Sensitivity analysis of agricultural inputs for large-scale soil organic matter modelling

The dynamics of Soil Organic Matter (SOM) and its relation to the carbon and nitrogen cycle affect many environmental problems (e.g. climate change, food security and water quality). The development of adaptation strategies requires model predictions, but for the necessary large-scale SOM dynamic studies, the quality of the input data is often limiting the reliability of the results. So we performed a uncertainty and sensitivity analysis at different sites of the federal state of Saxony, Germany, and assessed the importance of aggregated agricultural data, namely organic amendments, crop yields, area share of by-product incorporation, area share of conservation tillage and initial soil organic carbon (SOC) concentration (poram, p_yield, p_bp, pcons and p_soc respectively) on the result uncertainty by assuming an uniform error of +/- 10%. The agricultural data was regionalized from 717 long-term observation fields throughout the study region. We assessed the uncertainties of relative SOC stock change (Delta C-rel) and total nitrogen mineralisation from the organic matter (OM-N-min) and explored the changing sensitivities over the model period (1998-2014). Our results show that p_soc was the most important source of uncertainty for all sites of this study. For Delta C-rel, it is over the whole time constantly the by far most sensitive input parameter, with p_bp being the only factor of agricultural practice with some substantial influence on almost all sites. In the mountainous regions, p_cons ranks equal to p_bp, while for the sandy heathlands, none of them mark a substantial influence besides p_soc. For OM-N-min, p_soc loses its importance over time, being outranked by p_oram in the heathlands after 8 years and in the mountainous regions after 13 years. p_oram furthermore places second for all others but one other region, where p_cons is slightly more important. We therefore see the initial carbon content, the share of byproduct removal, and the amount of organic amendments as those factors, where improved data quality would bring the highest effect to reduce the uncertainty in regional SOM modelling.