Scientists from the University of Nebraska-Lincoln in the United States and Wageningen University in the Netherlands have used machine learning and large-scale data analysis to refine the Global Yield Gap Atlas (www.yieldgap.org), creating benefits for agricultural producers worldwide.
“This is very exciting. We are now able to estimate the yield potential for every single piece of cropland around the world,” says Patricio Grassini, professor with the University of Nebraska–Lincoln’s Department of Agronomy and Horticulture and one of the principal investigators of the Global Yield Gap Atlas.
The result, Grassini said, “offers a fantastic opportunity for farmers to benchmark their current productivity, and for orienting sustainable intensification of agricultural systems at the global level.”
Fernando Aramburu-Merlos, a research assistant professor with the University of Nebraska-Lincoln, led the efforts to develop a “metamodel” approach that takes advantage of machine learning’s ability to assemble and analyze a complex range of soil, climatic and cropping system information.
The resulting maps, focusing on corn, wheat and rice, provide highly detailed and accurate information on how much a specific location can potentially produce a crop under the best conditions. “With this metamodel, we take full advantage of the large amount of data collected by the Atlas over the last 12 years to get a global product that is ready and easy to use by scientists, farmers and companies,” said Aramburu-Merlos, an expert in crop simulation modeling.
Aramburu-Merlos and coauthors explain the refined analytical approach in a new paper recently published in the journal Nature Food. “Over the past decade,” they wrote, “substantial improvements in computing power, spatial information on soil and climate, and advancement in the use of machine learning for geospatial analysis have provided new tools that can help address the limitations of bottom-up and top-down approaches.”
The metamodel approach uses machine learning to build on the Global Yield Gap Atlas’s bottom-up methodology, bringing in a complex range of data sets and using a nuanced analytical approach that bolsters accuracy. As a result, an ag producer can find detailed, accurate yield potential data for his or her specific location.
With this information, farmers can diagnose their current performance and reflect on what can be done to close the gap between what they currently produce and what they can potentially produce with improved management, given their local climate and soil conditions.
“Applications go well beyond farm level,” Grassini said. “The refined capacity to estimate yield potential offers governments, international organizations and charitable foundations with a transparent and objective approach to understand where the largest opportunities for yield improvement exist.”
As a result, he said, the metamodel data “provide key inputs to inform their investments on agricultural research and development programs.”
Combining crop modeling and machine learning to help identify opportunities for yield intensification
