Project Description

Fellow: ESR6 - Sanda Dejanic Host institution: Eawag: Swiss Federal Institute of Aquatic Science and Technology Duration: 12 months Planned start date: March 2016
Actual start date: May 2016
Project title: Assessing the influence of model structure deficiencies and input errors on the uncertainty of model predictions (WP1, WP3, W6)
Supervisor name: Dr Jörg Rieckermann & Prof. Dr. Max Maurer PhD enrolment: N

Objectives:

  • Improve numerical implementations of Bayesian parameter estimation.
  • Test the enhanced numerical approach on a comprehensive dataset of sediment loads for the Sluzew catchment in Warsaw PL.
  • Apply the new approach to another case study to demonstrate generality.
Tasks and methodology: We will efficiently sample the posterior distribution with adaptive algorithms that generate stochastic processes based on sequences of transition kernels, where each transition kernel depends on past states. This includes numerical analyses and computational experiments, based partly on existing datasets. The enhanced numerical approach will be tested on a comprehensive dataset of sediment loads for the Sluzew catchment in Warsaw PL. ESR6 will be seconded to ULaval for transfer of Bayesian parameter estimation techniques with Markov Chain Monte Carlo and efficient implementation on a computer cluster.
Results: Journal papers on: prediction uncertainty of structurally deficient water quality models, the role of input uncertainty and efficient algorithms for posterior sampling (D6.3); Report on model structure errors and input uncertainty Eindhoven case study (D1.3).
Dissemination: Journal paper, Report on model structure error, two presentations as MC Ambassador.
Planned secondments: To TUDelft, (1 month), Application to Eindhoven case study. To ULaval (2 months), on transfer of Bayesian parameter estimation techniques with Markov Chain Monte Carlo and efficient implementation on a computer cluster.