Environmental predictability in phylogenetic comparative analysis: how to measure it and does it matter?
Data files
Jul 30, 2025 version files 5.16 MB
-
Codes_Periodicity-main.zip
5.15 MB
-
README.md
7.52 KB
Abstract
Aim
Abiotic environmental conditions shape ecological and evolutionary processes, yet quantifying their influence on organisms remains challenging due to variation among metrics and their intercorrelations. This study evaluates the utility of temporal environmental predictability measures and assesses their explanatory power in phylogenetic comparative analyses.
Innovation
We systematically compare widely used metrics of predictability and explore their correlations with environmental means and variances in a global meteorological dataset. Using cooperative breeding birds as a case study, we assess the impact of including predictability metrics in phylogenetic comparative analyses. We demonstrate the consequences of choosing specific metrics and the trade-offs between increased data inclusion and model interpretability.
Main conclusions
Predictability metrics, though intuitively meaningful, have been conceptualised and quantified with diverse approaches. We found that different measures of predictability can exhibit contrasting global patterns and strong correlations with other environmental quantities. Therefore, our findings caution against overloading statistical analyses with correlated predictors, highlighting the need for a thoughtful selection of environmental metrics to avoid spurious interpretations in ecological and evolutionary studies.
This dataset contains the raw and processed data, derived climatic measures, phylogenetic trees, and analysis scripts associated with the manuscript titled above. The analyses explore the role of environmental predictability in shaping organismal traits and adaptation patterns, combining global climate data, regional biological datasets, and phylogenetic comparative methods.
Description of the Data and File Structure
The dataset is structured into four main directories, each corresponding to a major analysis section of the manuscript:
1. GlobalAnalysis
-
/data/: Contains processed CRU TS climate datasets.These datasets retain the same structure as CRU TS:
- Temporal resolution: Monthly
- Spatial resolution: 0.5° latitude × 0.5° longitude
- Units:
- Temperature: degrees Celsius (°C)
- Precipitation: millimeters (mm)
Temperature and precipitation are provided in two separate files, each containing the following columns (values are presented in raw scale unless otherwise noted):
- Column names
lon,lat: geographic coordinatesavg,sd,med: mean, standard deviation, and median of the climate variablecv: coefficient of variation (unitless)prd_SpcEntropy: predictability based on normalized spectral entropy (unitless, 0–1)prd_ColwellFix,cst_ColwellFix,ctg_ColwellFix: Colwell's predictability, constancy, and contingency with fixed binning (unitless, 0–1)prd_ColwellDym,cst_ColwellDym,ctg_ColwellDym: same metrics calculated with dynamic binning (unitless, 0–1)
-
/makefiles/: Scripts and configuration files for generating processed climatic variables from raw CRU TS data. -
/rscripts/: Contains R scripts for generating global summary maps (used in main figures).
2. RegionalAnalysis
-
/data/: Contains modified species trait and distribution datasets originally based on Cornwallis et al. 2017. The dataset from the paper is publically available at this link. Climatic variables have been updated using the methods developed in this project and the species range provided by Cornwallis et al.- Column names
clade: passerine or non-passerineanimalandspecies: species namecare: the type of parental/cooperative carebreedsystem: breeding system typescoop: whether the species is cooperative (string format)COOP2: whether the species is cooperative (binary format)longitudeandlatitude: the coordinate of the field site
We added columns after 'latitude', following the same abbreviation rule as
GlobalAnalysiswith the prefix of TMP for temperature and PRE for precipitation. In addition, we also calculated similar summary stats for diurnal temperature range (DTP) and net primary production (NPP) from CRUTS.Abbreviations for variable prefixes:
TMP: Temperature (°C)PRE: Precipitation (mm)DTP: Diurnal Temperature Range (°C; not used in this project)NPP: Net Primary Productivity (g C/m²/month; not used in this project)
- Column names
/rscripts/: Includes scripts for correlation visualization and exploratory analysis of trait–climate relationships.
3. PhylogeneticRegression
/data/: Contains the trait data matrix and phylogenetic tree used in the regression analyses. The csv file is the same as inRegionalAnalysis, and the phylogenetic tree is provided from Cornwallis et al./rscripts/: Includes three key scripts:- Non-PCA regression: Used to produce results in Table 2.
- Regression with phylogenetic PCA: Corresponds to Table 4.
- Relative Variable Importance (RVI) analysis: Introduced in Section 3.1.3.
4. RandomForest
/Data/: Contains input datasets and the phylogenetic tree used in the Random Forest model. The data files are identical to those inPhylogeneticRegression./Output/: Contains model output including:- Bird_Clm_IC_04Sep.csv: Input trait–climate dataset after phylogenetically independent contrasts (used for some Random Forest models)
- CrossVal_GLM.csv: Cross-validation results for GLM
- CrossVal_GLMM.csv: Cross-validation results for GLMM with with cluster-based random effects to control for phylogeny
- CrossVal_RF.csv: Cross-validation results for RF with all variables and cluster
- CrossVal_GLM_IC.csv: Cross-validation results for GLM using phylogenetically independent contrasts (IC) based on species-level trait averages
- CrossVal_RF_IC.csv: Cross-validation results for RF with IC
- RF_cluster_main.rds: Trained RF model (cluster-based)
- RF_IC_main.rds: Trained RF model (IC-based)
- KernelShap_cluster_main.rds: SHAP object for cluster-based RF
- KernelShap_IC_main.rds: SHAP object for IC-based RF
/rscripts/: Includes 12 R scripts for:- 0_silhouette_scores.R: Determine optimal cluster number
- 1_MakeClusterData.R: Create cluster structure from trait data
- 2.1_CrossVal_GLM(means).R: GLM cross-validation (cluster)
- 2.2_CrossVal_GLMM(means_cluster).R: GLMM cross-validation
- 2.3_HyperParamTuning.R: RF hyperparameter search (cluster)
- 2.4_CrossVal_RF.R: RF cross-validation (cluster)
- 3_MakeRF_cluster_noise.R: Build RF with cluster and noise (reference) predictors
- 4_MakeDataset_IC.R: Construct datasets using phylogenetic independent contrasts (IC)
- 5.1_CrossVal_GLM(ICmeans).R: GLM cross-validation (IC)
- 5.2_HyperParams.R: RF hyperparameter search (IC)
- 5.3_CrossVal_RF_IC.R: RF cross-validation (IC)
- 6_MakeRF_IC_noise.R: Build RF with IC and noise (reference) predictors
- (Scripts are numbered to suggest the intended execution order.)
File Note
This dataset was submitted to Dryad as a single compressed archive file:
Codes_Periodicity-main.zip
This zip contains all the directories described above, preserving folder structure. No files are excluded from this README.
Sharing / Access Information
This dataset is also mirrored at the following public repository:
Note that the GitHub version may receive future updates or revisions, while the Dryad version represents the version archived at the time of manuscript publication.
Data sources used or adapted:
- CRU TS (Climatic Research Unit time-series data): https://crudata.uea.ac.uk/cru/data/hrg/
- Cornwallis et al. 2017 dataset: https://www.nature.com/articles/s41559-016-0057
All modifications or derived variables are documented in the makefiles directory and described in the manuscript.
Code / Software
All R scripts used for analysis are included in the respective /rscripts/ folders in each section. These scripts are compatible with R ≥ 4.2.1 and rely on standard packages including:
apephylolmrandomForestSHAPforxgboostclusterggplot2
All code is released under the CC0 waiver as part of this dataset. If you use the code, please cite the accompanying publication.