Multi-species occupancy model for investigating responses of amphibian metacommunities to environmental drivers in ponds along an urban-rural gradient in Hungary
Data files
Nov 05, 2026 version files 78.97 KB
-
Hamer_et_al_Ecol_Appl_detection_occupancy_model_1.R
14.39 KB
-
MSOM_occupancy_covariates.csv
9.56 KB
-
Multispecies_occupancy_models_species_specific_estimates.xlsx
36.97 KB
-
README.md
8.31 KB
-
species_data.zip
4.41 KB
-
survey_covariates.zip
5.33 KB
Abstract
Multiple stressors in urban ecosystems often generate complex and non-linear responses in ecological communities. However, empirical metacommunity studies across urban-rural gradients commonly focus on linear relationships, potentially oversimplifying species–environment relationships, which eventually might lead to overlooking important thresholds. We examined amphibian metacommunities at 100 ponds along an urbanisation gradient in central Hungary (spring and summer 2023), using multi-species occupancy modelling (MSOM) with Bayesian inference. We used a combination of linear, quadratic and interactive terms in four different models at local and landscape scales. We used covariates for the detection sub-model that were influential in an earlier study of the same 100 ponds. This dataset contains the site- and survey-specific covariate data that is required to run the occupancy and detection sub-models, as well as the MSOM code itself. The data file in XLSX format contains the parameter estimates for occupancy (α) and detection (β) covariates for the nine amphibian species obtained from four multi-species occupancy models. Occupancy increased with pond area, water level fluctuation (hydroperiod), aquatic vegetation cover, and decreased in ponds with fish. Urban land cover had a strong negative effect on occupancy. Surprisingly, occupancy increased with pond isolation. Our results highlighted the complexity in amphibian-habitat relationships in urban landscapes. We recommend incorporating these complex drivers of amphibian metacommunities into principles for urban pond management and conservation planning.
Dataset DOI: 10.5061/dryad.0vt4b8hfn
Description of the data and file structure
This dataset enables the user to obtain parameter estimates for relationships between the probability of occupancy of nine amphibian species at 100 ponds in Hungary, and eleven environmental covariates. The data is based on field surveys conducted in 2023 over a single breeding season (spring/summer) using three survey methods to detect all life stages: (1) call/ visual encounter surveys (VES), (2) dip-netting, and (3) Dewsbury newt traps. The number of dip-net sweeps and traps at a pond was pre-determined to be proportional to the pond surface area. Amphibians were identified and released at the point of capture. Environmental and climatic conditions were recorded during each survey that comprised site- and survey-specific covariates. Landscape-scale variables were generated using QGIS. Species data, and survey- and site-specific covariates, were entered into the model code to estimate parameter estimates and 95 % Bayesian credible intervals.
We used Multi-species Occupancy Models (MSOM) with Bayesian inference to estimate the probabilities of occupancy at the community and individual species levels. Models were used to examine relationships between the probabilities of occupancy of any life stage (i.e., eggs, larvae, juveniles or adults) and eleven covariates: (1) Distpond = distance to the nearest pond; (2) Area = pond surface area; (3) Hydro = hydroperiod (% water level fluctuation); (4) Aqveg = aquatic vegetation cover; (5) Fish = fish presence; (6) Urban_1000 = proportion of urban land cover within a 1000-m radius of a pond; (7) Canopy = proportion cover of canopy over a pond; (8) EC = electrical conductivity; (9) Forest_1000 = proportion of forest cover within a 1000-m radius of a pond; (10) Wetland_1000 = proportion of wetland cover within a 1000-m radius of a pond; (11) Chl_a = chlorophyll-a fluorescence.
For the detection sub-model, we used covariates that were previously found to have influential relationships with the probability of detection (see Hamer and Horányi, 2024): (1) Days = number of days since 19 March 2023 (survey day 0); (2) Days2 = quadratic effect of Days; (3) Effort = survey effort (number of minutes conducting call/ visual encounter surveys); (4) Temp = water temperature; (5) Fish = fish detected during a survey; (6) Emerg = % cover of emergent vegetation. We did not include wind speed, rain or water levels as detection covariates as they were not influential in the original detection only model.
The dataset contains individual species detection data saved as comma delimited (.csv) files. Each of the three survey columns in each species file is detection of the species using Visual Encounter Surveys and aural census (VES), dip-netting and newt trapping techniques combined. Detection of a species includes detection of either eggs, larvae or post-metamorphic individuals. The abbreviations of the individual species and survey-specific covariates are listed below. Further information on the study design, sampling method and modelling is provided in the methods section of the associated manuscript.
The occupancy sub-model contains code for the four individual sub-models that align with the hypotheses presented in Table 1 of the article in Ecological Applications.
The means, standard deviations and the 2.5th and 97.5th percentiles of the posterior distributions of the model coefficients were estimated from the MSOM, which represents 95% Bayesian credible intervals (BCI). Survey covariates consisting of continuous data were standardised (mean = 0, SD = 1) so that the relative importance of each covariate could be determined according to the magnitude of the coefficient. We modelled predictive relationships for influential covariates while holding the other covariates in the model at their mean values.
A summary of the parameter estimates for relationships between the probability of occupancy (α) and detection (β), and the eleven occupancy and six detection covariates for nine amphibian species from four separate multi-species occupancy models is included in this submission as an XLSX data file (Multispecies_occupancy_models_species_specific_estimates.xlsx).
Files and variables
File: Hamer_et_al_Ecol_Appl_detection_occupancy_model_1.R
Description: model code to run the Multi-species occupancy model
File: species_data.zip
Description: detection histories for nine species at 100 sites. Species data files correspond with the following species:
bombom = Bombina bombina; bufbuf = Bufo bufo; bufvir = Bufotes viridis; hylarb = Hyla arborea; lisvul = Lissotriton vulgaris; pelfus = Pelobates fuscus; pelcom = Pelophylax spp. complex; randal = Rana dalmatina; tridob = Triturus dobrogicus.
The columns "survey1", "survey2", "survey3" correspond with detection data from the three surveys (1 = detected; 0 = not detected).
Cells with "NA" at site T9 for survey1 correspond to no survey being conducted at that site during that particular survey.
File: survey_covariates.zip
Description: survey covariates recorded during each of the three surveys at the 100 sites. The column "site" corresponds to the site names of the 100 ponds surveyed. Individual survey-specific covariate data is saved as .csv files. The covariates include:
- days = number of days since 19 March 2023
- effort = survey effort expressed as minutes conducting VES
- water temp = water temperature (°C) recorded at the start of a survey at the shoreline
- fish = detection of fish from visual surveys, dip-netting and newt traps (binary, 0/1)
- emerg = estimated % cover of emergent vegetation over the pond surface
Continuous covariates (except fish) were standardised in the model code by subtracting the mean and dividing by one standard deviation.
File: MSOM_occupancy_covariates.csv
Individual site-specific covariate data is saved in "MSOM_occupancy_covariates.csv".
Description: contains the environmental covariate data for input into the occupancy sub-model.
Variables
- Code: individual codes for each site
- Distpond: distance to the nearest pond (m)
- Area: pond surface area (m2)
- Hydro: hydroperiod (% water level fluctuation)
- Aqveg: aquatic vegetation cover (%)
- Urban_1000: proportion of urban land cover within a 1000-m radius of a pond
- Canopy: proportion cover of canopy over a pond (%)
- EC: electrical conductivity (µS cm -1)
- Forest_1000: proportion of forest cover within a 1000-m radius of a pond
- Wetland_1000: proportion of wetland cover within a 1000-m radius of a pond
- Chl_a: chlorophyll-a fluorescence (ft units)
- Fish: detection or non-detection of fish at a pond.
File: Multispecies occupancy models_species specific estimates.xlsx
Description: Summary of species-specific estimates for occupancy (α) and detection (β) covariates for nine amphibian species from the four multi-species occupancy models. These are the parameter estimates from individual runs of the four models and form the basis of the results presented in the article. Please note that the estimates will vary slightly in subsequent model runs due to the stochastic nature of the MCMC sampling.
Variables: the same variables as described above.
Code/software
R Core Team, 2023. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/.
- Package R2jags
Su, Y., Yajima, M., 2021. Package ‘R2jags’: using R to run ‘JAGS.’ Version 0.7-1.
Plummer, M., 2017. JAGS Version 4.3.0 user manual. https://sourceforge.net/projects/mcmc-jags/
Access information
Other publicly accessible locations of the data:
- DOI: 10.5061/dryad.70rxwdc6w
-
this data forms the basis for the detection sub-model of the current model uploaded here.
-
Amphibian surveys were conducted at 100 ponds over a single breeding season (spring/summer) using three survey methods to detect all life stages: (1) call/ visual encounter surveys (VES), (2) dip-netting, and (3) Dewsbury newt traps. The number of dip-net sweeps and traps at a pond was pre-determined to be proportional to the pond surface area. During the surveys, we recorded water temperature, water level, wind, rain, detection of fish and estimated the percentage cover of emergent vegetation at ponds. Amphibians were identified and released at the point of capture.
We used Multi-species Occupancy Models (MSOM) with Bayesian inference to estimate the probabilities of occupancy at the community and individual species levels. Models were used to examine relationships between the probabilities of occupancy of any life stage (i.e., eggs, larvae, juveniles or adults) and eleven covariates: (1) Distpond = distance to the nearest pond; (2) Area = pond surface area; (3) Hydro = hydroperiod (% water level fluctuation); (4) Aqveg = aquatic vegetation cover; (5) Fish = fish presence; (6) Urban_1000 = proportion of urban land cover within a 1000-m radius of a pond; (7) Canopy = proportion cover of canopy over a pond; (8) EC = electrical conductivity; (9) Forest_1000 = proportion of forest cover within a 1000-m radius of a pond; (10) Wetland_1000 = proportion of wetland cover within a 1000-m radius of a pond; (11) Chl_a = chlorophyll-a fluorescence.
For the detection sub-model, we used covariates that were previously found to have influential relationships with the probability of detection (see Hamer and Horanyi, 2024; DOI: 10.5061/dryad.70rxwdc6w): (1) Days = number of days since 19 March 2023 (survey day 0); (2) Days2 = quadratic effect of Days; (3) Effort = survey effort (number of minutes conducting call/ visual encounter surveys); (4) Temp = water temperature; (5) Fish = fish detected during a survey; (6) Emerg = % cover of emergent vegetation. We did not include wind speed, rain or water levels as detection covariates as they were not influential in the original detection only model.
The occupancy sub-model contains code for the four individual sub-models that align with the hypotheses presented in Table 1 of the article in Ecological Applications. The models were run using the software program JAGS (version 4.3.1) called via the R2jags package from program R. Each model was run using three replicate Markov chain Monte Carlo (MCMC) iterations to generate 400,000 samples from the posterior distribution after discarding a ‘burn-in’ of 200,000 samples, with a thinning rate of 10.
The means, standard deviations and the 2.5th and 97.5th percentiles of the posterior distributions of the model coefficients were estimated from the MSOM, which represents 95 % Bayesian credible intervals (BCI). Survey covariates consisting of continuous data were standardised (mean = 0, SD = 1) so that the relative importance of each covariate could be determined according to the magnitude of the coefficient. We modelled predictive relationships for influential covariates while holding the other covariates in the model at their mean values.