Motivated by recent electromagnetic observations which suggest the existence
of compact binary populations in the Galactic Cluster M22 [Nature 490, 71
(2012)] and in the Galactic center [Nature 556, 70 (2018)], and considering
that eccentricity provides one of the cleanest signatures to identify these
compact binary populations, in this article we study the importance of
including higher-order waveform multipoles to enable gravitational wave
observations of eccentric binary black hole mergers. Using a catalog of
Einstein Toolkit numerical relativity simulations that describe eccentric,
non-spinning black holes mergers with mass-ratios $1\leq q \leq 10$, and
eccentricities $e_0\lesssim0.2$ ten cycles before merger, we determine the
mass-ratio, eccentricity and binary inclination angle combinations that
maximize the contribution of the higher-order waveform multipoles $(\ell, \,
|m|)= \{(2,\,2),\, (2,\,1),\, (3,\,3),\, (3,\,2), \, (3,\,1),\,
(4,\,4),\,(4,\,3),$ $(4,\,2),\,(4,\,1)\}$ for gravitational wave detection. We
then explore the implications of these results in the context of stellar mass
black holes that are detectable by LIGO detectors at design sensitivity, and
show that compared to models that only include the $(\ell, \, |m|)=(2,\,2)$
mode, the inclusion of higher-order waveform multipoles can increase the
signal-to-noise ratio of eccentric binary black hole mergers by up to
$\sim45\%$ for mass-ratio binaries $q\leq10$. Furthermore, building upon our
pioneering deep learning work [George & Huerta Phys. Rev. D 97, 044039 (2018);
George & Huerta Physics Letters B 778, 64 (2018)], we show for the first time
that machine learning can accurately reconstruct higher-order waveform
multipole signals from eccentric binary black hole mergers embedded in real
LIGO data.