A meta-model for soil carbon stock in agricultural soils

Abstract

Agricultural system models are frequently used to predict soil carbon (C) dynamics in agroecosystems at the field scale. Upscaling the results from those process-based models to assess the regional soil C change and sequestration potential is still a challenge due to the lack of detailed spatial information required. Meta-modelling could capture and summarize input-output relationships of systems models through simplified mathematical functions, and can significantly reduce the demand for input data required for initialization and parameterization of process-based models. Such a meta-model can be easily used for spatial assessment across regions. In this paper, we use the widely used Agricultural Production Systems sIMulater (APSIM) to generate simulation data, based on which we try to develop a simple meta-models for simulation of soil C changes across different regions in Australia. We run APSIM for 100 years (1907-2006) at 10 selected sites across the eastern Australia wheat-belt to generate soil C changes under a continuous wheat system. The 10 sites cover three annual rainfall patterns and a wide range of temperature and rainfall conditions, with mean annual rainfall in the wheat growth season ranging from 172 to 510 mm, and mean annual temperature ranging from 7.2 to 18.5°C. For each site, ten soils were used in the simulation to represent various soil types, with plant available water capacity (PAWC) ranging from 66 to 261 mm. The meta-models predict soil C content at the end of the 100-year simulation (C content) and soil C changes from 1907 to 2006 (C change). Our simulation results indicated that both C content and C change were significantly and positively related to the mean annual amount of retained residue. Residue production was significantly and positively related to PAWC and mean annual evapotranspiration, with the latter closely coupled with radiation (R), temperature (T) and precipitation (P). APSIM simulation outputs at five randomly selected sites from the ten sites and step-wise multiple regressions were used to develop the meta-model. The meta-models of C content and C change with a residual error ε are as follows: C content = 90.85 + 0.21C 0 + 0.043PAWC - 0.012R + 0.023P - 2.65T +ε , and C change = 90.85- 0.79C 0 + 0.043PAWC- 0.012R + 0.023P - 2.65T +ε . The two meta-models could explain 85% and 87% of the variation in C content and C change as simulated by APSIM, respectively. The two meta-models were validated using the modelled soil C values from APSIM at other five sites. It could explain 90% and 91% of the variation in C content and C change predicted by APSIM, respectively. This simulation study suggested that the meta-model based on basic soil and climate data could capture the soil C dynamics in agricultural soils as good as the agricultural systems model APSIM. The meta-model significantly reduces the requirement of input data, and can easily be applied for regional or continental assessment of soil C changes.