Uncertainty Analysis of Dynamic Models

You are here

Statistical Uncertainty Analysis of Dynamic Models

to be announced

Scope

Dynamic modelling plays a crucial role in life science research and a key feature of models is parameter uncertainty, arising from biological variation, or a lack of knowledge. It is generally hard to foresee how parameter uncertainty results in variation in the model predictions, especially when a model contains a lot of complex interactions. Some predictions may be sharp, others highly uncertain. Some parameter uncertainties make all predictions uncertain, whereas others may have no influence at all. The purpose of this course is to make the participants familiar with general statistical concepts describing uncertainty, and methods to compute prediction uncertainty coming from uncertain parameter values. We introduce dynamic input-state-output systems and methods to write your model in this format. Also methods are presented to obtain parameter uncertainty from input-output data in case some system states are not measurable, and in case of noise in the data. The methodology is illustrated with realistic examples.

Programme

Each day consists partly of lectures and partly of hands-on training. Exercises will be carried out in R.

  • Day 1: uncertainty: terminology and concepts, intro statistics, uncertain systems modelling, state space models, error propagation (Dr. Karel Keesman, Dr. Simon van Mourik)
  • Day 2: sampling based sensitivity analyses, random sampling and Latin hypercube sampling from probability distributions: normal, lognormal, gamma, beta and uniform (Ir. Saskia Burgers)
  • Day 3: work in teams on the first steps of a sensitivity and uncertainty analysis for your own model or a case study (Ir. Saskia Burgers, Dr Simon van Mourik)
  • Day 4: analysis of model output (based on sample), uncertainty analysis, top and bottom marginal variance, i.e. first order sensitivity index (Ir. Saskia Burgers)
  • Day 5: estimating (updating) the parameters and their uncertainty when (new) data come available, Bayesian analysis, adaptive Markov Chain Monte Carlo (MCMC), making predictions with their uncertainty from the MCMC output (Prof. Cajo ter Braak)
 
General information
 
Target Group The course is aimed at PhD candidates and other academics
Group Size Min. 15, max. 25 participants
Course duration 5 days
Language of instruction English
Frequency of recurrence Every two years
Number of credits 1.5 ECTS
Lecturers Prof. Cajo ter Braak and Ir. Saskia Burgers (Plant Research International, Wageningen UR), Dr. Simon van Mourik (Farm Technology group, Wageningen University), Dr. Karel Keesman (Biomass Refinery and Process Dynamics group, Wageningen University)
Prior knowledge Basic knowledge of mathematics and statistics
Location Wageningen University Campus

 

Fees 1

Generally, the following fees apply for this course, but note that the actual fees may be somewhat different for the next edition of this course.

  EARLY-BIRD FEE 2 REGULAR FEE 2
PE&RC / SENSE / WASS PhD candidates with an approved TSP € 310,- € 360,-
a) All other PhD candidates
b) Postdocs and staff of the above mentioned Graduate Schools
€ 660,- € 710,-
All others € 970,- € 1.020,-

1 The course fee includes a reader, coffee/tea, and lunches. It does not include accommodation
2 The Early-Bird Fee applies to anyone who REGISTERS AT LEAST 4 WEEKS PRIOR TO THE START OF THE COURSE

Note:

More information

Dr. Claudius van de Vijver (PE&RC)
Phone: +31 (0) 317 485116
Email: claudius.vandevijver@wur.nl

Dr. Lennart Suselbeek (PE&RC)
Phone: +31 (0) 317 485426
Email: lennart.suselbeek@wur.nl

Registration

This course is currently closed for registration, and will re-open once a new edition of this course is scheduled.