A digital soil map of Phytophthora cinnamomi in the Gondwana Rainforests of eastern Australia

摘要

The aim of this work is to map the presence the soil-borne fungal pathogen, Phytophthora cinnamomi, in the Gondwana Rainforests of eastern Australia. Logistic regression was used for modeling and the final model included mean and minimum temperature, distance to nearest drainage line, distance to nearest road/trail/lookout/visitor area, region, inside/outside Gondwana rainforests and northings as predictor variables. The model had good predictive ability as judged by an AUC value of 0.75 based on an independent validation set of one third (n = 311) of the total observations (n = 941). Two distinct groupings in the AUC existed when the validation set was divided further on regions, where one group had values of -0.85 and the other group had values of -0.60. Further work is needed to (i) explore alternate modeling approaches for improving the predictions (ii) consider local models where the region-specific AUC is poorer.