Improved prediction of soil organic carbon sequestration potentials in Austrian arable soils as simulated by multi-model ensembles

Abstract

<p>Soil organic carbon (SOC) constitutes the largest terrestrial biological carbon pool globally. SOC in croplands has declined by approximately 50% since the intensification of agriculture. In light of climate change due to rising greenhouse gas concentrations in the atmosphere, the 4p1000 initiative was launched, suggesting that anthropogenic CO<sub>2</sub> emissions could be offset by increasing SOC stocks in arable land by 0.4% per year by implementing more sustainable agronomic measures. In order to estimate the potential effect of different measures on SOC at the national scale, modelling approaches are required. In the last decades, a wide array of SOC models have been developed and validated for different soils, climate conditions and land uses across the globe. These models all have their own advantages, disadvantages, and sources of uncertainty. Carbon inputs into soil, a major driver of SOC dynamics, are an estimated quantity in all modelling procedures and represent an additional, large source of uncertainty. To reduce uncertainties, multi-model ensembles are suggested to outperform single model runs. The objective of this study is to determine the optimal SOC model ensemble to reduce estimation errors in future studies.</p><p>Therefore, a combination of four carbon turnover models (RothC, Yasso07, ICBM, and C-TOOL) and five published carbon input estimation methods was evaluated by comparing simulations to experimental data from six long-term experiments with 56 treatments on arable land in Austria, with durations from 10 to 32 years to obtain a possible optimal combination for future SOC modelling studies in Austria. Evaluation of model prediction was performed by calculating the absolute mean error (AME), Root Square Mean Error (RMSE) and coefficient of determination on yearly SOC changes to eliminate the effect of different experimental durations on model evaluation.</p><p>We show that obtained models strongly differ in their stock estimates, and our selected ensemble strongly improved the estimations of SOC against single model runs with significantly lower absolute mean errors and root mean square error. This is in accordance with literature results and presents a way forward towards a more accurate modelling. We thus argue that multi-model ensembles to estimate SOC stocks in arable soils in Austria should be preferred over single-model approaches due to improved accuracy.</p>