Artificial intelligence and soil carbon modeling demystified: power, potentials, and perils

Abstract

The global soil carbon pool has been estimated to exceed the amount of carbon stored in the atmosphere and vegetation, though uncertainties to quantify below-ground carbon and soil carbon fluxes accurately still exist. Modeling soil carbon using artificial intelligence (AI) - machine learning (ML) and deep learning (DL) algorithms - has emerged as a powerful force in the carbon science community. These AI soil carbon models have shown improved performance to predict soil organic carbon (SOC) storage, soil respiration (Rs), and other properties of the global carbon cycle when compared to other modeling approaches. AI systems have advanced abilities to optimize fits between inputs (geospatial environmental covariates) and outputs (e.g., SOC or Rs) through advanced pattern recognition, learning algorithms, latent variables, hyperparameters, hyperplanes, weighting factors, or multiple stacked processing (e.g., convolution and pooling). These machine-oriented applications have shifted focus from knowledge discovery and understanding of ecosystem processes, carbon pools and cycling toward data-driven applications that compute digital soil carbon outputs. The purpose of this review paper is to explore the emergence, applications, and progress of AI-ML and AI-DL algorithms to model soil carbon storage and Rs at regional and global scales. A critical discussion of the power, potentials, and perils of AI soil carbon modeling is provided. The paradigm shift toward AI modeling raises questions how we study soil carbon dynamics and what conclusions we draw which impacts carbon science research and education, carbon management, carbon policies, carbon markets and economies, and soil health.