Efficient crop model parameter estimation and site characterization using large breeding trial data sets

Lamsal, Abhishes; Welch, Stephen M.; Jones, James W.; Boote, Kenneth J.; Asebedo, Antonio R.; Crain, Jared; Wang, Xu; Boyer, W. P.; Giri, Anju; Frink, Elizabeth; Xu, Xuan; Gundy, Garrison James; Ou, Jiangtao; Arachchige, Pabodha Galgamuwe

doi:10.1016/j.agsy.2017.07.016

Cited by 18 publications

(14 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Buddhaboon et al [20] used GENCALC and GLUE to estimate GSPs of deep water rice using CERES-Rice model. Most recently Lamsal et al [21] used the independent component analysis (ICA) and separate factor approaches to estimate soybean GSPs from large breeding trial datasets in the USA.…”

Section: 0 Introductionmentioning

confidence: 99%

Options for calibrating CERES-maize genotype specific parameters under data-scarce environments

et al. 2019

View full text Add to dashboard Cite

Most crop simulation models require the use of Genotype Specific Parameters (GSPs) which provide the Genotype component of G×E×M interactions. Estimation of GSPs is the most difficult aspect of most modelling exercises because it requires expensive and time-consuming field experiments. GSPs could also be estimated using multi-year and multi locational data from breeder evaluation experiments. This research was set up with the following objectives: i) to determine GSPs of 10 newly released maize varieties for the Nigerian Savannas using data from both calibration experiments and by using existing data from breeder varietal evaluation trials; ii) to compare the accuracy of the GSPs generated using experimental and breeder data; and iii) to evaluate CERES-Maize model to simulate grain and tissue nitrogen contents. For experimental evaluation, 8 different experiments were conducted during the rainy and dry seasons of 2016 across the Nigerian Savanna. Breeder evaluation data were also collected for 2 years and 7 locations. The calibrated GSPs were evaluated using data from a 4-year experiment conducted under varying nitrogen rates (0, 60 and 120kg N ha -1 ). For the model calibration using experimental data, calculated model efficiency (EF) values ranged between 0.88–0.94 and coefficient of determination (d-index) between 0.93–0.98. Calibration of time-series data produced nRMSE below 7% while all prediction deviations were below 10% of the mean. For breeder experiments, EF (0.58–0.88) and d-index (0.56–0.86) ranges were lower. Prediction deviations were below 17% of the means for all measured variables. Model evaluation using both experimental and breeder trials resulted in good agreement (low RMSE, high EF and d-index values) between observed and simulated grain yields, and tissue and grain nitrogen contents. It is concluded that higher calibration accuracy of CERES-Maize model is achieved from detailed experiments. If unavailable, data from breeder experimental trials collected from many locations and planting dates can be used with lower but acceptable accuracy.

show abstract

Section: 0 Introductionmentioning

confidence: 99%

Options for calibrating CERES-maize genotype specific parameters under data-scarce environments

et al. 2019

View full text Add to dashboard Cite

show abstract

“…The high model complexity adds an additional challenge when parameterizing them for specific genotypes. One example of alternative approach was recently presented by Lamsal et al (2017), who used image analysis from historical field trials for a simultaneous estimation of crop growth model parameters and environmental quality in a soybean breeding programme. The modelling shown by Lamsal et al (2017) is a good example on how the simultaneous modelling of phenotypes, their proxies obtained through image analysis, crop dynamics and genetics can be used as a valuable tool to predict genotype adaptation and assist breeding programmes.…”

Section: Combining Crop Growth Models and Qtl Or Genomic Prediction Mmentioning

confidence: 99%

Modelling of genotype by environment interaction and prediction of complex traits across multiple environments as a synthesis of crop growth modelling, genetics and statistics

Bustos-Korts

View full text Add to dashboard Cite

Bustos -Korts, D. (2017). Modelling of Genotype by Environment Interaction and Prediction of Complex Traits across Multiple Environments as a Synthesis of Crop Growth Modelling, Genetics and Statistics. PhD thesis, Wageningen University, the Netherlands.The main objective of plant breeders is to create and identify genotypes that are well-adapted to the target population of environments (TPE). The TPE corresponds to the future growing conditions in which the varieties produced by a breeding program will be grown. All possible genotypes that could be considered as selection candidates for a specific TPE are said to belong to the target population of genotypes, TPG.Genotypes commonly show different sensitivities to environmental gradients and then genotype by environment interaction (GxE) is observed. GxE can lead to changes in genotypic ranking, complicating the breeding process. The main aim of this thesis was to investigate statistical models and the combination of statistical and crop growth models to improve phenotype prediction across multiple environments. One aspect that determines the quality of phenotype prediction is the set of genotypes used to train the prediction model, especially when the TPG is structured. We proposed a method that uniformly covers the genetic space of the TPG, leading to a larger prediction accuracy than random sampling. We produced positive results for wheat, maize and rice. A second aspect that influences the accuracy of phenotype predictions is the choice of environments used to train the prediction model, which should capture the heterogeneity in the TPE. When accounting for heterogeneity in environmental quality, it is important to distinguish between repeatable and well predictable elements in the environmental conditions from those that are badly predictable. We proposed statistical methods based on the AMMI model and on mixed models to identify groups of environments that show repeatable GxE, illustrating our ideas with multienvironment wheat data in North-Western Europe. The importance of training set construction strategies and multi-environment genomic prediction models was also demonstrated for barley data. If breeders are interested in identifying the genetic basis of the target traits, it is advantageous to have a higher SNP density. In this thesis, we used exome sequence data of the EU-Whealbi-barley germplasm, which corresponds to a unique set of genotypes with a diverse origin, growth habit and breeding history.vi For this diverse data, we assessed the effects of QTLs and haplotypes across multiple environments for awn length, grain weight, heading date and plant height. Our results show that the EU-Whealbi-barley collection possesses a large diversity of promising alleles regulating the four traits we analysed. The last major topic addressed in this thesis is the use of a combination of statistical-genetic models and crop growth models (APSIM) as a strategy to assess the traits and phenotyping schemes to improve the prediction accuracy of a target trait like yield...

show abstract

“…Fitting complex biological models to data remains a challenging problem, and a number of optimization strategies have been employed to obtain parameter estimates for crop growth models or their phenology models (Archontoulis et al, 2014;He et al, 2010;Lamsal et al, 2017;Messina et al, 2018;Sexton et al, 2016;Wallach et al, 2001). The approach used here assumes that there may be more than one equivalently good solution given the data (sometimes referred to as equifinality) and employs a highly-parallelized, multi-modal optimization strategy (Wong, 2015).…”

Section: ) Training Component Models: Fitting Knowledge-based Phenolmentioning

confidence: 99%

“…Like many biological models, these models are overparameterized, and they are unidentifiable given only the observations of calendar dates to developmental stages that are typically measured in applied field settings. This problem of fitting unidentifiable models with non-convex cost landscapes and limited data is not unique to the life sciences; while strategies exist to reparameterize the complex model to assist model fitting, crop growth modelers have generally employed various optimization approaches to regularize and fit the complex crop growth model (Archontoulis et al, 2014;He et al, 2010;Lamsal et al, 2017;Messina et al, 2018;Sexton et al, 2016;Wallach et al, 2001). The best optimization strategy will vary with the size and composition of the training dataset, sensitivity of the output to the selected set of input parameters, choice of priors and regularization schemes, and model runtime.…”

Section: Introductionmentioning

confidence: 99%

Intercontinental prediction of soybean phenology via hybrid ensemble of knowledge-based and data-driven models

McCormick

Truong

Rotundo

et al. 2020

Preprint

View full text Add to dashboard Cite

The timing of crop development has significant impacts on management decisions and subsequent yield formation. A large intercontinental dataset recording the timing of soybean developmental stages was used to establish ensembling approaches that leverage both discrete-time dynamical system models of soybean phenology and data-driven, machine-learned models to achieve accurate and interpretable predictions. We demonstrate that the knowledge-based, dynamical models can improve machine learning by generating expert-engineered features. Combining the predictions of the diverse component models via super learning resulted in a mean absolute error of 4.12 and 4.55 days to flowering (R1) and physiological maturity (R7), providing an improvement relative to the best benchmark model error of 6.90 and 15.47 days, respectively. The hybrid intercontinental model applies to a much wider range of management and temperature conditions than previous mechanistic models, enabling improved decision support as alternative cropping systems arise, farm sizes increase, and changes in the global climate continue to accelerate.

show abstract

Efficient crop model parameter estimation and site characterization using large breeding trial data sets

Cited by 18 publications

References 46 publications

Options for calibrating CERES-maize genotype specific parameters under data-scarce environments

Options for calibrating CERES-maize genotype specific parameters under data-scarce environments

Modelling of genotype by environment interaction and prediction of complex traits across multiple environments as a synthesis of crop growth modelling, genetics and statistics

Intercontinental prediction of soybean phenology via hybrid ensemble of knowledge-based and data-driven models

Contact Info

Product

Resources

About