Data Helps Stabilize Food Insecurity in Sub-Saharan Africa

Filling Agricultural Data Gaps Gives Farmers and Organizations Access to Powerful Tech

low-res_John-Poi-Namanjelie-on-his-farm-near-Bungoma-in-Western-Kenya.-He-has-insured-crops-including-beans-and-maize.DSC_5229
Eyes on the Ground project – John Poi Namanjelie on his farm near Bungoma in Western Kenya. He has insured crops including beans and maize.

Meridian Institute designs and implements collaborative processes with partners to solve complex problems that improve lives, the economy, and the environment. In partnership with The Rockefeller Foundation, they asked machine learning experts and data scientists, what are the most significant bottlenecks in the data science pipeline? The answer: incomplete or inaccurate data sets, especially in places like Africa, South Asia, and Latin America.

That’s when conversations began with a group of partners and donors—including co-founders Google.org, Canada’s International Development Research Centre (IDRC), and GIZ’s FAIR Forward on behalf of the German Federal Ministry of Economic Cooperation and Development—about setting up a collaborative fund to get resources to data scientists in low- and middle-income countries. The result was Lacuna Fund (lacuna means gap), which works to distribute funds to data scientists and researchers in underserved communities globally to help fill data gaps to make machine learning more equitable. 

One of Lacuna Fund’s first areas of focus was agriculture in Sub-Saharan Africa, where farmers have access to high-tech apps to identify crop yields, pests, and diseases but lack the relevant data to make the tools useful for them.

With over 100 applications from—or in partnership withorganizations across Africa, Lacuna Fund provided 11 projects with first-round funding to unlock the power of machine learning. These projects aim to alleviate food security challenges and spur economic opportunities by giving researchers, farmers, communities, and policymakers access to superior agricultural datasets.

Currently, the vast majority of datasets that fuel AI and machine learning applications contain information about North America and Europe. This means that some of the amazing applications developed to help healthcare workers, farmers, and governments, may not be applicable to large parts of the world because the underlying data may be missing at best, or inaccurate and misleading at worst.

JPM Jennifer Pratt Miles Partner Meridian Institute

The Challenge

 “Currently, the vast majority of datasets that fuel AI and machine learning applications contain information about North America and Europe,” says Jennifer Pratt Miles, partner at the Meridian Institute. “This means that some of the amazing applications developed to help health care workers, farmers, and governments, may not be applicable to large parts of the world because the underlying data may be missing at best, or inaccurate and misleading at worst.” 

In the agricultural context, many farmers today have access to a variety of high-power tech applications that can identify crop yield, pests, disease, and more—dramatically stabilizing and increasing production as well as economic benefit. These apps are often powered by satellite imagery, limiting their usefulness. 

This data by its very design leaves farmers in places like Sub-Saharan Africa at a distinct disadvantage, where food insecurity is already at an all-time high, affecting almost 60 percent of the population.

“Not only would having complete data provide a whole new service to farmers and make their livelihoods more sustainable,” notes Pratt Miles. “But it would allow for better information gathering and forecasting to help us act proactively to mitigate food security challenges.”  

Machine Learning Datasets for Crop Pest and Disease Diagnosis project led by the  Makerere University in Kampala, Uganda.

The Solution

When Lacuna Fund published a request for proposals targeting organizations in Africa with a focus on agriculture, they were pleasantly surprised by the depth and breadth of the more than 100 proposals that were submitted, and ultimately provided first-round funding to 11 groups across the continent. While the resulting projects and initiatives varied greatly, they all shared one primary goal: to create or complete agricultural data sets and most importantly, to make them widely available.

“Lacuna Fund’s IP policy states that data sets must be made public unless restrictions are needed to protect privacy or prevent harm,” said Pratt Miles. “But more and more the research community is finding ways to achieve both public access and privacy protection.”  

In addition, Lacuna Fund asked grantee organizations to engage with the local communities impacted by the data sets, create a sustainability plan for how the data would be collected and managed in the future, and perhaps most importantly, ensure that the data sets were actually utilized. To help ensure these conditions are met, Lacuna Fund requires that applicants be headquartered in—or have a substantial partnership with an organization located in—the region where the data will be collected

Multidisciplinary teams, from data scientists to agricultural experts, went to work.

In the Eyes on the Ground project, the team from ACRE Africa, an organization that provides insurance to small crop farmers, and the International Food Policy Research Institute (IFPRI) used smartphones to create a unique dataset of georeferenced crop images from 11 counties in Kenya.

“This is a novel concept that endeavors to provide smallholder farmers with risk mitigation and adaptation strategies through satellites and smart phones to ensure that they invest in high productivity agriculture,” says Lilly Waithaka, Agri-Climate Data Analyst at ACRE. “The ground pictures not only provide ACRE Africa the ability to fine-tune insurance products/models and minimize basis risk, but also to observe management practices that promote the adoption of productivity-enhancing yet resilient technologies.”

The University of Nigeria Nsukka, another group that received first-round funding, is using remote technology to monitor fish farming. These datasets will enable machine learning researchers to build models for predicting fish yield in terms of weight gain, water quality parameters, and feed consumption.

“We are hopeful this will open many opportunities to local fish farmers as it will shed tremendous light into what happens beneath the pond’s surface,” says Collins Udanor, Deputy Director at the University of Nigeria’s Education Innovation Centre. “This will indeed explain many things to the farmers and improve yields, as well as make available local datasets for the machine learning community.”

Participants of the Data Collection & Annotation for Open Science Convening, who worked at the Sensor Based Aquaponics Fish Pond Datasets project, at the University of Nigeria in Nsukka, Enugu, Nigeria.
Participants of the Data Collection & Annotation for Open Science Convening, who worked at the Sensor Based Aquaponics Fish Pond Datasets project, at the University of Nigeria in Nsukka, Enugu, Nigeria.

The Takeaway

Lacuna Fund recently brought grantee teams together in person for the first time to hear about work and outcomes from the 11 projects and to share learnings and information with each other.

“It was actually one of the most valuable parts of the entire process,” says Emma Heth, Project Associate and Ruckelshaus Fellow at Meridian Institute. “Grantee teams not only shared valuable knowledge and insights with one another, but Lacuna Fund learned a great deal about how we can improve this process in the future.”

For instance, it was clear that more support was needed to help teams move from the creation of data sets to developing and using machine learning models. “We’re now looking to scale our impact, so we support more than just the first point of the value chain in data science and move toward application,” says Heth.

And now that they have a proof of concept, they’re also looking at how to grow both funding and geography.  

In early 2023, Lacuna Fund will issue two calls for proposals for new partners. These calls will focus on data sets to better understand the relationship between climate change and forests, and sexual and reproductive health and rights.

“A key part of our plan to scale is forming regional or thematic hubs to identify partners in Africa, Asia, and Latin America, to lead and manage the grant-making process locally,” says Heth. Pratt Miles added, “This approach, paired with global, multi-disciplinary panels of experts that select grantees helps ensure the entire process is not only well informed and efficient, but that our guiding principles are always front of mind so that we’re funding datasets that have a transformational impact.”