Partitioning soil organic carbon into its centennially stable and active fractions with machine-learning models based on Rock-Eval® thermal analysis (PARTY_SOCv2.0 and PARTY_SOCv2.0_EU)

Partitioning soil organic carbon (SOC) into two kinetically different fractions that are stable or active on a century scale is key for an improved monitoring of soil health and for more accurate models of the carbon cycle. However,all existing SOC fractionation methods isolate SOC fractions that are mixtures of centennially stable and active SOC. If the stable SOC fraction cannot be isolated, it has specific chemical and thermal characteristics that are quickly (ca. 1 h per sample) measurable using Rock-Eval® thermal analysis. An alternative would thus be to (1) train a machine-learning model on the Rock-Eval® thermal analysis data for soil samples from long-term experiments for which the size of the centennially stable and active SOC fractions can be estimatedand (2) apply this model to the Rock-Eval® data for unknown soils to partition SOC into its centennially stable and active fractions. Here, we significantly extend the validity range of a previously published machine-learning model (Cécillon etal., 2018) that is built upon this strategy. The second version of this model, which we propose to name PARTY_SOC,uses six European long-term agricultural sites including a bare fallow treatment and one South American vegetation change (C₄ to C₃ plants) site as reference sites. The European version of the model (PARTY_SOCv2.0_EU) predicts the proportion of the centennially stable SOC fraction with a root mean square error of 0.15 (relative root mean square error of 0.27) at six independent validation sites. More specifically, our results show that PARTY_SOCv2.0_EU reliably partitions SOC kinetic fractions at its northwestern European validation sites on Cambisols and Luvisols, which are the two dominant soil groups in this region. We plan future developments of the PARTY_SOC global model using additional reference soils developed under diverse pedoclimates and ecosystems to further expand its domain of application while reducing its prediction error.