relative abundance of bird
ent (FRed, FEve, FRich), and predictive, variables (riparian forest width, disturbance index, NDVI mean, NDVI range, and NDVI Coefficient β) using simple linear models (lm R function) (R Core
species sampled in each riparian forest, and the key functional
Team 2014), and selected the best predictor variable based on
traits (Table 1) of those species, to calculate three components of
Akaike information criterion (AIC, see below) (Bolker, 2008).
Functional Diversity—Functional Redundancy, Functional Evenness,
Then, we looked for nonlinearity in the relationships among
and Functional Richness. We calculated all indices in R software (R
Functional Diversity components and riparian vegetation char-
Core Team 2014) using the “picante,” “FD,” and “ade4” packages. We
acteristics by comparing linear, saturating (monomolecular), and
used these three components to relate Functional Diversity with
null models using AIC values. Linear models show an increasing
environmental predictors because they did not show strong auto-
or decreasing constant trend in the relationships between de-
correlations between them (Spearman correlation, r
