Investigating the complementarity of thermal and physical soil organic carbon fractions

Abstract. Partitioning soil organic carbon (SOC) in fractions with different biogeochemical stability is useful to better understand and predict SOC dynamics and provide information related to soil health. Multiple SOC partition schemes exist, but few of them can be implemented on large sample sets and therefore be considered relevant options for soil monitoring. The well-established particulate organic carbon (POC) vs. mineral-associated organic carbon (MAOC) physical fractionation scheme is one of them. Introduced more recently, Rock-Eval® thermal analysis coupled with the PARTYSOC machine learning model can also fractionate SOC into active (Ca) and stable SOC (Cs). A debate is emerging as to which of these methods should be recommended for soil monitoring. To investigate the complementarity or redundancy of these two fractionation schemes, we compared the quantity and environmental drivers of SOC fractions obtained on an unprecedented dataset from mainland France. About 2000 topsoil samples were recovered all over the country, presenting contrasting land cover and pedoclimatic characteristics, and analysed. We found that the environmental drivers of the fractions were clearly different, the more stable MAOC and Cs fractions being mainly driven by soil characteristics, whereas land cover and climate had a greater influence on more labile POC and Ca fractions. The stable and labile SOC fractions provided by the two methods strongly differed in quantity (MAOC/Cs=1.88 ± 0.46 and POC/Ca=0.36 ± 0.17; n=843) and drivers, suggesting that they correspond to fractions with different biogeochemical stability. We argue that, at this stage, both methods can be seen as complementary and potentially relevant for soil monitoring. As future developments, we recommend comparing how they relate to indicators of soil health such as nutrient availability or soil structural stability and how their measurements can improve the accuracy of SOC dynamics models.