Simulation of Key Soil Properties in Two Long-Term Cropping Experiments in Subtropical Brazil: A Daycent Validation Exercise.

The objective of this study was to test the DayCent model using datasests from two long-term (±30 years) cropping experiments at Embrapa Wheat Research Center in Passo Fundo, Southern Brazil). Six treatments (3 rotations: wheat/soybean, wheat/soybean/vetch/corn and vetch/wheat/ wheat/soybeans, under conventional and no till management) from one experiment were used for calibration and 2 treatments  (2 rotations: wheat/soybean, wheat/soybean/corn/oats/soybeans, under no till for the last 24 years) from another experiment for validation. Both experiments have been sampled for soil properties, including multiple C and N stock analysis; since 2013, soil gas (CO₂, N₂O and CH4) fluxes.

DayCent was initialized with local environmental conditions, soil and crop parameters from both sites. Plant production parameters (PRDX) were iteratively adjusted to match observed data. Statistical analysis of observed and simulated crop biomass  showed coeficient of correlation (CC) of 0.8, and lack of fit (LOFIT) 0.05. Soil C stocks were initially overestimated by the model. Large increases (up to 200%) in the default CULT (cultivation) parameters were required to improve soil C stock simulation as compared with observed data (CC= 0.94). Parameters drainlag (# days between rainfall and drainage) and Ksat (saturated hydraulic conductivity ) were adjusted to achieve better model simulation of soil water.  Soil NO₃was poorly modeled, with an overall trend of overestimation.

 Greenhouse gas simulation focused on N₂O fluxes because CO2 and CH4 require  additional parametrization of the plant production submodel. After several iterations with adjustment of N gas parameters (N2Oadjust_fc, wfps_deni_adj, N2/N2Oadj), simulated and observed N₂O fluxes were still poorly correlated (0.21, p<0.05), but with a small root mean square error and without bias. Transient N₂O emision peaks and some N₂O influxes were not captured by the model. On the other hand, DayCent simulated total yearly N₂O fluxes adequately in comparison to the measured totals.

For the validation step conducted with an independent dataset, emphasis was on evaluating N₂O fluxes.  Site and crop parameters were slightly adjusted to reflect diferences between experiments. Statistical analysis of N₂O emissions indicated that Daycent performance was inadequate, with no correlation (0.06) and high RMSE (414,4) between simulated and measured fluxes. However, a visual (plot) analysis of  simulated N₂O fluxes showed that DayCent could capture the general trajectories of theses emissions. Cumulative annual fluxes were simulated adequately.

In conclusion, DayCent applications in Southern Brazilian Subtropical agroecosystems require additional calibration, preferably based on a comprehensive, multi-year dataset of soil properties and gas fluxes. Ongoing field sampling and modeling refinements in collaboration with DayCent developers at NREL-Colorado State University should improve model performance.