Light bridges are elongated and bright structures protruding into the umbra of sunspots. The presence of light bridges has a significant role in the evolution of sunspots and the heating of their overlying atmosphere. Therefore, investigating these structures is crucial to understanding fundamental aspects of sunspots.

By applying a novel code based on deep-learning algorithms called SICON to spectropolarimetric observations acquired with the Hinode satellite, we computed atmospheric parameters that allowed us to infer the variation of the physical properties of light bridges on a geometric height scale. We have also used the SIR inversion code, commonly used in solar physics, to cross-check our results, particularly in the inclination of the magnetic field vector. Despite being barely used in previous studies, the analysis of the stratification in terms of the geometric height offers a more realistic physical scenario due to the corrugation of the solar atmosphere.

We found that each light bridge shows a different physical scenario. The light bridge classified as filamentary shows properties possibly related to an enhanced chromospheric activity above it, which agrees with other investigations. On the other hand, regions with a granular morphology reveal more abrupt stratifications, having characteristics compatible with the injection of hot plasma through convective cells located in favoured positions with weaker magnetic fields. Our findings are usually incompatible with the presence of a magnetic canopy that wraps light bridges entirely, as proposed by other previous studies. The magnetic canopy might exist only at specific locations, which could align with the detection of structures similar to light bridges in the chromosphere and transition region by other authors. Thus, we emphasize the need for further investigations on light bridges and their magnetic canopy to fully understand these complex structures.