Modelled deer density and impact on vegetation across the Melbourne drainage and waterway extent, Victoria

This collection contains spatial predictions of deer density and deer impact on vegetation across the Melbourne drainage and waterway extent. Deer density is quantified in units of faecal pellets/m². Deer impacts on vegetation (whole plants) are quantified as a percentage, based on foliage browsed, stem damage or rubbing damage. The median response modelled for each variable is provided, alongside the 5ᵗʰ and 95ᵗʰ quantiles to characterise uncertainty in the predictions.\n\nPlease refer to the published manuscript under related links for further details (TBD). \nLineage: Deer density and deer impact models were fitted using quantile regression forests in R. Deer density was modelled with faecal pellet counts (pellets/m²) from 1,948 transects across the study region as the dependent variable. Deer impact on vegetation was modelled using vegetation impact scores (expressed as a percentage) from 23,144 individual plants across 343 transects co-located with faecal pellet counts. All field data was collected between 2005 and 2023. \n\nEach model was fitted using distance to waterbodies (> 10 ha), elevation, woody vegetation cover (mean canopy cover within 1 km), mean annual precipitation, mean annual temperature, slope, aspect and distance to streams as model covariates. In addition, deer impact on vegetation included the modelled deer density as a model covariate, which was the most important predictor overall of deer impact (relative influence of 21.0%). \n\nStatistical performance for each model, determined by 10-fold cross-validation, is reported below:\n\nDeer density (pellets/m²): RMSE = 1.45, MAE = 0.56, R² = 0.71, PBIAS = -28.0\n\nDeer impact (whole plant, %): RMSE = 8.61, MAE = 6.86, R² = 0.32, PBIAS = -2.65 \n\nwhere RMSE is the root mean squared Error, MAE is the mean absolute error, NSE is the Nash-Sutcliffe efficiency, R² is the coefficient of determination, and PBIAS is the percent bias. \n\nSpatial predictions were generated for the 5ᵗʰ (lower 90% prediction interval), 50ᵗʰ (median response) and 95ᵗʰ quantiles (upper 90% prediction interval). This was done to characterise uncertainty alongside the median response from the quantile regression forests.\n\nPlease refer to the published manuscript under related links for further details (TBD).