Training and testing of agrometeorological models based on machine learning for forecasting corn gray leaf spot

• Grant number: 23/04344-8
• Support Opportunities: Scholarships in Brazil - Master
• Start date: November 01, 2023
• End date: February 28, 2025
• Field of knowledge: Physical Sciences and Mathematics - Probability and Statistics - Applied Probability and Statistics
• Principal Investigator: Glauco de Souza Rolim
• Grantee: Rafael Fausto de Lima
• Host Institution: Faculdade de Ciências Agrárias e Veterinárias (FCAV). Universidade Estadual Paulista (UNESP). Campus de Jaboticabal. Jaboticabal , SP, Brazil

Abstract

The amount of data used for modeling diseases in agricultural crops are scarce. In this project to predict gray leaf spot of corn in corn we will have 24 years of data from experiments conducted by the São Paulo Agency of Agribusiness Technology (APTA) allowing the training and testing of machine learning models to predict the disease. Thus, the objective of this project is to train and test prediction models, with minimum anticipation for decision-making, for gray leaf spot of corn. The locations used will be: Capão Bonito, Cândido Mota, Palmital and Pedrinhas Paulista with experiments conducted in the period from 1998 to 2022 in the State of São Paulo. Meteorological data for the same period will be collected from the NASA/POWER (National Aeronautics and Space Administration/Prediction of Worldwide Energy Resources) platform. The components related to soil characteristics will be obtained from the SOILGRIDS platform maintained by the World Soil Information (ISRIC). In the first part of the project, corn hybrids will be classified into disease susceptibility groups: susceptible, medium susceptible, and tolerant. Regression analyses will be performed to verify the influence of the disease on crop development. In the modeling for disease prediction, different machine learning algorithms will be used, and implemented in Python programming environment. The meteorological variables will be selected 20 days before the crop bloom, in order to represent the incubation period and the beginning of the first symptoms of the disease. The models will be evaluated for accuracy, precision and trend, comparing the observed data in the field with the predicted data.