Objectives The DNDC and RothC models with different levels of complexity were used to simulate the dynamics of soil organic carbon (SOC) for dryland field under different amounts of straw return. The model results were compared and the causes for the differences of model simulations of two models were investigated, which provides references for model selections of cropland SOC modeling.

Methods The experimental data (climate, soil, crop, etc.) were obtained from the long-term field experiment at Changping Soil Quality National Observation and Research Station, which started in 2008 and included three treatments, namely, single application of chemical fertilizer (NPK), chemical fertilizer plus crop residue (NPKS) and chemical fertilizer plus organic fertilizer (NPKM). The DNDC and RothC models were used to simulate the dynamic changes of SOC in the dryland, and the measured 0−20 cm SOC density in the tillage layer was applied to calibrate and validate the models. Based on the validated models, the simulation results of SOC dynamics were compared and analyzed under the scenarios of different straw return rates (0, 2250 and 4500 kg/hm²).

Results Overall, both models achieved good simulation results for the SOC in the tillage layer, with the nRMSE less than 20%, the absolute value of nARE less than 15%, and the r ranging from 0.69 to 0.91, which indicated that both models were suitable for simulating the SOC dynamics of the dryland in this study area. With the increase of simulation years, the SOC densities under different straw return scenarios all showed a gradual increase, but the increase rate gradually slowed down, and the more straw return, the more obvious the increase in SOC density. The magnitude of the predicted changes in SOC varied between the two models due to the differences in the calculation of root carbon inputs, with the DNDC model predicting SOC density increase by 39%, 95% and 147% for the scenarios of no straw return, half straw return and full straw return, respectively, over a period of 190 years, and the RothC model predicting SOC density increase under the same scenarios by 104%, 206% and 307%.

Conclusions Both the DNDC and RothC models performed well in simulating SOC dynamics in dryland soils and showed similar trends in predicting SOC changes under different straw return scenarios. Due to its simpler structure and fewer input parameters, the RothC model is suitable for rapid SOC simulation and assessment. In dryland soils, it can replace the DNDC model for predicting SOC dynamics under straw return practices. On the other hand, the DNDC model simulates complex crop growth and soil biogeochemical cycles, making it more suitable for comprehensive simulation and assessment of the agroecosystem.