Abstract: Simulation models are frequently applied to determine the production potential of forage grasses under various scenarios, including climate change. Thorough calibrations and evaluations of forage grass models can help improve their applicability. This study evaluated the ability of the Light Interception and Utilization Simulator-GRAss (LINGRA) model to predict biomass yield of timothy (Phleum pratense L. cv. Grindstad) in the Nordic countries. Variety trial data for the first and second year after establishment were obtained for seven locations: Jokioinen, Finland (60 degrees 48 ‘ N; 23 degrees 29 ‘ E), Maaninka, Finland (63 degrees 09 ‘ N; 27 degrees 18 ‘ E), Korpa, Iceland (64 degrees 09 ‘ N; 21 degrees 45 ‘ W), Srheim, Norway (58 degrees 41 ‘ N; 5 degrees 39 ‘ E), Lillerud, Sweden (59 degrees 24’ N; 13 degrees 16 ‘ E), Ostersund, Sweden (63 degrees 15 ‘ N; 14 degrees 34 ‘ E) and Ulna Sweden (63 degrees 49 ‘ N; 20 degrees 13 ‘ E) from 1992 to 2012. Two calibrations of the LINGRA model were carried out using Bayesian techniques. In the first of these (SRrheim calibration), data on biomass yield and underlying variables obtained from independent field trials at Srheim were used. In the second (Nordic calibration), biomass data from the other locations were used as well. The model was validated against the remaining set of biomass yields from all locations not included in the Nordic calibration. The observed total seasonal yield the first and second year after establishment was 913 and 991 g DM m(-2) respectively on average across the locations. The corresponding average simulated yield after the Srheim calibration was 1044 (root mean square error (RMSE) 258) and 1112 g DM m(-2) (RMSE 312), respectively. After the Nordic calibration, the simulated average total seasonal yield was 863 (RMSE 242) the first year and 927 g DM m(-2) (RMSE 271) the second year after establishment. The differences between the observed and simulated first cut yield followed the same patterns, whereas the prediction accuracy for second cut yield did not differ substantially between the calibration approaches.Using the parameter set from the Nordic region decreased the model predictability at Srheim compared with only using model parameters derived from this location. These results show that using biomass data from several locations, instead of only one specific location, in the calibration of the LINGRA model improved the overall prediction accuracy of first cut dry matter yield and total seasonal dry matter yield across an environmentally heterogeneous region. To further analyse the usefulness of including multi-site data in forage grass model calibrations, other forage grass models could be evaluated against the same dataset.

Abstract: Process-based crop simulation models are increasingly used in regional climate change impact studies, but little is known about the implications of different calibration strategies on simulated yields. This study aims to assess the importance of region-specific calibration of five important field crops (winter wheat, winter barley, potato, sugar beet and maize) across 25 member countries of the European Union (EU25). We examine three calibration strategies and their implications on spatial and temporal yield variability in response to climate change: (i) calculation of phenology parameters only, (ii) consideration of both phenology calibration and a yield correction factor and (iii) calibration of phenology and selected growth processes. The analysis is conducted for 533 climate zones, considering 24 years of observed yield data (1983-2006). The best performing strategy is used to estimate the impacts of climate change, increasing CO2 concentration and technology development on yields for the five crops across EU25, using seven climate change scenarios for the period 2041-2064. Simulations and calibrations are performed with the crop model LINTUL2 combined with a calibration routine implemented in the modelling interface LINTUL-FAST. The results show that yield simulations improve if growth parameters are considered in the calibration for individual regions (strategy 3); e.g. RMSE values for simulated winter wheat yield are 2.36, 1.10 and 0.70 Mg ha(-1) for calibration strategies 1, 2 and 3, respectively. The calibration strategy did not only affect the model simulations under reference climate but also the extent of the simulated climate change impacts. Applying the calibrated model for impact assessment revealed that climatic change alone will reduce crop yields. Consideration of the effects of increasing CO2 concentration and technology development resulted in yield increases for all crops except maize (i.e. the negative effects of climate change were outbalanced by the positive effects of CO2 and technology change), with considerable differences between scenarios and regions. Our simulations also suggest some increase in yield variability due to climate change which, however, is less pronounced than the differences among scenarios which are particularly large when the effects of CO2 concentration and technology development are considered. Our results stress the need for region-specific calibration of crop models used for Europe-wide assessments. Limitations of the considered strategies are discussed. We recommend that future work should focus on obtaining more comprehensive, high quality data with a finer resolution allowing application of improved strategies for model calibration that better account for spatial differences and changes over time in the growth and development parameters used in crop models. (c) 2012 Elsevier B.V. All rights reserved.

Abstract: Crop models, used to make projections of climate change impacts, differ greatly in structural detail. Complexity of model structure has generic effects on uncertainty and error propagation in climate change impact assessments. We applied Bayesian calibration to three distinctly different empirical and mechanistic wheat models to assess how differences in the extent of process understanding in models affects uncertainties in projected impact. Predictive power of the models was tested via both accuracy (bias) and precision (or tightness of grouping) of yield projections for extrapolated weather conditions. Yields predicted by the mechanistic model were generally more accurate than the empirical models for extrapolated conditions. This trend does not hold for all extrapolations; mechanistic and empirical models responded differently due to their sensitivities to distinct weather features. However, higher accuracy comes at the cost of precision of the mechanistic model to embrace all observations within given boundaries. The approaches showed complementarity in sensitivity to weather variables and in accuracy for different extrapolation domains. Their differences in model precision and accuracy make them suitable for generic model ensembles for near-term agricultural impact assessments of climate change.