Solar physicists just got a much faster way to read the sun's chromosphere.
Researchers built a physics-informed neural network (PINN) that speeds up multilayer spectral inversion (MLSI), a technique for modeling two key chromospheric absorption lines: H-alpha at 6562.8 angstroms and Ca II at 8542.1 angstroms. Conventional MLSI fits these line profiles pixel by pixel with nonlinear least-squares math, which works well but is slow across large datasets. The new network instead predicts MLSI parameters directly from observed spectra, then feeds them through the same differentiable radiative-transfer model conventional MLSI uses, so the underlying physics stays intact. Trained and tested on Fast Imaging Solar Spectrograph data from the Goode Solar Telescope, covering both quiet-Sun and active-region targets, the network matched direct inversions with an average pixel-wise correlation of 0.933 across parameters, while cutting processing time from 3-5 minutes per raster down to just 5-15 seconds, a 12x to 60x speedup.
Solar chromosphere data volumes keep growing as imaging spectrographs get faster and telescopes like GST log ever larger datasets, and a method that only works on a handful of hand-picked images doesn't scale to that flood. This lets researchers run inversions across entire observing campaigns instead of cherry-picked frames, and because it still outputs the same physically interpretable parameters as conventional MLSI rather than an opaque prediction, the speedup doesn't come at the cost of trust in the results.
One caveat: the network still needs a conventional MLSI fit on a single reference image to fine-tune against, so this is a speed multiplier bolted onto the slow method, not a full replacement for it.