Soil organic carbon simulated with the AMG model in a high-organic-matter Mollisol

Abstract

Soil organic carbon (SOC) management requires a precise knowledge of how it is affected by soil use. Simulation models could help for this purpose. The AMG model is simple, requires information that is easily available, and uses few parameters. This model has neither been calibrated/adjusted nor validated for loamy soils with high SOC concentrations. We hypothesized that AMG would satisfactorily simulate SOC stock changes in soils with these characteristics. The aims of this work were: 1) to adjust and validate AMG for different tillage systems, nitrogen (N) fertilization levels and crop types for loamy-high-SOC Mollisols, and 2) to simulate future SOC changes under different production scenarios. We used SOC stocks (0-20 cm depth) from three long-term experiments (1976-2012) (tillage systems, crop rotations, and N fertilization) in the Southeastern Buenos Aires Province, Argentina (37º 45' S, 58º 18' W) on a complex of Mollisols. Data from two of those experiments was split into two groups to adjust unknown model parameters and for cross validation. Data from the third experiment was used for independent validation. The model was used to simulate SOC stock variation (30 yr) under different combinations of initial SOC stocks (SOCi, three levels) and crop rotations (six rotations regarding continuous cropping and crop-pasture rotations). Model performance was evaluated through statistical indicators based on observed-simulated value differences, and simple linear regression of observed on simulated values. Cross validation yielded promising indicators with the mean observed-simulated value differences close to 0 (<em>P </em>> 0.05). Root mean square error (RMSE) and RMSE as percentage of the mean of observed values (RMSEp) were 6.0 Mg C ha<sup>-1</sup> and 7.5%, respectively, which are acceptable. Simple linear regression of observed and simulated values was highly significant (<em>P </em>< 0.01) with intercept and slope not different from zero and one (<em>P </em>> 0.05), respectively, although R<sup>2</sup> was low. Indicators of model performance by groups of treatments were, in general, acceptable and did not show clear trends associated with any management type. However, model performance was poorer under no tillage (NT) and N fertilization probably because of little observed data available for that treatment factor combination. Validation with independent data confirmed that AMG simulated SOC change satisfactorily. Future scenario simulations showed that when the SOCi stock was high (close to SOC saturation), even rotations with high intensification and carbon input produced a SOC stock decrease. Conversely, when the SOCi stock was low (35% loss of SOC with respect to saturation) all scenarios led to a SOC stock increase. However, AMG failed to acceptably simulate the expected effect of pastures in the rotation. The AMG model satisfactorily simulated SOC stock changes due to different management techniques of soils with a loamy surface texture and high original SOC stock. Therefore, the model could be used as a tool to help management planning with an admissible simulation error (RMSEp ~6%).