Tree tops with mistletoe in winter, in Luxembourg.
Despite their vast differences, Luxembourg and the African continent share a common ground — the transformative power of AI in agriculture. However, the economical resources available in Europe are far greater.
Africa registers the highest percentage of smallholder farmers (0.5 to 3-hectare farms) and the highest number of people relying on agriculture for their livelihoods in the world. Additionally, 80% of employment in Africa comes from SMEs from the agricultural sector. Due to the effects of climate change, agriculture is no longer as reliable as it was in the past. As farming is used for subsistence rather than subsidence purposes, this change puts the life of the local population at high risk.
Luxembourg's and sub-Saharan Africa's plans for AI
Luxembourg's Government plans to build a reputation for itself as NVIDIA first European partner, following an agreement signed on July 5th, 2018. The country is committed to set up an AI laboratory — as per a collaboration between NVIDIA, University of Luxembourg’s LCSB (Luxembourg Centre for Systems Biomedicine) and SnT (Interdisciplinary Centre for Security, Reliability and Trust), and finally LIST (Luxembourg Institute of Science and Technology). Luxembourg then-Rector Stéphane Pallage commented: “This partnership will enable our researchers to tap into NVIDIA’s experience in addressing real-world problems using artificial intelligence.” The AI lab will open up to Luxembourg’s industry partners and startup scene, that are active in the finance, healthcare, space resources, and environmental research.
On the other hand, in Africa, researchers paired up with local stakeholders to build AI modules and make agriculture more resilient at a household level. In this context, Joyce Nabende, Head of Makerere AI Lab, in Kampala, Uganda’s capital city, outlined the importance of incorporating Indigenous knowledge with other data sources. Mobile phone cameras have proved to be the most common tool to collect data directly from the farm and the farmers, and create proxies for more advanced data collection tools that make use of satellite imagery. It is important to identify farm boundaries, to calculate areas of cultivation and know what is being cultivated within the plots (given intercropping agricultural practices, such as cultivating maize and beans at the same time).
Invasive, fall armyworm endangering African crops. Source: Pixabay.
AI applications wish to meet smallholder farmers’ main needs, in terms of detecting crop pests and diseases early on, as well as monitoring plant health at large. These apps speak directly to the producers. For example, they help farmers learn about soil: the right amount of soil moisture, its likelihood to dry out, and when to apply fertilisers or any other different input. Real-time or just-in-time information make it possible for farmers to intervene and remediate. Godliver Owomuguisha, Senior Lecturer and researcher at Busitema University, in eastern Uganda, explains how spectroscopy is used to detect plant diseases that visually manifest at a very late stage or up close to harvest, only 6 weeks in when the plant has been infected. This solution is able to prevent the loss of the crop, but is very expensive for the African context. Some low-cost prototypes for different crops are being designed.
Open-data networks for and by the farming community
Data remains fundamental to the building of these apps. The African open-data network allows for data sharing, reuse and repurposing for similar contexts. Still, communities are instrumental contributors to fill the data gap. Leo Matuku, researcher and AI Lead at the Local Development Research Institute (LDRI) in Kenya, describes radio as a unique dataset source, as the farming community is used to listen to and call into radio to get help — in her words, “agriculture is that kind of old talkshow”. A speech commission module is being built to process radio data. Matuku continues by telling about the developers’ responsibility to format datasets in a way that is representative of the smallholder farmers. With regard to this aspect, Joyce Nabende refers to community participation bearing in mind issues around ethics, bias, fairness, transparency and accountability.
Interestingly enough, climate-smart agriculture is made possible by the use of smart phones in Luxembourg as well: the Lux5GCloud initiative is an example. The project — a joint venture between SMC (Department of Media, Telecommunications and Digital Policy) and LIST — used 5G connectivity to develop a secure Cloud Hierarchy Database Platform (CHDP) for farmers, scientists and policy makers. In so doing, LIST integrated EO images collected from satellites with IoT data from in-situ sensors, and extrapolated relevant information according to end-user needs. The 5G network was provided by POST, the Cloud platform developed by InTech, while the satellite and IoT data managed by GOMSPACE and Frontier Connect respectively. New Machine Learning algorithms were capable of detecting anomalies in key parameters for accurate crop monitoring, with regard to soil moisture shortages, which is a pre-condition for managing drought adaptation and ecosystems resilience, such as foreseen by the EU Nature restoration plan of the EU Biodiversity strategy 2030. In this way, farmers can make better decisions about irrigation.
A workshop was organised to educate farmers about digitalisation, as well as to collect their feedbacks for improving the proposed application and ensuring future adoption by meeting their real needs. Farmers were invited to attend the event, along with farmers associations and advisors, representatives from the private and public sectors, institutional entities, academia, and any stakeholder involved in the smart agriculture value chain.
Carbon sequestration by land use category as per by the European Environment Agency.
Recovering EU organic soils for improved carbon sink
The European Commission explains “Restoration and rewetting¹ of organic soils in agricultural use (i.e. under grassland and cropland use) constituting drained peatlands, help achieve significant biodiversity benefits, an important reduction of green-house gas emissions and other environmental benefits, while at the same time contributing to a diverse agricultural landscape.” Peatlands — namely wetlands, such as marshes and swamps – cover only 3 per cent of the world’s land, yet they are responsible for storing twice as much carbon as all the forests. When disturbed or warmed (as in global warming), wetlands release the three greenhouse gases (GHGs) that contribute the most to global warming: carbon dioxide (CO2), methane (CH4) and nitrous oxide (N2O). In particular, CH4 emissions are linked to the decay of organic material under anaerobic conditions, which arise when drained soil is rewetted. While, N2O emissions are generally due to fertilisers, manures and crop residues.
To start with, organic soil has a high concentration of organic matter, as opposed to mineral soil. When a soil with a shallow layer of organic matter has been drained for a number of years and consequently loses carbon, it becomes a mineral soil. Organic soil is capable of higher levels of carbon sequestration;² wherein grasslands are more efficient than forests at storing carbon in the soil. As cultivation and drainage of organic soil causes significant CO2 emissions, preserving and restoring peatlands means keeping them wet, in favour of carbon sink³. Notably, Europe’s soils contain around 75 billion tonnes of organic carbon.
Dandelion shows heads on a field in Luxembourg.
Nutrition at the centre of multidimensional poverty
The circumstances of Luxembourg are clearly distinct from the ones of the African continent, in terms of food and nutrition security, not to mention the demographics. According to the 2023 global Multidimensional Poverty Index (MPI): Unmasking disparities by ethnicity, caste and gender which covers 42 sub-Saharan African countries — 534 millions (306 million of these are children) out of 1.1 billion poor people (that means half of all poor people in the world) live in Sub-Saharan Africa, accounting for 10 millions of the 12 million poor people with the highest deprivation scores (90–100%) in the world. In Sub-Saharan Africa, poverty affects an average of 49.5 percent of the population, but incidence and MPI values vary widely across countries.
Acute deprivations in health, education, and living standards are faced simultaneously by a high volume of the population. Nutrition is one of the two metrics taken into account under the health umbrella, along with child mortality. The number of poor people deprived in nutrition is similar in Sub-Saharan Africa and South Asia (around 245 million). In accordance with Zero Hunger SDG 2, feeding hungry people means meeting one of their most critical needs.
According to the World Bank, Nigeria is the largest contributor to poverty in sub-Saharan Africa, with 30.9% of the population living below the international extreme poverty line⁴ of 2.15$ per person per day in 2018/19, with poverty clustered in the country's north and in rural areas. In October 2022, headline inflation reached 21.1% year on year its highest rate in 17 years, while food inflation was at 23.7%. Michael Sunbola, founder of the 7-year-old Lagos Food Bank Initiative, aims to mitigate malnourishment in the most-populous city in the country (with 15,945,912 people, in 2023). The organisation impacts the life of 2,400,000 beneficiaries who earn 50 cents a day. By creating a database, it fights malnutrition through 8 different programmes. One of these supports backyard farming, providing seeds to families. It is the only operation of this kind in the whole country.
The structure of the Multidimensional Poverty Index by the Human Development Report Office.
Community participation in sub-Saharan African countries
Building trust within communities is pivotal when introducing outputs and recommendations from AI. Farmers in sub-Saharan Africa are used to learning about new science and technologies, and what to do about disease outbreaks, directly from government agricultural officers. Acting as information intermediary for new research the government wants to apply on a local context, they pass on information in a way that is digestible to farmers. Leo Matuku denounces the shortage of such extension informants: there is only 1 informant every 10,000 or even more farmers, while there should ideally be 1 extension informant-to-400 farmers ratio. In response to the scarcity of government agricultural officers, researchers built a network of key-farmer educators, identifying 1 farmer per village who can train on fellow farmers on new approaches and agricultural practices. Peer-farmer educators are in fact expected to step in when government officers are not available. As trusted members of the community, they can support data collection and introduce information bottom down.
Incidentally, researchers spend time with the farmers in order to get a grasp of what their challenges are. Understanding what is going on in the community, what approaches to best use to get the smallholder farmers to adopt new practices, is key to drafting messaging and recommendations with the support of stakeholders. Connecting directly with producers is fundamental to the success of the AI project — involving farmers early in the design and development of technologies through workshops and the set-up of a callcentre, secures early acceptability of the tools.
Increasing local knowledge transfer & plant health monitoring
Nabende tells about two applications that her AI Lab have been working on. Firstly, a recommended module, where a farmer in the network can come in and post a question. The app is able to automatically provide a response in the local languages farmers can understand, using NLP (Natural Language Processing). Additionally, a mobile app provides a connection between the so-called “agricultural expats” in Uganda (i.e. the government agriculture officers) and the extension farmers, those farmers who are more eligible to come in and help fellow farmers in their community.
Daniel Mutembesa, a member of Nabende’s AI Lab, informs us about how they evaluate the impact of the technology and the data crowdsourcing on community over a 3-year period of time. They hold sessions with the participants, and run quantitative and qualitative assessments to measure the economic, social, and intellectual impact of the project. Findings show that the mobile app mentioned above was able increase local knowledge transfer between and among farmers, uniting farmers with experts and fellow farmers. Half or more, of the 200+ smallholder farmers involved in the initiative, have become known or considered by fellow farmers as village information points today.
Monitoring different crops for subsistence, predicting on crop failure and the general plant health amidst climate variances, produces a major impact on the lives of smallholder farmers, who, in case of crop failure, have no income to purchase supplemental nutrition. Furthermore, market prices of crops once harvested from different year, along with an estimate of farmers’ income, make it possible for farmers to make plans around their farm and livelihood (including in terms of insurance and access to financial services). Propping up the yield can ultimately result in a return on harvest and thus, an increase of local income.
Dandelions in bloom in Luxembourg.
AI to address the farmers' needs
Predictive analysis has become more and more important to address today’s critical challenges — the global food crisis prominently qualifies as one. Continuous monitoring is required to accord traceability to supply chain as well as any LCA (life Cycle Analysis)]⁵, be it in the agricultural and farming sector or not. While locally designed and developed (recognising the relevant context and nuances), the AI modules developed in sub-Saharan Africa have the potential to benefit smallholder farmers across the globe. Whereas, Luxembourg’s AI research is very much designed for the European continent (where we primarily encounter intensive, large-scale agriculture), with the aim of addressing climate change mitigation. Additionally, the apps developed in Africa are intentionally designed to speak directly to the community of producers. We have a lot to learn from such inspiring and committed research work, that makes connecting with farmers a top priority. Farmers and policymakers are not only key to data collection. In truth, they are the ones who will adopt the new tools and agricultural practices, where AI is expected to produce a significant impact on their lives.
¹ Rewetting: Rewetting is the process of changing a drained soil into a wet soil.² Carbon sequestration: Carbon sequestration is the process of capturing and storing atmospheric carbon dioxide, the most commonly produced greenhouse gas. In the specific case of soil carbon sequestration, carbon is converted into stable organic compounds that are locked into the ground.³ Carbon sink: A carbon sink is anything that absorbs more carbon from the atmosphere than it releases – for example, plants, the ocean and soil.⁴ Extreme poverty line: Ever since September 2022, the international poverty line has been 2.15$ (in 2011, it was set at 1.90$), according to 2017 prices.⁵ LCA: LCA represents the internationally recognised technique to assess the environmental impacts of a particular product.
Sources (in alphabetical order): Africa Daily — BBC World Service; EU Commission; LIST; Practical AI: Machine Learning, Data Science; European Environment Agency; MIT; The Climate question — BBC World Service; UN.