Modelling the effects of stover harvest on soil organic carbon in the Pampas of Argentina

Our objective was to estimate the impact of harvesting stover from agricultural crops to generate biofuels or electricity on the soil organic carbon levels of the Pampean Region in Argentina. For this purpose, a carbon balance methodology based on artificial neural networks was used. Contrasting soil carbon scenarios for different subregions were constructed using a current map of organic carbon and statistical data for crop rotations. Average yields were also estimated using this information. The neural network methodology allowed calculating the annual carbon balance as the difference between estimating the contribution of carbon in crop residues (stover+roots) to the soil and losses as heterotrophic respiration. The model was run for each level of residue input until the soil carbon attained a steady-state. Current rotations were modelled, with predominance of soybean (Glycine max (L.) Merr.) and alternatives that included a greater proportion of wheat (Triticum aestivum L.) and corn (Zea mays L.). Only the stover of these latter two crops was considered to be partially harvested (30% and 60%). The input of carbon to soil was highly dependent on rotation, increasing as the proportion of wheat and corn in the rotation and the level of yield increased. In contrast, stover harvest had little impact on the carbon input due to the low proportion of both crops in the predominant current rotation. By increasing the proportion of cereal crops or the technological level and yield, it was possible to compensate for the effect of stover harvest on soil carbon. The carbon input from residue needed to maintain soil carbon ranged within 2.0–6.0 t C ha-1 year-1 depending on the initial soil carbon level. Retention efficiency of residue carbon was ~30% across different management scenarios. It is not recommended to harvest more than 30% of the stover in order to maintain the level of carbon in the soil organic matter of many Pampean soils.