FLC | Berkeley Lab machine-learning software has water management applications

Berkeley Lab machine-learning software has water management applications

June 12, 2020

Lawrence Berkeley National Laboratory researchers, led by Nicola Falco and Haruko Wainwright, are partnering with Park City, Utah-based Arva Intelligence to develop machine-learning software that integrates Department of Energy (DOE) environmental databases with local-scale monitoring and sensing for precision agriculture applications that include water management.

The project is one of three that are funded by the DOE and leverage Berkeley Lab’s strengths in artificial intelligence, sensors, and ecological biology. They aim to quantify and reduce the carbon intensity of agriculture while also increasing yield.

The DOE databases and facilities – such as AmeriFlux and ESS-DIVE, a repository for Earth and environmental science data – provide critical datasets for understanding agro-ecosystem functions on the regional and national scales. These functions include greenhouse gas fluxes, evapotranspiration, and soil biogeochemistry.

“The DOE’s databases and user facilities offer powerful assessments, yet they have been rarely used for ecosystem management,” Falco said.

In this project, the researchers will develop a scalable software system to couple the local-scale datasets – such as from sensors monitoring water, nutrients, and fertilizers – with the DOE datasets. The software will be able to, for example, couple evapotranspiration estimates derived from Ameriflux with local soil sensors and drone images to provide information on water management practices. The DOE has awarded Arva a small business grant of up to $200,000 for this project.

“We are excited to be working with the Department of Energy on another sustainable agriculture project," said Arva CEO Jay McEntire. "Building off the success we are seeing in artificial intelligence to support the health of farmland and raise farmer profits, we are proud to continue our work with Lawrence Berkeley National Laboratory to expand the horizons of precision agriculture.”

The prototype software being developed will demonstrate its utility through three tangible and measurable applications:

* water management based on in-situ soil sensors, geophysics, and unmanned aerial vehicle (UAV) images coupled with evapotranspiration estimates derived from AmeriFlux

* evaluation of farm practices, such as tilling versus no-tilling and water manipulation, on GHG fluxes including carbon and methane in rice fields based on high-resolution imagery and Ameriflux

* identification of soil biogeochemical properties and the link to soil functional types through Kbase and ESS-Drive.

Berkeley Lab’s efforts to leverage machine learning for sustainable agriculture started in 2018 on a farm in Arkansas. Falco and Wainwright have since developed a suite of algorithms, now available for licensing, to help farmers estimate sprout and plant density using images taken by drones. These estimates, in turn, enable real-time adjustments to boost productivity.

“Berkeley Lab has extensive expertise characterizing soil-plant interactions and other terrestrial ecosystem properties across scales,” Wainwright said. “By working with Arva and its machine-learning capabilities, we hope to transfer our tools and knowledge more quickly for operational management.”