Uncertainty quantification of EFIT reconstructions using the EFIT-AI database

The EFIT-AI database is a collection of tokamak discharges with multiple equilibrium reconstructions. This database is being developed for the EFIT-AI project because high quality data and open access to it are increasingly becoming a requirement for advancing tokamak science with Data Science methods. Currently, the database features the 2019 DIII-D campaign (approximately 2500 discharges) with 4 different EFIT reconstructions: 1. Magnetics-only EFIT, 2. Magnetics+MSE EFIT, 3. OMFIT-automated kinetic EFITs, and 4. CAKE-automated kinetic EFITs. Here we examine the sensitivity of the solution vector(s) from various EFIT reconstructions to the measurements, which enter EFIT as constraints. These constraints are used in the least-squares square minimization of EFIT to find the optimal fitting coefficients for the pressure and $F=RB_{\phi}$ profiles. We use correlation matrices and singular-value decomposition (SVD) to carry out the uncertainty quantification. Preliminary results show strong correlations between the magnetic measurements. SVD splits the time-dependent measurements into separate spatial and temporal parts, with the singular values indicating the significance of each eigenmode. SVD can be used to filter out the eigenmodes with low singular values.