Extreme Emission Line Galaxies (EELGs)

• Extreme Emission Line Galaxies are compact, low-mass systems marked by unusually high nebular emission line equivalent widths, indicating intense, bursty star formation and low metallicity.
• They are identified across surveys using diverse methods such as broad-band excess, narrow/medium band imaging, IFU spectroscopy, and JWST observations, each tailored to capture their line-dominated spectra.
• Studies show that EELGs exhibit high ionization parameters and subsolar metallicities, linking them to interaction-driven starbursts and making them valuable analogues of reionization-era galaxies.

Extreme emission line galaxies (EELGs) are galaxies whose spectra and broad-band spectral energy distributions are dominated by nebular emission lines with unusually large equivalent widths, most commonly in $[\mathrm{O\,III}] + \mathrm{H}\beta$ and $\mathrm{H}\alpha$. Across local, intermediate-redshift, and JWST high-redshift samples, they are consistently described as compact, low-mass, intensely star-forming, low-metallicity systems with high ionization conditions and bursty star-formation histories, and they are widely used as local or intermediate-redshift analogues of galaxies in the epoch of reionization (Gupta et al., 2023, Moral-Castro et al., 2024, Withers et al., 2023, Bonatto et al., 10 Apr 2026).

1. Operational definitions and taxonomic scope

There is no universal equivalent-width threshold for EELGs in the literature. Instead, different surveys adopt thresholds matched to their bandpasses, spectral resolution, and scientific goals (Moral-Castro et al., 2024). In low- and intermediate-redshift spectroscopic work, a common operational choice is an extreme $[\mathrm{O\,III}]\lambda5007$ equivalent width. In the zCOSMOS-bright survey, EELGs were defined by $$\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) \ge 100$$ Å over $0.11 \le z \le 0.93$, a cut that automatically selected galaxies with very high Balmer-line equivalent widths and very young bursts (Amorín et al., 2014). In SDSS-based stellar-population work, the class was identified with $\mathrm{EW}(\mathrm{H}\beta) > 30$ Å and $\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) > 100$ Å (Breda et al., 2022). In MUSE HUDF spectroscopy, the primary sample required $\mathrm{EW}_0 \ge 300$ Å and the extended sample $200 \le \mathrm{EW}_0 \le 300$ Å in at least one of $[\mathrm{O\,II}]$, $\mathrm{H}\alpha$0, or $\mathrm{H}\alpha$1 (Moral-Castro et al., 2024). In J-PLUS, the local sample was defined by $\mathrm{H}\alpha$2 Å at $\mathrm{H}\alpha$3 (Lumbreras-Calle et al., 2021).

In the early-universe context, the definition is often expressed photometrically rather than by a single line. One JWST-based formulation describes EELGs as systems where nebular emission contributes $\mathrm{H}\alpha$4–$\mathrm{H}\alpha$5 of the flux in certain photometric bands, with rest-frame $\mathrm{H}\alpha$6 equivalent widths of several hundred to $\mathrm{H}\alpha$7 Å (Gupta et al., 2023). In CEERS, the photometric definition was moved to the observed frame: $\mathrm{H}\alpha$8 Å in either $\mathrm{H}\alpha$9 or $[\mathrm{O\,III}]\lambda5007$0 (Davis et al., 26 Feb 2026).

Study	Operational definition	Context
zCOSMOS (Amorín et al., 2014)	$[\mathrm{O\,III}]\lambda5007$1 Å	$[\mathrm{O\,III}]\lambda5007$2 spectroscopic sample
MUSE HUDF (Moral-Castro et al., 2024)	Primary: $[\mathrm{O\,III}]\lambda5007$3 Å; extended: $[\mathrm{O\,III}]\lambda5007$4 Å	$[\mathrm{O\,III}]\lambda5007$5, $[\mathrm{O\,III}]\lambda5007$6, or $[\mathrm{O\,III}]\lambda5007$7
J-PLUS (Lumbreras-Calle et al., 2021)	$[\mathrm{O\,III}]\lambda5007$8 Å	Local $[\mathrm{O\,III}]\lambda5007$9 photometric sample
MOSEL/JADES (Gupta et al., 2023)	Nebular emission contributes $$\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) \ge 100$$0–$$\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) \ge 100$$1 of band flux; $$\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) \ge 100$$2 EWs of several hundred to $$\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) \ge 100$$3 Å	Early-universe/JWST context
CEERS (Davis et al., 26 Feb 2026)	$$\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) \ge 100$$4 Å observed frame	$$\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) \ge 100$$5 or $$\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) \ge 100$$6, $$\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) \ge 100$$7

This diversity of criteria does not imply a fragmented class. A plausible implication is that “EELG” functions as a physically coherent label for galaxies in a short-lived, line-dominated starburst phase, while the exact boundary depends on the survey and line complex used.

2. Survey strategies and selection methodologies

EELG samples are now assembled by a heterogeneous set of methods, each with distinct strengths. Broad-band excess selections target the line-dominated filter directly. At $$\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) \ge 100$$8, one method identified EELGs through a $$\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) \ge 100$$9-band excess relative to the best-fit stellar continuum, using $0.11 \le z \le 0.93$0 mag as the primary criterion; this corresponds to $0.11 \le z \le 0.93$1 Å (Onodera et al., 2020). Follow-up MOIRCS spectroscopy showed that $0.11 \le z \le 0.93$2 targets had clear emission lines and $0.11 \le z \le 0.93$3 were the intended $0.11 \le z \le 0.93$4 emitters at $0.11 \le z \le 0.93$5, validating the broad-band excess strategy (Onodera et al., 2020).

Medium- and narrow-band selections improve spectral localization. J-PLUS identifies local EELGs through an excess in the $0.11 \le z \le 0.93$6 filter, which covers $0.11 \le z \le 0.93$7 at $0.11 \le z \le 0.93$8, and its final sample contains 466 galaxies with $0.11 \le z \le 0.93$9 equivalent widths above 300 Å; the method reaches $\mathrm{EW}(\mathrm{H}\beta) > 30$0 completeness and $\mathrm{EW}(\mathrm{H}\beta) > 30$1 purity against SDSS spectroscopy (Lumbreras-Calle et al., 2021). J-PAS extends this logic to 56 optical bands. In a fully observed $\mathrm{EW}(\mathrm{H}\beta) > 30$2 region, a photometric method combining narrow-band equivalent widths with machine learning identified 917 EELGs up to $\mathrm{EW}(\mathrm{H}\beta) > 30$3, with $\mathrm{EW}(\mathrm{H}\beta) > 30$4 purity and $\mathrm{EW}(\mathrm{H}\beta) > 30$5 completeness for $\mathrm{EW}(\mathrm{H}\beta) > 30$6 mag (Giménez-Alcázar et al., 9 Dec 2025). miniJPAS demonstrates the same principle in pilot form, using a contrast criterion corresponding to rest-frame equivalent width $\mathrm{EW}(\mathrm{H}\beta) > 30$7 Å and a fully automated multiwavelength pipeline for source characterization (Breda et al., 2024).

Integral-field spectroscopy provides purely spectroscopic selection and spatial information simultaneously. In the MUSE HUDF survey, the search proceeded directly from rest-frame equivalent widths measured with pyPlatefit, followed by object-by-object inspection and remeasurement of line fluxes and EWs; the final sample comprises 13 EELGs at $\mathrm{EW}(\mathrm{H}\beta) > 30$8 (Moral-Castro et al., 2024). This removes dependence on broad-band color preselection and yields direct kinematic and chemical diagnostics from the same data cube.

JWST medium-band photometry has generalized the idea of “spectroscopy from photometry.” In CANUCS, NIRCam medium bands were used to select 118 EELGs over $\mathrm{EW}(\mathrm{H}\beta) > 30$9, with median $\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) > 100$0 Å and median $\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) > 100$1 Å; NIRSpec spectroscopy of 15 objects confirmed the redshifts and equivalent widths derived from the medium bands (Withers et al., 2023). In CEERS, a simple photometric method identified 1165 EELGs at $\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) > 100$2, and NIRSpec spectroscopy of 34 objects validated the photometric identification of extreme emission; the medium-band F410M filter was found to be particularly efficient for isolating these systems (Davis et al., 2023).

At the largest scale, DESI has enabled a classification-driven rather than line-threshold-driven census. The k-MENDEL sample used automatic $\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) > 100$3-means classification of DESI spectra to isolate rare, high-EW spectral classes, yielding 15,014 EELGs at $\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) > 100$4 after quality cuts; the selected classes correspond almost entirely to objects with $\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) > 100$5 Å, while the largest-$\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) > 100$6 Å objects occur only in the outlier classes (Bonatto et al., 10 Apr 2026).

3. Stellar populations, chemical abundances, and ionization conditions

Across these studies, EELGs occupy the low-mass, high-sSFR, metal-poor end of the star-forming population. In the MUSE HUDF sample, stellar masses lie in the dwarf regime, with $\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) > 100$7–8.60 and $\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) > 100$8 to $\mathrm{EW}([\mathrm{O\,III}]\,\lambda5007) > 100$9, implying very high specific star-formation rates (Moral-Castro et al., 2024). The zCOSMOS sample spans $\mathrm{EW}_0 \ge 300$0 with median $\mathrm{EW}_0 \ge 300$1, and its specific star-formation rates reach up to $\mathrm{EW}_0 \ge 300$2 (Amorín et al., 2014). The DESI k-MENDEL sample broadens the demographic baseline to $\mathrm{EW}_0 \ge 300$3, $\mathrm{EW}_0 \ge 300$4, and $\mathrm{EW}_0 \ge 300$5, with EELGs systematically above the local star-forming main sequence (Bonatto et al., 10 Apr 2026). In CANUCS, the median stellar mass is $\mathrm{EW}_0 \ge 300$6, with median metallicity $\mathrm{EW}_0 \ge 300$7, median $\mathrm{EW}_0 \ge 300$8 mag, and median $\mathrm{EW}_0 \ge 300$9 (Withers et al., 2023).

Gas-phase metallicities are consistently subsolar and often extremely low. The MUSE HUDF sample has $200 \le \mathrm{EW}_0 \le 300$0, with four galaxies below $200 \le \mathrm{EW}_0 \le 300$1 (Moral-Castro et al., 2024). The zCOSMOS EELGs have median $200 \le \mathrm{EW}_0 \le 300$2, including a handful of extremely metal-deficient systems below $200 \le \mathrm{EW}_0 \le 300$3 solar (Amorín et al., 2014). In the DESI sample, direct-$200 \le \mathrm{EW}_0 \le 300$4 metallicities span $200 \le \mathrm{EW}_0 \le 300$5, with a median near 7.85, while the full strong-line scale extends to $200 \le \mathrm{EW}_0 \le 300$6 (Bonatto et al., 10 Apr 2026). Dust attenuation is typically low: the MUSE HUDF sample has $200 \le \mathrm{EW}_0 \le 300$7–0.45 mag (Moral-Castro et al., 2024), the $200 \le \mathrm{EW}_0 \le 300$8 broad-band-selected sample has $200 \le \mathrm{EW}_0 \le 300$9 mag (Onodera et al., 2020), and the J-PLUS local sample has $[\mathrm{O\,II}]$0 (Lumbreras-Calle et al., 2021).

A defining feature of EELGs is their high-ionization nebular spectrum. In the $[\mathrm{O\,II}]$1 sample, $[\mathrm{O\,II}]$2 values are typically $[\mathrm{O\,II}]$3–3, with several sources at $[\mathrm{O\,II}]$4 and lower limits as high as $[\mathrm{O\,II}]$5; the authors infer $[\mathrm{O\,II}]$6–8.5 and ionization parameters higher by $[\mathrm{O\,II}]$7 dex than local star-forming galaxies (Onodera et al., 2020). The ionization parameter is conventionally written as

$[\mathrm{O\,II}]$8

and interaction-driven starbursts can plausibly raise $[\mathrm{O\,II}]$9 while compressing the gas, thereby increasing $\mathrm{H}\alpha$00 and line equivalent widths (Gupta et al., 2023).

At the most extreme end, EELGs exhibit a very-high-ionization component that exceeds the reach of standard three-zone H II region models. Detailed UV and optical spectroscopy of two nearby archetypes shows strong nebular He II, C IV, $\mathrm{H}\alpha$01, and $\mathrm{H}\alpha$02, motivating a four-zone ionization model with an added He$\mathrm{H}\alpha$03 zone above 54.4 eV (Berg et al., 2021). In that framework, traditional three-zone estimates can under-estimate the average $\mathrm{H}\alpha$04 by up to 0.5 dex, while the total nebular abundances remain nearly unchanged (Berg et al., 2021). The same study identifies a model-independent abundance dichotomy in which $\mathrm{H}\alpha$05 abundances are consistent but N/H, C/H, and Fe/H are relatively deficient, implying $\mathrm{H}\alpha$06 enhancement by $\mathrm{H}\alpha$07 times (Berg et al., 2021). It also concludes that current photoionization models still cannot reproduce the observed very-high-ionization lines, leaving a high-energy ionizing photon production problem unresolved (Berg et al., 2021).

4. Structure, kinematics, and triggering mechanisms

EELGs are usually compact, but their morphology is not uniform. In zCOSMOS, the median half-light radius is $\mathrm{H}\alpha$08 kpc, and $\mathrm{H}\alpha$09 of the galaxies show non-axisymmetric morphologies, including clumpy and tadpole systems; $\mathrm{H}\alpha$10 show additional low-surface-brightness features that strongly suggest recent or ongoing interactions (Amorín et al., 2014). The local J-PLUS sample similarly finds that most objects have compact morphologies, while $\mathrm{H}\alpha$11 are more extended dwarfs with clumpy or disturbed structure, underscoring that “extreme [O III]” denotes a physical state rather than a single morphological class (Lumbreras-Calle et al., 2021).

Integral-field observations show that compactness does not imply kinematic simplicity. In the MUSE HUDF EELGs, four of the seven primary-sample galaxies and five of the six extended-sample galaxies are spatially resolved; three of the resolved primary EELGs show a clear rotating-disk pattern, whereas none of the resolved extended-sample objects do (Moral-Castro et al., 2024). The same study cautions that mergers can mimic rotation at MUSE resolution, so higher-resolution IFU data are required to separate rotating disks, unresolved mergers, and outflows (Moral-Castro et al., 2024).

Environmental work has increasingly linked EELGs to interactions. The MOSEL survey used JADES NIRCam imaging to examine spectroscopically confirmed strong $\mathrm{H}\alpha$12 emitters at $\mathrm{H}\alpha$13. Compared with control galaxies, the EELGs show a median brightest-companion mass ratio $\mathrm{H}\alpha$14 and total companion mass ratio $\mathrm{H}\alpha$15, versus $\mathrm{H}\alpha$16 and $\mathrm{H}\alpha$17 in the full control sample, with KS-test probabilities $\mathrm{H}\alpha$18 and $\mathrm{H}\alpha$19 (Gupta et al., 2023). Even after matching in both stellar mass and specific SFR, EELGs retain $\mathrm{H}\alpha$20–4 times higher brightest-companion mass ratios and $\mathrm{H}\alpha$21 times higher total companion mass ratios than the matched controls (Gupta et al., 2023). TNG100 tests show that with the same projected-separation and velocity cuts, $\mathrm{H}\alpha$22 of galaxies have at least one companion sharing the same descendant by $\mathrm{H}\alpha$23, supporting the interpretation that these close massive companions represent real mergers or strong interactions rather than chance projections (Gupta et al., 2023). A plausible implication is that interaction-driven gas cooling and inflow are major triggers of the EELG phase at $\mathrm{H}\alpha$24.

The role of active galactic nuclei is now better constrained in the JWST era. A spectroscopic study of CEERS EELGs with observed-frame EW $\mathrm{H}\alpha$25 Å finds that $\mathrm{H}\alpha$26 of photometrically selected EELGs have broad Balmer lines, increasing to $\mathrm{H}\alpha$27 in deep spectroscopy, so AGN are not negligible within the class (Davis et al., 26 Feb 2026). However, many AGN selected photometrically as EELGs have incorrectly high photometric equivalent widths, and in sources that remain true high-EW EELGs spectroscopically, the narrow $\mathrm{H}\alpha$28 component dominates over the broad component; the same study concludes that CEERS EELGs are powered primarily by star formation rather than AGN (Davis et al., 26 Feb 2026). It also finds that $\mathrm{H}\alpha$29 morphology becomes more compact at higher equivalent width, consistent with increasingly centralized ionizing sources in the most extreme systems (Davis et al., 26 Feb 2026).

5. High-redshift analogues, reionization, and contamination of galaxy selection

One of the principal scientific uses of EELGs is as analogues of the galaxies that dominated the early Universe. In JWST observations at $\mathrm{H}\alpha$30, $\mathrm{H}\alpha$31 of galaxies have rest-frame $\mathrm{H}\alpha$32 equivalent widths above 800 Å, almost three times the equivalent width of a typical star-forming galaxy at $\mathrm{H}\alpha$33 (Gupta et al., 2023). The HUDF EELGs occupy the same loci as $\mathrm{H}\alpha$34–8 galaxies in the mass–metallicity relation, lying well below the local $\mathrm{H}\alpha$35 relation and close to the $\mathrm{H}\alpha$36 MOSDEF and JWST high-redshift sequences (Moral-Castro et al., 2024). The DESI k-MENDEL sample similarly follows a shallower mass–metallicity relation offset by $\mathrm{H}\alpha$37–0.5 dex from local relations and closely resembling young galaxies observed with JWST at $\mathrm{H}\alpha$38–10 (Bonatto et al., 10 Apr 2026). Local and low-redshift photometric samples make the same point from the opposite direction: J-PLUS and J-PAS explicitly frame their EELGs as nearby analogues of reionization-era galaxies, and in J-PAS most sources exceed the canonical ionizing-efficiency threshold often associated with sustaining reionization (Lumbreras-Calle et al., 2021, Giménez-Alcázar et al., 9 Dec 2025).

Ionizing-photon production efficiency is one of the clearest bridges between low-redshift EELGs and high-redshift galaxies. In the $\mathrm{H}\alpha$39 broad-band-selected sample, $\mathrm{H}\alpha$40 is typically $\mathrm{H}\alpha$41–25.9, with a median around 25.5 and a clear positive correlation with $\mathrm{H}\alpha$42; galaxies with $\mathrm{H}\alpha$43 Å typically have $\mathrm{H}\alpha$44 (Onodera et al., 2020). J-PAS finds the same qualitative behavior, with

$\mathrm{H}\alpha$45

and a significant fraction of its EELGs above the often-quoted $\mathrm{H}\alpha$46 threshold (Giménez-Alcázar et al., 9 Dec 2025). This suggests that the EELG phase is closely tied to high ionizing efficiency.

The same properties that make EELGs valuable analogues also make them problematic contaminants in photometric searches for the highest-redshift galaxies. CLASH showed that EELGs with rest-frame $\mathrm{H}\alpha$47 equivalent widths $\mathrm{H}\alpha$48 Å, and in some cases $\mathrm{H}\alpha$49–3700 Å, can mimic dropout colors when broad-band coverage is incomplete or signal-to-noise is limited (Huang et al., 2014). That study identified 52 candidates in cluster-lensed HST fields and concluded that the fraction of EELGs in future high-redshift galaxy selections cannot be neglected (Huang et al., 2014). JWST has sharpened the issue rather than removed it: the CEERS census reports examples of EELGs that could be incorrectly classified at ultra-high redshift, $\mathrm{H}\alpha$50, owing to extreme $\mathrm{H}\alpha$51 emission blended across the reddest filters (Davis et al., 2023). Medium-band imaging mitigates the problem. In CEERS, F410M is especially effective because it captures the line complex while bracketing filters better constrain the continuum (Davis et al., 2023); in CANUCS, medium-band color selection is explicitly advantageous because it selects by equivalent width rather than requiring strong continuum emission, and therefore recovers faint-continuum or red-continuum EELGs that broad-band selections can misclassify (Withers et al., 2023).

6. Modeling challenges, systematics, and open problems

The strongest methodological lesson from EELG research is that nebular emission cannot be treated as a perturbation. Self-consistent spectral modeling of 414 SDSS EELGs shows that stellar mass and mean age estimates from STARLIGHT are systematically biased toward higher values, and that adequate recovery of EELG stellar properties is only possible when nebular continuum emission is included (Breda et al., 2022). The same paper finds that the discrepancies between stellar-only and stellar-plus-nebular synthesis correlate with specific SFR and with the summed flux of the strongest emission lines, implying that the systematic error is itself a function of star-formation intensity (Breda et al., 2022). This point also appears in broad-band SED work: the zCOSMOS analysis found that ignoring nebular line contamination in photometry would overestimate $\mathrm{H}\alpha$52 by a median of $\mathrm{H}\alpha$53 dex and by factors of 3–5 in the most extreme cases (Amorín et al., 2014).

Selection systematics remain substantial. In the MOSEL/JADES environment study, companion identification depends on photometric redshifts with $\mathrm{H}\alpha$54 and $\mathrm{H}\alpha$55 catastrophic outliers, so the fiducial $\mathrm{H}\alpha$56 cut is necessarily broad (Gupta et al., 2023). In the MUSE HUDF sample, continuum estimation is intrinsically difficult because EELGs have very faint continua; varying the continuum window changed at least one object from above-threshold to borderline status, showing how sensitive equivalent widths are to small continuum errors (Moral-Castro et al., 2024). In J-PAS, the quoted $\mathrm{H}\alpha$57 values depend on the assumed attenuation law and on the choice $\mathrm{H}\alpha$58; the authors note that an SMC-like law would systematically increase $\mathrm{H}\alpha$59 by up to $\mathrm{H}\alpha$60 dex, and that the reported values are lower limits if ionizing photons escape (Giménez-Alcázar et al., 9 Dec 2025). The CEERS AGN study adds a further caveat: broad-line AGN can acquire spuriously large photometric equivalent widths when continuum slopes are misestimated, so photometric “ultra-EELG” samples require spectroscopic vetting (Davis et al., 26 Feb 2026).

Beyond measurement systematics, there are unresolved physical problems. In the DESI sample, the mass–metallicity relation shows large intrinsic scatter and the scatter is not reduced by projection along the fundamental metallicity relation, indicating strong departures from simple equilibrium “bathtub” models and suggesting stochastic metal-poor inflows plus strong feedback (Bonatto et al., 10 Apr 2026). In the nearby two-object study, even after invoking $\mathrm{H}\alpha$61 enhancement and very high $\mathrm{H}\alpha$62, photoionization models still fail to reproduce the observed very-high-ionization lines, leaving the high-energy ionizing photon production problem open (Berg et al., 2021). This suggests that the hardest ionizing radiation in EELGs may require ingredients beyond conventional stellar population models, or at least beyond their current calibration in the low-metallicity, high-sSFR regime.

Taken together, these results define EELGs not as a narrowly bounded phenomenological class but as a recurrent, non-equilibrium phase in galaxy evolution. The phase is marked by very large nebular equivalent widths, compact and often centrally concentrated star formation, low metallicity, hard radiation fields, and in at least some samples a strong association with interactions or rapid gas accretion. It is now observed from the nearby Universe to $\mathrm{H}\alpha$63, and the convergence of DESI, J-PAS, MUSE, and JWST results suggests that EELGs are simultaneously a local laboratory, a high-redshift selection challenge, and a key empirical route to the physics of the galaxies that dominated the first billion years of cosmic history.