Articles
Recent developments and challenges in crop growth modelling: uncertainty analysis, global sensitivity analysis and data assimilation
Article number
1154_17
Pages
129 – 136
Language
English
Abstract
Following the method of systems theory, discussion in this paper is mainly focused on crop systems modeling and analysis.
Techniques applied to crop growth models are mostly local sensitivity analysis and parameter estimation.
Unfortunately, more powerful and already available global sensitivity analysis methods and data assimilation are scarcely used.
Thus, firstly, the state of the art of uncertainty analysis, global sensitivity analysis, and new data assimilation approaches applied to crop growth models is summarized.
Secondly, future directions to improve the overall modeling process of crop growth are addressed.
In dynamic models of crops, analyzing the uncertainties related to model parameters, model structure, and input variables is critical.
An uncertainty analysis can be carried out before a sensitivity analysis or once a model evaluation (validation) is performed.
In any case, it relies on the definition of probability distribution functions (PDFs) to input factors.
Nowadays, despite the fact that a local sensitivity analysis is generally performed on crop growth dynamic models, the modeler should use so-called global sensitivity methods such as the elementary effects method, Monte Carlo filtering, standardized regression coefficients, scatterplots, and variance-based methods.
In addition, several researchers have agreed that the parameter estimation process of crop models is still an open research problem and that new solutions are needed.
However, Bayesian parameter estimation, in particular, and Bayesian modeling in general are approaches that are still in their infancy in the crop modeling community.
Finally, there is already some evidence for the improvement of model quality that can be achieved by applying data assimilation methods such as nonlinear Kalman filters, particle filtering, and variational data assimilation.
The main conclusion of this work is that the quality of crop growth models can be raised considerably by improving either current practice of sensitivity analysis or current practice of data assimilation.
Techniques applied to crop growth models are mostly local sensitivity analysis and parameter estimation.
Unfortunately, more powerful and already available global sensitivity analysis methods and data assimilation are scarcely used.
Thus, firstly, the state of the art of uncertainty analysis, global sensitivity analysis, and new data assimilation approaches applied to crop growth models is summarized.
Secondly, future directions to improve the overall modeling process of crop growth are addressed.
In dynamic models of crops, analyzing the uncertainties related to model parameters, model structure, and input variables is critical.
An uncertainty analysis can be carried out before a sensitivity analysis or once a model evaluation (validation) is performed.
In any case, it relies on the definition of probability distribution functions (PDFs) to input factors.
Nowadays, despite the fact that a local sensitivity analysis is generally performed on crop growth dynamic models, the modeler should use so-called global sensitivity methods such as the elementary effects method, Monte Carlo filtering, standardized regression coefficients, scatterplots, and variance-based methods.
In addition, several researchers have agreed that the parameter estimation process of crop models is still an open research problem and that new solutions are needed.
However, Bayesian parameter estimation, in particular, and Bayesian modeling in general are approaches that are still in their infancy in the crop modeling community.
Finally, there is already some evidence for the improvement of model quality that can be achieved by applying data assimilation methods such as nonlinear Kalman filters, particle filtering, and variational data assimilation.
The main conclusion of this work is that the quality of crop growth models can be raised considerably by improving either current practice of sensitivity analysis or current practice of data assimilation.
Authors
I.L. López-Cruz
Keywords
dynamic model, data assimilation, uncertainty, greenhouse environment, model calibration
Online Articles (33)