Speaker
Description
Agricultural soils are a primary source of anthropogenic $\text{N}_{2}\text{O}$, a potent greenhouse gas (GHG). It is emitted from the cultivation of crops, especially those with low nitrogen (N) use efficiency such as winter oilseed rape (WOSR) wherein high amounts of N fertilizer are added to soils to ensure high crop productivity. WOSR is an important crop cultivated for fuel, food and fodder in Germany. In order to reduce $\text{N}_{2}\text{O}$ emissions from its cultivation, N inputs must be optimized and suitable cultivars need to be developed with better N use efficiency (Stahl et al., 2017). $\text{N}_{2}\text{O}$ fluxes also depend on soil texture, weather and N availability (Ruser et al., 2017) which are assessed using agroecosystem models to develop management strategies for low emissions and high yields. Soils, in particular, influence plant N availability, leaching etc., which depends on soil-specific characteristics. They exhibit spatial heterogeneity and their properties are uncertain at regional scales (Folberth et al., 2016). Not accounting for their uncertainty leads to unreliable model predictions. Bayesian inference is used to account for difference sources of uncertainty including those in model inputs such as soils. In this study, we aim to account for soil uncertainty while using Bayesian inference to calibrate a crop model to observations of WOSR grown in Germany in order to improve crop model predictions and obtain representative uncertainty estimates of $\text{N}_{2}\text{O}$ emissions.
We use observations from 5 most extensively grown WOSR varieties in the official variety trials performed by the Federal Plant Variety Office (Bundessortenamt), from 9 sites across Germany between 2009 and 2020. For each variety, 30 growing seasons are used for calibrating the DSSAT CROPGRO model to phenology, seed yield as well as seed oil and protein content. The considered site-specific soil profiles and respective horizon-wise properties, which are model inputs, are based on the BUEK200 soil map (Bundesanstalt für Geowissenschaften und Rohstoffe). The map includes soil profile descriptions along with a percentage of their areal coverage in each spatial unit. For each trial site, three profiles with the highest areal coverage are chosen. The percentage of areal coverage are interpreted as prior information about their probability of occurrence at the given site and are assigned as prior weights for each profile. During Bayesian calibration of the model to observations, these weights are updated along with the crop model parameters. The resultant posterior soil profile weights indicate which soil profile is more likely to occur at the trial sites, given the crop observations. To assess the impact of accounting for soil uncertainty, results from this calibration scenario are compared with those in which only the crop model parameters are estimated: one with a single soil profile per site and the other with top three soils weighted as per their areal coverage in the BUEK200. The resultant model output variables of crop yield and $\text{N}_{2}\text{O}$ are expressed as probability distributions.
The prior uncertainty in soil profiles at each site is constrained by calibrating the model to data from multiple varieties grown at the same set of trial sites. This study could provide relevant insights into the importance of soil type on yield and $\text{N}_{2}\text{O}$ simulations. Furthermore, accounting for soil uncertainty could also result in a more representative estimate of crop model parameters, which would otherwise compensate for errors in soil inputs. Bayesian calibration is used to account for uncertainties in soil input, model parameters and observations, resulting in improved estimates of uncertainties in yield and $\text{N}_{2}\text{O}$ emissions. These results could feed into life cycle assessment-based GHG accounting studies.
Stahl A., Pfeifer M., Frisch M., Wittkop B., Snowdon R.J. (2017), Recent Genetic Gains in Nitrogen Use Efficiency in Oilseed Rape, Front. Plant Sci, 8:963, doi: 10.3389/fpls.2017.00963.
Ruser R., Fuß R., Andres M., Hegewald H., Kesenheimer K., Köbke S., Räbiger T., Suarez Quinones T., Augustin J., Christen O., Dittert K., Kage H., Lewandowski I., Prochnow A., Stichnothe H., Flessa H. (2017), Nitrous oxide emissions from winter oilseed rape cultivation, Agriculture, Ecosystems & Environment, Volume 249, pp. 57-69, doi: 10.1016/j.agee.2017.07.039.
Folberth, C., Skalský, R., Moltchanova, E., Balkovic J., Azevedo L.B., Obersteiner M., van der Velde M. (2016), Uncertainty in soil data can outweigh climate impact signals in global crop yield simulations, Nat Commun 7, 11872 (2016), doi: 10.1038/ncomms11872.
| Keywords | winter oilseed rape; DSSAT CROPGRO model; Bayesian calibration; N2O emissions; soil uncertainty |
|---|
