Guest researcher Giulio Genova focuses on deep learning for soil mapping
A new guest researcher, Giulio Genova, joins ISRIC for five months in order to advance efforts applying deep learning to digital soil mapping.
Giulio arrived in September from Italy where he is pursuing a PhD at the University of Bozen-Bolzano in partnership with the EURAC Research Institute for Alpine Environment. His area of focus is to provide better maps of heavy metals concentrations in the soil. The heavy metals he is mapping, such as copper and zinc, are the result of land management practices associated with vineyards and orchards.
Here at ISRIC, Giulio is collaborating with Laura Poggio and Luís de Sousa to experiment with applying deep learning methods at a global scale that may improve digital soil maps. This involves analysing nearly one million soil observations and the use of high-performance computing facilities.
“Better soil maps empower people with better base products that they can use for research,” Giulio explained. “If you think about soil carbon stocks, for instance, they are an important part of the carbon cycle which we need to map in order to manage climate change.”
The digital soil map improvement which Giulio, Laura and Luís are experimenting with applies convolutional neural networks to create maps. This type of approach is a powerful tool to assist a computer to recognize patterns across map layers. This is the same type of pattern-recognition technology that allows a smartphone to perform facial recognition but here it is applied to soil maps at a global scale.
Making better soil map products means making better models which means better understanding our world.
“These convolutional neural network algorithms, they can better understand the landscape context for soil data,” Giulio said. “It’s worth a try because it might give us really interesting insights. Making better soil map products means making better models which means better understanding our world.”
Giulio is at ISRIC from September 2020 – February 2021.