Conference Agenda

The European Green Deal demand geospatial data for climate monitoring. A field trial was conducted in INIAV Elvas innovation hub, Elvas, Portugal, to compare two distinct digital workflows for nitrogen fertilisation in winter fodder crops: a heavy-infrastructure UAV-to-cloud approach and a lightweight, on-the-go (OTG) edge computing system via ISOBUS. The UAV workflow involved flight, photogrammetry, and prescription map generation, resulting in a data footprint of 3 GB per hectare and significant time lags. Conversely, the OTG system utilised local AI algorithms for instantaneous processing, requiring only 106 MB per hectare and enabling immediate variable rate application. Results demonstrate that while the UAV showed stronger correlations with crude protein (R²=0.58), the OTG system autonomously reduced nitrogen application by 15.23 percent. By reducing the data volume nearly 30-fold, the edge computing approach minimises storage costs and energy consumption, aligning with EU environmental goals. Furthermore, the local AI successfully filtered environmental noise, embedding agronomic expertise directly into the machinery. This compensates for the lack of skilled labour and eliminates dependence on external cloud servers. Embedded Edge AI is a vital pathway for European agriculture, protecting farmer data privacy and promoting local self-sufficiency through a minimal digital footprint and interoperable hardware.

Purpose: Digital agriculture increasingly depends on data‑driven systems such as FMISs, autonomous machinery and cross‑chain data platforms. Yet semantic fragmentation remains a major barrier to interoperability. This study examines why standardisation has not delivered practical, scalable data exchange in agrifood systems despite the availability of numerous standards.

Methods: We mapped the agrifood data‑standards landscape across different domains. The analysis integrates EU interoperability frameworks, sector‑driven initiatives and international efforts including ISO TC‑347 on Data‑Driven Agrifood Systems. Additional insights stem from the EU HASHTAG project, which focuses on safe and reliable geographic data exchange for precision‑agriculture applications.

Results: While standards such as AGROVOC, DCAT, SAREF4AGRI, and rmAgro exist, they remain unevenly distributed, misaligned across semantic and technical layers and seldom implemented in operational software. Sector initiatives, including the Dutch eCrop JSON working group, show promise but remain fragmented. Persistent discrepancies between national code lists and inconsistent uptake of EU‑promoted schemes such as EPPO further hinder harmonisation.

Conclusions: Progress requires machine‑executable, workflow‑aligned and cross‑domain standards, supported by coordinated action across research, industry, policy and standardisation bodies. These standards must be well‑documented, openly published and accompanied by clear implementation guidance to enable integration into agricultural software.

The evolution towards Agriculture 5.0 requires real-time management of heterogeneous data generated by IoT sensors, interconnected agricultural machinery, and robotic systems. In this context, this study proposes an IT architecture based on Apache Kafka, used not only as a message broker but as a central platform for agricultural data integration. The aim is to demonstrate how the Kafka ecosystem, integrated with domain-specific tools such as direct communication with ROS2, can support timely Apache Kafka as a data crawling and ingestion layer. In rural environments characterized by intermittent connectivity, the architecture acts as a resilient bridge, ensuring that field data (e.g., weed presence and water stress indicators) are reliably transmitted to a central data cluster. A multi-broker architecture is introduced, including an offline Apache Kafka broker deployed on agricultural and robotic vehicles, enabling continuous sensor data acquisition and synchronization with a remote broker when a stable connection is available. The results show that Apache Kafka is a robust and scalable solution for digital agriculture. The proposed implementation, optimized for the integration of tractors, operators, robotic platforms, and onboard sensors, enables system standardization and supports the adoption of a microservices-based paradigm in smart farming.

Conventional science advances by accumulating validated findings, while datasets without
clear conclusions are often abandoned. AI-driven research can set a testable end goal
(behavior or value) and let AI automate data rescue and statistical checks, so incomplete
clues can be recombined across domains.
We analyze a ~40-year-old octopus patch-clamp gating-current file and propose an
ion-channel quantum-network model. Among many quantum-brain ideas, we treat
invention/discovery, self-awareness, and qualia as convergence to a global optimum in a
search tree; recursion strains classical LLM context limits, while quantum superposition
suggests an alternative. Ion selectivity may arise from quantum tunneling and resonance
in a multi-peaked pore potential; simulations of an AI-generated toy model reproduced
selective transmission.
AI inferred the legacy format (ASCII+binary headers, data offset, int16 little-endian, 1 kHz,
110 s). In the extracted open-onset event train, sub-Poisson-like behavior appeared at
specific counting bin widths (Dt) but was not significant against classical surrogate models,
implying weak quantum signatures. Weak correlations could still scale: if a fraction f of N
channels is correlated, additive readout gives SNR » fÖN; with N»10^17, macroscopic
amplification is plausible. Coupled with our tunneling-and-stochastic-resonance
global-optimization algorithm, this may extend classical LLMs toward consciousness-like
functions.

Common lands in Portugal, known as baldios, are administered by local communities supported by agricultural and forestry associations, which are still dependent on slow-paced administrative procedures, including a predominantly human-centred management model and a heavy reliance on manual fieldwork, limiting spatial and temporal monitoring coverage. To adress this chalenges, this work proposes a digital platform for systematic vegetation monitoring that integrates the satellite-derived spectral Normalised Difference Vegetation Index (NDVI) with historical fire records. The platform was developed in Power BI and processes rasterised satellite imagery covering the entire North of Portugal at 100-metre resolution, with fire data spanning the last five years. Three common lands located in Peneda-Gerês and Montesinho Natural Parks were selected as case studies. The platform enables visualisation of historical NDVI trends, identification of fire-affected and at-risk areas, regional comparisons, and projection of future vegetation trends. Centralising these data into a single, accessible platform will significantly reduce the time and effort required for land monitoring. Field technicians responsible for baldios management can access relevant environmental information in one place, supporting faster and more informed decision-making. This contributes to a more efficient, transparent, and proactive governance of common lands, with direct benefits for biodiversity conservation and ecosystem management.

This study presents an integrated machine learning framework developed for the Syngenta Digital Agtech Cropwise Operations platform to automate large-scale agricultural monitoring. The purpose is to overcome the limitations of manual digitization and cloud-induced data gaps in digital agronomy. The methodology employs deep learning semantic segmentation (U-Net–based architectures) for automated field boundary detection and temporal machine learning models for crop type identification and phenological growth stage detection. In addition, a separate machine learning service was developed to generate synthetic NDVI by fusing Sentinel-1 SAR and Sentinel-2 optical imagery, enabling vegetation monitoring during periods of persistent cloud cover. Results indicate that the system successfully generates comprehensive "Country Crops Maps", providing automated field boundaries in vector format, crop classifications, and early-season yield estimations across entire regions. Furthermore, the synthetic NDVI service enables cloud-independent vegetation monitoring with increased observation density. The study concludes that integrating multimodal satellite fusion with advanced computer vision provides a scalable, resilient solution for global crop monitoring. This architectural approach enhances the precision of agronomic recommendations and strategic supply forecasting, representing a significant advancement in biosystems engineering and digital transformation for the agricultural sector.