Per-spaxel fitting leaves the spatial information in LVM data unused, and binning destroys it. I present progress on a framework that forward-models many pointings jointly, enabling weak-line science while also improving strong-line and kinematic measurements.
We built and demonstrated a new class of models for IFU analysis that use accelerated Gaussian processes to incorporate the spatial information. We applied the technique to Local Volume Mapper data of the Rosette nebula, and presented maps of emission lines well below the (per-spaxel) noise floor.
LVM is mapping the ionised gas of the Milky Way in unprecedented detail. However, the recombination lines that would settle the abundance discrepancy are faint, and there are 55 million spectra. I present a model that fits hundreds of thousands of them at once.
Composable latent-field models for IFU spectroscopy.
The exoALMA survey presented the unprecedented opportunity to map gas motions induced in the natal discs of newborn planets. However, such measurements are biased by an instrumental effect called beam smearing, leading to the underestimation of velocity perturbations and planet masses. We built an applied a probabilistic tool to correct for this.