The COVID-19 pandemic has had far-reaching effects on not only human health, but also economic impacts around the world due to the severe disruption of our daily lives. Individuals in regions where smallholder farms comprise most of the agricultural production sector, such as many sub-Saharan African countries, are at greater risk of economic and food insecurity due to disruptions to supply chains and local food markets. However, up-to-date, and high-resolution spatial information about croplands (often called cropland or crop type maps) are often lacking in these regions, making it difficult for governments to locate farmers and assess food security and production.
Planet, in partnership with NASA Harvest, NASA’s Food Security and Agriculture Program, run out of the University of Maryland (UMD), aims to enable and advance the adoption of satellite Earth observations to benefit food security, agriculture, and human and environmental resiliency, which has become increasingly necessary in the midst of COVID-19.
NASA Harvest scientists recently supported the Government of Togo in their innovative COVID-19-related food security relief efforts by creating a cropland map, at 10-meter resolution, of the entire country. One of the West African nation’s efforts is “YOLIM” an interest-free digital loan program designed to boost food production across smallholder farms by funding the cost of farming essentials, and providing access to an e-wallet which qualified growers can use to withdraw funds towards fertilizers, pesticides or renting tractors. The Togolese Government is also experimenting with ways of utilizing the maps within the context of its flagship social protection program “NOVISSI” which is a digital cash transfer scheme aimed at providing a social safety net for vulnerable groups affected by the COVID-19 pandemic. Their aim here is to understand which communities have significant concentrations of smallholder farming activities for critical crops. These areas could then be prioritized for cash transfers under the program or may be included in a dedicated social protection campaign to protect farmers from shocks triggered by the pandemic which may negatively impact national food security.
This is where satellite data can help fill in the gaps. Thankfully, NASA Harvest maps derived from satellite data alongside poverty and census data are providing more complete insights instrumental to identifying priority areas rapidly and effectively, where programs like YOLIM and NOVISSI have the most impact.
The map produced by NASA Harvest “provides unmatched clarity into the nature and distribution of agricultural land nationwide,” states Cina Lawson, Minister of Posts, Digital Economy, and Technological Innovation of Togo. “On top of this map, we are overlaying data from poverty maps that we have developed in collaboration with UC Berkeley’s Data-Intensive Development Lab and Innovations for Poverty Action (IPA). Together, they provide decisive knowledge being used to design social protection policies aimed at improving the livelihoods of agrarian rural communities.”
“When rapid action was needed and mobility across the country was limited due to the COVID-19 outbreak, satellite data offered an effective and accelerated means to map the country’s distribution of croplands and characterize the nature of agricultural fields during the pandemic,” adds Dr. Inbal Becker-Reshef, NASA Harvest Program Director.
Earth observations can reveal where croplands are located as well as provide insight into crop conditions, yields and production forecasts, giving early warning of impending food shortages to policymakers and farmers alike. In Togo—and across Sub-Saharan Africa—many smallholder farms are less than one hectare in size, making them difficult or impossible to resolve in most satellite images. Because of this, cropland mapping has traditionally required a wealth of ground-truth data for training and verifying machine learning classifiers. However, publicly available ground-truth data is often sparse in smallholder-dominated regions and is the main barrier to developing machine learning methods to support agricultural monitoring in smallholder regions. The Togolese Ministry of Agriculture was progressive and proactive in the face of the pandemic by looking to satellite imagery to face unforeseen challenges and bolster food security throughout the country.
NASA Harvest created a new method for rapidly generating crop maps over a large heterogeneous area by harnessing the power of machine learning and high-resolution SkySat imagery from Planet and data coming from the European Space Agency’s Copernicus Sentinel-2 and from NASA-USGS Landsat satellite to map Togo’s croplands without the need for ground-truth data—all in the span of only 10 days from the initial request for the map. The generated cropland map, alongside poverty and census information, enabled the Togolese Government team to identify priority areas rapidly and effectively for its relief programs. The cropland map is now also featured in the NASA, European Space Agency, and Japan Aerospace Exploration Agency’s joint COVID-19 impact-monitoring dashboard as a demonstration of the utility of earth observations for food security applications.
With the cropland maps, Togolese government officials had trustworthy information on the physical size and geographic location of agricultural lands that census data might have missed. This collaborative effort not only emphasizes the utility of satellite-driven information in times where ground access is limited and information is needed quickly, such as during the ongoing pandemic, but also illustrates the benefits of public and private institutions working together towards a common goal. This enables timelier responses to much-needed information in support of agricultural policy decisions.
“Cropland maps are critical in times of crisis when decision makers need to rapidly design and enact agriculture-related policies and mitigation strategies, including providing humanitarian assistance, disbursing targeted aid, or boosting productivity for farmers,” says Dr. Hannah Kerner, Assistant Research Professor at UMD and lead on the Togo mapping project for NASA Harvest. For these maps to be of maximum utility for aid in response to COVID-19, they needed to be created using up-to-date imagery. Previously available maps were at least two years old and created from lower resolution imagery that did not sufficiently detect smallholder farms in the country. Dr. Kerner was pleased with the ready action from Planet commenting that, “After hearing that we needed high-resolution recent data in Togo, Planet quickly mobilized to get our team access in less than 48 hours, and were immediately responsive when we needed support during our mapping sprint.”
Kerner and her team used Planet’s SkySat imagery to manually map out the locations of hundreds of smallholder farms across Togo. These labels were then ingested into a machine learning pipeline as a training dataset, combined with thousands of crowdsourced labels covering croplands globally, to help the algorithm automatically identify pixels in Sentinel-2 imagery containing cropland. According to Dr. Kerner, “Being able to stream high-resolution data for the period we needed was critical for collecting training data for our model and validating the resulting cropland map. It would not have been possible with the Sentinel-2 open data alone as many of the smallholder farms are difficult to resolve visually in Sentinel-2 images.” The resulting maps in this work show the distribution of croplands across Togo in remarkable detail compared to previous lower-resolution maps. “This is a powerful example of combining the power of satellite imagery with machine learning for a rapid response effort to address a humanitarian crisis,” Kerner states.
The NASA Harvest team has made the code, data, and maps from this work publicly available for other scientists to use in their own research.