Modeling perspectives on soil carbon sequestration in Mediterranean regions: a comparison of process-based and statistical models in the croplands of Northern Greece

Soil organic carbon (SOC) models are utilized to extrapolate our knowledge of SOC dynamics over time and space, allowing us to evaluate SOC stocks for entire regions of interest. Numerous research works have been implemented following different approaches to evaluate SOC dynamics on a regional, national, and international scale.

In the agricultural regions of Northern Greece, two fundamentally different approaches to SOC sequestration modeling have been employed and evaluated. Public EO data from the GEE geoprocessing platform, concerning temperature and precipitation variables (ERA5), Land Cover information from the Copernicus Global Land Service (CGLS) products, MODIS-based annual NPP and GPP (Net and Gross Primary Production), and WorldClim bioclimatic variables, representing seasonality and annual trends in climatic parameters, were utilized for this purpose. The deterministic modeling approach, which is based entirely on process-based models of multi-compartment carbon decomposition and accumulation (RothC) combined with local soil observation data from 2009 soil-mapping surveys, was used to analyze SOC spatiotemporal dynamics during the decade 2009-2018. This approach was compared with the statistical, data-driven modeling approach, which was applied to the revisited points of the LUCAS Land Use/Land Cover soil observation database in the area during the same decade. A collection of global environmental covariates, selected to reflect a variety of soil-forming factors and soil-change drivers, was assembled using GEE platform resources. These covariates were utilized in the data-driven modelling approach, to generate spatial predictions of SOC, by modelling the relationship between target and auxiliary environmental variables.

The distribution of SOC dynamics in the study area was found significantly different between the two modeling approaches. In some locations, the data-driven model that was built with LUCAS data identified substantial SOC stock losses, while RothC model simulated steadily increasing SOC stocks. This discrepancy can be attributed to the inherent limitations of the RothC process-based modelling approach.

This work was co-financed by the Interreg Euro-MED Programme within the project “Capturing and Storing Atmospheric CO2 for Improvement of Soil Quality - CARBON 4 SOIL QUALITY”.