COMBINING DATABASE, VIS-NIR SPECTROSCOPY AND MACHINE LEARNING TO PROPOSE MODELS AND SOFTWARE FOR ESTIMATION OF NUTRIENTS IN LEAVES AND PRODUCTIVITY OF FRUITS

• Grant number: 24/09878-3
• Support Opportunities: Regular Research Grants
• Start date: November 01, 2024
• End date: October 31, 2025
• Field of knowledge: Agronomical Sciences - Agronomy - Soil Science
• Principal Investigator: Danilo Eduardo Rozane
• Grantee: Danilo Eduardo Rozane
• Host Institution: Faculdade de Ciências Agrárias. Universidade Estadual Paulista (UNESP). Campus do Vale do Ribeira. Registro , SP, Brazil
• Associated researchers: Jean Michel Moura Bueno ; José Alexandre Melo Demattê ; Juliana Domingues Lima ; Leandro Hahn ; Luciane Almeri Tabaldi ; Simone Rodrigues da Silva ; Tadeu Luis Tiecher ; William Natale
• Associated scholarship(s): 25/05122-4 - Proposal of models and software for assessing nutritional status and productivity in citrus and grapevines, BP.TT

Abstract

Brazil is one of the largest fruit producers in the world. The largest areas cultivated with vines and citrus are located in the states of Rio Grande do Sul (RS) and São Paulo (SP). In recent decades, improvements in management technologies for both fruit trees have been observed, which has led to an increase in production and an increase in values related to fruit quality variables. However, there is a demand from producers and the production chain regarding the use of non-destructive technologies to determine nutrients in the leaves of these crops. In this context, the challenge is to propose more sustainable analysis techniques that allow the measurement of nutrients; in sheets, quickly, non-destructively, without generating chemical waste and at low cost. Furthermore, there is a need to predict fruit productivity, as it is a variable also used to define the need and doses of nutrients in vineyards and orchards. Added to this, there is a demand for software/applications, where these models are stored, enabling the use, handling and application of these models by technicians and producers at the field level. This may all be possible with the strategy of combining databases, spectroscopy techniques in the visible (Vis) and near-infrared (NIR) regions, machine learning techniques and software development. Using this strategy, it will be possible to calibrate and propose models for predicting the nutrient content in fruit leaves and fruit productivity, which will be presented in software format for the production chain. With all this, it will be possible to improve the nutritional diagnosis of fruit trees through rapid and non-destructive analyses. Likewise, fertilization recommendations will be more assertive, as the expected productivity of vineyards and orchards can be predicted in advance, helping to refine the doses of fertilizers to be applied in each situation. This way, it will even be possible to reduce the amount of fertilizers to be used in orchards, avoiding excess nutrients in soil, which will prevent contamination of soil and water. But with guaranteed productivity and quality fruits. All of this will be possible, because the coordinator and his team in recent years have carried out and continue to carry out studies on techniques for nutritional diagnosis of fruit trees, but also studies related to the use of the Vis-NIR spectroscopy technique to estimate nutrients in leaves and prediction models of fruit productivity. The institutions where the coordinator and his team are linked have the structure and modern equipment necessary to properly carry out the project. Thus, the project aims to combine databases and the use of Vis-NIR spectroscopy and machine learning techniques to propose models for predicting nutrient content in leaves and grape and citrus productivity in the South and Southeast regions of Brazil. (AU)