Data from: Resilience metrics are robust across data qualities but sensitive to community size models
Data files
Sep 28, 2023 version files 207.33 GB
-
motif1_15_invasive_FI_model.rds
-
motif1_15_invasive_multiAR_model.rds
-
motif1_15_invasive_multiJI_model.rds
-
motif1_15_invasive_MVI_model.rds
-
motif1_15_invasive_slope_posteriors.RData
-
motif1_15_invasive_slope_ranges.RData
-
motif1_15_invasive_uniJI_model.rds
-
motif1_25_invasive_FI_model.rds
-
motif1_25_invasive_multiAR_model.rds
-
motif1_25_invasive_multiJI_model.rds
-
motif1_25_invasive_MVI_model.rds
-
motif1_25_invasive_slope_posteriors.RData
-
motif1_25_invasive_slope_ranges.RData
-
motif1_25_invasive_uniJI_model.rds
-
motif1_5_invasive_FI_model.rds
-
motif1_5_invasive_multiAR_model.rds
-
motif1_5_invasive_multiJI_model.rds
-
motif1_5_invasive_MVI_model.rds
-
motif1_5_invasive_slope_posteriors.RData
-
motif1_5_invasive_slope_ranges.RData
-
motif1_5_invasive_uniJI_model.rds
-
motif1_threshold_multiAR.rds
-
motif1_threshold_multiJI.rds
-
motif1_threshold_posteriors.RData
-
motif1_threshold_ranges.RData
-
motif1_threshold_uniJI.rds
-
README.md
Abstract
Modern biodiversity monitoring is generating increasingly multidimensional representations of wildlife populations and ecosystems. It is therefore appealing for conservation and environmental governance to combine that information into single measure of ecosystem or population health.
The Jacobian matrix is a common characteristic used to identify the sensitivity of simulated/mathematical systems to perturbation (a.k.a. its resilience) and predict its near future dynamics. Jacobians have therefore been suggested as a theoretically grounded measure of ecosystem resilience. Whilst historically it has been challenging to estimate the Jacobian from empirical data, recent work has proposed a suite of metrics capable of reconstructing it for real-world community using multi-species time series data.
Here we assess the robustness of five resilience metrics influenced by varying time series lengths and data qualities based on that seen in real-world wildlife time series. We generate data using multispecies Lotka–Volterra equations and simulate stressed and unstressed communities of varying species number. These data were then corrupted through the introduction of sampling error (to mimic varying search efforts) and truncating time series (to match the typical time series lengths reported in global biodiversity datasets such as the Living Planet Index and BIOTIME).
The robustness of all resilience metrics improved with time series length, whilst the amount of sampling error had little effect on their performance. However, community size (number of species) dramatically altered metric capability, with larger communities decreasing the reliability of resilience metric trends.
Overall, resilience metrics behave predictably across realistic data corruptions. Generic resilience estimation is therefore possible from abundance time series alone, and we suggest that, given the increasing availability of multivariate community data, focussing on Jacobian estimates for resilience is a promising avenue of research. However, we also show it is prudent to apply ecological knowledge when selecting which species to contribute.
README: Resilience metrics are robust across data qualities but sensitive to community size - models
https://doi.org/10.5061/dryad.00000008d
This repository contains the model files and summary statistics associated with the publication 'Resilience metrics are robust across data qualities but sensitive to community size' and the software archived on Zenodo (https://doi.org/10.5281/zenodo.8341499) and GitHub (https://github.com/duncanobrien/lpi-multivariate-res). The files require archiving here due to file size restrictions on those platforms. Models were fitted to test whether the temporal trend of resilience metrics varied with across stressed vs unstressed communities, increasing time series length and increasing search effort (equivalent to sampling error).
Resilience metrics were calculated from raw abundance time series simulated via generalised Lotka-Volterra models of random community structures and sizes, which were subsequently corrupted by trimming the time series and introducing search effort error.
Description of the data and file structure
20 INLA R language models are present, all sharing a consistent naming structure that describes the simulation type, community size, and resilience metric.
For example 'motif1_15_invasive_FI_model.rds' indicates a generalised linear mixed effect model of the resilience metric FI (Fisher information) through time. The remainder of the naming convention states that FI was calculated for 15 species communities simulated using an Lotka-Volterra model with a species interaction matrix of motif "1" which contains an invasive species.
Abbreviations include:
- FI - Fisher information
- multiAR - multivariate autocorrelation Jacobian index
- mulitJI - multivariate Smap Jacobian index
- MVI - multivariate index of variability
- uniJI - univariate Smap Jacobian index
Additional binomial models are provided for whether certain resilience metrics exceeded an instability threshold.
File list
- motif1_5_invasive_slope_posteriors.RData
- motif1_5_invasive_slope_ranges.RData
- motif1_5_invasive_FI_model.rds
- motif1_5_invasive_multiJI_model.rds
- motif1_5_invasive_multiAR_model.rds
- motif1_5_invasive_MVI_model.rds
- motif1_5_invasive_uniiJI_model.rds
- motif1_15_invasive_slope_posteriors.RData
- motif1_15_invasive_slope_ranges.RData
- motif1_15_invasive_FI_model.rds
- motif1_15_invasive_multiJI_model.rds
- motif1_15_invasive_multiAR_model.rds
- motif1_15_invasive_MVI_model.rds
- motif1_15_invasive_uniiJI_model.rds
- motif1_25_invasive_slope_posteriors.RData
- motif1_25_invasive_slope_ranges.RData
- motif1_25_invasive_FI_model.rds
- motif1_25_invasive_multiJI_model.rds
- motif1_25_invasive_multiAR_model.rds
- motif1_25_invasive_MVI_model.rds
- motif1_25_invasive_uniiJI_model.rds
- motif1_threshold_posteriors.RData
- motif1_threshold_ranges.RData
- motif1_threshold_multiAR.rds
- motif1_threshold_multiJI.rds
- motif1_threshold_uniJI.rds
Data specific information for: **_model.rds
Single R object containing an INLA (https://www.r-inla.org) bayesian linear mixed effect model. Models were fitted with interactions between stress, time series length and search effort, with a random intercept and slope for community identity, and an autoregression term per simulation. Models are presented as single files due to their large file sizes - ~10-14GB
Data specific information for: **_posteriors.RData
Posterior draws for each Bayesian model of a certain community size (5, 15 or 25 species), categorised by whether the model was fitted to data simulated with specific stress levels, time series length and search effort error.
- Number of R objects: 4
- Number of variables within each object: 4
- Variable list:
- .draw - label indicating the random sample/draw from the Bayesian model's posterior distribution (numeric: 1 - 10000)
- .value - the posterior value returned by the draw
- stressed - binary classification of whether the model was fitted to unstressed (0) or stressed (1) time series.
- ts_length - what length of time series was the model was fitted to in years (numeric: 10 - 70)
- search_effort - probability of encountering an individual (numeric: 0.1 - 1.0)
Data specific information for: **_slopes.RData
- Number of R objects: 4
- Number of variables within each object: 4
- Variable list:
- stressed - binary classification of whether the model was fitted to unstressed (0) or stressed (1) time series.
- ts_length - what length of time series was the model was fitted to in years (numeric: 10 - 70)
- search_effort - probability of encountering an individual (numeric: 0.1 - 1.0)
- .value - median value of the posterior distribution (numeric)
- .lower - lower credible interval value (numeric)
- .upper - upper credible interval value (numeric)
- .width - quartile that credible interval covers (numeric: 0.5, 0.8, 0.95)
- .point - "median"
- .interval - "quartile range"