Edge-Computing for Smart Agriculture: A Winery Case Study

Thesis Abstract

The rapid digitization of agricultural operations and the increasing demand for grape quality and environmental sustainability necessitate robust, low-latency data processing to support decision-making in winemaking facilities. This study investigates the deployment of edge-computing architectures within a mid-sized winery to enhance real-time sensing, processing, and decision support across vineyard and winery operations. The problem addressed is the latency, bandwidth, and reliability limitations of cloud-centric analytics for time-sensitive tasks such as irrigation control, pest and disease detection, microclimate monitoring, and fermentation process control. The aim is to design, implement, and evaluate an edge-enabled smart agriculture framework tailored to oenology supply chains, with specific objectives to (i) architect an edge-first data pipeline that integrates vineyard, winery, and climate data; (ii) evaluate the performance gains in latency, bandwidth usage, and energy consumption compared with cloud-centric approaches; (iii) validate the accuracy of edge-driven predictive models for irrigation scheduling, disease risk, and fermentation status; (iv) assess system reliability and fault tolerance under network variability; and (v) derive governance and operational guidelines for scalable deployment. The methodology adopts a mixed-methods, single-site case study over twelve months at Vinova, a winery employing precision viticulture and automated fermentation control. The population comprises on-field sensors, climate stations, IoT devices, and fermentation sensors. A purposive sample of 120 sensors (soil moisture, leaf temperature, vine vigor, ambient climate, and fermenter parameters) plus 15 edge gateways and 5 cloud endpoints is analyzed. Data collection instruments include calibrated multispectral imagery, gas and temperature probes, soil moisture meters, fermentation pH and?? (Brix) sensors, and system logs from edge devices. The study employs a quasi-experimental design to compare edge-based versus cloud-based analytics, with replication across three vineyard blocks and two fermentation tanks. Analytical techniques include time-series analysis (ARIMA, Prophet) for microclimate and irrigation forecasting, machine learning classifiers (Random Forest, Gradient Boosting) for disease risk detection from imagery and sensor fusion, and regression models to map edge latency to control performance. Model validation uses k-fold cross-validation, RMSE, AUC, and confusion matrices. A thematic analysis of operator interviews (n=20) provides insights into usability, reliability, and governance, guided by the Technology Acceptance Model and Affective Events Theory to interpret adoption dynamics. The theoretical lens integrates Resource-Based View and Distributed Systems Theory to evaluate capabilities and constraints of edge ecosystems in agro-industrial contexts. Expected findings indicate that edge-enabled processing reduces end-to-end latency by 70% for irrigation decisions, cuts data transfer to cloud by 60%, and lowers energy consumption of field devices by 25% due to localized preprocessing. Predictive models at the edge achieve an average F1-score of 0.88 for disease risk classification and a 0.92 R^2 for fermentation status forecasting, with observed improvements in grape quality indicators (sugar concentration stability, phenolic consistency) and water-use efficiency. The study may reveal trade-offs between edge computation complexity and gateway resource constraints, emphasizing the need for lightweight inference techniques and model partitioning strategies. The contribution to knowledge encompasses (i) a rigorous, replicable edge-centric framework for integrated vineyard-winery operations; (ii) empirical evidence on performance, reliability, and cost-benefit aspects of edge versus cloud processing in a real-world agro-industrial setting; and (iii) actionable guidelines for deployment, governance, and scaling in similar viticultural contexts. The main conclusion posits that edge-computing markedly enhances real-time decision support and operational efficiency in smart agriculture within winery environments, while maintaining data fidelity and resilience under network variability. Recommendations include adopting a tiered data architecture with edge-first analytics for critical control loops, implementing adaptive sampling to balance latency and energy use, and developing standardized metrics for evaluating edge deployments. Suggestions for further research include multi-site replication across diverse terroirs, exploration of federated learning for cross-vineyard model sharing, and extended cost-benefit analyses incorporating carbon footprint considerations.