Simulating soil organic carbon dynamic changes in dryland and paddy field of Northeast China using RothC Model

<p dir="ltr"><b>Objective</b>: This study explored the applicability of the RothC model for simulating soil organic carbon (SOC) dynamics in dryland and paddy fields in Northeast China and evaluated the impact of various calibration methods on simulation performance.</p><p dir="ltr"><b>Method</b>: This study selected one typical dryland and one typical paddy field as long-term experimental sites. The dryland experiment was conducted at the Heilongjiang Agricultural Ecology Experimental Station of the Chinese Academy of Sciences (2004-2015), and the paddy field experiment utilized data from the 850 Farm (2010-2017). At each experimental site, two treatments were selected for model simulation validation and performance evaluation: one with fertilization only, without straw return (NPK), and the other with both fertilization and straw returning (NPKS). For the paddy field soil, in addition to the RothC model, two modified versions, including RothC_p and RothC_0.6, were also selected for suitability evaluation. Three different model calibration methods were employed: the equilibrium method, parameter optimization method, and transfer function method, to analyze the impact of these calibration methods on model simulation performance. Normalized root mean square error (<i>nRMSE</i>), mean difference (<i>MD</i>), and the index of agreement (<i>d</i>) were selected as model evaluation metrics.</p><p dir="ltr"><b>Result</b>: At the Heilongjiang station, organic carbon input exhibited a significant fluctuating trend, with the average annual carbon input under NPK and NPKS treatments being 1.71 and 3.52 t·hm^-2, respectively. In contrast, organic carbon input at the 850 Farm was relatively stable, with the average annual carbon input for NPK and NPKS treatments being 1.89 and 5.90 t·hm^-2, respectively. The simulation validation results from the Heilongjiang station showed that, under different model calibration methods, the <i>nRMSE</i> was consistently below 5%, and the index of agreement (<i>d</i>) ranged from 0.60 to 0.74. This indicated that the model performance was excellent across all calibration methods, and RothC was able to accurately simulate the SOC stock changes for both NPK and NPKS treatments in the dryland. When using the M2 method, the <i>nRMSE</i> for NPK and NPKS was the smallest, at 3.46% and 3.09%, respectively. The simulation validation results for the 850 Farm showed that the <i>MD</i> for RothC and RothC_p ranged from -1.47 to -13.41, with <i>nRMSE</i> values between 2.90% and 26.48% and <i>d</i>-values all below 0.1. This indicated that both models significantly overestimated the increase in SOC stocks and were unable to accurately simulate the changes in SOC stocks in the paddy field. For the RothC_0.6 model under the NPK treatment, the <i>MD</i> ranged from -0.08 to 0.44, with <i>nRMSE</i> values between 0.24% and 0.85% and <i>d</i>-values ranging from 0.31 to 0.76. Under the NPKS treatment, the <i>MD</i> ranged from -5.71 to -6.22, with nRMSE values between 11.21% and 12.12% and d-values between 0.12 and 0.13. These results indicated that RothC_0.6 could accurately simulate the dynamic changes in SOC stocks under the NPK treatment but significantly overestimate the changes in SOC stocks under the NPKS treatment.</p><p dir="ltr"><b>Conclusion</b>: RothC and RothC_0.6 were suitable for studying the dynamic changes in SOC stocks under dryland and paddy field conditions without straw returning in the Northeast region, respectively, and could accurately simulate the trends in SOC stocks. The impact of different model calibration methods on simulation performance was not significant. However, the transfer function method was simpler to compute, saved model running time, and provided better simulation performance. Therefore, this study recommended prioritizing the use of the transfer function method for model calibration.</p>