Sowing the Seeds of Accuracy: United States Department of Agriculture and USAID Train University Students to Map Crops in Tanzania
This post was written by Rory Nealon, senior geographic information system (GIS) analyst, USAID.
The transformation of societies brought on by rapid innovations in the digital economy has resulted in a digital skills and digital data divide. Countries at the forefront of this digital transformation can develop and adapt skills, infrastructure, data and policies for their needs. With this race to shift economies to digital, not all countries are able to evenly benefit. Through its Digital Strategy and forthcoming Geospatial Strategy, USAID supports countries as they adapt their economies to this new digital age.
The USAID GeoCenter launched YouthMappers in 2015 to address gaps in publicly accessible digital map data and to invest in the professional development of university students. The YouthMappers program prepares students with the skills needed to succeed in a modern digital economy while also mapping communities that have historically been overlooked or under-mapped. Since its beginning, the program has evolved into a comprehensive development effort that supports women’s economic empowerment, digital literacy and custom mapping campaigns focused on climate adaptation, public safety and health services.
New geospatial data created through the YouthMappers program helps countries’ digital economies develop. A well-mapped road and transportation network supports efficiency in delivering goods to markets. Access to business location information helps consumers benefit from educational and health care services, as well as find places to shop and eat. By training YouthMappers students to use modern mapping tools, they directly support the development of their own local communities.
The broader agricultural industry is already adapting to the digital economy. It has incorporated machine learning, satellite imagery, the internet of things and mobile phone apps. However, adoption of machine learning requires very large sets of data to train its models. Gaps in geospatially referenced agricultural data in developing countries has constrained the ability to apply the benefits of machine learning techniques in many places where USAID works. Detailed crop maps are needed to create agriculture production estimates and to help understand the effects of climate change. In the United States, highly detailed maps of agricultural fields have been generated from remote sensing and are more than 90% accurate for major crops such as corn (95%), rice (96%), and cotton (91%). In Africa, however, the accuracy of satellite-derived crop maps is lower than 70%. According to the authors of an article in Frontiers in Sustainable Food Systems, “a major bottleneck to the application of machine learning tools to satellite data for African farms is the lack of high-quality ground truth data.”
To help address this gap in ground truth information (data that is validated by physically visiting the agricultural fields and confirming the crop varieties that are being grown), the Foreign Agricultural Service of the United States Department of Agriculture (USDA) teamed up with the USAID GeoCenter to engage YouthMappers to pilot a solution. They trained university students to identify crops on site at the field level and use mobile phones to collect specific information about the crops. The crop data collected in the field would be combined with satellite images of the same agricultural fields to inform the machine learning process. The ultimate goal was to improve the accuracy of local crop maps, despite the data-poor environment, while teaching the students new skills.
The interagency team chose the East African country of Tanzania to apply this approach because it has strong YouthMappers chapters, a well-developed open mapping community, and many of the crops that are important to the USDA Foreign Agricultural Service. To collect a wide breadth of data and engage as many YouthMappers students as possible throughout the country, the team used a “training of the trainers” method. This meant students who attended the initial training workshop would train and conduct further field surveys with their own YouthMappers chapters after returning to their home universities.
Tanzanian YouthMappers from Arusha, Dodoma and Mwanza were selected because these areas possessed variation in the types of crops grown across a diverse geographic landscape. The students were able to survey more than 1,700 fields with different crop types and the quality of the data is currently being tested with machine learning models at George Washington University in Washington, D.C.
This pilot project is helping to improve applicability of cutting-edge technology in data-poor countries while exposing university students to the agricultural industry that is so critical to their local economies. USAID’s forthcoming Geospatial Strategy prioritizes the need to support partnerships that improve developing country capacity to use geospatial data and tools. And because the data created for this project was shared on OpenStreetMap, a web-based open mapping platform, it becomes a digital public good for others to reuse. The project tested an approach for harnessing the strengths of two different U.S. government agencies in partnership with an international academic community to advance the agriculture sector. It demonstrated that these relationships can yield valuable insights about how to apply evolving geospatial technologies to international development.