Integração de tecnologias de agricultura de precisão e digital aplicadas na gestão da cafeicultura

Abstract

The factors that most influence crop yield is interconnected and encompass climatic, agronomic, and technological aspects. The evaluation and monitoring of these factors become necessary, as, when well investigated, they can lead to increased productivity and fruit quality, in addition to making the production cycle more sustainable. This study aims to investigate the use of Precision Agriculture (PA) and Digital Agriculture (DA) technologies in coffee cultivation, applied in an experimental field of EPAMIG, located in Três Pontas, MG, Brazil. The study area comprises 1.2 hectares of Coffea arabica plants, cultivar Topázio MG1190. Between 2020 and 2022, data were collected on yield, soil moisture, leaf water potential, soil fertility, leaf nutrition, along with data obtained through flights with a Remotely Piloted Aircraft (RPA). Based on this dataset, the research was divided into three parts.The first part provides a literature review on the importance of coffee cultivation in Brazil, the production challenges, and the role of emerging technologies such as PA and DA in optimizing productivity and sustainability. In the second part, three scientific articles are presented. The first article evaluated the spatial and temporal variability of coffee yield over three harvests using geostatistical methods. The results revealed high heterogeneity in yield, with significant variations in both space and time, even within a relatively small area.The second article analyzed the spatial variability of the NDVI vegetation index and chlorophyll content obtained through active and passive remote sensing sensors. Through geostatistical analysis, spatial variability of the evaluated attributes was observed. Pearson's correlation coefficient (R) was used to quantify the correlation level between the data from different sensors, and the t-test revealed significant correlations among them.In the third article, the machine learning algorithms Random Forest (RF), Gradient Boosting (GB), Multi-Layer Perceptron (MLP), K-Nearest Neighbors (KNN), and Decision Tree (DT) were tested for predicting coffee yield. The predictive models were built using input variables such as soil moisture, leaf water potential, soil fertility, leaf nutrition, vegetation indices, plant diameter and height, and leaf area index. The output variable corresponds to the observed yield for the years 2021 and 2022. Model implementation and manipulation were conducted in Python, and performance evaluation was carried out using R² (coefficient of determination) and MAPE (mean absolute percentage error) metrics, allowing an analysis of prediction accuracy.The third part of this work presents final considerations on the studies carried out, highlighting the importance of Precision and Digital Agriculture in coffee farming.