Predicting fine-scale distributions and emergent spatiotemporal patterns from temporally dynamic step selection simulations

This script is a walkthrough to build intuition around fitting models with harmonic interaction terms (harmonic regression)

Inputs

• Data file that contains random steps, sampled covariate values and KDE density at the end of each used and random step (when using memory parameters that were estimated for all individuals simultaneously)
  • buffalo_popn_GvM_covs_ST_KDEmem1000_allOPTIM_10rs_2024-02-05.csv

Outputs

• Fitted model objects (2 pairs of harmonics as an example)
  • model_twostep_2p_harms_dry.rds

Generate a track object to sample random step lengths and turning angles, and sample covariate values at the end of each step.

Inputs

• Buffalo GPS data
  • buffalo.csv
• spatial covariate rasters
  • canopy_cover.tif
  • ndvi_GEE_projected_watermask20230207.tif
  • slope_raster.tif
  • veg_herby.tif

Outputs

• Data file that contains random steps and sampled covariate values at the end of each used and random step
  • buffalo_parametric_popn_covs_GvM_10rs_2024-09-04.csv

This script fits step selection models with varying numbers of harmonic terms to the buffalo's dry (a) or wet (b) season data.

Inputs

• Data file that contains random steps, sampled covariate values and KDE density at the end of each used and random step (when using memory parameters that were estimated for all individuals simultaneously)
  • buffalo_parametric_popn_covs_GvM_10rs_2024-09-04.csv

Outputs

• Fitted model objects (0, 1, 2 or 3 pairs of harmonics)
• Tables of hourly coefficient values from fitted models 

This script takes the hourly coefficients of the fitted models and simulates (dynamic) animal movement trajectories.

Inputs

• Tables of hourly coefficient values from fitted models (change the input file to change which model's coefficients are used to simulate trajectories) 
• Spatial covariate rasters
  • ndvi_GEE_projected_watermask20230207.tif
  • canopy_cover.tif
  • veg_herby.tif
  • slope_raster.tif

Outputs

• Simulated data (animal movement trajectories)

This script takes animal movement trajectories (observed and simulated), and calculates the landscape-scale prediction maps by aggregating the simulated locations into raster. We ran this script on the QUT HPC.

This script assesses the convergence of the simulated trajectories, to ensure that the landscape-scale prediction maps are stable. We ran this script on the QUT HPC.

We calculate hourly summary statistics which we used to compare between the observed data and simulated data.

Inputs

• Simulated data (animal movement trajectories)
• Spatial covariate rasters
  • ndvi_GEE_projected_watermask20230207.tif
  • canopy_cover.tif
  • veg_herby.tif
  • slope_raster.tif
• Data file of buffalo GPS data that contains random steps and sampled covariate values

Outputs

• Hourly summary statistic values for observed and simulated data (with number of harmonics depending on which trajectories are read in)

This script assesses the long-term and hourly prediction maps, and validates them against the observed buffalo data using the continuous Boyce index, which is appropriate for presence-only GPS data.

Inputs

• Prediction maps from the DynamicSSF_4a_Aggregating_simulations.R script.

Outputs

• Plots of the landscape-scale predictions for the late dry season (~ 4 months) and for each hour of the day.
• Validation results from the continuous Boyce index

For more details, don't hesitate to get in contact at scott.forrest@qut.edu.au.