Machine learning applied to detect of pre-emergence photospheric magnetic field patterns

   Magnetic flux generated within solar convection zone rises to the surface forming active regions (ARs) and sunspots. Early detection of ARs will be useful for gaining warning time against a variety explosive events - such as flares and coronal mass ejections - that lead to severe space weather consequences.   Evidence of any pre-emergence signatures will also shed light on subsurface processes responsible for emergence. Here we use deep convolutional neural networks (CNN) to analyse SDO/HMI line-of-sight magnetograms of pre-emerging ARs. The CNN classifies pre-emerging ARs (PEs) from a control set of non-emerging ARs (NEs) with a True Skill Statistic score  (TSS) of ~ 90%, 3 hrs prior to emergence and ~ 40%, 24 hrs  prior to emergence.  We also develop techniques to "open up" the trained CNN and highlight detected pre-emergence magnetic field patterns.