Soybean productivity forecast with machine learning models

• Grant number: 23/00166-8
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: June 01, 2023
• End date: December 31, 2023
• Field of knowledge: Physical Sciences and Mathematics - Probability and Statistics - Applied Probability and Statistics
• Principal Investigator: Glauco de Souza Rolim
• Grantee: Maria Gabriela de Sousa
• Host Institution: Faculdade de Ciências Agrárias e Veterinárias (FCAV). Universidade Estadual Paulista (UNESP). Campus de Jaboticabal. Jaboticabal , SP, Brazil

Abstract

Soy is an oleaginous plant rich in proteins, the main source of vegetable oil cultivated in different regions of the planet and its world production has doubled in the last twenty years. The objective is to carry out a productivity forecast, with greater anticipation of the harvest, for the main soybean producing regions in Brazil: Nova Mutum (MT), Sorriso (MT), Formosa do Rio Preto (BA), São Desidério (BA), Itapeva (SP) and Itaberá (SP), using meteorological data and machine learning models. The models are tools that can help both rural producers in economic strategies in the use of agricultural inputs, but also for companies, commodity traders, among other services that deal with the production and trade of soy. Daily meteorological data from the study regions from 1989 to 2020, such as air temperature, precipitation, relative humidity, wind speed and radiation will be obtained from the NASA-POWER platform. The productivity forecast will be carried out using the Multiple Linear Regression (RLM) models, Artificial Neural Networks (ANNs) and the Random Forest (RF) learning machine, considering the climatological characteristics and the dependent variable productivity as independent variables. To evaluate the performance of the model, the adjusted coefficient of determination (adjusted R2), unsystematic root mean square error (RMSEu) and systematic root mean square error (RMSEs) will be used. It is expected to obtain an agrometeorological model that is sufficiently accurate to be routinely applied to predict soybean productivity, allowing for more assertive decisions and strategies in relation to the agricultural system and national agribusiness.