Fuzzy clustering methods reveal structurally discrete and graded alarm calls in male Red-winged Blackbirds (Agelaius phoeniceus)

Alarm calls, the vocalisations given during encounters with threats, vary tremendously in acoustic structure across species. Many alarm calls are structurally graded, which can convey information about predator type, urgency or proximity. However, graded variation is challenging to analyse with traditional categorical approaches. Here, we used fuzzy clustering to quantitatively classify and describe the alarm call repertoire of male Red-winged Blackbirds (Agelaius phoeniceus) in Michigan, U.S.A. We recorded calls from 74 males, generating a library of more than 13,000 vocalisations. We then selected 1,521 high-quality, non-overlapped calls for acoustic analysis and extracted 50 acoustic parameters per call. Using fuzzy k-means clustering, we identified the most stable number of call types, calculated membership values for primary and secondary clusters and developed thresholds that distinguish discrete, intermediate and graded call structures. Male Red-winged Blackbirds produced six alarm calls, check, cluck, zeer, seee, se-er and ti-er, which varied in the extent to which they share acoustic features. Cluck calls were discrete, whereas zeer, seee, se-er and ti-er formed a graded continuum; check calls were intermediate. These findings reveal previously unmeasured variation and unrecognised complexity in the species’ alarm signalling system and demonstrate the value of fuzzy clustering for classifying graded vocal repertoires in animals.