When faced with the challenge of maintaining healthy agricultural soil, Mississippi Agricultural and Forestry Experiment Station, or MAFES, scientists are digging into the problem by looking at it from a distance-and drawing on satellites to develop a high-tech soil sampling tool.

Mississippi State received $300,000 from a National Institute of Food and Agriculture $9 million grant to develop projects to help maintain or improve soil productivity, environmental health, and sustainability.

MSU's research team combines the expertise of three assistant professors from the Department of Agricultural and Biological Engineering-principal investigator Dr. Vitor S. Martins and co-investigators Dr. Nuwan Wijewardane and Dr. Xin Zhang, all MAFES scientists. Postdoctoral fellow Dr. Lucas Ferreira is also working on the project, supporting all aspects of the research.

"Satellites are not a new tool in agricultural research, but advancing technology is now allowing us to collect and analyze satellite data from crop fields on an incredibly large scale," Martins said.

The practice of soil sampling is critical to maintaining healthy, productive fields. Regular sampling provides farmers with data to support field management decisions, with samples taken at random points or in locations based on historical data and then analyzed in a lab. While this practice allows farmers to make evidence-based management decisions, it is time-consuming and expensive, especially for small and medium-sized farms. Moreover, data taken from a limited number of locations does not always represent the entire field, as soil properties may vary widely from point to point and from season to season.

"There can be so much variation-even in a small field-in terms of the soil's moisture and other characteristics. Samples taken only one meter apart may yield very different data," said Wijewardane. "It's not practically possible to take as many samples as you need to assess the whole field."

The research team is designing a new Satellite-based Soil Sampling Design, or S3DTool, to extract data from historic satellite images using a deep learning algorithm. Where conventional methods depend on previous knowledge about the area, the S3DTool quickly provides field-wide information from multiple points in time using historical satellite images. From this data, S3DTool finds recommended sampling locations to maximize the soil sampling heterogeneity throughout the field.

"There are satellites capable of capturing a new image of a field almost every day, and from the plant canopy signal, you can evaluate how well the crops are growing," said Martins. "Since crop growth is an indicator of plant-soil feedback, we hope to use our tool to provide feedback on the soil of an entire field, with all of its variabilities."

In the early phase of creating S3DTool, the scientists turned to artificial intelligence, or AI, technology to map out the boundaries of their test site on a soybean field at the MAFES R.R. Foil Plant Science Research Center. Next, they compiled the highest quality satellite images of the site from the last 10 to 20 years, applying AI machine learning techniques to create clusters indicating optimal geographic locations to test the soil. This year, they collected 132 soil samples from locations pinpointed by the S3DTool and will conduct manual soil tests on those samples to confirm the accuracy of S3DTool's data.

"We've spent the last year building S3DTool and working out computational challenges," said Martins. "In the next year, we will validate it with interpolated maps from soil sampling results-using values from known locations to estimate values in unknown locations-and complete the tool development with demonstrations across the state."

Establishing the best possible soil maintenance practices today will allow agriculture to flourish now and for generations of farmers to come. If the team is successful, the S3DTool will one day be in the hands of farmers all over the world, providing targeted data to help them manage healthy cropland.

"We've made good progress on developing the S3DTool, but there's still a lot of work ahead," said Martins. "We're excited about the technology and its potential, and if it works, we can expand our model nationally and, possibly, even globally.

Automated Soil Sampling

The MAFES team is also developing an uncrewed ground vehicle capable of automating the time-consuming process of soil sampling. Funded by USDA-NIFA, the project uses spectroscopic technology that measures the interaction of different wavelengths, or colors, of the chemical compounds found in soil. The scientists are integrating different sensors, real-time kinematics, and GPS, providing coordinates for the autonomous vehicle.

The spectroscopic sensing will measure soil health on-site and in one machine, saving time and money.

This research is funded by the National Institute of Food and Agriculture.