Project Description

Fellow: ESR2 - Alexandre Wadoux Host institution: Wageningen University Duration: 36 months on QUICS (48 months PhD) Planned start date: November 2014
Actual start date: September 2015 (Original fellow resigned)
Project title: Sampling design optimisation for uncertainty analysis of integrated water quality models (WP3, WP1)
Supervisor name: Dr Gerard Heuvelink & Dr Dick Brus PhD enrolment: Y

Objectives:

  • Develop and test statistical methods for sampling design that minimise the space-time interpolation error of model inputs.
  • Develop statistical methods for sampling design for optimisation of model outputs used for Bayesian calibration of various water quality models.
  • Investigate optimal sampling designs for minimisation of input, calibration and validation uncertainty.
  • Use multi-criteria analysis techniques to merge these into a unified optimal sampling strategy for water quality based models
Tasks and methodology: Create model that combines geostatistical modelling methods (e.g. regression kriging) with numerical optimisation (e.g. simulated annealing). Bayesian calibration is attractive for estimating uncertainties in model parameters, because it can handle a variety of models and show the full multivariate distribution of the uncertain model parameters. Create a method to allow Bayesian calibration of spatially complex dynamic models. Devise model for sample design optimisation to reduce the variance of model parameters. Apply to the above methods to catchment data from TUDOR.
Results: Rule-base and associated software that generates the optimal sampling design for minimisation of input, calibration and/or validation uncertainty of integrated water quality models (D3.1)
Dissemination: Four journal articles on sampling design optimisation, Blog and social media. Two presentation/year as MC Ambassador. Annual conference presentation.
Planned secondments: Secondment to University of Bristol, 3 months duration, to study rainfall estimation errors at different temporal/spatial scales. Secondment to TUDOR, 5 months duration, use TUDOR case study data to test different design optimisation methods.