README for v1.0 of LoTSS DR2 source classifications with DESI DR1
By Marina Arnaudova, Luke Holden & Dan Smith, July 2026
m.i.arnaudova@gmail.com l.holden@herts.ac.uk d.j.b.smith@herts.ac.uk
Description
This README refers to the contents of classfn_v1.0.fits, and should be referred to alongside Arnaudova, Holden, Smith et al. (2026) to understand how to use the probabilistic LoTSS + DESI classifications.
This is the first public release (version 1.0) of the probabilistic source classifications for LoTSS DR2 using WL-SLAYER (Arnaudova et al. 2026, and references therein) to produce spectral fits to select emission lines (Hbeta, [O III], Halpha, [N II]) in DESI DR1 optical spectroscopy. The method builds on the previous work of Drake et al. 2024 for this new sample, including extending to all sources at z < 0.947 using a modified MEx method in a manner similar to Arnaudova et al. (2025). The full method is described in Arnaudova et al. (2026); an abbreviated version is given below.
Quick start guide: Example use cases
This catalogue includes specific columns enabling the user to quickly define classes of objects (e.g. SFGs, LERGs) with all the recommended flags applied, and at a confidence level tailored to their selection to specific scientific needs. These include high-confidence samples at >99% (Class_99) or >90% (Class_90), as well as for more complete samples by including sources with >50% confidence (Class_50) or by adopting the most likely classification (Class_ML). Examples of these selections are shown below.
Selecting a sample of high confidence SFGs:
To identify a high-purity sample of star forming galaxies, you can proceed as follows : Class_SFG_BPT > 0.99 selects SFGs classified by the BPT and radio-excess (RX) diagnostic (at redshifts z<0.483 where all of the necessary lines are available) at a confidence level of 99%. Similarly, `Class_SFG_MEX > 0.99` selects high-confidence SFGs using the modified MEx method, alongside the RX diagnostic which works out to z < 0.98. It is therefore possible to use both methods, such that if the BPT method is inconclusive, the MEx method can be used instead, e.g:
High purity SFGs == (Class_SFG_BPT > 0.99 & BPT_warning_frac < 0.1 & zscore_BPT < 2.5) & (Class_SFG_BPT <= 0.99 & Class_SFG_MEx > 0.99 & MEx_warning_frac < 0.1 & zscore_MEx<2.5) & (FLAG_MASS == True) & (FLAG_SPEC == True) & (RX_warning_frac<0.1)
Due to the length of this example, we have produced convenience columns to enable common sample definitions more easily. This SFG example is fully equivalent to defining a subset using CLASS_99==1 (see description of all columns, including the CLASS_XX convenience columns, below).
Of course at z > 0.483, only the MEx method is possible since the BPT lines are not visible. Note that the recommended flags for the stellar masses and classifications have been applied. See the list of columns, below, for definitions of what each column means.
The same conditions can be applied for the rest of the classes at different thresholds with the exception of LERGs, where we have an extra requirement (see below).
Selecting a sample of high confidence LERGs
To select LERGs at a given confidence threshold, we apply the criteria described in the previous example (here adopting a 90% threshold), and additionally include sources identified by the [O III] EW method - namely, those that exhibit radio excess (RX) and [O III] EW < 3 Å, each with >90% probability.
High confidence LERGs ==
(Class_LERGs_BPT > 0.90 & BPT_warning_frac<0.1 & RX_warning_frac<0.1 & zscore_BPT<2.5)
&
(Class_LERGs_BPT <= 0.90 & Class_LERGs_MEx > 0.90 & MEx_warning_frac<0.1 & RX_warning_frac<0.1 & zscore_MEx<2.5 & FLAG_MASS == True)
&
(Class_LERGs_BPT <= 0.90 & Class_LERGs_MEx <= 0.90 & RX>90% & EW_frac>90% & RX_warning_frac<0.5)
&
(FLAG_SPEC == True)
Here, EW_frac refers to the fraction of Monte Carlo realisations (or equivalently, the probability) for which [O III] EW < 3 Å. Note that when using the EW_frac we allow the RX_warning_frac to reach 50%. These code snippets to identify a high-confidence LERG class are fully equivalent to using the convenience column: Class_90==3.
Selecting a sample of most likely HERGs
The combined DESI and LoTSS sample selection has a big impact on the number of sources falling in each class. Irrespective, it may be of interest to identify a sample of sources for which the most likely classification is that of a HERG. This can be accomplished using e.g:
BPT_HERGS == (Class_HERG_BPT > Class_SFG_BPT) & (Class_HERG_BPT > Class_LERG_BPT) & (Class_HERG_BPT > Class_RQAGN_BPT) & (FLAG_SPEC == True) & (RX_warning_frac<0.1) & (BPT_warning_frac<0.1) & (zscore_BPT<2.5)
Or if the modified MEx method is preferred instead of the BPT:
MEx_HERGS == (Class_HERG_MEX > Class_SFG_MEX) & (Class_HERG_MEX > Class_LERG_MEX) & (Class_HERG_MEX > Class_RQAGN_MEX) & (FLAG_MASS == True) & (FLAG_SPEC == True) & (RX_warning_frac<0.1) & (MEx_warning_frac<0.1) & (zscore_MEx<2.5)
The convenience column Class_ML (see below) combines the two methods, where the BPT+RX method is chosen for all sources at z<0.483, whereas the MEx+RX method is done for all sources at 0.483<z<0.947. The example above is equivalent to setting Class_ML==4. Similar workflows can be applied to the rest of the classes (i.e. for SFGs, RQ AGN and/or LERGs), and these are equivalent to using the Class_ML column.
High-confidence BPT-SFG and BPT-LINER LERGs
This example indicates a way to select a simple sample of LERGs located in the BPT-SFG or BPT-LINER region at 90% confidence:
BPT_SFG_LERGS == (Class_LERG_BPT > 0.9) & (BPT_SFG > 0.9) & (FLAG_MASS == True) & (FLAG_SPEC == True) & (RX_warning_frac<0.1) & (BPT_warning_frac<0.1)& (zscore<2.5)
BPT_LINER_LERGS == (Class_LERG_BPT > 0.9) & (BPT_SFG < 0.1) & (FLAG_MASS == True) & (FLAG_SPEC == True) & (RX_warning_frac<0.1) & (BPT_warning_frac<0.1)& (zscore<2.5)
Note that the columns BPT_SFG and CLASS_BPT_SFG differ; while BPT_SFG contains the estimated probability of a source being located in the BPT_SFG region, the Class_BPT_SFG column contains the probability of a source being located in the BPT_SFG region AND not identified as having a radio excess according to the RX diagnostic (see Arnaudova et al. 2026 for further details).
Identifying and excluding probable type 1 AGN
Some of the sources in this sample are likely to be type 1 AGN. We have implemented a flag to identify some of the most likely candidates (FLAG_TYPE1), specifically if all the following criteria are met:
- If the Hbeta total FWHM > 500 km/s.
- The total FWHMs for Hbeta and [O III] are not consistent within 3sigma.
- Total FWHM for [O III] subtracted from the total FWHM for Hbeta is less than 3sigma.
You will undoubtedly wish to adapt any of these examples to match your particular interests.
Methodology
Sample
The sample was produced by a 1.5 arcsecond positional crossmatch using a nearest-neighbour algorithm between the positions (RA and Dec) of DESI DR1 optical spectra and the LoTSS DR2 optical IDs of Hardcastle et al. 2023). Our classification scheme requires the redshifted Hbeta and OIII emission lines to fall within the response curve of the DESI Spectrograph. Therefore, considering objects with non-zero DESI redshift and quality flags, aperture corrections (aper_corr) with factors 1<aper_corr<5 and redshifts z<0.947, we obtain a total sample size of 279,700 objects (at z<0.483 there are 207,657 objects, for which Halpha and [NII] are also visible; at 0.483<z<0.947 there are 72,043 objects).
Emission-line fitting and line fluxes
Emission-line fitting was performed using the WEAVE-LOFAR Spectral Analyser (WL-SLAYER) code (Arnaudova et al. 2026; see also Arnaudova et al. 2024, Holden & Tadhunter 2025, Arnaudova et al. 2025, Holden et al. 2026). WL-SLAYER uses a Markov Chain Monte Carlo routine to first fit the stellar continuum with pPXF (https://pypi.org/project/ppxf/) and then fits each required emission line with N number of Gaussian profiles, where N is determined by iteratively increasing the number of Gaussian profiles until the additional model complexity outweighs the improvement of the fit (as determined using the Bayesian Information Criterion). The emission-line quantities such as fluxes, EWs and FWHM for each emission line were determined by generating 1000 realisations of the emission-line profile from the MCMC chains and calculating the area underneath the line profile for each realisation and taking the 50th percentile of the resulting flux distribution; 1sigma errors were estimated using the 16th and 84th percentiles.
Where possible, extinction values (needed to correct line fluxes) were derived from the Halpha and Hbeta Balmer decrement. Otherwise, we used a diagnostic involving the rest-frame U, V, and J photometric magnitudes (the UVJ diagram), based on a correlation between UVJ colours and the total extinction measured from the Balmer decrement. This is discussed in Arnaudova et al. (2026).
Classification scheme
We apply classifications based on two methods - the BPT-NII (for the subset where all the necessary lines are available) and the `MEx’ classification (based on a modified version of the well-known Mass Excitation method; Juneau et al. 2011, 2014) for all sources where both Hbeta and [OIII] are detectable. In the catalogues, the results for the two schemes are identified based on the "BPT" and "MEx" subscripts. The BPT classification is similar to the one presented in Drake et al. (2024), where a radio-excess (based on the dust-corrected Halpha luminosity) and BPT-NII diagnostic are used in a probabilistic manner (however we use the line from Cid Fernandes et al. 2010 to distinguish between LINERs and Seyferts). The modified MEx classification is presented in Arnaudova et al. (2025), where a radio excess diagnostic (based on the Hbeta flux and dust extinction derived from the UVJ diagram) and the modified MEx diagram are used. In addition, to identify LERGs where we are confident that they have a radio excess, but which have low-significance emission lines, we use the OIII EW to select radio sources for which the OIII EW is below 3A (roughly where the distribution of OIII EW of HERGs drops off).
Column descriptions
In addition to the values used for the classifications, below, we also provide two velocity parameters calculated from the emission-line profiles: total FWHM (the full width at half maximum for the total emission-line profile) and W80 (the velocity width that contains 80 per cent of the total line flux), as well as equivalent widths (EW).
| Column | Units | Description |
|---|---|---|
source_name |
Object identifier (ILT name from the LoTSS DR2 catalogue) | |
targetid |
DESI target ID | |
RA |
deg | Right ascension of the DESI target |
DEC |
deg | Declination of the DESI target |
z |
Redshift value from DESI | |
ngauss |
The number of Gaussian components fitted to the emission lines | |
desi_target |
DESI target selection and fiberassign data | |
sersic |
Power-law index for the Sersic profile model from DESI | |
r_50 |
arcsec | The half-light radius taken from Hardcastle et al. 2023 |
S144 |
Jy | 144 MHz total flux density |
S144_err |
Jy | Uncertainty for 144 MHz total flux density |
L144 |
W/Hz | The 144 MHz luminosity assuming a spectral index of -0.7 |
logm_CG |
dex(sol mass) | Log mass estimate from DESI’s AGN Host Galaxies Physical Properties VAC |
logm_err_CG |
dex(solar mass) | Error on the log mass estimatea |
Av_Balmer |
magnitudes | Maximum likelihood total extinction as calculated from the Halpha and Hbeta Balmer decrement |
Av_UVJ_p16 |
magnitudes | 16 per cent lower confidence bound for extinction calculated from the UVJ diagnostic |
Av_UVJ_p50 |
magnitudes | Median likelihood extinction value for extinction calculated from the UVJ diagnostic |
Av_UVJ_p84 |
magnitudes | 84 per cent lower confidence bound for extinction calculated from the UVJ diagnostic |
aper_corr |
Aperture corrections taken from DESI’s FastSpecFit VAC | |
flux_Hb_4861_p16 |
10^-17 erg/s/cm^2 | 16 per cent lower confidence bound for Hbeta line flux |
flux_Hb_4861_p50 |
10^-17 erg/s/cm^2 | Median likelihood value for Hbeta line flux |
flux_Hb_4861_p84 |
10^-17 erg/s/cm^2 | 84 per cent upper confidence bound for Hbeta line flux |
flux_OIII_5007_p16 |
10^-17 erg/s/cm^2 | 16 per cent lower confidence bound for [O III]5007 line flux |
flux_OIII_5007_p50 |
10^-17 erg/s/cm^2 | Median likelihood value for [O III]5007 line flux |
flux_OIII_5007_p84 |
10^-17 erg/s/cm^2 | 84 per cent upper confidence bound for [O III]5007 line flux |
flux_NII_6583_p16 |
10^-17 erg/s/cm^2 | 16 per cent lower confidence bound for [N II] 6583 line flux |
flux_NII_6583_p50 |
10^-17 erg/s/cm^2 | Median likelihood value for [N II] 6583 line flux |
flux_NII_6583_p84 |
10^-17 erg/s/cm^2 | 84 per cent upper confidence bound for [N II] 6583 line flux |
flux_Ha_6563_p16 |
10^-17 erg/s/cm^2 | 16 per cent lower confidence bound for Halpha line flux |
flux_Ha_6563_p50 |
10^-17 erg/s/cm^2 | Median likelihood value for Halpha line flux |
flux_Ha_6563_p84 |
10^-17 erg/s/cm^2 | 84 per cent upper confidence bound for Halpha line flux |
EW_Hb_4861_p16 |
10^-17 erg/s/cm^2 | 16 per cent lower confidence bound for Hbeta line EW |
EW_Hb_4861_p50 |
10^-17 erg/s/cm^2 | Median likelihood value for Hbeta line EW |
EW_Hb_4861_p84 |
10^-17 erg/s/cm^2 | 84 per cent upper confidence bound for Hbeta line EW |
EW_OIII_5007_p16 |
10^-17 erg/s/cm^2 | 16 per cent lower confidence bound for [O III]5007 line EW |
EW_OIII_5007_p50 |
10^-17 erg/s/cm^2 | Median likelihood value for [O III]5007 line EW |
EW_OIII_5007_p84 |
10^-17 erg/s/cm^2 | 84 per cent upper confidence bound for [O III]5007 line EW |
EW_NII_6583_p16 |
10^-17 erg/s/cm^2 | 16 per cent lower confidence bound for [N II] 6583 line EW |
EW_NII_6583_p50 |
10^-17 erg/s/cm^2 | Median likelihood value for [N II] 6583 line EW |
EW_NII_6583_p84 |
10^-17 erg/s/cm^2 | 84 per cent upper confidence bound for [N II] 6583 line EW |
EW_Ha_6563_p16 |
10^-17 erg/s/cm^2 | 16 per cent lower confidence bound for Halpha line EW |
EW_Ha_6563_p50 |
10^-17 erg/s/cm^2 | Median likelihood value for Halpha line EW |
EW_Ha_6563_p84 |
10^-17 erg/s/cm^2 | 84 per cent upper confidence bound for Halpha line EW |
fwhm_total_Hb_4861_p16 |
km/s | 16 per cent lower confidence bound for Hbeta line total FWHM |
fwhm_total_Hb_4861_p50 |
km/s | Median likelihood value for Hbeta line total FWHM |
fwhm_total_Hb_4861_p84 |
km/s | 84 per cent upper confidence bound for Hbeta line total FWHM |
fwhm_total_OIII_5007_p16 |
km/s | 16 per cent lower confidence bound for [O III]5007 line total FWHM |
fwhm_total_OIII_5007_p50 |
km/s | Median likelihood value for [O III]5007 line total FWHM |
fwhm_total_OIII_5007_p84 |
km/s | 84 per cent upper confidence bound for [O III]5007 line total FWHM |
fwhm_total_NII_6583_p16 |
km/s | 16 per cent lower confidence bound for [N II] 6583 line total FWHM |
fwhm_total_NII_6583_p50 |
km/s | Median likelihood value for [N II] 6583 line total FWHM |
fwhm_total_NII_6583_p84 |
km/s | 84 per cent upper confidence bound for [N II] 6583 line total FWHM |
fwhm_total_Ha_6563_p16 |
km/s | 16 per cent lower confidence bound for Halpha line total FWHM |
fwhm_total_Ha_6563_p50 |
km/s | Median likelihood value for Halpha line total FWHM |
fwhm_total_Ha_6563_p84 |
km/s | 84 per cent upper confidence bound for Halpha line total FWHM |
w80_Hb_4861_p16 |
km/s | 16 per cent lower confidence bound for Hbeta line W80 |
w80_Hb_4861_p50 |
km/s | Median likelihood value for Hbeta line W80 |
w80_Hb_4861_p84 |
km/s | 84 per cent upper confidence bound for Hbeta line W80 |
w80_OIII_5007_p16 |
km/s | 16 per cent lower confidence bound for [O III]5007 line W80 |
w80_OIII_5007_p50 |
km/s | Median likelihood value for [O III]5007 line W80 |
w80_OIII_5007_p84 |
km/s | 84 per cent upper confidence bound for [O III]5007 line W80 |
w80_NII_6583_p16 |
km/s | 16 per cent lower confidence bound for [N II] 6583 line W80 |
w80_NII_6583_p50 |
km/s | Median likelihood value for [N II] 6583 line W80 |
w80_NII_6583_p84 |
km/s | 16 per cent upper confidence bound for [N II] 6583 line W80 |
w80_Ha_6563_p16 |
km/s | 16 per cent lower confidence bound for Halpha line W80 |
w80_Ha_6563_p50 |
km/s | Median likelihood value for Halpha line W80 |
w80_Ha_6563_p84 |
km/s | 84 per cent upper confidence bound for Halpha line W80 |
The following “convenience columns” include all the recommended flags applied, as well as the relevant thresholds reached for each class:
| Column | Units | Description |
|---|---|---|
Class_ML |
The most probable/likely (best guess) classification, where SFGs=1, RQAGN=2, LERGs=3, HERGs=4 and Unclassified=-1 | |
Class_50 |
Classification using a 50% reliability threshold, where SFGs=1, RQAGN=2, LERGs=3, HERGs=4 and Unclassified=-1 | |
Class_70 |
Classification using a 70% reliability threshold, where SFGs=1, RQAGN=2, LERGs=3, HERGs=4 and Unclassified=-1 | |
Class_90 |
Classification using a 90% reliability threshold, where SFGs=1, RQAGN=2, LERGs=3, HERGs=4 and Unclassified=-1 | |
Class_99 |
Classification using a 99% reliability threshold, where SFGs=1, RQAGN=2, LERGs=3, HERGs=4 and Unclassified=-1 | |
Class_SFG_BPT |
Probability of source classification being a star forming galaxy (SFG) based on the radio excess and BPT diagnostic | |
Class_RQAGN_BPT |
Probability of source classification being a radio quiet AGN (RQAGN) based on the radio excess and BPT diagnostic | |
Class_LERG_BPT |
Probability of source classification being a low excitation radio galaxy (LERG) based on the radio excess and BPT diagnostic | |
Class_HERG_BPT |
Probability of source classification being a high excitation radio galaxy (HERG) based on the radio excess and BPT diagnostic | |
Class_SFG_MEx |
Probability of source classification being a star forming galaxy (SFG) based on the radio excess and MEx diagnostic | |
Class_RQAGN_MEx |
Probability of source classification being a radio quiet AGN (RQAGN) based on the radio excess and MEx diagnostic | |
Class_LERG_MEx |
Probability of source classification being a low excitation radio galaxy (LERG) based on the radio excess and MEx diagnostic | |
Class_HERG_MEx |
Probability of source classification being a high excitation radio galaxy (HERG) based on the radio excess and MEx diagnostic | |
BPT_SFG |
Probability of source having line ratios within the SFG region of the BPT diagram | |
BPT_CLIN |
Probability of source having line ratios within the composite LINER (CLIN) region of the BPT diagram | |
BPT_CSeyf |
Probability of source having line ratios within the composite Seyfert (CSeyf) region of the BPT diagram | |
BPT_LIN |
Probability of source having line ratios within the LINER region of the BPT diagram | |
BPT_Seyf |
Probability of source having line ratios within the Seyfert region of the BPT diagram | |
MEx_SFG |
Probability of source having line ratios and a stellar mass within the SFG region of the BPT diagram | |
MEx_CLIN |
Probability of source having line ratios and a stellar mass within the composite LINER (CLIN) region of the BPT diagram | |
MEx_CSeyf |
Probability of source having line ratios and a stellar mass within the composite Seyfert (CSeyf) region of the BPT diagram | |
MEx_LIN |
Probability of source having line ratios and a stellar mass within the LINER region of the BPT diagram | |
MEx_Seyf |
Probability of source having line ratios and a stellar mass within the Seyfert region of the BPT diagram | |
RX |
Probability that source has a radio excess, i.e. that the 144 MHz to the Halpha luminosity ratio is larger than predicted values for 99% of SFGs in the entire sample. | |
RX_warning_frac |
The fraction of realisations for which “poorly-behaved values” occur in the RX diagnostic | |
BPT_warning_frac |
The fraction of realisations for which “poorly-behaved values” occur in the BPT diagnostic | |
MEx_warning_frac |
The fraction of realisations for which “poorly-behaved values” occur in the MEx diagnostic | |
EW_frac |
The fraction of realisations for which the OIII 5007 EW <3A | |
Av_flag |
The fraction of times for which the Av was randomly drawn from the best-fit Av values for sources with 5sigma in Halpha and Hbeta | |
zscore_BPT |
z-score for the maximum-likelihood classification from the BPT diagnostic | |
zscore_MEx |
z-score for the maximum-likelihood classification from the MEx diagnostic | |
ML_BPT |
Maximum likelihood classification from BPT diagnostic ('Unc' = unclassified) | |
ML_MEx |
Maximum likelihood classification from MEx diagnostic ('Unc' = unclassified) | |
flag_mass_CG |
True if mass is measured and reliable | |
flag_spec |
True if emission-line fits are reliable, i.e. flux uncertainties are not zero or very large and the chi-squared value for the fit is below the 99th percentile of the chi-squared distribution fitted for the entire sample. | |
flag_type1 |
True if source is a possible Type 1 AGN, i.e. that Hbeta total FWHM > 500 km/s, is not consistent with the [O III]5007 total FWHM within 3sigma, and that [O III]5007 total FWHM - Hbeta total FWHM < 3sigma. |