Artificial neural networks and neighborhood techniques for predicting phosphorus and potassium requirements for industrial tomato crops

The adoption of innovative technologies, such as artificial neural networks (ANNs), has emerged as an economical and viable approach to solving agricultural problems. The aim was to adapt the neighborhood technique to reduce the data collected for possible phosphorus (P) and potassium (K) mapping. The neighborhood technique was adapted to define the points that would provide the input and output data for the ANN. The training was conducted using the backpropagation algorithm. Geostatistical analysis was then conducted, and the semivariogram model was adjusted, isoline maps were drawn up, and accuracy was assessed using the Kappa and Global Accuracy indexes. The trained model showed a good fit in the external validation phase (R² > 0,90), a correlation of over 91%, and excellent performance from the ANN, where the t-test confirmed accuracy. It was possible to estimate P and K requirements, and the adaptation of the neighborhood technique enabled a 40% reduction in the number of samples needed for field data collection. By reducing the number of samples, the network can obtain valid estimates of P and K.