Once experimental field data has been obtained, calibrating crop growth models is a labour intensive tasks requiring detailed domain knowledge. Machine learning techniques such as stochastic gradient descent in computational graphs and Bayesian optimization can potentially be time saving and might find more accurate solutions.
The purpose of this project is to investigate whether machine learning techniques can improve the accuracy of the calibration of crop growth models to experimental field data. The student will implement a computational graph of the crop growth model LINTUL3 and calibrate this model for winterwheat using experimental field data . We will compare the results to the results from conventional calibration methods , and see how we can incorporate expert knowledge in our calibration process using Bayesian optimization .
- Translate the process-based crop growth model LINTUL3 into a computational graph in Pytorch
- Optimize the model parameters using stochastic gradient descent and Bayesian optimization
- Compare the results to results from conventional calibration procedures.
- Obtaining winter wheat parameters for LINTUL from a field experiment, Wiert Wiertsema, master thesis WUR 2015
- Simulation of nitrogen limited growth of winter wheat in The Netherlands and the importance of the use of recent data sets, H. Berghuijs et al., to be published
- FTE-35306 Machine Learning
- GRS-34806 Deep Learning
Theme(s): Modelling & visualisation