WORKING GROUP "Inverse modelling and optimization"


The working group consolidates knowledge of three different disciplines from three different departments: Computational Hydrosystems , Ecological Modelling and Monitoring&Exploration Technologies . The group meets regularly about every three weeks. It discusses problems in inverse modelling and optimisation common to all disciplines and examines the transfer of field-specific solution strategies to the other disciplines.
Researcher from all departments of the UFZ are invited to join the group meetings, present their specific problem and discuss solution strategies. Everybody can, of course, also contact the members directly, who can then point to the right contact person.

Sebastian Lehmann (OESA), Email

Expertise

  • stochastic derivative-free optimization (Differential Evolution DE, Simulated Annealing SA, Covariance Matrix Adaptation CMA)
  • markov chain monte carlo methods (adaptive and approximative techniques ABC)
  • metamodels (Radial Basis Functions RBF, Gaussian Processes GP)
  • error modelling (iterative and functional)
  • identifiability, sensitivity and uncertainty analysis

Current projects

Calibration and sensitivity analysis of FORMIND model


Juliane Mai (CHS), Email

Webpage

Expertise

  • stochastic derivative-free optimization (Dynamically dimensioned search DDS, Modified DDS, Simulated Annealing SA, Covariance Matrix Adaption CMA, Shuffled Complex Evolution SCE, Nelder-Mead Algorithm)
  • sensitivity analysis (Sobol index, Elementary Effects, Efficient Elementary Effects, Parameter Importance, DELSA)
  • metamodels (RS-HDMR)

Current projects

  • efficient methods for identification of non-informative parameters in complex models
  • calibration and sensitivity analysis of mesoscale hydrologic model mHM over Germany and Europe ( mHM model )
  • identification of parameter sensitivities under groundwater drought conditions


Hendrik Paasche (MET), Email

Webpage


Expertise

  • Derivative-based (deterministic) local search optimization, e.g. Newton-Raphson types, resolution analyses on data and model parameter,
  • Derivative-free (stochastic) global search optimization e.g. population or trajectory-based algorithms, ambiguity and risk appraisal
  • Multi-objective optimization strategies, e.g. based on aggregation concepts, Pareto optimality, game theory
  • Error modelling


Current projects

  • Multi-objective geophysical inversion
  • Probabilistic parameter estimation


Matthias Cuntz (CHS), Email

Webpage


Expertise

  • Identification of informative model parameters
  • Derivative- and variance-based sensitivity analysis of complex environmental models
  • High-dimensional Model Representation
  • Optimisation strategies (derivative-based, simplex algorithms, derivative-free)
  • Error structure in model calibration

Current projects

  • Efficient methods for identification of non-informative parameters in complex models
  • Using Eddy-covariance data for the adaptation and calibration of a forest gap model
  • Generalised likelihood estimation with heteroscedastic errors of a mesoscale hydrologic model
  • Multi-output sensitivity analysis of a land surface model in distinctly different river catchments


Sebastian Lehmann










 

Juliane Mai













 

Hendrik_Pasche












 

Matthias Cuntz