Time and matrix quality increase the relative habitat value of smaller patches in fragmented landscapes
Data files
Jan 17, 2025 version files 21.35 MB
-
alldat.RData
82.87 KB
-
df.lst85.RData
129.17 KB
-
mod.std.RData
4.53 KB
-
README.md
8.45 KB
-
source_metadata_R1.csv
30.83 KB
-
topmod
10.85 MB
-
topmreg
10.24 MB
Abstract
The landscape-scale impacts of habitat subdivision (fragmentation per se) on biodiversity are not fully resolved. While smaller patches usually contain more species for equal total area, many implications of this remain unclear. For example, do equal areas of smaller and larger patches provide equivalent habitat value? How might this change over time and under differing matrix conditions?
To help address these knowledge gaps, we developed an indicator of relative habitat value based on a species-individual null model. We used the indicator to compare sets of patches ordered in small-to-large and large-to-small orders, building hierarchical Bayesian regression models to test the role of time since patch creation and contrasting matrix conditions. This allowed us to assess habitat value for 85 metacommunities inhabiting fragmented landscapes (1354 patches, > 4500 species). We expected comparable habitat value following patch creation due to unpaid extinction debts, and that the matrix would determine the direction of change over time, a harsher matrix increasing the relative value of larger patches.
Averaged over time and matrix quality, the indicator probability density was mostly negative, suggesting slightly greater habitat value among larger patches. This pattern was consistent across taxonomic groups, although amphibians and reptiles were the most affected, and invertebrates least so.
Larger patches were of greater value within 20 years of patch creation, but over time smaller patches increased in value under any matrix type, whereas value remained constant for larger patches. Matrix conditions mediated the relative difference: after 100 years under a light matrix, patches of all sizes were essentially of equivalent value, while larger patches were still favored under a harsh matrix.
Policy implications. In long-fragmented, light-matrix landscapes, small and large patches apparently offer comparable per-unit-area habitat value. Conversely, under a harsh matrix, larger patches retain slightly greater habitat value. We speculate that this reflects a ‘colonization credit’ which might occur within smaller patches after the initial loss of fragmentation-sensitive species in disturbed landscapes. Overall, analyses support the need to maintain and enhance total habitat area – regardless of configuration – for biodiversity conservation, especially in long-fragmented landscapes with light matrices.
README: Data from Time and matrix quality increases the relative habitat value of smaller patches in fragmented landscapes
https://doi.org/10.5061/dryad.8sf7m0czn
Description of the data and file structure
Raw data for the study were all obtained from online databases or other published sources and consisted of 85 studies from fragmented landscapes i.e., isolated patches of the original natural habitat that are now surrounded by human land uses. The areas dominated by human land uses in fragmented landscapes are referred to as the 'matrix', and this is thought to affect the diversity of species found in the remaining fragments of natural habitat. Understanding how the nature of the matrix, and the time since the landscape became fragmented affect the diversity of species in fragments of different size was the aim of the study.
Each study included information on the quality of the matrix (i.e., how strongly the human uses dominating the landscape differed from the native habitat fragments), the time since the fragmented landscape was created (typically when the landscape was first cleared), the number of individuals of each species observed in each fragment and the area of each fragment.
Data from each study were first arranged in a consistent wide-format dataset. Rows held information about individual fragments, while columns provide the area of the fragment (in hectares) in the first column, with all other columns giving the number of individuals for each species found in that fragment.
Analysis proceeded first via a pre-processing stage, where the 85 individual datasets (provided in df.lst85.RData) were analysed twice, first combining fragments in small-to-large order and then repeated in large-to-small order. The results of this stage of analysis for all datasets were then collated into a single long-format dataframe (alldat.RData) retaining information on the order fragments were combined and the identity of the study. This dataframe was used in modelling. Two models were fit using these data and the model objects are included. Model object 'topmod' was used to infer the effects of the matrix and time on the relative diversity of the fragments in small and large fragments, while model object 'topmreg' modelled only the effect of fragment size without accounting for other predictors.
Also included is a summary file (mod.std.RData), which gives a summary of the results at the level of the study, rather than at the level of individual fragments used in modelling, and a file on the sources of information (metadata.csv).
All analyses used the R programing language (as detailed in the R script stored at the Zenodo repository). We have included our raw data (df.lst85.RData) d six different data files five of which are specific to the R environment (3 x .RData files, 2 x model objects) and a .csv giving metadata on the data sources. Detailed description of these data files and where they fit within the analysis workflow are presented below.
Files and variables
File: alldat.RData
Description: An R data object in the form of a data frame, where each row gives the value for the response variable in all 2538 possible unique combinations of fragments in small-to-large and large-to-small order. Contains post processed data based on the file df.lst85.RData. These data were used in regression modelling.
Variables in alldat.RData:
- "rd" = residual deviation, the response variable
- "ord" = order of fragment combination: stl is small-to-large; lts is large-to-small
- "df" = dataframe, a unique indicator of the source data
- "patch" = name of fragment (not used)
- "ind" = number of individuals observed
- "paper" = code for the source paper from which the study derived
- "taxa" = broad taxonomic group (levels: "amphibians & reptiles", "birds","invertebrates","mammals", "plants")
- "ftime" = time since fragment (patch) creation (levels: <20 years, 20-100 years, > 100 years)
- "mathost" = matrix hosility (quality) (levels: light, moderate, harsh)
- "npatcomb" = number of patch combinations in small-to-large or large-to-small order
- "numrat" = numerical ratio (alternative response, not used)
- "lograt" = natural logarithm of the numerical ratio (also an alternative response, not used)
- "nsamp" = number of samples in original dataset
- "nspp" = total number of species in original dataset
- "ord_df" = indicator concatenating the two orders of combination (stl, lts) and the dataframe
- "se_ord" = standard error calculated for each dataset in each order (alternative covariate not used in model)
- "nsamp.c" = centred number of samples (not used)
- "nspp.c" = centred number of species (not used)
File: df.lst85.RData
Description: An R data object in the form of a list. Each element in the list is a sites x species data frame (n = 85) which gives the abundance of species sampled in each patch. Note patch area is given in the first column. Names of each dataset in the list reference the 'source' field of the metadata file (source_metadata_R1.csv)
File: mod.std.RData
Description: An R data object in the form of a data frame, giving metadata on each study system.
Variables in mod.std.RData:
- "source" = coded name of the study (see metadata file)
- "nsite" = total number of sites (fragments) in the dataset
- "nsamp" = total number of samples in the dataset (most fragments have > 1 sample)
- "nspp" = total number of species in the dataset
- "rd" = mean value of the response variable delta RD
- "paper" = primary source from which data were obtained
- "taxa" = as per alldat
- "timefrag" = time since fragment (patch) creation (levels: <20 years, 20-100 years, > 100 years)
- "mathost" = matrix hostility
- "biome" = Type of vegetative biome where data were collected (e.g. grassland, shrubland, forest).
- "ftime" = time since fragment (patch) creation (levels: <20 years, 20-100 years, > 100 years)
- "df" = dataframe
- "ms_sd" = standard deviation in rd for that source
- "res" = results of whether smaller (stl) or larger (lts) fragments were more diverse or there was no evidence of an effect of order (NoDiff)
- "bc" = combination of time and matrix used as a 'best case' scenario
- "wc" = combination of time and matrix used as a 'worst case' scenario
File: topmod
Description: An R data object created by R package brms, which contains the model output for the multivariate regression model. This object can only be used within the R environment using brms and other packages designed to work with the model.
File: topmreg
Description: An R data object created by R package brms, which contains the model output for the meta-regression model. This object can only be used within the R environment using brms and other packages designed to work with the model.
File: source_metadata_R1.csv
Description: A comma separated values file giving the name and source publication for each of the datasets used in analysis.
Variables in source_metadata_R1.csv
- source: a unique identifer for the different datasets
- description: a broad description of the ecological context for each study.
- citation: Full publication details from which data were obtained.
- country: Country where data were collected.
- climate: Broad climatic zone where data were collected.
- continent: continent where data were collected.
- biome: Type of vegetative biome where data were collected (e.g. grassland, shrubland, forest).
- taxa: Broad taxonomic group that was the focus of the study (e.g., birds, mammals).
- time.since.fragmentation: A three level factor variable describing how long since fragments were created (<20 years, 20-100 years, >100 years).
- Matrix.category:A three level factor variable describing how different the habitat is with respect to the landscape matrix (harsh, intermediate, light).
- LatN: latitude of centroid of study location (datum: WGS84)
- LongE: longitude of centroid of study location (datum: WGS84)
Code/software
All R data objects require the freely available programming language R and code to run the analyses is provided. These files were created using R version 4.4.1.
Access information
Other publicly accessible locations of the data:
- FragSAD database.
Data was derived from the following sources:
- Primary literature (publication details are given in the metadata).
Methods
This dataset was derived from 85 published datasets, which were processed using a common methodology.