Figure 5. Partial dependence plot from the Random Forest (RF) framework showing the relationship between the probability of Bd infection and GI₃₀ (simulated pathogen growth in the 30 days prior to sampling, from the CLIMEX process-based model) for adult male Litoria pearsoniana.

The response line is a lowess smoother. Data were derived from four field sites across three years of sampling (2006–2009). In a model containing the 30 day growth index (GI₃₀), body length (SUL) and Rain (mm on the day of sampling), positive PCR results for Bd infection were correctly predicted in ∼72% of cases (see results).