Predicting the Seasonal Dynamics of Fruit Fly Anastrepha fraterculus Populations in Apple Orchards Using Artificial Neural Networks.

The fruit fly Anastrepha fraterculus (Wiedemann) (Diptera: Tephritidae) is one of the main pests in apple orchards. Artificial neural networks (ANNs) are tools with good ability to predict phenomena such as the seasonal dynamics of pest populations. Thus, the objective of this work was to determine a prediction model for the seasonal dynamics of A. fraterculus in apple orchards using ANNs. Insect densities and climatic elements were monitored for 8 years in two commercial apple orchards. Of the 1452 ANNs determined, the one with meteorological data with a 35-day lag was selected. This ANN presented the highest correlation (0.693) between predictions and pest densities, the lowest square root mean validation error (0.066), and five neurons in the hidden layer. Among the model's predictors, wind speed and relative humidity showed positive correlations with pest density, while precipitation was negatively correlated. The predicted population curves, based on the fruiting period and temperature, reached a peak in the number of A. fraterculus individuals per trap per day. ANN was able to adequately predict pest density in different orchards, plant fruiting stages, and years. Therefore, this ANN model is promising for predicting A. fraterculus densities in apple orchards.