WASHINGTON, Dec. 09, 2019 (GLOBE NEWSWIRE) -- To make geospatial information more accessible to data scientists who are working on global priorities like food insecurity, Radiant Earth Foundation has launched Radiant MLHub, the world’s first cloud-based open library dedicated to Earth observation training data for use with machine learning algorithms, it announced today.

Over a quarter of the global population — 26.4%, or approximately 2 billion people — experienced moderate or severe levels of food insecurity in 2018, according to the United Nations (UN), which cites as major causes increasing food prices and decreasing food production, not to mention trends like climate change and population growth, which could boost agricultural demand by 50% by 2050.

Despite significant challenges, data from satellite imagery could help the global development community achieve the UN’s goal of zero hunger by 2030 by supporting the flow of social and economic resources to individuals and communities that need them most.

There’s just one major problem: Although there’s an abundance of satellite imagery, there’s a shortage of training data and tools to advance applications of machine learning on them.

Enter Radiant MLHub, an open digital data repository that allows anyone to discover and access high-quality Earth observation (EO) training datasets and machine learning models. In addition to discovering others’ data, individuals and organizations can use Radiant MLHub to register or share their own training data, thereby maximizing its reach and utility. Furthermore, Radiant MLHub maps all of the training data that it hosts so stakeholders can easily pinpoint geographical areas from which more data is needed.

Distributed under the Creative Commons license (CC BY 4.0), training datasets hosted on Radiant MLHub give data scientists and other machine learning enthusiasts benchmarks they can use in order to train and validate their algorithms and improve its performance.

Radiant MLHub debuts today with “crop type” training data for major crops in Kenya, Tanzania and Uganda. Based on multispectral data from the European Space Agency’s Sentinel-2 mission, including temporal data captured by Sentinel-2 during the growing season, these datasets contain information on wheat, maize, sorghum and various vegetables, which are supplied by Plant Village, Dalberg Data Insights and the Great African Food Company. The “crop type” training datasets are the first of many global training datasets that Radiant Earth Foundation will release independently and in collaboration with its partners encompassing a range of issues. Future plans, for example, include datasets on global land cover types based on Sentinel-2 observations and surface water based on Sentinel-1’s.

Radiant Earth Foundation chose “crop type” as its first dataset because crop classification is a building block for understanding macro-level trends in food crops. Traditionally, agricultural classification is achieved by visiting farmers directly in order to identify the types of crops they’re growing, their diversity and their growth cycles, among other data. Administering such surveys is time-consuming and requires a large workforce. As such, surveys typically are only carried out once every five years by wealthy economies like the United States. In developing regions like Africa, agricultural censuses typically are completed just once every decade. Although individual researchers and NGOs may have carried out their own surveys, only 57% of African countries completed at least one government-sponsored agricultural survey between 2007 and 2017, according to the World Bank.

Because doing so automates data collection and analysis, applying machine learning algorithms to satellite imagery allows agricultural surveys to be completed faster, cheaper and therefore more often.

“ Projections of population growth in sub-Saharan Africa predict an increase of 1.3 billion to 4.3 billion people in the next 80 years,” said Anne Hale Miglarese, founder and CEO of Radiant Earth Foundation. “Given the high rate of food insecurity that already is facing people in this region, and which will continue to be exacerbated by climate change, we must drive innovation to continuously monitor sustainable supply and distribution channels and re-orient our agricultural production systems around them. The training datasets that we are releasing on Radiant MLHub are a first step toward building open machine learning models that can dynamically identify crop type and in the future yield to inform future food production.”

This first training dataset released on Radiant MLHub is funded by The Patrick J. McGovern Foundation.

“We are thrilled about the launch of Radiant MLHub, a significant development for the global effort to end widespread food insecurity,” said The Patrick J. McGovern Foundation Trustee Dr. Liz McGovern. “Open data and machine learning innovations like those being developed by Radiant Earth hold the promise of driving new solutions to the most urgent challenges facing humanity.”

To gain access to the data, download this how-to-guide.

About Radiant Earth
Founded in 2016, Radiant Earth Foundation is a nonprofit organization whose mission is empowering organizations and individuals with open AI and EO data, standards and tools to address the world’s most critical international development challenges. Radiant Earth’s goal is to make Radiant MLHub the primary repository for geospatial training data that can be used by machine learning algorithms to conduct satellite imagery analysis. Visit us on Twitter, LinkedIn, Facebook, Instagram, Medium and GitHub.

Media Contact:
Louisa Nakanuku-Diggs
Louisa@radiant.earth
+1.202.596.3603