Abstract:

Agriculture worldwide is increasingly practised under uncertainty — erratic rainfall, intensifying heat and water scarcity — much of it on smallholdings in remote, low-power, low-connectivity settings where reliable data network and Cloud computing hosted services access cannot be assumed. This talk argues that resilience in such environments depends on a combination of technology and people, built around an IoT–edge–cloud continuum (often referred to as the "computing continuum") in which as much processing as possible happens on the farm itself. Drawing on our work in Rural AI, this talk will discuss how inexpensive field-side units (single-board computers, low-power sensor nodes and cameras) capture soil, microclimate, weather and images (hyperspectral & thermal cameras), and run analytics locally so that actionable results reach the farmer without continuous connectivity. Using a cross-farm weed-detection example, we show how federated learning over a serverless computing continuum offers a valuable middle ground between cloud-only and edge-only approaches: each farm builds a local model, shares only the model rather than raw data, and contributes to an aggregated model that outperforms any single site while preserving data sovereignty and tolerating intermittent participation. Understanding how federated learning approaches can be utilised in such resource constrained environments also provides opportunities for use in other areas.  We also describe ForestPulse, a configurable "forest health" composite-index platform that fuses heterogeneous datasets into an interpretable, site-specific health score for a forest area, with per-indicator visualisations. Its built-in rated feedback makes it a natural mechanism for engaging farmers and field workers — addressing the reluctance to share data and the governance and adoption challenges that ultimately determine whether such decision support tools succeed. The ForestPulse project involves collaboration with the Danau Girang Field Centre (DGFC) in Malaysia.

Biography:

Omer Rana is Professor of Performance Engineering in the School of Computer Science & Informatics at Cardiff University, UK. His research spans distributed systems, edge and cloud computing, and the application of AI across the computing continuum, with a particular focus on how these technologies can be deployed in resource-constrained environments. He has led and contributed to a range of projects applying edge computing, IoT and federated learning to real world domains including precision agriculture, environmental monitoring and at the interface of energy and transport systems (e.g. for electrified transport). His current interests include serverless architectures for federated learning, the fusion of geo-spatial data sets in sustainable transport, and the design of tools that make such data interpretable and actionable for non-technical users. Much of this work is pursued through international collaborations spanning the UK, Australia, New Zealand the USA and Southeast Asia.