CLASH: EXTREME EMISSION-LINE GALAXIES AND THEIR IMPLICATION ON SELECTION OF HIGH-REDSHIFT GALAXIES

The CLASH data were obtained with three HST cameras: Advanced Camera for Surveys (ACS)/WFC, WFC3/IR, and WFC3/UVIS. All 25 clusters have now been observed. The images and catalogs are processed with APLUS (Zheng et al. 2012a), which is an automatic pipeline modified from the APSIS package (Blakeslee et al. 2003) with an enhanced capability of processing WFC3 data and aligning them with the ACS data. APLUS processes the calibrated images from the HST instrument pipelines, namely, the flc images for ACS (corrected for the detector's charge transfer efficiency) and the flt images for WFC3/IR. Recently, APLUS has been updated so that images of individual exposure are aligned using DrizzlePac (Gonzaga et al. 2012), and the accuracy can achieve 1/5 pixel (∼0015) or better.

In the APLUS process, images of different filter bands and cameras were further aligned, resampled, and combined, with a common pixel scale of 0065. We created detection images from the weighted sum of ACS/WFC and WFC3/IR images and ran SExtractor (Bertin & Arnouts 1996) in a dual mode for all 16 bands. mag_iso were chosen in the color selections. We also verified the photometry by comparing with public catalogs, ²¹ which were processed with a modified version of the Mosaicdrizzle pipeline (Koekemoer et al. 2003, 2011), and no systematic deviation was found between the two pipelines. In this paper, we focus on two ACS/WFC filters (F814W, F850LP) and five WFC3/IR filters (F105W, F110W, F125W, F140W, F160W), which are hereafter called the I₈₁₄, Z₈₅₀, Y₁₀₅, YJ₁₁₀, J₁₂₅, JH₁₄₀, and H₁₆₀ bands.

A color–color selection has been successfully used for identifying EELGs in VDW11. We carried out two selections with two sets of filter bands. First, we follow the selection criteria of VDW11, namely,

$$\begin{equation*} J_{125} - I_{814} < -0.44 - \sigma \wedge J_{125} - H_{160} < -0.44 - \sigma. \end{equation*}$$

In the other selection, we use the Y₁₀₅ band as the central band, namely,

$$\begin{equation*} Y_{105} - I_{814} < -0.44 - \sigma \wedge Y_{105} - H_{160} < -0.44 - \sigma, \end{equation*}$$

where the σ refers to the 1σ error of the color. We also require that the three bands in each selection are detected above 3σ to ensure good EW measurements. After these preliminary selections (Figure 1), we check the images and photometry of the WFC3/IR bands for each object. Sources contaminated by cosmic-ray events, nearby bright sources, and detector-edge effects are excluded. We build two samples with 40 and 12 candidates named as the "J" sample and "Y" sample. Note that the color excess of 0.44 mag in J₁₂₅ and Y₁₀₅ corresponds to a rest-frame EW of about 600 Å. The composite color images of these galaxies are shown in Figure 2. We include the apparent angular sizes, which are measured through full width at half-maximum (FWHM) by SExtractor in Tables 1 and 2.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To illustrate the boosting effect in different bands and different redshifts, we simulate model spectra with a simple power-law continuum plus emission lines and obtain the observed magnitudes using the throughputs of HST filters (lower panel in Figure 3). The index of the power-law continuum is fixed at β = 2, which is defined as F_λ   ∼   λ^−β. Such a continuum is a constant in different wavelengths with AB magnitude system and is set to 28 mag. In the model, we choose metal-line lists from galaxies with subsolar metallicity of Z = 0.2 Z_☉ = 0.004 in Anders et al. (2003) and only include emission lines with relative line intensities $F_{{\rm line}}/F_{H_{\beta }}$ larger than 0.1. The H_α line is included by assuming the ratio H_α/H_β = 2.86 from case B recombination (Storey & Hummer 1995).

Zoom In Zoom Out Reset image size

In the model, the EW([O iii] λ5007) is set to 2000 Å. The simulated magnitudes and colors are shown with thick blue lines in Figure 4. Based on this model, those EELGs in the redshift ranges ∼1.57–1.79 (the J sample) and ∼0.93–1.14 (the Y sample) are selected (gray regions in Figure 4). The redshift ranges would not change significantly if we use a different EW in the model. The upper panel of Figure 3 shows the wavelength ranges where the strongest emission lines, Hα λ6563, [O iii] λ5007, and [O ii] λ3727 impact the observed flux. For the Y sample, the [O ii] λ3727 falls into the I₈₁₄ band; thus EELGs with stronger emission lines can be selected.

Zoom In Zoom Out Reset image size

The spectral slopes of star-forming galaxies have been shown with a 1σ dispersion of 0.4 (Bouwens et al. 2009). In the second model, we take into consideration the effect of spectral slopes and vary them between β = 1.5 and 2.5. The effect is shown in a blue shadow region in the right panel of Figure 4. The changes in spectral slope would affect the color excess lower than 0.25 mag. Some EELGs with blue slopes will not be selected. However, as the continuum is calculated with the average of the two bands at different wavelengths, the change of slope is not a problem in the following EW estimates.

Other emission lines, especially Hα λ6563 and [O ii] λ3727, may also contribute to the broadband photometry, but the relative flux to [O iii] λ5007 will vary due to the difference in metallicity, star formation history, and extinction (Anders et al. 2003; Kewley et al. 2004; Salzer et al. 2005). In order to show the boosting effect, we build a model with line intensities from metallicity Z = 0.02 Z_☉ galaxy (green dots in Figure 4). In another model, we remove all other emission lines to show the contribution of [O iii] λλ4959, 5007 only (red lines in Figure 4). While most emission lines do not affect the photometry as significantly as [O iii] λλ4959, 5007, Hα λ6563 also boosts the magnitudes, as its strength is similar to [O iii] in typical star-forming galaxies, and the [O ii] λ3727 cannot be ignored.

As shown with the red lines in Figure 4, EELGs with redshift 1.14–1.57 would also be selected into the Y sample or J sample, but only if Hα λ6563 is extremely weak. The [O iii] λ5007 lines can be excited by massive stars as well as active galactic nuclei (AGNs). The relatively strong [O iii] λ5007 and weak Hα λ6563 can be due to either metal-poor star-forming galaxies or the contribution of AGNs (Kauffmann et al. 2003). To exclude these EELGs from the J sample is not possible with current observations. In the Y sample, these EELGs can be identifiable by comparing the magnitudes of JH₁₄₀ and H₁₆₀. As the Hα λ6563 falls in both the two bands in these galaxies, the two bands should be observed with similar magnitudes as shown in the left of Figure 4. We find 5/12 candidates in the Y sample have similar JH₁₄₀ and H₁₆₀ magnitudes. If parts of these weak H_α EELGs with low $H_{\alpha }/[\rm O\,\scriptsize{III}]$ ratios are AGNs, the upper limit of the AGN fraction in our Y sample is about 42% (5/12), which is still consistent with the AGN fraction (∼17%) estimated with a larger spectral emission-line galaxy sample in Atek et al. (2011). As our analysis is not sensitive to the redshift, we assume that the redshift ranges are z   ∼   1.57–1.79 and 0.93–1.14 for the J and Y samples, respectively.

The major advantage of CLASH observations is that cluster lensing provides a powerful tool to enable us to discover intrinsically faint galaxies. The magnification maps for all 25 clusters are made based on the strong lensing model of Zitrin et al. (2009, 2011). As magnification factors are redshift-dependent, we use the median redshift 1.03 for the Y sample and 1.68 for the J samples in the calculations. However, the following EW calculations are based on the color excesses and are thus independent of magnifications. The uncertainties of magnifications are lower than 10% for the ranges of redshifts. The estimated magnifications are likely consistent with the true value at 68% confidence for magnifications lower than 5 (Figure 11 in Bradley et al. 2014). The uncertainties are only significant for those highly magnified candidates.

The source magnification factors are listed in Tables 1 and 2. Candidate J22 has a magnification as large as 8.4 and shows an apparent extended structure. The delensed magnitude is estimated as ∼28 mag. Its EW is higher than 2000 Å, as confirmed with the color excess in YJ₁₁₀, J₁₂₅, and JH₁₄₀. The magnification distributions of our two samples are shown in Figure 5 in red color. We also calculate the magnifications for field galaxies that are within the same two redshift ranges as our samples in all 25 clusters and show the distributions in filled blue histograms.

Zoom In Zoom Out Reset image size

The EW is estimated using two continuum bands and one central peak band through

$$\begin{equation} {\rm EW} = \frac{F_{{\rm total}}-F_{c}}{F_{c}} \frac{{{W}}}{{1+z}}, \end{equation} \tag{ 1 }$$

where F_total is the flux of the central peak band, which is the total flux of emission lines and continuum, and F_c is the continuum flux in the I₈₁₄ and H₁₆₀ bands. W is the effective width of the central peak bands. z is the redshift of the EELGs and is set to the median redshifts, which are 1.03 and 1.68 for the Y and J samples, to translate EW to the rest frame. An accurate continuum analysis is important to estimate the EW. In the procedure, we assume the spectral slopes equal two to estimate the continuum based on the I₈₁₄ and H₁₆₀ bands, and use the weighted averages as the continuum at the central peak bands. In the redshifts of our two samples, the rest-frame UV continuums are observed by the ACS/WFC bands; thus the UV slopes can be obtained by linear fits of these bands (Column 11 in Tables 1 and 2). The average slopes of the J and Y samples are 2.06 ± 0.02 and 2.07 ± 0.03, consistent with our assumption and other studies of strong emission-line galaxies (van der Wel et al. 2011, 2013).

The uncertainties in the EW measurement are due to the I₈₁₄ and H₁₆₀ photometry and cause an overestimate of the continuum level and subsequently an underestimate of the EW. Within our model of EW = 2000 Å, the continuum is boosted by less than 0.1 mag for the J sample but larger for the Y sample, in which the [O ii] λ3727 line falls into the I₈₁₄ bands. Therefore, the EW should be considered as a lower limit in this situation. Furthermore, the continuum bands are considerably fainter than the central peak bands. Even if the continuum flux is estimated by averaging two continuum bands, the continuum errors are still the main factors affecting the EW measurements. For some faint EELGs, their continuum flux cannot be well constrained by photometric data due to these facts, which limit the accuracy of EW calculations.

As shown in Figures 3 and 4, all the YJ₁₁₀, J₁₂₅, and JH₁₄₀ are covered by the [O iii]+H_β for the J sample; thus the EWs can be constrained from the color excesses of these three bands. The measurements can then be verified by comparing EWs calculated from different central peak bands (Figure 6). The uncertainties in EW measurements are slightly higher for broader bands such as YJ₁₁₀. Nonetheless, the consistency of EWs inferred from three broad bands makes our results robust.

Zoom In Zoom Out Reset image size

At the redshift of the Y sample (Figure 4), Hα λ6563 has moved into the wavelength ranges of J₁₂₅ and JH₁₄₀; both [O iii]+H_β and Hα λ6563 are covered by the YJ₁₁₀. The Y₁₀₅ band is boosted by [O iii]+H_β only. The EWs derived from the color excess are included in Table 2.

In the J sample, the magnitudes in Y₁₀₅, YJ₁₁₀, and JH₁₄₀ are all boosted by [O iii] λλ4959, 5007, and the derived EWs can be averaged with the errors as weights to make a better constraint. Two EELGs (J2 and J29) display exceptionally large EWs of 3737 ± 726 Å and 3332 ± 439 Å, respectively. As seen in Figure 7, both of these EELGs are extremely compact. The FWHM of the instrument point-spread function (PSF) is 014 for the J₁₂₅; hence the EELGs are all resolved. After subtracting the instrument PSF and the lensing effect, the measured sizes correspond to physical sizes of about 1.7 and 0.9 kpc.

Zoom In Zoom Out Reset image size

In addition, the EW of Y9 is also measured to be 3307 ± 1293 Å. As only the Y₁₀₅ is free from contaminations by other emission lines in the Y sample, the EW can only be derived from the excess of the Y₁₀₅ band and thus contains higher uncertainty. Y9 is also selected as a LBG in Bradley et al. (2014) (B13 hereafter), and we will discuss this candidates in the next section.

The same selection method as our J sample is also utilized for the CANDELS field in VDW11, in which 69 candidates were identified in a total area of 279 arcmin². The typical area coverage over each cluster field in CLASH is ∼4 arcmin², indicated from the area available within the WFC3/IR field of view. To count for the influence of the bright galaxies in the fields, we counted the areas covered by galaxies based on the segmentation images and subtracted these areas from the total areas indicated from the exposure time images. Then the total areas to search EELGs in CLASH fields is ∼112 arcmin². The VDW11 used the data from the Ultra Deep Survey (UDS) field in the wide program and the data from the GOODS–South Deep (GSD) field at four-epoch depth in the deep program. The VDW11 sample includes 40 and 29 EELGs from these two fields, named as the UDS sample and GSD sample hereafter. The number density for the J sample is 0.36 armin⁻², compared to 0.19 armin⁻² in the UDS sample and 0.39 armin⁻² in the GSD sample. The high detection rates in the J sample and the GSD sample are due to the deeper observations. As shown in Figure 8, the J sample is the deepest and can reach about 28.5 mag after correcting the lensing effect. The lensing effect increases the depth and reduces the volume of observations in the meantime, in addition of the small samples, we do not find significant different detection rate between the CLASH fields and the GSD field.

Zoom In Zoom Out Reset image size

It is apparent in Figure 8 that the EELGs with higher EWs are more common in fainter EELGs. The three EELGs with extremely high EWs are fainter than most candidates. In the VDW sample, only two candidates have EWs larger than 2000 Å, which are 2304 ± 515 Å and 2002 ± 849 Å.²² Both of these candidates are selected from the GSD field, which is deeper than the UDS field. All these samples support the idea that we can detect EELGs with stronger emission lines in deeper observations. The three extreme candidates are unlikely due to noise fluctuations. We examine the probability that one EELG with EW([O iii]+Hβ λ4861) = 2000 Å at the same redshift is measured to be EW ⩾ 3000 Å. In our spectral model, fluctuations with the level of 1/3 continua, which are the worst cases in our samples, are added to each band. We simulate the photometry for 10,000 times and measure the EWs with the same method. The possibility of spurious large EW (⩾3000 Å) is lower than 20%. The rate will be reduced to lower than 4% if the EW are measured with flux excesses in two bands, and the rate will be negligible if there are three bands to constrain the EW. As the spectra in Figure 7 and the results in Table 1 show, the large EWs are confirmed with three bands for the J2 candidate and two bands for the J29 candidate. Both the two EELG candidates show similar spectral shapes; moreover, the significant boosts in Z₈₅₀ and Y₁₀₅ due to the [O ii] λ3727 line are also in agreement with the strong [O iii] 5007 Å line. The EW of the Y9 candidate can only be estimated with the boost of Y₁₀₅; therefore, the noise fluctuation cannot be totally excluded, and the strength could be confirmed with further observations.

VDW11 found these candidates are low-mass (∼10⁸ M_☉), starbursting (∼5 M_☉ yr⁻¹), young (5   ∼   40 Myr) galaxies. Their nature is also consistent with the spectral observations of two highly lensed EELGs at z = 1.85 and 3.12 (Brammer et al. 2012; van der Wel et al. 2013, respectively). We estimate the age and stellar mass following the same method as VDW11. In general, we use the Starburst99 model (SB99, Leitherer et al. 1999) with continuous star formation and a Chabrier (2003) IMF with a high-mass cutoff at 100 M_☉ and metallicity 0.2 Z_☉. The EW(H_β) declines as time and is used as an age indicator. EW(H_β) is assumed to contribute 1/8 to the combined EW. The upper limit of EW([O iii]+H_β) is 4336 Å from this model and will decrease to 3838 Å if we use a solar metallicity. VDW11 estimated stellar masses based on the rest-frame V band luminosity from the H₁₆₀ band photometry. To increase the accuracy, we use the weighted average photometry of the continue bands instead. The inferred masses are shown in Tables 1and 2. The median value of 7.1 × 10⁶ M_☉ is one order lower than the VDW sample. Maseda et al. (2013) found the dynamic mass and young stellar mass ratios of the VDW sample to be similar to that of other star-forming galaxies in z   ∼   2 and confirmed that the low stellar masses are dominated by the intense starbursts. These EELGs provide insight into the evolution of the dwarf galaxies and provide evidence that the starburst phase plays a key role in the mass buildup for at least some low-mass galaxies.

First, we test whether any sources in our samples would satisfy the color selections for LBGs. From Figure 1, both the terms of J₁₂₅ − I₈₁₄ in the J sample and Y₁₀₅ − I₈₁₄ in the Y sample can reach more than 1 mag. In the color–color selections of LBGs, a decrement of 1 mag is adopted (Bouwens et al. 2012; Zheng et al. 2012b; Oesch et al. 2012, 2013). Furthermore, a candidate at z > 7 must not be detected in the optical bands. For the bright EELGs in our samples, their continuum from near-infrared to UV bands is detected with high confidence. However, there are still a few EELGs with faint continua, and the bands bluer than I₈₁₄ fall bellow the detection limit, resulting in mimic LBGs with redshift around six. When the signal-to-noise ratio is low, it becomes difficult to distinguish whether the color excess is due to the Lyman break, as seen in LBGs, or the boost by strong emission lines in EELGs.

Recently, B13 reported a considerable number of galaxy candidates at z   ∼   6–8 in 18 CLASH clusters (A1423, A209, CLJ1226.9+3332, MACS 0429.6–0253, MACS 1311.0–0310, MACS 1423.8+2404, and RXJ 2129.7+0005 are not included compared to the total of 25 clusters). Their selections are based on the redshifts calculated from the Bayesian photometric redshift (BPZ) code (Benítez 2000). This method also identifies high-redshift galaxy candidates primarily based on the Lyman break feature, and the results are generally in very good agreement with the common color–color selection method. Possible contaminations of EELGs at the high-redshift samples are shown by matching the samples to our EELGs. Two EELGs, Y8 and Y9, are also selected as m1115–0352 (z = 6.2) and m1720–1114 (z = 5.9) in the z ∼ 6 sample of B13. Another galaxy in A209 (R.A.: 22.954264, decl.: −13.611176), which is removed from the Y sample because the I₈₁₄ band detection is below 3σ, is also included for its high photometric redshift (5.89) from BPZ and is called Y0. These candidates have similar spectral shapes, as shown with solid black circles in Figure 9. The best BPZ results are shown with open orange boxes. We also fit the I₈₁₄ band and redward bands with our strong emission-line model which is shown with blue boxes. In the emission-line model, we fix the spectral slope to two and the EW([O iii]+H_β) to be the value estimated with the Y₁₀₅ band excess. Emission lines with flux ratios from a 0.2 Z_☉ galaxy (Anders et al. 2003) are considered. The only two free parameters left are the redshift and the normalization. While the high-redshift assumption often fails to fit the H₁₆₀ band, our emission line model can explain the drop of flux in the H₁₆₀ bands in three galaxies. The emission-line model also overestimates the flux in the Y₈₁₄ band. The χ² values are shown with the same colors in the figure. For all the three candidates, the spectra favor the emission-line models for the slightly lower χ² values.

Zoom In Zoom Out Reset image size

If such candidates are chosen as LBGs, the UV-continuum slopes derived from infrared bands would be misleading. We estimate the slopes for the three candidates with a linear-fitting method for the bands redder than the I₈₁₄ band. The derived UV slopes are 3.7 ± 0.5, 3.6 ± 0.5, and 3.5 ± 0.4 for Y0, Y8, and Y9, respectively. Bouwens et al. (2010) obtain the UV-continuum slopes for redshift-6 galaxies with Y₁₀₅, J₁₂₅, and H₁₆₀. Using the same method, we get even bluer slopes, which are 4.5 ± 0.8, 4.9 ± 0.9, and 4.4 ± 0.6. These values are all at least 1σ bluer than the mean UV-continuum slope in redshift 6 (Bouwens et al. 2010). Nevertheless, the current CLASH data and Spitzer/IRAC data are not deep enough to confirm the nature of these galaxies. Our EELG color–color selections are limited by the luminosity of the continuum. Many more EELGs with stronger emission lines are faint and below our detection limit, according to the EW and luminosity trend. What is more, our samples only include EELGs in specific redshift ranges. Therefore, the total contamination of EELGs to the high-redshift sample of B13 can be higher.

Recently, the search for LBGs has reached z   ∼   10 (Zheng et al. 2012b; Bouwens et al. 2013; Ellis et al. 2013; Coe et al. 2013; Oesch et al. 2013). One of them, UDFj-39546284, has already been found to be possibly an EELG at redshift ∼2 (Brammer et al. 2013; Capak et al. 2013). More candidates are found in the ongoing Hubble Frontier Fields (Zheng et al. 2014; Zitrin et al. 2014; Ishigaki et al. 2014). As there are not enough spectral observations to confirm the redshifts, properly considering the contamination of EELGs is quite necessary for these samples.

For galaxies with z   ∼   8.7–10.5, the Lyα-break shifts into the JH₁₄₀ band and could be selected as J₁₂₅ drop out. If the redshift is higher than 10.5, the Lyα-break has shifted out of the J₁₂₅ band, and only H₁₆₀ and parts of JH₁₄₀ cover the continuum. EELGs with strong emission lines falling in the H₁₆₀ band can also mimic such a spectral feature. Considering a power-law spectrum with EW([O iii]+H_β) = 2000(3000) Å in our model, the dropout J₁₂₅ − H₁₆₀   ∼   1.1(1.4) would be observed for EELGs at redshift 1.9. Taking into account the deviation of the slopes, which would increase the observed color difference in EELGs, we propose that, unlike current HST WFC3/IR surveys in CLASH or UDF12, at least 1 mag deeper observations in the bluer bands will be required to constrain the flux of the continuum and draw a distinction between these two scenarios. In EELGs at the matched redshift, the Hα λ6563 has moved out of the H₁₆₀ band, but the impact of [O ii] λ3727 on J₁₂₅ cannot be neglected; therefore the deeper observations in the Y₁₀₅ or YJ₁₁₀ bands will be helpful. If a deeper observation is not available, a significant detection in Spitzer observation is required to confirm the high-redshift LBGs (Zheng et al. 2012b; Oesch et al. 2013). Real high-redshift LBGs have a flat UV continuum, and the detection from the H₁₆₀ band to the midinfrared bands cannot be mimicked by any type of EELGs.

While it is almost impossible to distinguish EELGs and high-redshift galaxies without spectroscopic observations, we should consider that there are a certain number of EELGs in the high-redshift galaxies. The spectral features of our J and Y samples are similar to the I₈₁₄ and Y₁₀₅ dropout galaxies at redshift range 5.6–8.0. The B13 sample includes a total of 206 galaxies in this redshift range, and we find two galaxies common to our EELG samples. If both of these galaxies are EELGs, the contamination fraction for the B13 sample is at least 2/206   ∼   1%. However, the real contamination fraction could be higher due to the limit of the observations. Our EELG samples only include candidates such that the continua are detected higher than 3σ in I₈₁₄. There are substantially fainter EELGs, which probably include stronger emission lines(See the trend in Figure 8).

Recently, the frontier of high-redshift samples are redshift higher than nine, selected based on the J₁₂₅ dropout. The fraction of EELGs in these J₁₂₅ dropout samples will be more significant than the contamination in the B13 sample. First, these J₁₂₅ dropout samples are in redshift range 8.7–10.5 and can be contaminated by EELGs in redshift range 1.8–2.2. The comoving volume of the redshift range 1.8–2.2 is 1.1 times the total volume of the Y and J samples. Meanwhile, the comoving volume of the J₁₂₅ dropout samples is 0.6 times the I₈₁₄ dropout samples. In the same projected area, there will be 1.1/0.6   ∼   1.8 times more contaminations due to EELGs. Second, the buildup of star-forming galaxies peaks at near redshift two, which means there is a substantial population of EELGs in that redshift. From the median redshift of our EELG samples 1.5 to redshift 1.9, the space density of star-forming galaxies increases by 2.3 times (Oesch et al. 2010). Moreover, the space densities of star-forming galaxies decline since redshift around two (Bouwens et al. 2014). From redshift 6 to redshift 9, the space density of galaxies decreases about 10 times. All of these will result in that the EELG contamination fraction for the J₁₂₅ dropout samples will be 1.8 × 2.3 × 10 ≈ 41 times higher. Pirzkal et al. (2013) develops a Markov Chain Monte Carlo fitting method with the ability to accurately estimate the probability density function of the redshift for each object. After the analysis of the redshift 8–12 galaxy sample, they report that there is an average probability of 21% that these well-defined high-redshift galaxies are low-redshift interlopers.

Without infrared or deeper HST observations, these EELG contaminations cannot be effectively excluded. Therefore, it is necessary to properly include the contamination fraction of EELGs for the analysis of high-redshift samples. However, the contamination fraction is still largely uncertain due to the lack of observations of EELGs. The ongoing Hubble Frontier Fields program devotes a total of 560 orbits to observe four clusters along with four parallel blank fields. This initiative can reach the 5σ magnitude limit of ∼28.7 mag, which is 1.2 mag deeper than the CLASH field. A systematical selection of EELGs in the Hubble Frontier Fields will help us to constrain the number density of faint EELGs and resolve the EELG contamination fraction for high-redshift galaxies.