Using artificial intelligence for predicting the seasonal dynamics of a primary pest in essential crops.

Abstract

Bemisia tabaci (Gennadius) (Hemiptera: Aleyrodidae) is a primary pest in essential crops for food security, such as tomato, potato, and soybean. Soybean (Glycine max) (L) (Merr) is the leading source of animal protein in the world. B. tabaci remains a relevant pest in soybean crops; on average, one whitefly per sample yields 30 kg. ha^-1 of losses. Determining the seasonal dynamic of B. tabaci is helpful in controlling the pest in time, avoiding losses. This investigation aims to model the seasonal dynamics of B. tabaci in soybean crops through neural networks. This research tracked whitefly density, climatic elements and soybean age in 100 soybean fields in the Brazilian Cerrado to build seasonal dynamic models along 2 years. Features were selected according to correlation analysis and biological meaning. ANNs structures were investigated to forecast whitefly through the years and the model presenting the highest Pearson correlation and lowest root mean square error were chosen. Feature importance was analyzed to examine these attributes' effect on B. tabaci. Then, the model was validated by comparing whitefly observed and fit data densities during the study. The ANN selected has five entries (soybean age, average temperature, rainfall, wind speed, and atmosphere pressure) and four neurons in the hidden shell. Average temperature and wind speed are key features in the model presenting the most elevated relative importance index to predict whitefly adult population. Therefore, this study highlighted artificial intelligence's power in modelling a key pest's seasonal dynamic upon a range of attributes seven days in advance.