AgroHPC

Agriculture Empowered by Supercomputing

Objective

Present HPC as part of a solution to complex interactions between e.g. climate change and adaptation; weather forecasting and high resolution climate models; financial and commodities markets; trade, supply chain and transportation; new algorithms and data science; plant breeding and genomics research; all intertwined with the strategic and operational decisions to be made at the farm. We hope that this workshop will inform end-users (farmers, corporations, governments) on how supercomputing could empower the primary sector while challenging technology colonisation concepts and discussing modular solutions adaptable to local conditions. 

Confirmed Keynotes.

• Prof. Ewa Deelman. Research Professor and Research Director, University of Southern California, Information Sciences Institute. Recipient of the IEEE Computer Society Sidney Fernbach Memorial Award (2025). AAAS, IEEE, and USC/ISI Fellow. 
• Prof. Manish Parashar. Inaugural Chief AI Officer; Executive Director, Scientific Computing & Imaging Institute; and Presidential Professor, University of Utah’s Kahlert School of Computing. 2025 ACM Distinguished Service Award. AAAS, ACM, and IEEE Fellow. 
• Prof. Dhabaleswar K (DK) Panda. Professor and University Distinguished Scholar of Computer Science and Engineering at the Ohio State University. Director of the NSF-AI Institute, Intelligent Cyberinfrastructure with Computational Learning in the Environment (ICICLE). 2022 IEEE Charles Babbage Award. ACM and IEEE Fellow. 

Workshop Keynote Speaker

Abstract: Artificial intelligence (AI) is transforming every sector of society. Digital Agriculture is one of the prime candidate domains to be transformed through AI during the coming decade. However, this requires a set of tools and services which can be integrated together in a plug-and-play manner for various crops and common activities in agriculture (such as weed/disease detection, precision spraying, understanding crop development and non-uniform emergence, timely harvesting, etc.) In this talk, I will provide an overview of tools and services designed by the NSF-AI Institute ICICLE (https://icicle.osu.edu/) to help agriculture communities to adopt AI with ease for the above-mentioned activities for diverse crops. End-to-end workflows and solutions using sensors, drones and tractors to carry out multiple tasks in AI-driven digital agriculture pipelines (such as data labeling and annotation, image segmentation, data transfers, model training, inferencing) will be highlighted. The effectiveness of these tools and services from field-level deployments will be demonstrated.

Abstract: Selective spraying represents one of the most significant disruptions in modern agriculture since the shift to no-till farming, and stands among the first artificial intelligence applications on the edge to reach commercial scale. SprAI, DeepAgro’s retrofit system, performs real-time green-on-green weed detection with deep neural networks running on embedded hardware mounted directly on sprayer booms at speeds up to 18 km/h. Deployed on over 100 machines across Argentina, Brazil, Uruguay, and the US, it has covered over one million hectares with ~70% average herbicide savings. This talk presents the full edge-to-HPC loop behind that deployment: (1) neural network inference under hard latency, power, and environmental constraints; (2) data logistics from rural fleets, image storage, and IoT telemetry feeding a proprietary multi-year image corpus; (3) embedding-based similarity search over that corpus for dataset curation; and (4) GPU training and over-the-air model updates closing the cycle. We conclude by outlining the open challenges in bringing large-scale computing to agriculture and where the HPC community is best positioned to contribute.

Abstract: Abstract: Precision agriculture generates growing volumes of data, yet their agronomic value depends on preserving, interpreting, and linking them throughout the production cycle. Within a single season, one farmer may rely on different contractors for planting, spraying, and harvesting, using machinery, displays, and platforms from multiple manufacturers. The resulting records differ in proprietary formats, semantics, units, spatiotemporal resolutions, and identifiers. This fragmentation prevents advisors from reconstructing a trustworthy field history and limits the downstream use of advanced analytics and artificial intelligence.
This work examines the problem through the Argentine production model and proposes an interoperable, offline-first reference architecture spanning the edge-to-HPC continuum. The approach includes adapters for heterogeneous sources and open standards; local validation, geospatial harmonization, and provenance capture at the edge; asynchronous synchronization under intermittent connectivity; and consolidation into a common repository supporting parallel workflows for quality control, multi-season integration, modelling, and machine learning. It preserves the farmer-farm-field-season-operation-machine-contractor relationship, supports farmer data sovereignty, and decouples field operations from any single vendor or platform.

Abstract: Agriculture has no line in our vertical taxonomy, so it sits outside every vendor roadmap and every funding line. The technology arrived anyway. A modern sprayer is an edge inference cluster, and independent field trials have beaten the manufacturer’s own published savings claims. Capability is not the constraint. Access, agronomic skill and data ownership are. Our research puts a large share of HPC and AI spending in smaller systems rather than flagship machines, and Latin America already owns the expensive layers. The data layer on top of them is not missing. It belongs to someone else. My recommendation: put it under agricultural cooperatives rather than vendors or ministries, on public compute the region already has, with data portability as a condition of sale.

Abstract: Plant breeders are eager adopters of HPC and advanced large-scale data analysis techniques because these technologies have been shown to accelerate the selection of plants with advantageous traits. Crop improvement must keep pace with rising yield demands and simultaneous mounting environmental pressures, including floods, drought, heat, pests, and salty soils. This talk describes how plant breeders use HPC to find associations between genotypes and phenotypes, to predict plant phenotypes from genotype, and to simulate crop yield under varying environmental conditions. In particular, we will discuss use cases where HPC is currently being applied to crop improvement in Latin America, as well as opportunities for AI-enabled cloud computing to relieve bottlenecks in the breeding pipeline, such as with high-throughput phenotypic processing. These examples demonstrate resource-efficient approaches that make a practical difference in plant breeding programs.

Abstract: “Listen to the Land” (Whakarongo ki te whenua) is a design for a nation-wide sensor and analytics platform for fine-scale, continuous weather monitoring to support agriculture and climate resilience. Its scale requires supercomputer-level computation and communication. However, geographical dispersal and exposure to weather (winds up to 250 km/h, temperatures from -26C to 42C, rainfall up to 200 mm/hour) make failures possible that typical supercomputers will never see. This presentation describes the proposed system, surveys some of the failure modes and proposes mitigations for them.