Influences of climate, landscape and vegetation proximity on the spatial distribution of the Tasmanian Eucalyptus leaf beetle, Paropsisterna bimaculata in Tasmanian Eucalyptus plantations
The distribution of the beetle Paropsisterna bimaculata was related to 33 environmental covariates distributed across in Tasmania. This beetle is an economic pest in Eucalyptus plantations, episodically occurring in large aggregations that damage tree canopies through overgrazing. A number of hypotheses relating to beetle outbreaks have been proposed by various researchers. Random forest modelling, a powerful non-parametric statistical approach that has not been frequently applied before in ecology, was used to assess specific predictions of these hypotheses. The models were developed using landscape layers and proximity to vegetation layers that were created using GIS for multiple combinations of climate variables and districts. The climate combinations assessed included: (1) mean summer maximum temperature and summer rainfall for survey years, (2) 30 year average for mean summer maximum and summer rainfall, (3) 30 year average for climate variables, (4) 30 year average for climate variables and mean summer maximum temperature and summer rainfall for survey years. The district combinations included all districts, Bass only, and all districts excluding Bass. Locations less than 10 kilometres from Poa grasslands showed highest beetle numbers, while increasing elevation also showed a strong positive relationship with P. bimaculata populations, and mean annual summer rainfall and survey year were important in the district of Bass. Thus, in terms of prior hypotheses, Poa appears to be an important overwintering site for P. bimaculata; beetle populations increase with elevation; beetle populations increase with age of plantation between 2 and 8 years; and Bass has greater variability than other districts in beetle density. Model outputs did not support, or only weakly supported, hypotheses that suggest beetle numbers are highest at highly productive sites, and are affected by the shape and size of forestry coupes and by plantation species. The random forests model with least error was based on all districts except Bass for thirty-year climate average data plus summer mean rainfall and summer mean maximum temperature data. A cost-benefit analysis nevertheless indicated that directing the ongoing beetle survey effort on the basis of model outputs would not reduce costs compared to the current broad-scale monitoring program. The model did, iv however, generate state-level maps of value to forestry workers when assessing the likelihood of beetle outbreaks in their districts.