Adaptive frequency classification: a new methodology for pest monitoring and its application to European red mite (Panonychus ulmi, Acari: Tetranychidae)

Abstract

We developed a new method for monitoring pests over a growing season and applied the method to monitoring the European red spider mite (Panonychus ulmi Koch) in apples. We evaluated the performance of the monitoring method in this system by simulation experiments and a field test in 28 orchard blocks in New York State, USA in 1995. The method is based on serially linking sample occasions (bouts) in time. At each sample bout, the monitoring procedure decides between intervening or not intervening. In the case of no intervention, the procedure schedules the next sample bout on the basis of an estimate of the current density, a descriptive population growth model (exponential for mites) and intervention thresholds. The next sample bout is scheduled when the risk of the pest density becoming greater than a future intervention threshold exceeds a specified tolerance. The name of the sampling method reflects this adaptiveness of the sampling frequency. The sampling protocol is constructed by combining a sequential probability ratio test for taking time-efficient invervention decisionsat high density with a fixed sample size estimation at low density. The low-density estimates are used for calculating the waiting times until the next sample. The probabilities of the intervene or wait decisions and the expected sample sizes for these sampling protocols were calculated as functions of the pest density by Monte Carlo simulation. The expected performance characteristics of the monitoring method for a given population trajectory are estimated by integrating the performance criteria of the sampling protocols over a population trajectory. We simulated the performance of the monitoring method for two sets of 400 simulated population trajectories. The trajectories were calculated with an exponential polynomial equation with random parameters. The trajectories in the first set were characterized by low growth rates and a maximum density below the intervention threshold while the trajectories in the other set had high growth rates and a maximum density above the threshold. As performance criteria we used the probability of intervening, the number of scheduled sample bouts, the total number of sample units, the cumulative mite density per leaf (mite days) and the mite density at the time of intervention. Simulation indicated that the adaptive frequency classification would have similar performance characteristics to existing monitoring methods when monitoring rapidly growing populations, while substantial savings on the number of sample bouts would be expected with slowly growing populations. The field test confirmed these predictions and demonstrated the savings on sample bouts that are obtained by using the adaptive frequency classification.