Implementation of an automatic time‐series calibration method for the DSSAT wheat models to enhance multi‐model approaches

Abstract

Multi-modeling (MM) approaches allow increasing modeling accuracy through a combination of different modeling structures for the simulation of plant growth and yield. The Decision Support System for Agrotechnology Transfer (DSSAT) 4.7 modeling platform currently includes three different wheat (Triticum aestivum L.) models (CERES, N-Wheat, and Cropsim). However, the main obstacle for using an MM approach is the calibration procedure. Calibration is time consuming and complex, especially if the user is not familiar with all three models. It results in a subjective calibration optimum and might discriminate models if the user is less trained. To avoid these conflicts, an automated calibration program which optimizes cultivar coefficients based on the root means square error (RMSE) of time-series data was developed to ensure objective calibration results across three different wheat models and to highlight the potential of MM approaches for decision support in the future. Model calibration was performed on a 4-yr nitrogen wheat fertilizer trial (0-240 kg ha −1 N) in southwest Germany. The evaluation mean showed satisfying results for the calibration (d-index = .93) and evaluation dataset (d-index = .81). By comparing different years, the MM approach improved modeling accuracy in most cases. Especially in the drought season of 2018, the MM approach revealed higher modeling accuracy for yield (d-index = .61) in contrast to a single simulation of CERES (d-index = .34) and Cropsim (d-index = .39). This demonstrated the advantage of an MM approach as different modeling structures could compensate for errors that occur in single modeling approaches.

Comprehensive evaluation of the DSSAT‐CSM‐CERES‐Wheat for simulating winter rye against multi‐environment data in Germany

“…The DSSAT shell offers two separate tools: GENCALC (Hunt et al., 1993) and GLUE (J. W. Jones et al., 2011), both of which calculate the genotypic coefficient based on end‐season observation data from experiments provided in File‐A. To be able to also consider in‐season growth data from File‐T in calibrating CSM‐CERES for rye, we used the newly developed automated time‐series calibration tool (TSE) (Memic et al., 2021; Röll et al., 2020).…”

“…Contrary to the common practice of reassessing a limited number of site-years in crop modeling studies, there is a tendency to build new models or adapt existing ones based on a relatively small number of experiments that cover a narrow range of environmental conditions. For instance, Röll et al (2020) simulated CSM-CERES-Wheat (where CERES is Crop and Environment Resource Synthesis) for 4 years at a single location in Germany, while Anar et al (2019) adapted CSM-CERES-Sugarbeet using five site-years from two locations. Similarly, Anapalli et al (2010) developed the CSM-CROPGRO-Canola model using six experiments from two locations (six site-years), and Paff and Asseng (2019) created the CSM-Nwheat-Teff model using 10 site-years from nine locations.…”

“…For instance, Röll et al. (2020) simulated CSM‐CERES‐Wheat (where CERES is Crop and Environment Resource Synthesis) for 4 years at a single location in Germany, while Anar et al. (2019) adapted CSM‐CERES‐Sugarbeet using five site‐years from two locations.…”

“…I G U R E A 1 CSM-CERES-Wheat, where CSM is cropping system model and CERES is crop and environment resource synthesis, comparison of calibration scenarios (a-e),Röll et al's (2020) approach (f-j), and default settings (k-o) for growth stages, tiller density (tiller m −2 ), grain yield (kg ha −1 ), grain density (grain m −2 ), and 1000-kernel mass (TKM). BBCH, Biologische Bundesanstalt, Bundessortenamt and CHemische industrie; RMSE, root mean square error.F I G U R E A 2 Simulated (lines) and observed (symbols) leaf area index (LAI) response of CERES-wheat, where CERES crop and environment resource synthesis, to various nitrogen management treatments over a 3-year experiment at Hohenschulen Experimental Farm in Northern Germany (Sieling et al, 2016).…”

Yield Response of Spring Maize under Future Climate and the Effects of Adaptation Measures in Northeast China

“…The GLUE (Generalized Likelihood Uncertainty Estimation) program is used to estimate genotype-specific coefficients for the DSSAT crop models. It is a Bayesian estimation method that uses Monte Carlo sampling from prior distributions of the coefficients and a Gaussian likelihood function to determine the best coefficients based on the simulated and observed yield values [27]. The uncertainties in predictions of the impact of future climate change on the yield of spring maize in the main production areas of China were evaluated using the existing field corn genotype coefficient and soil parameter database contained within the DSSAT and field data collected from the agrometeorological stations near the study sites [28].…”