-
- B22A-01: Using Machine Learning and generative AI to constrain uncertainties in soil organic carbon storage & dynamics predictions (invited)
-
NOLA CC
Author(s):Generic 'disconnected' Message
Umakant Mishra, Sandia National Laboratory (First Author, Presenting Author)
Soil organic carbon (SOC) is critical for healthy soils, productive agriculture, and climate regulation. It affects how soil holds water and nutrients and how it supports plants and microbes. Changes in SOC can influence how ecosystems function and respond to environmental stress. Despite its importance, there is still a lot of uncertainty about how much carbon is stored in soils globally and how it changes over time. We combined thousands of soil observations with environmental data and advanced approaches like machine learning (ML) and artificial intelligence (AI) to better understand how environmental factors control SOC storage, how global models represent those controls, and how we can reduce uncertainty in SOC stock estimates. We found that ML could capture nonlinear relationships between SOC and environmental factors with high accuracy. However, the climate models we examined did not represent these relationships correctly. Our results suggest that using ML-based insights can help improve these models. We also used generative AI to create realistic soil data in poorly sampled areas, helping fill data gaps. Together, our work shows that combining global soil data with AI/ML can improve predictions of soil carbon storage and guide better climate and land-use decisions.
Scientific DisciplineSuggested ItinerariesNeighborhoodType
Enter Note
Go to previous page in this tab
Session
