Search

Agricultural systems models are complex and tend to be over-parameterized with respect to observational datasets. Practical identifiability analysis based on local sensitivity analysis has proved effective in investigating identifiable parameter sets in environmental models, but has not been applied to agricultural systems models. Here, we demonstrate that identifiability analysis improves experimental design to ensure independent parameter estimation for yield and quality outputs of a complex grassland model. The Pasture Simulation model (PaSim) was used to demonstrate the effectiveness of practical identifiability analysis in designing experiments and measurement protocols within phenotyping experiments with perennial ryegrass. Virtual experiments were designed combining three factors: frequency of measurements, duration of the experiment. and location of trials. Our results demonstrate that (i) PaSim provides sufficient detail in terms of simulating biomass yield and quality of perennial ryegrass for use in breeding, (ii) typical breeding trials are insufficient to parameterize all influential parameters, (iii) the frequency of measurements is more important than the number of growing seasons to improve the identifiability of PaSim parameters, and (iv) identifiability analysis provides a sound approach for optimizing the design of multi-location trials. Practical identifiability analysis can play an important role in ensuring proper exploitation of phenotypic data and cost-effective multi-location experimental designs. Considering the growing importance of simulation models, this study supports the design of experiments and measurement protocols in the phenotyping networks that have recently been organized. [ABSTRACT FROM AUTHOR]

The formalization of novel equations explicitly modelling the impact of extreme weather events into the crop model WOFOST (EMS: existing modelling solution; MMS: modified modelling solution) is proposed as a way to reduce the uncertainty in estimations of crop yield. A sensitivity analysis (SA) was performed to assess the effect of changing parameters values on the yield simulated by the model (both EMS and MMS) for different crops (winter and durum wheat, winter barley, maize, sunflower) grown under a variety of conditions (including future climate realisations) in Europe. A two-step SA was performed using global techniques: the Morris screening method for qualitative ranking of parameters was first used, followed by the eFAST variance-based method, which attributes portions of variance in the model output to each parameter. The results showed that the parameters related to the partitioning of assimilates to storage organs (FOTB) and to the conversion efficiency of photosynthates into storage organs (CVO) generally affected considerably the simulated yield (also underlying tight correlation with this output), whereas the parameters involved with respiration rate (Q10) or specific leaf area (SLA) became influential in case of unfavourable weather conditions. Major differences between EMS and MMS (which includes a component simulating the impact of extreme weather events) emerged in extreme cases of crop failure triggered by markedly negative minimum temperatures. With few exceptions, the two SA methods revealed the same parameter ranking. We argue that the SA performed in this study can be useful in the design of crop modelling studies and in the implementation of crop yield forecasting systems in Europe. [ABSTRACT FROM AUTHOR]

Predicting regional and global carbon (C) and water dynamics on grasslands has become of major interest, as grasslands are one of the most widespread vegetation types worldwide, providing a number of ecosystem services (such as forage production and C storage). The present study is a contribution to a regional-scale analysis of the C and water cycles on managed grasslands. The mechanistic biogeochemical model PaSim (Pasture Simulation model) was evaluated at 12 grassland sites in Europe. A new parameterization was obtained on a common set of eco-physiological parameters, which represented an improvement of previous parameterization schemes (essentially obtained via calibration at specific sites). We found that C and water fluxes estimated with the parameter set are in good agreement with observations. The model with the new parameters estimated that European grassland are a sink of C with 213 g C m −2 yr −1 , which is close to the observed net ecosystem exchange (NEE) flux of the studied sites (185 g C m −2 yr −1 on average). The estimated yearly average gross primary productivity (GPP) and ecosystem respiration (RECO) for all of the study sites are 1220 and 1006 g C m −2 yr −1 , respectively, in agreement with observed average GPP (1230 g C m −2 yr −1 ) and RECO (1046 g C m −2 yr −1 ). For both variables aggregated on a weekly basis, the root mean square error (RMSE) was ∼5–16 g C week −1 across the study sites, while the goodness of fit (R 2 ) was ∼0.4–0.9. For evapotranspiration (ET), the average value of simulated ET (415 mm yr −1 ) for all sites and years is close to the average value of the observed ET (451 mm yr −1 ) by flux towers (on a weekly basis, RMSE∼2–8 mm week −1 ; R 2 = 0.3–0.9). However, further model development is needed to better represent soil water dynamics under dry conditions and soil temperature in winter. A quantification of the uncertainties introduced by spatially generalized parameter values in C and water exchange estimates is also necessary. In addition, some uncertainties in the input management data call for the need to improve the quality of the observational system. [ABSTRACT FROM AUTHOR]