Evolving Connectionist Systems can predict outbreaks of the aphid Rhopalosiphum padi

Modeling of insect pest outbreaks is important for the protection of economically significant crops. This paper describes an attempt to model the outbreaks of the aphid Rhopalosiphum padi in the Canterbury region of New Zealand. Outbreaks were predicted using two representations of weather variables: Firstly, from moving time windows over the variables; Secondly, from the gradient or rate of change of the variables, which is presented here for the first time. Two artificial neural network types were used in this modeling, Multi-Layer Perceptrons (MLP) and Simple Evolving Connectionist Systems (SECoS). The results show that while SECoS are able to predict outbreaks of R. padi from either approach, MLP are unable to do so. Also, the results show that there is no significant difference in the modeling accuracy of SECoS between either modeling approach. These results indicate that the rate of change of weather variables is as important to the prediction of aphid outbreaks as the values of those variables. This work represents the first steps towards an outbreak prediction system that can assist with the management of these crop pests.