AUEB Stats Seminars 10/6/2021: Monte-Carlo Statistical Methods for Parameter Estimation of the GreenLab Plant Growth Model by S. Trevezas

Monte-Carlo Statistical Methods for Parameter Estimation of the GreenLab Plant Growth Model

Samis Trevezas
Department of Mathematics, University of Athens

ΠΕΜΠΤΗ 10/6/2021, 12:30

Link is available here

ΠΕΡΙΛΗΨΗ

The last decades advanced plant growth models have been proposed in the literature. Among these models, the GreenLab model has been proved to be very generic. The problem of parameter estimation is very challenging in this type of complex models, including a large variety of plant and tree structures. In this talk, we focus on a certain class of plants with known organogenesis (structural development) and we present different approaches for making parameter estimation feasible. The resulting model can be cast into the framework of non-homogeneous hidden Markov models (or state space models). The hidden states correspond to a sequence of unknown biomasses (masses measured for living organisms) produced during successive growth cycles and the observed variables correspond to organ masses. For the parameter estimation, in the frequentist approach, we compare two stochastic variants of an appropriate Expectation-Maximization algorithm, one based on Markov Chain Monte Carlo (MCMC) and the other based on sequential importance sampling with resampling (SISR), which correspond to two different ways of approximating the E-step. Several technical issues are also discussed, including optimal resampling strategies and the development of an automated version of the MCMC-EM. A Bayesian MCMC approach is also developed for the GreenLab model and the pros and cons of these estimation methods are also discussed. The results obtained from all these different approaches are tested and compared on simulated and real data from the sugar-beet plant.