FORECASTING CITRUS (Citrus sinensis (L.) Osbeck) QUALITY USING MACHINE LEARNING MODELS BASED ON CLIMATE VARIABILITY

• Grant number: 24/19045-9
• Support Opportunities: Scholarships in Brazil - Master
• Start date: April 01, 2025
• End date: March 31, 2027
• Field of knowledge: Agronomical Sciences - Agronomy - Agricultural Meteorology
• Principal Investigator: Glauco de Souza Rolim
• Grantee: Pedro Antonio Lorençone
• Host Institution: Faculdade de Ciências Agrárias e Veterinárias (FCAV). Universidade Estadual Paulista (UNESP). Campus de Jaboticabal. Jaboticabal , SP, Brazil

Abstract

The orange (Citrus sinensis (L.) Osbeck) is one of the most widely produced and consumed fruits in the world, playing an essential role in the agricultural economy of many countries, especially Brazil. Brazil is the world's largest producer of oranges and the citrus belt (CCI), located mainly in the state of São Paulo and the southwestern regions of Minas Gerais and Triângulo Mineiro, accounts for more than 80% of national production. The quality of oranges, both for fresh consumption and for juice production, is a crucial factor that determines the market value and competitiveness of the sector, requiring strict control and accurate forecasts to optimize the production chain. This project aims to predict the quality of oranges using machine learning models, taking into account climate variability and the differences between maturity groups: early, mid-season and late, and in production systems: irrigated and rainfed in the JRC region. Citrosuco, the world's largest orange juice producer, will provide the data used for this study, which covers the period from 2012 to 2024. 2,121 plots distributed over 27 properties located in the JRC will be analyzed. Citrosuco, the world's largest orange juice producer, will provide the data used for this study, which covers the period from 2012 to 2024. 2,121 plots distributed over 27 properties located in the JRC will be analyzed. Citrosuco has a consistent history of monitoring fruit quality, guaranteeing reliable data for training and testing the predictive models. The predictive models will be developed using machine learning techniques, integrating climate data obtained from 27 weather stations installed on the properties and data from the NASA POWER system to fill in gaps. The quality parameters to be considered will be soluble solids content (BRIX), total titratable acidity (ATT), RATIO, kilo of soluble solids per box (KGSS) and percentage of juice (%SUCO). As a result, this study aims to develop robust models for predicting fruit quality, identifying the climatic variables that most influence fruit quality and helping to optimize management and harvesting strategies. This will enable strategic decision-making by producers and the juice industry. The results will also be used to generate maps of expected quality, helping with agricultural planning in the region.