Dataset DOI: 10.5061/dryad.z08kprrsx

Description of the data and file structure

This dataset supports the article "Landscape configuration shapes bird–plant multilayer networks in urban green spaces: Implications for urban biodiversity management".

This dataset was collected to investigate bird–plant interactions and seed dispersal processes in urban green spaces in Tokyo. Bird fecal samples were analyzed using DNA metabarcoding, and seed dispersal events were identified through field surveys combined with Sanger sequencing. The resulting interaction matrices were used to construct multilayer bird–plant networks, quantify network metrics, and assess their relationships with landscape configuration at multiple spatial scales.

The dataset contains study-generated sequencing reads, processed bird–plant interaction matrices, seed identification records, bird-node network metrics, landscape configuration tables at three spatial scales, a GenBank reference accession manifest, and an R Markdown analysis script. Bird trait data from AVONET and reference nucleotide sequences from GenBank are accessed through their original data sources and are not included among the deposited files.

Files and variables

The descriptions below list all included data files along with their associated variables. Missing values are indicated as blank cells or “NA” where applicable.

File: feces_data.csv

Description: Fecal metabarcoding read-count table aggregated by sampling plot and bird species and proportionally scaled to approximately 10,000 assigned reads per plot–bird group when the unscaled group total exceeded that threshold.

Variables:

• Plot – Site/plot code (categorical identifier; unitless)
• Name – Bird species name (taxonomic identifier; unitless)
• sum_Genus – Normalized assigned-read count for the specified plant genus (e.g., sum_Prunus or sum_Cornus; read-count equivalents per plot–bird group; target maximum approximately 10,000 after rounding).

File: birds_multidegree.csv

Description: Bird-node multidegree values calculated from the overall multilayer bird–plant interaction network.

Variables:

• Species1 – Scientific name of the bird species (taxonomic identifier; unitless).
• multidegree – Node degree of each bird species in the overall multilayer network, calculated as the sum of distinct links across the foraging and seed-dispersal layers (number of links; count).

Note: Trait-based analyses use species-level AVONET data obtained from the original source and joined by bird scientific name (see “External data sources and licensing”).

File: feces_data_matrix_cleaned.csv

Description: Cleaned fecal interaction matrix used in the construction of overall bird–plant networks.

Variables:

• Name – Bird species name (taxonomic identifier; unitless)
• Plant genus columns (95 columns, e.g., Prunus, Cornus, Zelkova) – standardized integer-valued sequence-read abundance weights (read-count equivalents). Each of the 12 bird-species rows totals 20,000 across plant genera. This overall interaction matrix is provided as the input for bird–plant network analyses.

Files: feces_data1.csv / feces_data2.csv

Description: Processed plant-genus read-count tables from the two sequencing runs (distinct from the raw FASTQ read files).

Variables:

• Plot – Site/plot code (categorical identifier; unitless)
• Name – Bird species name (taxonomic identifier; unitless)
• Plant genera columns – Counts of sequencing reads assigned to the corresponding plant genus (reads in the corresponding input record).

File: multilayer_network_metrics.csv

Description: Summary statistics for multilayer bird–plant networks by site.

Variables:

• Site – Site code (K1–K13; categorical identifier, unitless)
• Density – Network density (dimensionless)
• APL – Average path length (number of network steps; dimensionless)
• Edges – Total number of links (count)
• Centrality – Degree centralization calculated from the dietary layer with igraph centr_degree() in the deposited R script (dimensionless index)
• Modularity – Modularity value (dimensionless)

File: plant_centrality_metrics.csv

Description: Node centrality metrics for plant genus in the multilayer network.

Variables:

• Plant – Plant genus (taxonomic identifier; unitless)
• WeightedDegree – Weighted node strength calculated by summing edge weights from the two layers after each interaction matrix was separately min–max scaled (dimensionless weighted sum)
• Degree – Unweighted node degree (number of links; count)

File: seed_data_matrix_cleaned.csv

Description: Cleaned seed-dispersal interaction matrix.

Variables:

• Birds – Bird species name (taxonomic identifier; unitless)
• Plant genus columns – Numbers of bird-dispersed seeds by plant genus (count); this processed matrix is provided as input to the seed-dispersal network analyses.

File: seed_data.csv

Description: Raw seed dispersal event records.

Variables:

• Area – Sampling site identifier (categorical; unitless)
• Birds – Disperser species name (taxonomic identifier; unitless)
• Plants, Plants2 – Identified seed taxa (taxonomic labels; unitless)

Files: configuration500.csv / configuration1000.csv / configuration1500.csv

Description: Landscape configuration tables for 13 parks (K1–K13), each with 19 columns. They contain identical site-level network metrics and park attributes, along with scale-specific landscape variables for buffers extending 500 m, 1000 m, or 1500 m beyond each park boundary. Filenames used in the R script omit export timestamps.

Variables:

• ID – Site code (categorical identifier; unitless)
• Park – Park name (text identifier; unitless)
• APL, Centrality, Modularity, Density – Network indices (dimensionless; APL expressed as average number of steps between nodes)
• NDVI – Park-level normalized difference vegetation index (dimensionless)
• Dis.Center – Distance to the city center (km)
• Area – GIS-delineated park area (m²). The recorded values range from 117,616.63 to 1,318,770.33 m² across the 13 parks.
• A.Shanoon – Park-level Shannon diversity index of birds (dimensionless; original column name retained for compatibility with the analysis script).
• P.num – Number of plant taxa recorded at each site (count)
• NL_500 / NL_1000 / NL_1500 – Annual mean VIIRS nighttime-light radiance in 2023 for each buffer (nW·cm⁻²·sr⁻¹).
• Num.H – Annual visitor traffic intensity per unit green-space area, based on park-management records (people·ha⁻¹·year⁻¹).
• NDVI_500 / NDVI_1000 / NDVI_1500 – NDVI calculated for each buffer scale (dimensionless)
• GP_500 / GP_1000 / GP_1500 – Proportion of green-space area within each buffer (%)
• LPI_500 / LPI_1000 / LPI_1500 – Largest patch index of green-space patches (% of buffer landscape occupied by the largest patch).
• PD_500 / PD_1000 / PD_1500 – Green-space patch density (patches per 100 ha of buffer landscape).
• ENN_MN_500 / ENN_MN_1000 / ENN_MN_1500 – Mean edge-to-edge Euclidean nearest-neighbor distance among green-space patches (m).
• AI_500 / AI_1000 / AI_1500 – Green-space patch aggregation index (%; 0 = most disaggregated, 100 = maximally aggregated according to the index).

File: seeddispersalnetwork_revised_CC0.Rmd

Description: R Markdown workflow for sequence processing, taxonomic assignment, fecal-data normalization, multilayer network construction, bird-trait generalized linear models, multiscale landscape modeling, β-diversity, PERMANOVA, dbRDA, and figure generation. The script reads birds_multidegree.csv and the three configuration*.csv files. External AVONET traits, rbcL reference files, and local sequencing read directories are supplied through configurable paths.

External data: rbcL reference FASTA files

Taxonomic assignment used two custom rbcL reference FASTA files prepared from GenBank records and obtained separately from this dataset:

• rbcL_final_2.fasta – locally prepared reference file used with DADA2 assignTaxonomy().
• rbcL_final.fasta – locally prepared reference file used with DADA2 addSpecies().

For taxonomic assignment, the R Markdown script reads these files from the RBCL_ASSIGN_FASTA and RBCL_SPECIES_FASTA paths (default filenames: rbcL_final_2.fasta and rbcL_final.fasta).

File: rbcL_reference_accessions.csv

Description: Reference manifest listing 54,408 unique GenBank accession identifiers, reference fragment lengths, and checksums associated with the rbcL reference library. The two reference FASTA files used the same sequence collection with distinct taxonomic headers. This manifest contains metadata only, without nucleotide sequences.

Variables:

• Reference_order – 1-based index of a record in the original reference library (integer; unitless).
• GenBank_accession – Original GenBank identifier (text accession; unitless).
• Reference_length_bp – Length of the locally curated reference fragment (base pairs, bp).
• Reference_SHA256 – SHA-256 checksum of the uppercase nucleotide string without header or line breaks (hexadecimal digest; unitless).

Most curated reference fragments were 334 bp long. The accession identifiers provide a route to the source records, while fragment lengths and checksums enable comparison with locally prepared reference sequences. Taxonomic reassignment from raw reads uses the corresponding curated reference FASTA files. Source: https://www.ncbi.nlm.nih.gov/genbank/about/.

These reference databases are distinct from study-generated raw paired-end metabarcoding reads, which are deposited separately in NGS_Rawdata.zip and also available from DDBJ under DRA023607.

File: NGS_Rawdata.zip

Description: This archive contains raw paired-end next-generation sequencing reads generated from fecal plant metabarcoding. The archive contains 944 gzipped FASTQ files (.fq.gz) and does not include nested subfolders.

The raw reads correspond to fecal samples collected from 13 urban green spaces in Tokyo and were used for ASV inference, taxonomic assignment, and downstream construction of bird–plant foraging interaction networks.

File naming convention:

FASTQ filenames include site identifiers, survey-round identifiers, sample identifiers, sequencing project or run identifiers, lane information, and paired-end read direction.

For example:

K1_1_S3_FKDL240023550-1A_HW5GNDRX3_L2_1.fq.gz

K1_1_S3_FKDL240023550-1A_HW5GNDRX3_L2_2.fq.gz

K1 – site identifier. In this example, K1 represents the first urban park site.

K1_1 – site and survey-round identifier. In this example, K1_1 represents the first survey round at site K1.

S3 – sample identifier. In this example, S3 represents sample 3 from that site and survey round.

FKDL240023550-1A – sequencing project or order identifier.

HW5GNDRX3 – sequencing run or flow-cell identifier.

L2 – sequencing lane.

_1.fq.gz – forward read file.

_2.fq.gz – reverse read file.

Files with the same prefix but ending in _1.fq.gz and _2.fq.gz form a paired-end read pair for the same sample.

MD5.txt – MD5 checksums for all 944 FASTQ files, provided to allow users to verify file integrity after download.

Related public sequence archive:

Metabarcoding reads are also available from the DDBJ Sequence Read Archive under accession number DRA023607. Sequence data have been deposited in the DDBJ database under accession numbers LC902028–LC902538.

Code/software

All analyses were conducted in R version 4.3.1 (R Core Team, 2024).

The following open-source R packages were used in the workflow:

DADA2 – sequence quality assessment, filtering and trimming, error-rate learning, ASV inference, read merging, chimera removal, and taxonomic assignment
phyloseq – OTU/ASV object construction and rarefaction
dplyr, tidyr, purrr, tibble, stringr, tidyverse – data wrangling, reshaping, iteration, and table assembly
igraph – network construction, plotting, and node-level degree/strength analyses
multinet – multilayer network construction and metric calculation
betalink – estimation of β-diversity among multilayer interaction networks
vegan – PERMANOVA and dbRDA
ggplot2, ggrepel, patchwork – figure generation, annotation, and plot assembly
BiMultiNetPlot – alternative multilayer bipartite network visualization
broom – model tidying and coefficient extraction
car – multicollinearity diagnostics
MASS, MuMIn – stepwise model selection and AICc-based comparison

Workflow

Sequence processing – Raw FASTQ files from two metabarcoding runs were processed separately in DADA2, including quality assessment, filtering/trimming, error-rate learning, ASV inference, read merging, sequence-length filtering, chimera removal, and read-count tracking.

ASV integration and taxonomic assignment – Non-chimeric sequence tables from the two runs were merged, and taxonomy/species assignment was performed using custom rbcL reference files prepared locally from external sources (not deposited). A phyloseq object was then constructed and rarefied to 10,000 reads per sample, yielding abundance and taxonomy tables.

Fecal-data processing – Feeding interaction data from feces_data1.csv and feces_data2.csv were combined by plot and bird species. Plot–bird totals above 10,000 were scaled to approximately 10,000 read-count equivalents for feces_data.csv. The separately standardized overall matrix feces_data_matrix_cleaned.csv contains 20,000 read-count equivalents per bird species and is provided as input to the overall network analysis.

Site-level interaction matrices – Feeding and seed-dispersal interaction matrices were generated separately for each site and reformatted for multilayer network construction.

Site-level multilayer networks – Site-specific multilayer networks linking foraging and seed-dispersal layers were constructed with multinet, and network metrics including density, average path length, edge number, centralization, and modularity were calculated.

Network visualization and overall network assembly – Representative multilayer network plots were generated, and cleaned feeding and seed-dispersal matrices were consolidated to construct the overall multilayer interaction network.

Node importance and trait modeling – Plant and bird node-level degree or multidegree metrics were calculated from the overall network, and bird multidegree was analyzed against morphological and ecological traits obtained separately from the AVONET data source using generalized linear models, stepwise model selection, and coefficient visualization. The analysis requires matching species names to externally sourced traits.

Multiscale landscape modeling – Landscape tables at 500 m, 1000 m, and 1500 m were analyzed by comparing three predictor sets (land, land+park, and land+urban). For each network metric, the best model and spatial scale were identified using stepwise selection and AICc comparison against null models, and VIF diagnostics were calculated.

β-diversity and ordination – Multilayer β-diversity matrices among sites were estimated with betalink, and relationships with multiscale landscape predictors were evaluated using PERMANOVA and visualized using dbRDA.

Provided scripts

seeddispersalnetwork_revised_CC0.Rmd – Annotated R Markdown workflow for the analyses and figures. The trait and taxonomic-assignment stages use locally obtained AVONET data and rbcL reference files, respectively.

Note: Adjust file paths to your local environment before running.

External data sources and licensing

The deposited research data are released under Dryad's CC0 waiver. AVONET species-level trait values and GenBank reference nucleotide sequences are obtained from their original sources and are not included in the archived files. The original sources are cited below.

• AVONET: Tobias, J. A., Sheard, C., Pigot, A. L., et al. (2022). AVONET: morphological, ecological and geographical data for all birds. Ecology Letters, 25, 581–597. https://doi.org/10.1111/ele.13898. Original data: https://figshare.com/s/b990722d72a26b5bfead (CC BY 4.0). For trait analyses, the script matches the AVONET species-mean table to birds_multidegree.csv by Species1. The original GLM uses AVONET Mass (g), Beak.Length_Culmen (mm), Beak.Width (mm), Habitat.Density (unitless categorical code), and Primary.Lifestyle (unitless categorical class). Other AVONET morphology variables, when used outside the deposited dataset, are measured in mm except body mass (g); Hand-Wing.Index is dimensionless.
• NCBI GenBank: Public sequence access and usage information: https://www.ncbi.nlm.nih.gov/genbank/about/. Accession identifiers and reference-fragment hashes are provided in rbcL_reference_accessions.csv; source nucleotide sequences can be accessed from GenBank.
• Study-generated sequence records: Raw paired-end sequencing reads are archived in this dataset and independently available from DDBJ DRA023607; study sequence accession range LC902028–LC902538 is documented above. These are separate from external GenBank-derived reference sequences.
• Landscape and radiance units: The landscape processing methods described in Tang and Hirao (2026, Urban Forestry & Urban Greening, https://doi.org/10.1016/j.ufug.2026.129357) identify the landscapemetrics R package and 2023 VIIRS Nighttime Lights. Metric unit definitions: https://r-spatialecology.github.io/landscapemetrics/ ; VIIRS band documentation: https://developers.google.com/earth-engine/datasets/catalog/NOAA_VIIRS_DNB_ANNUAL_V22.

Data availability and reproducibility

All deposited files and their variables are described above. To run bird-trait analyses, obtain the AVONET species-mean table from its original source and set the AVONET_TRAITS_CSV environment variable to the local CSV path; the script joins traits to birds_multidegree.csv by species name. For sequence-based taxonomic assignment, prepare the corresponding rbcL reference FASTA files using the GenBank source accessions, and configure RBCL_ASSIGN_FASTA and RBCL_SPECIES_FASTA. The deposited reference manifest supports provenance and fragment comparison; it is not a replacement for the curated FASTA files. Local FASTQ directories and output paths can be configured in the script. The processed fecal and seed-dispersal matrices are provided for direct use in network analyses.