1. R&S= Patrick Raanes & Andreas Stordal (NORCE, Norway) – Ensemble methods, Theory and Phyton exercisesA broad introduction to the EnKF (Bayesian perspective), with Python code and exercises in Jupyter notebooks. We’ll cover:
   • DA&EnKF
   • Bayesian Inference
   • Univariate Kalman filtering
   • Multivariate Kalman
   • Time series analysis
   • Dynamical Systems, chaos, Lorenz
   • Ensemble representation
   • Writing your own EnKF

Time permitting, we will also do a history matching tutorial (https://github.com/patnr/HistoryMatching )

The students should have a computer and a Google account (to run the notebooks in the cloud). If they want, they can also work locally, following the installation instructions here: https://github.com/nansencenter/DA-tutorials

2. Femke Vossepoel (TU Delft) – Particle filters
3. Arnold Heemink (TU Delft) – Variational methods
   • Introduction and formulation of a data assimilation problem as a nonlinear optimization problem.
   • The simple scalar case: Estimating parameters using different optimization algorithms
   • The large scale case: Estimating the initial condition of a linear numerical model using the Lagrangian approach
   • Calculus of variations and the adjoint formalism
   • Variational data assimilation
   • Revisit of the large scale case: Estimating the initial condition of a linear numerical model using variational data assimilation
   • The continuous case: Estimating parameters in a partial differential equation using variational data assimilation
   • The variational data assimilation algorithm for large scale non-linear numerical models
   • Special topics: Regularization, Automatic Differentiation, comparison with Deep Neural Networks, Model Order Reduction to obtain adjoint-free implementations, and Check pointing
4. A&M = Anca Hanea and Mark Burgman (CEBRA, UniMelb, Australia) – Risk modelling, uncertainty quantification & structured expert judgment
   • rational decision making and how does probabilistic modelling fit in
   • why and where we need experts
   • experts’ biases when expert judgement (EJ) is needed
   • EJ for models, EJ for parameters
   • workshop – 4 groups —  build a BN from a narrative problem
   • consensus (qualitative) modelling
   • IDEA for parameters; calibration and target questions
   • Workshop (same groups) second round IDEA
   • scores in the classical model (CM) for EJ
   • the consensus BN quantified, EJ scores for performance
5. Tijana Janic : In summer school I would like to focus on use of physical constraints in EnKF. Hybrid EnKF as in this paper and several others following it: https://journals.ametsoc.org/view/journals/mwre/142/2/mwr-d-13-00056.1.xml?tab_body=abstract-display