Skip to content

Data scienceFAU Erlangen-Nürnberg · NASA Fermi observations

Finding hidden structure in spatial data

Built a Python pipeline using clustering and model comparison to investigate whether neighbouring catalogue detections belong to one extended object.

Result

Eight regions shortlisted from 48 candidate groups.

The challenge

One extended astronomical object can appear as several catalogue detections. Nearby detections can also be unrelated. The task was to distinguish plausible extended sources from chance groupings and effects of the background model.

My contribution

  • Built a Python pipeline combining DBSCAN clustering, statistical model comparison and diagnostic visualisations.
  • Compared the existing catalogue models with single extended-object alternatives.
  • Tested the results against alternative background models and independent catalogues.

Results and validation

Grouped 124 catalogue entries into 48 candidate clusters. Eight passed the statistical and quality checks. Three were new extended-source candidates. The catalogue analysis was published in Astronomy & Astrophysics (2026).

Technical details

Applied to NASA Fermi observations. Five selected regions overlap known extended objects. The other three require follow-up, and the inferred shape of one changes with the background model.