Predicting the potential distribution of 'Sirex Noctilio' infestations in Kwazulu-Natal, South Africa: comparisons between classification trees and random forest classifers

Abstract

Reducing the impact of the siricid wasp, Sirex noctilio is crucial for the future productivity and sustainability of commercial pine resources in South Africa. In this study we present an alternative modeling framework that accurately identifies existing commercial pine forests that are susceptible to S.noctilio infestation. Using maps that show the potential distribution of S.noctilio infestations, forest managers can now adopt the most appropriate course of intervention before the wasp population reaches epidemic proportions. Two machine learning methods were used to predict the potential geographical distribution of S.noctilio infestations and to examine the relationship between the siricid and its environment. More specifically, classification trees (CT) and the random forest classifier (RF) were used to examine the relationship between 1301 pine forest compartments (showing the absence or presence of S.noctilio) and 72 environmental variables (consisting of historical climatic and topographic datasets). The results obtained from this study are very encouraging and show that RF was the most accurate predictive model and had several advantages over CT. The overall model accuracy (kappa statistic) was 0.71 for CT whereas RF produced accuracies of 0.77. Additionally, accuracy assessments by an independent dataset might be unnecessary as RF provides a reliable internal estimate of accuracy as determine by the out of bag error estimate. The highest ranked environmental variables as determined by RF were the median rainfall for February followed by the evapotranspiration during August and September. The results concur with previous studies and indicate that pine forests that are experiencing some form of stress are more susceptible to S.noctilio attack. The RF prediction model when used in conjunction with geographical information systems provides a useful and robust tool that can assist with current forest pest management initiatives.