Modelling soil organic carbon 1. Performance of APSIM crop and pasture modules against long-term experimental data

We used the APSIM model to explore the ability to simulate soil organic carbon (SOC) from three long term agricultural experiments (LTAE) over 24-44years in eastern Australia (Hamilton Victoria, Wagga Wagga New South Wales and Warwick Queensland). The model was initialized for each location soil type using a 20-year spin-up period to reach stable SOC fractions. The model was not tuned to any of the data at each site. The cropping systems model simulated reasonably well the calculated changes in SOC over the 24-44-year period at the three different sites under various agronomic management practices. At Hamilton calculated SOC did not change significantly over 32years and the APSIM-Agpasture model represented this satisfactorily. Root mean square error was 4.9tCha (0-30cm) (5.2%) ranging from 1.5 to 10.9tCha (1.7 to 11.8%) over time. Similarly, over twenty-four years of continuous cropping at Wagga Wagga under a Wheat-Lupin crop rotation using APSIM-Wheat and APSIM-Lupin under stubble retention and zero tillage showed a largely neutral trend over time with RMSE of 3.4tCha (0-30cm) (9.5%) ranging from 1.4 to 7.8tCha (4 to 21%) over time. The SOC was much lower at Wagga Wagga compared to that at Hamilton. Forty-four years of continuous cropping at the Queensland site resulted in significant declines in SOC irrespective of the farming practices applied. However, the rates of decline were different with the least decline achieved from high N application with stubble retention. The modelled changes, using APSIM-Wheat and APSIM-Barley, in SOC matched reasonably well the calculated behaviour with the greatest decline under zero N application with overall RMSE of 4.1tCha (0-30cm) (4.5%) ranging from 1.4 to 8.7tCha (2 to 7%) over time. The biophysical simulation model exemplified by the APSIM model explained well the observed changes in SOC at the various locations without specifically fitting the model to the observed data, despite large variation in the calculated data. This indicates that the SOC model was robust over the moderate to long term period. As such the model is suitable to extrapolate a simulated response beyond these locations under various treatment combinations that should predict realistic SOC stocks within the expected mean error of less than 10% (3 to 6tCha 0-30cm)