Electron temperature relations and the direct N, O, Ne, S and Ar abundances of 49 959 star-forming galaxies in DESI Data Release 2 2026 Electron temperature relations and the direct N, O, Ne, S and Ar abundances of 49 959 star-forming galaxies in DESI Data Release 2 – B D. Scholte, F. Cullen, J. M. Moustakas, H. Zou, A. Saintonge, K. Z. Arellano-Cordova, T. M. Stanton, B. Andrews, J. Sui, J. Aguilar, S. Ahlen, D. Bianchi, D. Brooks, F. J. Castander, T. Cheng, T. Claybaugh, A. de la Macorra, B. Dey, P. Doel, K. Douglass, S. Ferraro, J. E. Forero-Romero, E. Gaztañaga, S. Gontcho A Gontcho, G. Gutierrez, R. Joyce, A. Kremin, O. Lahav, M. Landriau, L. Le Guillou, P. Martini, A. Meisner, R. Miquel, W. J. Percival, C. Poppett, F. Prada, I. Pérez-Ràfols, G. Rossi, E. Sanchez, D. Schlegel, Z. Shao, J. Silber, D. Sprayberry, G. Tarlé and B. A. Weaver Author affiliations are listed at the end of the paper Abstract We present the largest direct-method abundance catalogue of galaxies to date, containing measurements of 49 959 star-forming galaxies at z < 0.96 z<0.96 from DESI Data Release 2. By directly measuring electron temperatures across multiple ionization zones, we provide constraints on a number of electron temperature relations finding good consistency with previous literature relations. Using these temperature measurements, we derive reliable abundances for N, O, Ne, S and Ar and measure the evolution of abundances and abundance ratios of as a function of metallicity and other galaxy properties. Our measurements include direct oxygen abundances for 49 766 galaxies, leading to the discovery of the two most metal-poor galaxies in the nearby Universe, with oxygen abundances of 12 + log ⁡ ( O / H ) = 6.77 − 0.03 + 0.03 ​ dex \rm 12+\log(O/H)=6.77_{-0.03}^{+0.03}~\rm dex (1.2% Z ⊙ \rm Z_{\odot} ) and 12 + log ⁡ ( O / H ) = 6.81 − 0.04 + 0.04 ​ dex \rm 12+\log(O/H)=6.81_{-0.04}^{+0.04}~\rm dex (1.3% Z ⊙ \rm Z_{\odot} ). We identify a rare outlier population of 139 galaxies with high N/O ratios at low metallicity, reminiscent of galaxy abundances observed in the early Universe. We find these high N/O galaxies are more massive than typical galaxies at the same metallicity. We find the Ne/O ratio is constant at low metallicity but increases significantly at 12 + log ⁡ ( O / H ) > 8.105 ± 0.004 \rm 12+log(O/H)>8.105\pm 0.004 dex. We show that the S/O and Ar/O abundance ratios are strongly correlated, consistent with the expected additional Type Ia enrichment channel for S and Ar. In this work we present an initial survey of the key properties of the sample, with this dataset serving as a foundation for extensive future work on galaxy abundances at low redshift. 1 Introduction The chemical enrichment of galaxies through nucleosynthesis in successive generations of stars is a key process in galaxy evolution. The vast majority of elements, except for hydrogen, helium and some lithium ( Alpher et al. 1948 ) , are synthesised through various processes in different types of stars ( Burbidge et al. 1957 ) . The relative abundance ratios of different elements in galaxies are a product of their respective production pathways in e.g., massive stars and intermediate mass stars or SNe Ia ( Kobayashi et al. 2020 , e.g.,) and therefore sensitive to the star-formation history. The gas-phase abundances of elements in galaxies are set through the interplay between their production, dilution through the inflows of pristine, unenriched gas and removal through processes such as stellar winds and supernovae; this complex interplay of processes is often referred to as the baryon cycle of galaxies ( Tinsley 1980 ) . The gas phase abundances of N, O, Ne, S and Ar can be measured through their emission lines in the rest-frame optical spectra of star forming galaxies. The measurement of gas phase abundances has been an ongoing effort over successive generations of spectroscopic surveys ( Tremonti et al. 2004 ; Izotov et al. 2006 ; Guseva et al. 2011 ; Curti et al. 2017 ; Curti et al. 2020 ; Berg et al. 2020 ; Rogers et al. 2022 ; Arellano-Córdova et al. 2024 ; Scholte et al. 2024 ; Esteban et al. 2025 ; Sanders et al. 2025 , see e.g.,) . Most studies focus on the abundance of oxygen, which is the most abundant metal in the ISM and therefore serves as a key tracer for understanding the chemical evolution of galaxies ( Tremonti et al. 2004 , see) . However, detailed studies of multiple abundances provide a more complete picture, since abundance ratios are sensitive to the different production pathways of elements. Elements with similar production pathways exhibit fairly constant abundance ratios over a wide range of metallicities (and other galaxy properties). This is the case for elements produced in the triple-alpha process in massive stars ( Woosley et al. 2002 , M_{\star}>8\rm~M_{\odot} , see e.g.,) , so called alpha-elements such as O, Ne, S and Ar, of which the relative abundances are approximately constant ( Izotov et al. 2006 ; Berg et al. 2019 ; Arellano-Córdova et al. 2024 ; Esteban et al. 2025 , e.g.) . However, S and Ar each have a secondary production pathway through SNe Ia, due to which there is a delayed release of a significant fraction of the abundance of these elements ( Matteucci & Chiappini 2005 ; Arnaboldi et al. 2022 ; Rogers et al. 2024 ; Stanton et al. 2025b ; Bhattacharya et al. 2025a ; Foley et al. 2025 ) . Elements with different production pathways, such as nitrogen, exhibit a more complex behaviour compared to alpha-elements. Nitrogen is primarily produced in intermediate mass stars ( {\rm 4~M_{\odot}}=40 . The redshifts of DESI galaxies are measured using the RedRock template-fitting pipeline to derive classifications and redshifts for each targeted source ( Anand et al. 2024 , Bailey et al. in prep.) . The detection of faint auroral emission lines is crucial for the analysis in this work. Therefore, we perform a signal-to-noise ( \rm S/N ) selection where we only select galaxies with \rm S/N>5 detection of the line flux of at least one of the following auroral lines or doublets in the FastSpecFit emission line catalogue of the full DESI DR2 data: [ Oiii ] \lambda 4363, [ Nii ] \lambda 5755, [ Siii ] \lambda 6312 or [ Oii ] \lambda\lambda 7320,7330. We also require galaxies with \rm EW(H\beta)_{rest}>20 , to ensure the galaxies in the sample have a good detection of emission lines that are not significantly affected by uncertainties in the continuum subtraction. This selection criterion also ensures that the selected galaxy sample is not strongly affected by emission line contamination from diffuse ionised gas ( Vale Asari et al. 2019 ; Mannucci et al. 2021 , DIG; see) . The spectra of all the selected galaxies are refit with FastSpecFit , using a custom set of SED templates, a custom emission line list, and an improved uncertainty measurement using 100 Monte Carlo simulations of the fit to each spectrum. These updates tailor the SED fitting to the sample of galaxies selected here and ensure the reliable detection of particularly faint emission lines. The SED templates are derived using a Chabrier 2003 initial mass function, MIST isochrones ( Choi et al. 2016 ) and FSPS population synthesis ( Conroy et al. 2009 ; Conroy & Gunn 2010 ) . We parametrise the star formation history using 8 variable-width age bins with constant star formation in each bin: 0-30 Myr, 30-100 Myr, 100-259 Myr, 259-670 Myr, 670 Myr -1.73 Gyr, 1.73-4.50 Gyr, 4.50-11.6 Gyr and 11.6-13.7 Gyr. We also include templates in three metallicity bins: 10% \rm Z_{\odot} , 50% \rm Z_{\odot} and 100% \rm Z_{\odot} . Dust emission is modelled using the Draine & Li 2007 dust model where we have adopted the following values: q_{\rm PAH}=1\% , U_{\rm min}=1 and \gamma=0.01 ( Draine et al. 2007 , see also) . Our custom emission line list (see Table 1 ) includes a large number of faint emission lines, e.g., we include measurements of the [ Sii ] \lambda\lambda 4068, and the [Ar IV ] \lambda 4740 emission lines which are essential for our electron temperature and abundance measurements. The results from the custom FastSpecFit analysis are used for measurements of emission line fluxes and stellar masses used throughout this work. Table 1: Custom emission line list. Line name Rest wavelength Comment [NeV] \lambda 3346 3346.79 [NeV] \lambda 3426 3426.85 [OII] \lambda 3726 3727.10 [OII] \lambda 3729 3729.86 H \theta 3798.978 H \eta 3836.478 [NeIII] \lambda 3869 3869.86 H \zeta 3890.166 [NeIII] \lambda 3968 3968.593 H \epsilon 3971.198 [SII] \lambda 4068 4069.750 [SII] \lambda 4076 4077.500 H \delta 4102.892 [FeII] \lambda 4288 4288.599 H \gamma 4341.692 [FeII] \lambda 4360 4359.387 [OIII] \lambda 4363 4364.436 HeI \lambda 4471 4472.735 HeII \lambda 4686 4687.02 [ArIV] \lambda 4713 4713.574 [ArIV] \lambda 4740 4741.495 H \beta 4862.71 [OIII] \lambda 4959 4960.295 Flux tied to [OIII] \lambda 5007 using the flux ratio from Dimitrijević et al. 2007 . [OIII] \lambda 5007 5008.240 [NII] \lambda 5755 5756.191 HeI \lambda 5876 5877.249 [OI] \lambda 6300 6302.046 [SIII] \lambda 6312 6313.81 [OI] \lambda 6364 6365.535 [NII] \lambda 6548 6549.861 Flux tied to [NII] \lambda 6584 using the flux ratio from Dojčinović et al. 2023 . H \alpha 6564.60 [NII] \lambda 6584 6585.273 [SII] \lambda 6716 6718.294 [SII] \lambda 6731 6732.674 [ArIII] \lambda 7135 7137.77 [OII] \lambda 7320 7321.94 [OII] \lambda 7330 7332.21 [SIII] \lambda 9069 9071.1 [SIII] \lambda 9532 9533.2 Figure 1: Figure of an example DESI spectrum of DESI J031.5272+08.5595 with the fitted FastSpecFit ( Moustakas et al. 2023a ) model. The spectrum is shown in black with uncertainties in the flux measurements shown by the grey band. The fitted model is shown in red. We highlight the fitted emission lines in each panel with pink vertical bands and the names of the fitted lines. In the bottom panel we show five inset figures showing the fitted emission lines of the [ Sii ] \lambda\lambda 4068,4076, [ Oiii ] \lambda 4363, [ Nii ] \lambda 5755, [ Siii ] \lambda 6312 and [ Oii ] \lambda\lambda 7320,7330 auroral lines. This spectrum was specifically chosen as all five auroral lines/doublets are detected. The inset image in the top panel shows the Legacy Survey imaging of this target with the 1.5" aperture of the DESI optical fiber at the pointing location of the observation as well as a 10" aperture for scale. After refitting the spectra, we reapply the same quality cut on our final custom emission line catalog; we select galaxies with \rm S/N>5 detection of the line flux of at least one of the auroral lines or doublets, this time including the [ Sii ] \lambda 4069 auroral line. We also exclude any auroral line measurements where 1.5\times{A_{\rm line}}>|r^{90}_{\rm line}| , where r^{90}_{\rm line} is the 90 th percentile residuals within a 60 Å window around the emission line and A_{\rm line} is the amplitude of the emission line. This selection removes line fits in noisy spectra or spectra with sky subtraction or continuum subtraction issues. Additionally, we exclude auroral line measurements where the line centre falls within 10Å of the edge of one of the DESI cameras. Faint emission lines can be affected by inaccurate measurements in these regions. We only include [ Oiii ] \lambda 4363 measurements where the 1- \sigma linewidth is less than 2Å as [ Oiii ] \lambda 4363 is difficult to deblend from [Fe ii ] \lambda 4360 when the emission lines are broader ( Curti et al. 2017 ) . To ensure we can constrain dust attenuation, we require a S/N>3 detection of H \beta and either H \alpha or H \gamma . The bulk of the targets in our sample selection are star-forming galaxies, however, there are two other categories: galaxies with emission from active galactic nuclei (AGN) and individual Hii -regions of nearby galaxies. We identify AGN in the [ Nii ]-BPT ( Baldwin et al. 1981 ) emission line diagram as galaxies which are not identified as star-forming using the criterion defined by Kewley et al. 2001 and also not star-forming based on the criterion by Kauffmann et al. 2003 . At z\gtrsim 0.45 the H \alpha and [ Nii ] \lambda 6584 emission lines are not observed by DESI, therefore at these redshifts we identify AGN using the mass-excitation diagram ( Juneau et al. 2014 , see equation 1 in) . We also exclude any galaxies which have broad H \alpha or H \beta lines with \sigma^{\rm broad}_{\rm H\alpha}>1000\rm~km~s^{-1} or \sigma^{\rm broad}_{\rm H\beta}>1000\rm~km~s^{-1} . Additionally, we exclude galaxies where \sigma^{\rm narrow}_{\rm H\beta}>2\times\sigma^{\rm narrow}_{\rm[OIII]\lambda 5007} , this selection excludes galaxies where the broad and narrow components of the Balmer lines are not fitted correctly. This occurs almost exclusively at z\gtrsim 0.48 , where the H \alpha emission line is not observed and therefore the narrow and broad components of Balmer lines are more difficult to constrain. We identify individual Hii -regions of nearby galaxies using the SGA-2020 ( Moustakas et al. 2023b ) . We mask targets as likely resolved Hii -regions when located within the radius of the major-axis of the 26 mag arcsec -2 r -band isophote; we conservatively use the circular radius and not the full elliptical shape of the isophote. We do not mask targets where z>0.15 or where the distance of the targeted DESI fiber from the centre of the SGA-2020 galaxy is less than 1.5 arcsec. Throughout this work we only include the data of star-forming galaxies, not including AGN or resolved Hii -regions of nearby galaxies. This final sample contains 49 959 star-forming galaxies. The redshift distribution of our sample is shown in Figure 2 . Figure 2: The redshift distribution of the galaxies with temperature measurements in our sample (grey). The vertical lines show the maximum redshift where the T_{\rm e} can be determined based on the listed ionic species. The coloured histogram outlines in matching colours show the measurement distribution of electron temperatures of each ion. 3 Fiducial model for nebular physical conditions and abundances Our analysis of the nebular physical conditions and abundances is based on forward modelling using the PyNeb emission line analysis software ( Luridiana et al. 2015 , version 1.1.25) . Based on atomic physics, PyNeb allows us to derive model emission line flux ratios based on a wide range of ionised gas conditions in star-forming regions. The atomic data used in our analysis are summarised in Table 2 . We use our forward model to infer the physical properties of the ionised gas that produces the emission lines observed in our spectra ( Scholte et al. 2025 ; Cullen et al. 2025 , see also) . We sample the parameter-space using the UltraNest nested sampling algorithm ( Buchner 2021 ) . In this section, we describe the assumptions in our fiducial model, including electron temperature relations and ionisation correction factors (ICFs) used throughout the analysis. We have made choices that reflect some of the most common assumptions in the current literature. Therefore, our results can be readily compared to other literature studies. In Section 4 , we discuss these assumptions and compare to alternatives in more detail. Table 2: Atomic data used in our analysis using PyNeb . Ion Transition probabilities ( A_{\rm ij} ) Collisional strengths ( Y_{\rm ij} ) H + Storey & Hummer 1995 — N + Froese Fischer & Tachiev 2004 Tayal 2011 O + Froese Fischer & Tachiev 2004 Kisielius et al. 2009 O ++ Froese Fischer & Tachiev 2004 Storey et al. 2014 Ne ++ Galavis et al. 1997 McLaughlin & Bell 2000 S + Rynkun et al. 2019 Tayal & Zatsarinny 2010 S ++ Froese Fischer et al. 2006 Tayal & Gupta 1999 Ar ++ Munoz Burgos et al. 2009 Munoz Burgos et al. 2009 Ar 3+ Mendoza 1983 , Kaufman & Sugar 1986 Ramsbottom & Bell 1997 Table 3: The observables, parameters and priors used in the pyneb analysis. The first step is to determine the electron density, temperature and dust attenuation, which are then used as priors in the second step to determine the ionic abundances. The priors are uniform distributions, \mathcal{U}(a,b) , where a and b are the lower and upper limits of the distribution. Observables Parameters Priors Step 1: electron density, temperature and dust attenuation \textsc{[Oii]}\lambda 3726/\textsc{[Oii]}\lambda 3729 n_{\rm e} \mathcal{U}(10,1000){~\rm cm^{-3}} \textsc{[Nii]}\lambda 5755/\textsc{[Nii]}\lambda 6584 T_{\rm e}(\rm N^{+}) \mathcal{U}(0.5,3.5)\times 10^{4}{~\rm K} \textsc{[Oii]}\lambda\lambda 7320,30/\textsc{[Oii]}\lambda\lambda 3726,29 T_{\rm e}(\rm O^{+}) \mathcal{U}(0.5,3.5)\times 10^{4}{~\rm K} \textsc{[Oiii]}\lambda 4363/\textsc{[Oiii]}\lambda 5007 T_{\rm e}(\rm O^{++}) \mathcal{U}(0.5,3.5)\times 10^{4}{~\rm K} \textsc{[Sii]}\lambda 4068/\textsc{[Sii]}\lambda\lambda 6716,31 T_{\rm e}(\rm S^{+}) \mathcal{U}(0.5,3.5)\times 10^{4}{~\rm K} \textsc{[Siii]}\lambda 6312/\textsc{[Siii]}\lambda 9069 T_{\rm e}(\rm S^{++}) \mathcal{U}(0.5,3.5)\times 10^{4}{~\rm K} {\rm H}\beta/{\rm H}\alpha A_{\rm V} \mathcal{U}(0.0,4.0){~\rm mag} {\rm H}\gamma/{\rm H}\beta Step 2: ionic abundances \textsc{[Nii]}\lambda 6584/{\rm H}\beta {\rm log(N^{+}/H^{+})} \mathcal{U}(-9,-2) dex + \begin{array}[]{c}{\rm Using~posterior~distributions}\\ {\rm determined~in~step~1}\\ \left(\begin{array}[]{c}n_{\rm e}\\ T_{\rm high}\\ T_{\rm mid}\\ T_{\rm low}\\ A_{\rm V}\end{array}\right)\end{array} \textsc{[Oii]}\lambda\lambda 3726,3729/{\rm H}\beta {\rm log(O^{+}/H^{+})} \mathcal{U}(-9,-2) dex \textsc{[Oiii]}\lambda 5007/{\rm H}\beta {\rm log(O^{2+}/H^{+})} \mathcal{U}(-9,-2) dex \textsc{[}{\rm Ne}\textsc{iii]}\lambda 3869/{\rm H}\beta {\rm log(Ne^{2+}/H^{+})} \mathcal{U}(-9,-2) dex \textsc{[Sii]}\lambda\lambda 6716,6731/{\rm H}\beta {\rm log(S^{+}/H^{+})} \mathcal{U}(-9,-2) dex \textsc{[Siii]}\lambda 9069/{\rm H}\beta {\rm log(S^{2+}/H^{+})} \mathcal{U}(-9,-2) dex \textsc{[}{\rm Ar}\textsc{iii]}\lambda 7135/{\rm H}\beta {\rm log(Ar^{2+}/H^{+})} \mathcal{U}(-9,-2) dex \textsc{[}{\rm Ar}\textsc{iv]}\lambda 4740/{\rm H}\beta {\rm log(Ar^{3+}/H^{+})} \mathcal{U}(-9,-2) dex 3.1 Nebular physical conditions The forward modelling analysis is split into two steps. In the first step, we jointly derive the nebular physical conditions: electron density, n_{\rm e} , electron temperature, T_{\rm e} , and dust attenuation, A_{\rm V} . In the top half of Table 3 we show the observable line ratios (column 1) used to constrain these parameters (column 2) together with the priors (column 3) placed on the physical parameters. We use non-informative, uniform priors for all parameters. The electron density is constrained using the [ Oii ] \lambda 3726/[ Oii ] \lambda 3729 doublet ratio which we choose over the [ Sii ] \lambda 6731/[ Sii ] \lambda 6716 ratio as the [ Sii ]-doublet is not in the observable wavelength range beyond z\sim 0.45 , whereas the [ Oii ]-doublet is available for our full sample. As shown in the bottom half of Table 3 , we have measurements of auroral lines from N + , O + , O ++ , S + and S ++ ions, whenever detected with sufficient \rm S/N . We only report electron temperatures where there is a \rm S/N>3 detection 2 2 2 Our sample selection requires the detection of at least one of the auroral lines in a spectrum at \rm S/N>5 , however, any subsequent auroral lines are still included in our analysis if \rm S/N>3 . The more stringent sample selection criterion is to remove false positive auroral line detections. However, when one auroral line has reliably been detected at \rm S/N>5 the chance of spurious subsequent auroral lines is very low, and therefore a lower threshold is justified. of the auroral line (e.g., [ Oiii ]4363) and a \rm S/N>3 detection of the accompanying strong emission line of the adjacent transition of the same species (e.g., [ Oiii ]5007 in case of the [ Oiii ]4363 auroral line). Dust attenuation is jointly constrained through the H \beta /H \alpha and H \gamma /H \beta line ratios using a Cardelli et al. 1989 attenuation prescription with R_{\rm V}=3.1 assuming case B recombination ( Baker & Menzel 1938 ) . 3.2 Abundances Once the physical conditions are constrained, we derive ionic abundances of the \rm N^{+} , \rm O^{+} , \rm O^{2+} , \rm Ne^{2+} , \rm S^{+} , \rm S^{2+} and \rm Ar^{2+} and \rm Ar^{3+} ions. We assume a three-zone temperature structure of the ionised gas. Each ion occupies a ionisation zone: \rm O^{2+} , \rm Ne^{2+} and \rm Ar^{3+} are in the high ionisation zone ( T_{\rm high} ); \rm S^{2+} and \rm Ar^{2+} are in the intermediate ionisation zone ( T_{\rm mid} ); \rm N^{+} , \rm O^{+} and \rm S^{+} are in the low ionisation zone ( T_{\rm low} ) ( Mingozzi et al. 2022 ; Berg et al. 2022 ) . The temperature of each zone can be constrained directly if the auroral line ratio of an ion in this zone is detected. Therefore, in this work T_{\rm high} is constrained by T_{\rm e}(\rm O^{++}) and T_{\rm mid} by T_{\rm e}(\rm S^{++}) . In the low ionisation zone we have three possible constraints. We assume T_{\rm low} is given by T_{\rm e}(\rm O^{+}) as this line ratio is detected for the majority of observations, otherwise we use T_{\rm e}(\rm S^{+}) or T_{\rm e}(\rm N^{+}) . If the temperature in an ionisation zone is not constrained, then we infer the temperature from a different zone using electron temperature relations ( T_{\rm e}-T_{\rm e} ). We assume the temperature relation defined by Campbell et al. 1986 and Garnett 1992 based on H ii -region models by Stasińska 1982 which provides a relation between the low- and high-ionisation zones: T_{\rm low}=0.7\times T_{\rm high}+3000\rm K . We assume the relation defined by Croxall et al. 2016 for the relation between the intermediate- and high-ionisation zones: T_{\rm mid}=1.265\times T_{\rm high}-2320\rm K . A more detailed analysis of T_{\rm e}-T_{\rm e} relations is provided in Section 4.1 . We constrain the ionic abundances of elements using the ratios of the respective strong emission line strength compared to H \beta combined with the posterior distributions (derived in step 1) on the electron density, electron temperature in each zone and dust attenuation. The line ratios used to constrain the abundance of each ion are listed in Table 3 . We only report ionic abundances where there is a reliable temperature measurement (based on the S/N criteria of the relevant emission lines) and the strong emission lines of the ion are detected at \rm S/N>3 . The total oxygen abundances are derived using the sum of the abundances of the singly ( \rm O^{+}/H^{+} ) and doubly ( \rm O^{++}/H^{+} ) ionised states. The contribution of higher ionised states is considered to be negligible, even at high electron temperatures the contribution of this state is small ( Berg et al. 2021 ; Rickards Vaught et al. 2025 ; Cullen et al. 2025 , e.g.,) . To calculate the total abundances of each other element we use the ionisation correction factors by Izotov et al. 2006 . These ionisation correction factors allow us to calculate the total abundances from measurements of an incomplete subset of ionic abundances. For N and Ne we only measure one ionic abundance each ( \rm N^{+}/H^{+} and \rm Ne^{++}/H^{+} , respectively). Therefore, the total abundances are estimated only from these single ionic abundances (equations 18 and 19 in Izotov et al. 2006 , respectively) . For S we measure both the singly ( \rm S^{+}/H^{+} ) and doubly ( \rm S^{++}/H^{+} ) ionised state; we only calculate the total S abundances when both of these ionic abundances are constrained ( Izotov et al. 2006 , equation 20 in) . For Ar we measure the doubly ( \rm Ar^{++}/H^{+} ) and triply ( \rm Ar^{3+}/H^{+} ) ionised states. We calculate the total Ar abundance using both ionised states where each is constrained ( Izotov et al. 2006 , equation 23 in) ; however, if \rm Ar^{3+}/H^{+} is not constrained, we calculate the total Ar abundance from just the \rm Ar^{++}/H^{+} ionic abundances ( Izotov et al. 2006 , equation 22 in) . We discuss the validity of the chosen ICFs in more detail in Section 4.2 . The total number of galaxies with abundance measurement of each ion/element is listed in Table 4 . Nitrogen, oxygen, Neon and Argon abundances are each available for the majority of galaxies with an electron temperature measurement. The sulfur abundances are only available for a smaller subset as the [ Siii ] \lambda 9069 emission line is only observable in DESI spectra at z\lesssim 0.08 . 3.3 Star formation rates We derive the star formation rates from the aperture corrected and dust corrected H \alpha flux (or H \beta flux if H \alpha is not available). We use the aperture corrections derived using FastSpecFit in Section 2 . Our star-formation rates are calculated using the metallicity-dependent calibration by Korhonen Cuestas et al. 2025 which is defined as \log\left(\frac{\dot{M}_{\star}}{{\rm M}_{\odot}~{\rm yr}^{-1}}\right)=\log\left(\frac{\mathcal{L}_{\rm H\alpha}}{\textrm{erg~s}^{-1}}\right)-\log(C(Z_{\star}/{\rm Z}_{\odot})), (1) where \mathcal{L}_{\rm H\alpha} is the dust corrected H \alpha luminosity and C(Z_{\star}/{\rm Z}_{\odot}) is the metallicity-dependent conversion factor. The conversion factor is defined as outlined in section 2.4 of Korhonen Cuestas et al. 2025 . We infer the stellar metallicity, Z_{\star}/{\rm Z_{\odot}} , from the gas phase oxygen abundances where we assume \log(Z_{\star}/{\rm Z_{\odot}})=\rm\log(O/H)-\log(O/H)_{\odot} . This assumption implies that there is no enhancement of alpha-elements as the stellar metallicity is defined in terms of iron abundances. At low-metallicity, the luminosity to SFR conversion factor of this calibration is lower than a typical conversion factor not accounting for metallicity ( Kennicutt 1998 ; Hao et al. 2011 ) . 4 Results and discussion In this work we present a large sample of electron temperature measurements and abundance measurements of N, O, Ne, S, Ar in star-forming galaxies. In this section, we present our results, and discuss implications and connections to previous work. The DESI DR2 dataset provides an unprecedented number of auroral line detections in star-forming galaxies. Due to this large sample size we are not only able to derive the average relations between electron temperatures and abundances, but also to explore the scatter and trends in more detail. The large sample size means we are more likely to detect outlier populations with rare properties. In this work we focus on analysing the electron temperature relations and provide a cursory overview of our abundance measurements. A more detailed analysis of the abundance measurements and their implications for the chemical evolution of galaxies will be presented in a follow-up paper (Scholte et al. in prep.). The total number of galaxies for which each measurement is available is summarised in Table 4 . 4.1 Electron temperature measurements and temperature relations We derive electron temperature measurements using the emission lines of 5 different ions with varying ionisation potentials. Figure 3 shows the histograms of the electron temperature measurements derived from each of these ions. The [ Oiii ] \lambda 4363 auroral line (corresponding to T_{\rm e}\rm(O^{++}) ) is brightest at high temperatures, the [ Siii ] \lambda 6312 emission line (corresponding to T_{\rm e}\rm(S^{++}) ) is brightest at intermediate temperatures and [ Nii ] \lambda 5755, [ Oii ] \lambda\lambda 7320,7330 and [ Sii ] \lambda\lambda 4068,4076 (corresponding to T_{\rm e}\rm(N^{+}) , T_{\rm e}\rm(O^{+}) and T_{\rm e}\rm(S^{+}) , respectively) at low temperatures. Due to this fact, these auroral emission lines are detected in different subsets of galaxies. Additionally, due to temperature inhomogeneities in/between star-forming regions of galaxies, the temperatures measured using these different probes deviate when multiple auroral lines are detected in a galaxy. These effects are reflected in the median electron temperatures measured using each auroral line (red dashed lines in Fig. 3 ). The histograms also show that, whilst our models allow electron temperatures up to 35,000 K, electron temperatures above \sim 25,000 K are extremely rare in star-forming galaxies ( \sim 0.05\% in our sample). Figure 3: The histograms of the electron temperature measurements derived from the auroral emission lines of different ions. The red dashed lines show the median electron temperature measured using each auroral line. The [ Oiii ] \lambda 4363 auroral line (corresponding to T_{\rm e}\rm(O^{++}) ) is brightest at high temperatures, the [ Siii ] \lambda 6312 emission line (correcponding to T_{\rm e}\rm(S^{++}) ) is brightest at intermediate temperatures and T_{\rm e}\rm(O^{+}) , T_{\rm e}\rm(N^{+}) and T_{\rm e}\rm(S^{+}) are brightest at low temperatures. The number of objects with temperature constraints using each ion are indicated in the top-right corner of each panel. We investigate the relations between the electron temperature measurements of different probes when multiple temperature measurements are available for any given galaxy. We fit temperature relations using orthogonal distance regression as implemented in Scipy ( Boggs et al. 1981 ; Virtanen et al. 2020 , ODR;) . We include all temperature measurements of star-forming galaxies where the respective auroral and strong emission lines are detected at \rm S/N>3 and where T_{\rm e}^{84}-T_{\rm e}^{16}<5000\rm~K (where T_{\rm e}^{x} is the x {}^{\rm th}~ percentile of the posterior distribution of T_{\rm e} ). The uncertainty in the T_{\rm e} measurements are included as the weights, w , in the ODR fitting, where w=((T_{\rm e}^{84}-T_{\rm e}^{16})/2)^{-2} . We measure the scatter around the fitted relations as \sigma_{\rm tot}=(r^{84}-r^{16})/2 for residuals, r . The intrinsic scatter is measured as \sigma_{\rm int}=\sqrt{\sigma_{\rm tot}^{2}-\widetilde{\sigma}_{\rm meas}^{2}} , where \widetilde{\sigma}_{\rm meas} is the median propagated measurement uncertainty in the residuals. Table 4: Summary of electron temperature and abundance measurements from the PyNeb analysis. The table lists the parameters and the number of galaxies for which each parameter was measured. † The total number of galaxies for which at least one electron temperature was measured. Electron temperature measurements Parameter Number of galaxies T_{\rm e}\rm(N^{+}) 340 T_{\rm e}\rm(O^{+}) 33 521 T_{\rm e}\rm(O^{++}) 26 228 T_{\rm e}\rm(S^{+}) 421 T_{\rm e}\rm(S^{++}) 5 836 T_{\rm e} 49 959 † Abundance measurements Parameter Number of galaxies \rm log(N^{+}/H^{+}) 42 434 \rm log(N/H) 42 410 \rm log(O^{+}/H^{+}) 49 768 \rm log(O^{++}/H^{+}) 49 957 \rm log(O/H) 49 766 \rm log(Ne^{++}/H^{+}) 47 939 \rm log(Ne/H) 47 747 \rm log(S^{+}/H^{+}) 41 401 \rm log(S^{++}/H^{+}) 15 892 \rm log(S/H) 15 698 \rm log(Ar^{++}/H^{+}) 39 982 \rm log(Ar^{3+}/H^{+}) 1 260 \rm log(Ar/H) 39 933 Figure 4: The electron temperature relation between the high-ionisation zone, T_{\rm e}(\mathrm{O}^{++}) , and the intermediate-ionisation zone, T_{\rm e}(\mathrm{S}^{++}) . The red line shows the best fit relation derived from our observations. The shaded region shows the 1 \sigma uncertainty in the relation. The total scatter in the relation is \sigma_{\rm tot}(\mathrm{S}^{++})=1700~\mathrm{K} and the intrinsic scatter is \sigma_{\rm int}(\mathrm{S}^{++})=900~\mathrm{K} . The total scatter in T_{\rm e}(\mathrm{O}^{++}) is \sigma_{\rm tot}(\mathrm{O}^{++})=1600~\mathrm{K} and with intrinsic scatter \sigma_{\rm int}(\mathrm{O}^{++})=900~\mathrm{K} . The literature relations by Izotov et al. 2006 , Croxall et al. 2016 and Rogers et al. 2021 and Méndez-Delgado et al. 2023 are shown for comparison as indicated in the legend. All data points are coloured by the Gaussian kernel density of the plotted distribution. Figure 5: The electron temperature relation between the high-ionisation zone, T_{\rm e}(\mathrm{O}^{++}) , and the low-ionisation zone, T_{\rm e}(\mathrm{O}^{+}) . The red line shows the best fit relation derived from our observations. The shaded region shows the 1 \sigma uncertainty in the relation. The total scatter in the relation is \sigma_{\rm tot}(\mathrm{O}^{+})=2100~\mathrm{K} and the intrinsic scatter is \sigma_{\rm int}(\mathrm{O}^{+})=1700~\mathrm{K} . The total scatter in T_{\rm e}(\mathrm{O}^{++}) is smaller at \sigma_{\rm tot}(\mathrm{O}^{++})=1400~\mathrm{K} and with intrinsic scatter \sigma_{\rm int}(\mathrm{O}^{++})=1100~\mathrm{K} . The literature relations by Campbell et al. 1986 , Izotov et al. 2006 , Andrews & Martini 2013 and Cataldi et al. 2025 are shown for comparison as indicated in the legend. All data points are coloured by the Gaussian kernel density of the plotted distribution. Figure 6: The electron temperature relation between the intermediate-ionisation zone, T_{\rm e}(\mathrm{S}^{++}) , and the low-ionisation zone, T_{\rm e}(\mathrm{O}^{+}) . The red line shows the best fit relation derived from our observations. The orange line is our fiducial model based on the Campbell et al. 1986 and Croxall et al. 2016 relations. The shaded region shows the 1 \sigma uncertainty in the relation. The total scatter in the relation is \sigma_{\rm tot}(\mathrm{O}^{+})=1400~\mathrm{K} and the intrinsic scatter is \sigma_{\rm int}(\mathrm{O}^{+})=1000~\mathrm{K} . The total scatter in T_{\rm e}(\mathrm{S}^{++}) is larger at \sigma_{\rm tot}(\mathrm{S}^{++})=2100~\mathrm{K} and with intrinsic scatter \sigma_{\rm int}(\mathrm{S}^{++})=1600~\mathrm{K} . All data points are coloured by the Gaussian kernel density of the plotted distribution. 4.1.1 The T_{\rm e}(\rm O^{++}) versus T_{\rm e}(\rm S^{++}) relation The constraints on the temperatures in the high- and intermediate-ionisation zones are constrained by the T_{\rm e}(\rm O^{++}) and T_{\rm e}(\rm S^{++}) measurements, respectively. In Figure 4 , we show the measured electron temperatures and the relation derived from our observations, which is given by T_{\rm e}({\rm S^{++}})=(1.062\pm 0.022)\times T_{\rm e}({\rm O^{++}})-(0.06\pm 0.03)\times 10^{4}~\rm K. (2) The total scatter in the relation is \sigma_{\rm tot}(\rm S^{++})=1700~\rm K and the intrinsic scatter is \sigma_{\rm int}(\rm S^{++})=900~\rm K , where the residual is defined as T_{\rm e}(\rm S^{++})_{\rm obs}-T_{\rm e}(\rm S^{++})_{\rm inferred} . The total scatter in T_{\rm e}(\rm O^{++}) is \sigma_{\rm tot}(\rm O^{++})=1600~\rm K and with intrinsic scatter \sigma_{\rm int}(\rm O^{++})=900~\rm K . The average temperature relation we derive is very close to a one-to-one relation, showing that the \rm O^{++} and \rm S^{++} ions probe similar ionisation zones within the ionised gas in individual galaxies. Our electron temperature relation is less steeply increasing than the relations derived by Croxall et al. 2016 and Rogers et al. 2021 . Simultaneously, our T_{\rm e}({\rm S^{++}}) are slightly higher at fixed T_{\rm e}({\rm{O++}}) than expected from the relations by Izotov et al. 2006 , based on photoionisation models, and Méndez-Delgado et al. 2023 , based on individual H ii -regions. The differences are smallest in comparison to the relation derived by Croxall et al. 2016 which we used in our fiducial model and the relation derived by Méndez-Delgado et al. 2023 . Some differences we observe might also be due to the fact that the relations by Croxall et al. 2016 , Rogers et al. 2021 and Méndez-Delgado et al. 2023 are derived based on individual H ii -regions and our measurements are of integrated galaxy spectra. The intrinsic scatter in our temperature relations is significantly larger than the scatter in the relations derived through individual H ii -region measurements. This is most likely due to the integrated galaxy measurements, which leads to the blending of line emission from multiple H ii -regions with a range of physical properties. 4.1.2 The T_{\rm e}(\rm O^{++}) versus T_{\rm e}(\rm O^{+}) relation The relation between T_{\rm e}(\rm O^{++}) and T_{\rm e}(\rm O^{+}) has been widely reported in previous literature ( Campbell et al. 1986 ; Garnett 1992 ; Izotov et al. 2006 ; Andrews & Martini 2013 ; Yates et al. 2020 ; Cataldi et al. 2025 , e.g.,) . In Figure 5 we show the measured electron temperatures and the relation derived from our observations, which is given by T_{\rm e}({\rm O^{+}})=(0.648\pm 0.015)\times T_{\rm e}({\rm O^{++}})+(0.327\pm 0.018)\times 10^{4}~\rm K, (3) with a total scatter of \sigma_{\rm tot}(\rm O^{+})=2100~\rm K and an intrinsic scatter of \sigma_{\rm int}(\rm O^{+})=1700~\rm K . The total scatter in T_{\rm e}(\rm O^{++}) is smaller at \sigma_{\rm tot}(\rm O^{++})=1400~\rm K and with intrinsic scatter \sigma_{\rm int}(\rm O^{++})=1100~\rm K . Our relation differs slightly from the relation derived by Campbell et al. 1986 and Garnett 1992 based on models by Stasińska 1982 , which we used in our fiducial model. At fixed T_{\rm e}(\rm O^{++}) , our relation predicts a slightly lower T_{\rm e}(\rm O^{+}) . However, this difference is smaller than the typical scatter in the relation. Our average relation is also closely aligned with the relation derived by Cataldi et al. 2025 , derived using integrated galaxy spectra. This large scatter is due to a combination of factors such as the different regions of ionised gas traced by these ions due to different ionisation potentials but also possible contamination such as the [ Oii ] \lambda\lambda 7320,7330 measurements due to telluric emission lines. Additionally, temperature fluctuations and gradients within H ii regions and the presence of emission from multiple star-forming regions with different temperatures can further increase the observed dispersion (see e.g., Peimbert 1967 ; Stasińska 2005 ; Bresolin 2008 , for detailed discussion on temperature measurements and also the resulting effect on abundance measurements) . These effects highlight the challenges in using a single temperature relation to infer T_{\rm e}(\rm O^{++}) from T_{\rm e}(\rm O^{+}) (and vice versa) and therefore, this emphasizes the importance of direct measurements of individual ionisation zones whenever possible. 4.1.3 The T_{\rm e}(\rm S^{++}) versus T_{\rm e}(\rm O^{+}) relation The relation between T_{\rm e}(\rm S^{++}) and T_{\rm e}(\rm O^{+}) is less well studied in the literature. In Figure 6 we show the measured electron temperatures and the relation derived from our observations, which is given by T_{\rm e}({\rm O^{+}})=(0.654\pm 0.017)\times T_{\rm e}({\rm S^{++}})+(0.326\pm 0.018)\times 10^{4}~\rm K. (4) The total scatter in T_{\rm e}(\rm O^{+}) -residuals is \sigma_{\rm tot}(\rm O^{+})=1400~\rm K and the intrinsic scatter is \sigma_{\rm int}(\rm O^{+})=1000~\rm K , where the residual is defined as T_{\rm e}(\rm O^{+})_{\rm obs}-T_{\rm e}(\rm O^{+})_{\rm inferred} . The total scatter in T_{\rm e}(\rm S^{++}) is \sigma_{\rm tot}(\rm S^{++})=2100~\rm K and with intrinsic scatter \sigma_{\rm int}(\rm S^{++})=1600~\rm K . The relation we derive is slightly steeper than our fiducial model derived using the combination of the Campbell et al. 1986 and Garnett 1992 and Croxall et al. 2016 relations (see Fig. 6 ). However, the differences are small in comparison to the typical scatter in the relation. Figure 7: A compilation of the electron temperature relations between measurements from different ions. The red lines show the best fit relations derived from our observations. The shaded regions show the 1 \sigma uncertainties in the relations. All data points are coloured by the Gaussian kernel density of the plotted distribution. 4.1.4 Other temperature relations The remaining combinations of electron temperature measurements comprise significantly smaller samples. Therefore we do not discuss them in detail, however, the derived relations are provided below: \begin{split}&T_{\rm e}({\rm N^{+}})=(1.1\pm 0.4)\times T_{\rm e}({\rm O^{++}})-(0.0\pm 0.4)\times 10^{4}~\rm K,\\ &T_{\rm e}({\rm N^{+}})=(1.11\pm 0.16)\times T_{\rm e}({\rm S^{++}})+(0.04\pm 0.13)\times 10^{4}~\rm K,\\ &T_{\rm e}({\rm N^{+}})=(1.01\pm 0.21)\times T_{\rm e}({\rm O^{+}})+(0.08\pm 0.18)\times 10^{4}~\rm K,\\ &T_{\rm e}({\rm S^{+}})=(1.01\pm 0.11)\times T_{\rm e}({\rm O^{++}})-(0.04\pm 0.11)\times 10^{4}~\rm K,\\ &T_{\rm e}({\rm S^{+}})=(0.75\pm 0.08)\times T_{\rm e}({\rm S^{++}})+(0.21\pm 0.08)\times 10^{4}~\rm K,\\ &T_{\rm e}({\rm S^{+}})=(0.79\pm 0.10)\times T_{\rm e}({\rm O^{+}})-(0.18\pm 0.10)\times 10^{4}~\rm K,\\ &T_{\rm e}({\rm S^{+}})=(1.02\pm 0.22)\times T_{\rm e}({\rm N^{+}})-(0.15\pm 0.22)\times 10^{4}~\rm K.\end{split} (5) The fitted relations are also shown in Figure 7 . We note that the T_{\rm e}({\rm N^{+}}) temperatures of a significant fraction of galaxies is higher than expected assuming \rm N^{+} and \rm O^{+} trace the same gas conditions. This is visible in the number of high T_{\rm e}({\rm N^{+}}) outliers in the relevant panels of Figure 7 . This is a similar finding to, e.g. Arellano-Córdova et al. 2024 where systematically high temperatures of \rm N^{+} were also found. The T_{\rm e}({\rm S^{+}}) measurements are similar to T_{\rm e}({\rm O^{+}}) , as is expected due to the similar ionisation potential of these ions. We note that particularly in these diagrams with smaller samples, the fitted relations do not always pass through the highest density region of the data. This is due to the fact that the ODR fitting includes the uncertainties in the measurements, where more uncertain measurements are given less weight in the line fitting. Figure 8: The ionic abundance ratios as a function of the degree of ionisation, defined as \rm O^{++}/O , shown together with the relations for ionisation correction factors by Izotov et al. 2006 , Pérez-Montero et al. 2007 , Dors et al. 2013 and Amayo et al. 2021 as indicated by the legend. Due to systematic evolution of the \rm N/O ratio the data is not expected to coincide with the plotted ICFs in the top-left panel as those plotted lines are based on a constant N/O. The ICF functions are scaled to the median total abundance ratios of the plotted data using the Izotov et al. 2006 ICFs as described in Section 4.3 . All data points are coloured by the Gaussian kernel density of the plotted distribution. Typical uncertainties are shown by the black error bars in the top-right corner of each panel. 4.2 Ionisation correction factors We derive ionic abundances of the \rm N^{+} , \rm O^{+} , \rm O^{++} , \rm Ne^{++} , \rm S^{+} , \rm S^{++} , \rm Ar^{++} and \rm Ar^{3+} ions. To derive total elemental abundances we use ionisation correction factors to account for unobserved ionisation states. In Figure 8 we show the ionic abundance ratios as a function of the degree of ionisation, defined as \rm O^{++}/(O^{+}+O^{++}) . We compare our measurements to the ICFs by Izotov et al. 2006 which we use in our fiducial model. The ICFs by Izotov et al. 2006 are based on photoionisation models and are widely used in the literature. We also compare to the ICFs derived by Pérez-Montero et al. 2007 , Dors et al. 2013 and Amayo et al. 2021 . Due to systematic evolution of the \rm N/O ratio the data is not expected to coincide with the plotted ICFs in the top-left panel as those plotted lines are based on a constant N/O. Typically, different ICFs are in close agreement with each other for N, S and Ar. Some of the largest deviations are observed for Ne at low degrees of ionisation where the different ICFs predict different contributions of unobserved Ne + . Our observations generally cover the full range covered by the different ICFs, however, the high-metallicity ICF by Izotov et al. 2006 provides a good match to the average of our measurements at low \rm O^{++}/(O^{+}+O^{++}) , where most of the galaxies in this regime are in the high-metallicity selection of this ICF. We find that the total argon abundances are on average 0.13_{-0.09}^{+0.15} dex higher when the the \rm Ar^{3+} abundances are included. The \rm Ar^{3+} abundances are only constrained for a small subset of 1 259 out of 39 917 galaxies with argon abundance measurements. The [Ar iv ] \lambda 4740 emission line is almost exclusively detected in galaxies with a high degree of ionisation (high \rm O^{++}/O ) where a larger fraction of the Ar is in the Ar 3+ ionised state. 4.3 Elemental abundances In this section we present our total abundance measurements of N, O, Ne, S and Ar. Throughout this section we only present the abundances of galaxies where the \rm log(X/H)^{84}-log(X/H)^{16}<0.6~dex . See Table 4 for a summary of the number of abundance measurements of each element. Figure 9: Oxygen abundances as a function of electron temperature of the high ionisation zone. We show T_{\rm e}(\rm O^{++}) (purple squares), for objects where this is measured, and show the converted values of T_{\rm e}(\rm S^{++}) (teal diamonds) or T_{\rm e}(\rm O^{+}) , T_{\rm e}(\rm N^{+}) or T_{\rm e}(\rm S^{+}) (lime-green circles) using the fiducial temperature relations if the preceding are not available. The data points are shaded by the Gaussian kernel density of the plotted distributions. Typical uncertainties are shown by the coloured error bars in the bottom-left corner of each panel. We also highlight the two lowest metallicity galaxies in our sample, DESI J169.0571+14.1514 and DESI J211.9086+28.2461 (purple stars). For a detailed analysis of the lowest-metallicity galaxies in this sample we refer to Moustakas et al. (in prep.). The inset panel shows a comparison of our measurements to the direct abundance measurements of galaxies in the CLASSY survey ( Berg et al. 2022 ) and measurements of DESI extremely metal poor galaxies (EMPGs) and EMPG candidates in DESI early data by Zou et al. 2024 . 4.3.1 Oxygen abundances Oxygen is the most abundant metal in the ISM of galaxies and its abundance is typically used to quantify the gas-phase metallicity of galaxies. In Figure 9 we show the oxygen abundance as a function of electron temperature of the high ionisation zone. There is a strong inverse relation between the oxygen abundance and electron temperature. This is mainly due to the increased efficiency of gas cooling in more metal rich gas. We show T_{\rm e}(\rm O^{++}) , for objects where this is measured, and show the converted values of T_{\rm e}(\rm S^{++}) , T_{\rm e}(\rm O^{+}) , T_{\rm e}(\rm N^{+}) or T_{\rm e}(\rm S^{+}) using the fiducial temperature relations if the preceding are not available. Our measurements cover a wide range of oxygen abundances, from the lowest metallicity galaxies in the local Universe to high-metallicity galaxies with 99% of our measurements between \rm 7.42<12+log(O/H)<8.85 dex. We generally find good agreement with literature measurements as shown in the inset panel in Figure 9 . We compare to the direct abundance measurements of galaxies in the CLASSY survey ( Berg et al. 2022 ) as well as measurements of DESI extremely metal poor galaxies (EMPGs) and EMPG candidates in DESI early data by Zou et al. 2024 . These comparisons show that our measurements are consistent with previous literature. Due to the large number of galaxies observed by DESI, and the increased survey depth in comparison to e.g., the SDSS Main Galaxy Survey ( York et al. 2000 ) , we are able to detect a much larger number of galaxies with measurable auroral lines, and therefore direct-metallicity constraints, than has previously been possible. An example of this is the number of EMPGs detected in our dataset. Our sample contains 3 114 galaxies with metallicities below 10% solar, which is a significant increase in the known sample of EMPGs in the nearby Universe (see e.g., Sánchez Almeida et al. 2016 ; Zou et al. 2024 , for a detailed study of SDSS and DESI EMPGs) . This new sample will make more detailed studies of the chemical evolution of galaxies in pristine environments possible. This is important in particular to improve our understanding of the chemical evolution of galaxies in the early Universe, where the gas is known to be more metal-poor than in the local Universe on average ( Sanders et al. 2021 ; Jain et al. 2025 ; Stanton et al. 2025a ; Zahid et al. 2011 ) . We discover two galaxies with metallicities lower than any previous confirmed direct measurements to date. Our lowest metallicity galaxy, DESI J211.9086+28.2461, has an oxygen abundance of \rm 12+\log(O/H)=6.77_{-0.03}^{+0.03}~\rm dex or \sim 1.2\%~\rm Z_{\odot} . We also highlight DESI J169.0571+14.1514, for which only the \rm O^{++} abundance is measured due to a non-detection of the [ Oii ] \lambda\lambda 3726,3729 doublet, suggesting a negligible fraction of the oxygen abundance in the \rm O^{+} ionic state. Based on \rm O^{++} , we find an oxygen abundance of \rm 12+\log(O/H)=6.81_{-0.04}^{+0.04}~\rm dex or \sim 1.3\%~\rm Z_{\odot} . The metallicities are lower than previous extremely metal-poor record holders with direct-metallicity measurements such as HSC J1631+4426 at \rm 12+\log(O/H)=6.90\pm 0.03~\rm dex ( Kojima et al. 2020 ) in the local Universe and EXCELS-63107 at \rm 12+\log(O/H)=6.89_{-0.21}^{+0.26}~\rm dex ( Cullen et al. 2025 ) in the early Universe ( Vanzella et al. 2025 ; Cai et al. 2025 ; Hsiao et al. 2025 , there are some more metal-poor candidates at high-redshift, however, not confirmed by direct measurements due to the extreme observational challenge;) . The detection of these galaxies highlights the power of the DESI survey to detect rare objects with extreme properties. For a more detailed analysis of the lowest-metallicity galaxies in this sample we refer to Moustakas et al. (in prep.). Whilst the increased depth of DESI has resulted in a significant increase in the number of galaxies for which direct metallicities can be derived, it is still the case that the auroral lines are only detected in a small fraction of individual galaxy spectra. Due to this the sample of galaxies included here is not a representative sample of the galaxy population. Instead, the sample is biased towards strongly star forming, bright galaxies. These biases are reflected in the scaling relations of the galaxies in this sample, e.g., the mass-metallicity relation in the left panel of Fig. 11 is biased towards systematically lower metallicities as compared to more representative samples ( Scholte et al. 2024 , see e.g.,) . In the right panel of Fig. 11 , we also show that the galaxies in our sample have systematically higher star formation rates than typical z=0 galaxies; the average star formation rates are more similar to high redshift galaxies ( Popesso et al. 2023 , see e.g.,) . Figure 10: The abundance ratios, \rm log(X/O) , versus metallicity for nitrogen (top-left), neon (top-right), sulphur (bottom-left) and argon (bottom-right). The blue lines shows literature relations derived by Izotov et al. 2006 , Andrews & Martini 2013 , Nicholls et al. 2017 and Berg et al. 2012 ; Berg et al. 2019 . The orange lines show the predicted relations from chemical evolution models for the Solar neighbourhood, Milky Way halo and bulge stars by Kobayashi et al. 2020 . The cyan \odot -sign shows the solar abundance ratio ( Asplund et al. 2021 ) . The pink datapoints in the top-left panel show a selection of galaxies with low metallicity and high \rm N/O ratios as discussed in the text. All data points are coloured by the Gaussian kernel density of the plotted distribution. Typical uncertainties are shown by the black error bars in the top-left corner of each panel. 4.3.2 Nitrogen abundances and ratios Contrary to oxygen, which is mostly produced in massive stars, the main production channel of nitrogen in galaxies is through intermediate mass stars ( Kobayashi et al. 2020 , e.g.,) . Due to these different production channels there is a large scatter in the observed nitrogen to oxygen abundance ratio, \rm N/O , in star-forming galaxies. In the top-left panel of Figure 10 we show the \rm N/O ratio as a function of oxygen abundance. We also note that the dynamic range is twice as large in the y-axis for the \rm N/O panel than any of the other panels. Similar to previous literature, we fit a broken linear relation to the \rm log(O/H) versus \rm log(N/O) relation. We use the orthogonal distance regression as implemented in Scipy ( Boggs et al. 1981 ; Virtanen et al. 2020 ) to fit the relation. The uncertainty in the \rm O/H and \rm N/O measurements are included as the weights, w , in the ODR fitting, where w_{x}=(\log(\rm O/H)^{84}-\log(\rm O/H)^{16})/2)^{-2} and w_{y}=(\log(\rm N/O)^{84}-\log(\rm N/O)^{16})/2)^{-2} . The fitted relation is given by \log({\rm N/O})=\begin{cases}c_{\rm plateau},&\text{if }x8.0 , however, at lower metallicity we get better agreement with the relation of Berg et al. 2019 who find \rm\log(N/O)=-1.41\pm 0.09 dex at low-metallicity, where we find \rm\log(N/O)=-1.391\pm 0.003 dex. This value is also in agreement with the \rm\log(N/O) ratio of blue compact galaxies derived by Izotov & Thuan 1999 at -1.46\pm 0.14 dex. Generally, there is still some disagreement on the \rm\log(N/O) plateau at low metallicities, with some studies finding a plateau at \rm\log(N/O)\sim-1.4 dex ( Izotov & Thuan 1999 ; Berg et al. 2019 , e.g.,) and others finding a plateau at lower values of \rm log(N/O)\sim-1.6 dex ( Berg et al. 2012 ; Nicholls et al. 2017 , e.g.,) . The higher plateau of our \rm N/O ratio at lower metallicities could be due to sample incompleteness due to the faintness of the [ Nii ] \lambda 6584 line at low metallicities, particularly for galaxies with low \rm N/O ratios. However, as the higher plateau is also detected in the stacks of galaxy spectra from Andrews & Martini 2013 it may be a real feature of the galaxy population. There are also differences in the literature about the slope of the O/H versus N/O relation in the high metallicity regime as well as the transition point where N/O starts to increase, e.g., at 8.142\pm 0.014 dex in this work compared to \sim 8.50 dex in Andrews & Martini 2013 . These differences are most likely due to the dependencies of other parameters such as the stellar mass and SFR of the galaxies in the selected sample ( Boardman et al. 2024a ; Boardman et al. 2024b ; Boardman et al. 2025 , see also) . In particular, the comparison to Andrews & Martini 2013 highlights the effect of sample selection. The detection requirement of auroral emission lines in individual galaxies means that our sample contains more strongly star forming galaxies (as shown in the right panel of Fig. 11 ), whereas Andrews & Martini 2013 are sensitive to a more representative sample of galaxies as they use stacked measurements to detect auroral emission lines. Therefore, at fixed metallicity their sample will include more faint, low SFR and low mass galaxies which have systematically lower N/O (see middle panel of Fig. 11 in this work, or Fig. 14 in Andrews & Martini 2013 ). The chemical evolution models reveal that the \rm N/O ratio is strongly dependent on the modelled region in the MW chemical evolution models of Kobayashi et al. 2020 . This is a reflection of the different nucleosynthesis processes and distribution timescales into the ISM of these elements. The observed scatter in the \rm N/O ratio at fixed metallicity is likely due to variations in the star formation history of galaxies. More detailed modelling of the chemical evolution is able to reproduce the observed relation based on these processes, see e.g., Vincenzo & Kobayashi 2018a and Vincenzo & Kobayashi 2018b . We show the mass-metallicity relation (MZR), mass- \rm N/O relation and star formation main sequence in Figure 11 . Similar to other studies we find a strong relation between the stellar mass and the oxygen abundance, with more massive galaxies having higher oxygen abundances ( Tremonti et al. 2004 ; Berg et al. 2012 ; Andrews & Martini 2013 , see e.g.,) . We find that at fixed stellar mass our sample has systematically lower metallicities than the general population of galaxies with strong emission line detections ( Scholte et al. 2024 , as shown by the MZR of DESI DR1 galaxies derived using strong line calibrations that are calibrated to the direct metallicity scale;) , which is expected from the selection criteria which preferentially select more strongly star-forming galaxies where auroral emission lines are detected. The \rm N/O ratio is also strongly correlated with the stellar mass. Our sample shows a similar trend to the relations derived by Andrews & Martini 2013 and Hayden-Pawson et al. 2022 . There is significant interest in nitrogen-rich galaxies due to the detection of high \rm\log(N/O) galaxies in the early Universe such as GN-z11 with \rm log(N/O)>-0.25 dex ( Cameron et al. 2023 ) , CEERS-1018 with \rm log(N/O)=-0.18\pm 0.11 dex ( Marques-Chaves et al. 2024 ) and RXCJ2248-ID with \rm log(N/O)=-0.39_{-0.08}^{+0.10} dex ( Topping et al. 2024 ) . Our sample contains a significant number of galaxies with high \rm\log(N/O) ratios at low \rm\log(O/H) . We selected a sample of 139 outliers at \rm 12+\log(O/H)<8.0 and \rm\log(N/O)>-0.75 as shown by the pink datapoints and outlined parameter space in the top-left panel of Figure 10 . The high metallicity selection boundary is chosen to exclude the metallicity regime where high N/O is common. We choose the N/O threshold to select galaxies with N/O similar to extreme nitrogen emitters observed in the early Universe. We find that the high N/O galaxies selected are significantly more massive than the average population of galaxies at similar metallicity (see Figure 11 ). This is consistent with findings by Arellano-Córdova et al. 2025 who find that high \rm\log(N/O) galaxies are more massive than typical low-metallicity galaxies. They find that this is the case for both local and high-redshift galaxies ( Bhattacharya & Kobayashi 2025 , see also) . This finding is consistent with the general interpretation of gas inflows diluting the metallicity, resulting in a reduction of O/H at fixed N/O ( Andrews & Martini 2013 ; Curti et al. 2020 ; Scholte & Saintonge 2023 ; Scholte et al. 2024 ; Scholte et al. 2025 , e.g.,) . However, we do not find that the high N/O galaxies are significantly more star forming than other galaxies in our sample at the same mass as would be expected if gas dilution is the mechanism behind the high-N/O outliers. It is possible that the high N/O galaxies are tracing an early period of star formation shortly after the infall of pristine gas where the metallicity has been reduced but the high N/O abundance ratio, produced by AGB stars after the previous burst of star formation, has been retained ( McClymont et al. 2025 , see e.g.,) . The question still remains whether the high N/O abundance pattern is indeed an observed phase after significant gas accretion or whether there is another mechanism producing this peculiar abundance pattern. An alternative may be the rapid production and release of large amounts of nitrogen by Wolf-Rayet stars ( Berg et al. 2011 ; Rivera-Thorsen et al. 2024 ; Berg et al. 2025 ) or even very massive stars ( Vink 2023 ) following a burst of star formation. This will be discussed in more detail in a further study on this sample (Scholte et al. in prep.). Figure 11: The mass-metallicity relation (left), the mass- \rm N/O relation (middle) and the star formation main sequence (right). The data points are coloured by the Gaussian kernel density of the plotted distribution. Typical uncertainties are shown by the black error bars in the bottom-right corner of each panel. The red data with error bars show the median measurements and 16 {}^{\rm th}~ to 84 {}^{\rm th}~ percentiles at fixed stellar mass. The blue lines show the best fit relations derived from our observations. The orange lines show measured relations from the literature as indicated in the legend ( Berg et al. 2012 ; Andrews & Martini 2013 ; Hayden-Pawson et al. 2022 ; Popesso et al. 2023 ; Scholte et al. 2024 ) . The pink datapoints in the both panels show a selection of galaxies with low metallicity and high \rm N/O ratios as discussed in the text. 4.3.3 Neon abundances and ratios As neon and oxygen are both alpha-elements, their abundances are expected to be strongly correlated. In the top-right panel of Figure 10 we show the \rm Ne/O ratio as a function of oxygen abundance. We find that the median \rm log(Ne/O) ratio is -0.74_{-0.08}^{+0.11} dex. We fit a linear function to the relation using orthogonal distance regression: \log({\rm Ne/O})=(0.0970\pm 0.0021)x-(1.512\pm 0.017), (7) where x=12+\log(\rm O/H) . This relation is consistent with the relation derived by Izotov et al. 2006 and with the expected relation derived from chemical evolution models ( Kobayashi et al. 2020 ) . The average \rm O/Ne ratio is also consistent with the solar value ( Asplund et al. 2021 ) at solar metallicities. At metallicities higher than \rm 12+\log(O/H)\sim 8.1 , we find an increasing trend in the \rm Ne/O ratio, compared to a mostly flat trend at lower metallicities. Therefore, we also fit the \rm Ne/O ratio using a broken linear relation (see Eq. 6 ) providing the following fitted parameters: c_{\rm plateau}=-0.7499\pm 0.0006 , c_{\rm break}=8.105\pm 0.004 and c_{\rm slope}=0.250\pm 0.004 . The best fit relation is plotted in grey in Figure 10 . This result is similar to the findings by e.g., Amayo et al. 2021 , Miranda-Pérez & Hidalgo-Gámez 2023 , Arellano-Córdova et al. 2024 and Esteban et al. 2025 , who each also find a more sharply increasing trend. The nature for the increasing trend in \rm Ne/O at high metallicity is still unclear. As shown in Section 4.2 , there is a large spread in the ionisation correction factors of neon which complicates the interpretation of these results. The differences between these ICFs is large enough to suggest that the sharp increase in the \rm Ne/O ratio could be due to inaccurate ICFs. However, this trend could also reflect real changes in the \rm Ne/O abundance ratio of the ionized gas at high metallicity. This could be due to the increased depletion of oxygen onto dust grains at higher metallicities, which would lead to a higher gas phase \rm Ne/O ratio. Dust-to-metal ratios in the ISM of galaxies are expected to increase with metallicity ( Galliano et al. 2018 , e.g.,) , which would lead to a higher depletion of oxygen onto dust grains at higher metallicities. Particularly, dust evolution models and observations suggest that the dust-to-metal ratio may increase sharply at \rm 12+log(O/H)\sim 8.2 dex ( De Vis et al. 2017 ; De Vis et al. 2019 ) . This would lead to a higher \rm Ne/O ratio in the gas phase at high metallicities, as neon is not depleted onto dust grains but oxygen is depleted. Compared to a dust free scenario the expected increase in the \rm\log(Ne/O) ratio is \sim 0.2 dex, assuming typical ISM depletion fractions of oxygen ( Jenkins 2009 ) . This is consistent with the observed increase in the \rm\log(Ne/O) ratio at high metallicities. 4.3.4 Sulphur and argon abundances and ratios Similar to neon, the elements sulphur and argon are alpha-elements, therefore, their abundances are expected to be strongly correlated with the oxygen abundance. However, there is a more complex set of production pathways of these elements, where a significant fraction of the S and Ar is produced in SNe Ia ( Kobayashi et al. 2020 , e.g.,) . This leads to a more complex relation between the sulphur, argon abundances and the oxygen abundances as has been confirmed observationally in both planetary nebulae ( Arnaboldi et al. 2022 ) and star-forming galaxies ( Stanton et al. 2025b ; Bhattacharya et al. 2025a ; Bhattacharya et al. 2025b , e.g.,) . In the bottom-left panel of Figure 10 we show the \rm S/O ratio as a function of oxygen abundance. We find a median \rm log(S/O) ratio of -1.70_{-0.11}^{+0.12} dex. We fit a linear function to the relation using orthogonal distance regression: \log({\rm S/O})=(0.13\pm 0.05)x-(2.74\pm 0.04), (8) where x=12+\log(\rm O/H) . This is broadly consistent with the relation derived by Izotov et al. 2006 , although with a different slope, and in agreement with the expected relation derived from chemical evolution models ( Kobayashi et al. 2020 ) . Our average \rm S/O ratio at solar metallicities is lower than the solar value ( Asplund et al. 2021 ) , however, this is within the scatter of our measurements. In the bottom-right panel of Figure 10 we show the \rm Ar/O ratio as a function of oxygen abundance. We find the \rm log(Ar/O) ratio has a median value of -2.48_{-0.11}^{+0.11} dex. We fit a linear function to the relation using orthogonal distance regression: \log({\rm Ar/O})=-(0.112\pm 0.004)x-(1.55\pm 0.04), (9) where x=12+\log(\rm O/H) . Our average \rm Ar/O ratio is similar to the values found by Izotov et al. 2006 , and Arellano-Córdova et al. 2024 . However, particularly at low metallicity, our average \rm Ar/O ratio is significantly higher than the expected relation derived from chemical evolution models ( Kobayashi et al. 2020 ) . Similar to our results for sulphur, our average \rm Ar/O ratio at solar metallicities is lower than the solar value ( Asplund et al. 2021 ) , however, within the scatter of our measurements. Finally, as both sulphur and argon have a production channel through SNe Ia, the \rm S/O and \rm Ar/O abundance ratios are expected to be correlated. In Figure 12 we show that this is indeed the case. Where the dashed black line is the expected relation with unity slope corrected for the relative difference in abundance between sulphur and argon as measured through the median values of our sample: \rm log(Ar/O)=log(S/O)-0.78~dex . We also show the Pearson correlation coefficient between the two ratios, which is r=0.612 with a p -value less than 10^{-3} . This indicates a significant correlation. Figure 12: The \rm S/O ratio as a function of the \rm Ar/O ratio. The dashed black line shows the expected relation with unity slope corrected for the relative difference in abundance between sulphur and argon as measured through the median values of our sample: \rm log(Ar/O)=log(S/O)-0.78~dex . The strong correlation between the two ratios is expected due to the similar production channels of sulphur and argon, this is also shown by the tracks of chemical evolution models in orange, as indicated by the legend on Figure 10 . The pearson correlation coefficient and associated p -value are shown in the top-left corner of the panel. The orange lines show the predicted relations from chemical evolution models for the Solar neighbourhood, Milky Way halo and bulge stars by Kobayashi et al. 2020 . All data points are coloured by the Gaussian kernel density of the plotted distribution. Typical uncertainties are shown by the black error bars in the bottom-right corner of each panel. 5 Summary and conclusions In this work we presented a comprehensive analysis of the electron temperatures and abundances of N, O, Ne, S and Ar in star-forming galaxies using the DESI DR2 dataset. The main findings of our analysis are summarised below: • Our dataset contains electron temperature and abundance measurements for 49 959 galaxies. The exact breakdown of the number of galaxies with electron temperature and abundance measurements is shown in Table 4 . • We find that the average relation between T_{\rm e}(\rm O^{++}) and T_{\rm e}(\rm S^{++}) is close to the one-to-one relation. This suggests that the electron temperatures derived from these ions likely trace similar physical conditions in the ionized gas (Figure 4 ). • We find a large scatter in the relation between T_{\rm e}(\rm O^{++}) and T_{\rm e}(\rm O^{+}) , which is consistent with previous literature (Figure 5 ). • We provide constraints on a number of other temperature relations as discussed in Section 4.1 . • We find a large sample of 3 114 extremely metal poor galaxies (EMPGs) with metallicities less than 10% solar. These EMPGs are ideal candidates for studying the chemical evolution of galaxies in pristine environments such as present in the early Universe. • We discover two of the most metal-poor galaxies in the local Universe, DESI J211.9086+28.2461 and DESI J169.0571+14.1514, with oxygen abundances of \rm 12+\log(O/H)=6.77_{-0.03}^{+0.03}~\rm dex and \rm 12+\log(O/H)=6.81_{-0.04}^{+0.04}~\rm dex , respectively. For a detailed analysis of the lowest-metallicity galaxies in this sample we refer to Moustakas et al. (in prep.). • We find that the average \rm N/O ratio of the galaxies in our sample is consistent with previous literature results ( Izotov & Thuan 1999 ; Berg et al. 2012 ; Nicholls et al. 2017 ; Berg et al. 2019 , e.g.,) . • We also find a significant number of galaxies with high \rm N/O ratios at low \rm O/H . We find that this population of 139 high \rm N/O galaxies are more massive than typical galaxies at low metallicity and that their \rm N/O ratios are roughly consistent with the expected values at their respective mass. This will be investigated in more detail in a further study (Scholte et al. in prep.). • We find that the average \rm Ne/O ratio is consistent with the relation derived by Izotov et al. 2006 and with the expected relation derived from chemical evolution models ( Kobayashi et al. 2020 ) . However, we also find a more sharply increasing trend in the \rm Ne/O ratio at high metallicities. This elevated trend may be caused by systematic effects in the measurements or could reflect real changes in the physical conditions of the ionized gas at high metallicities. A likely explanation for this trend is the increased depletion of oxygen onto dust grains at higher metallicities, which would lead to a higher gas phase \rm\log(Ne/O) ratio. • We find that the average \rm S/O and \rm Ar/O ratios are consistent with the relation from chemical evolution models ( Kobayashi et al. 2020 ) . We find a strong correlation between the \rm S/O and \rm Ar/O ratios, which is expected due to the similar production channels of these elements. This new dataset provides a wealth of information on the physical conditions and chemical abundances in star-forming galaxies. The number of galaxies with auroral line detections allows us to perform statistical studies of the chemical evolution of galaxies, including outlier populations, which has previously been very challenging. We will study the abundance relations of galaxies including outlier populations in more detail in a follow-up paper (Scholte et al. in prep.). There are several other key areas where this dataset will allow significant progress. Electron temperature relations may be improved by including additional properties ( Langeroodi & Hjorth 2024 , as shown by e.g.,) . This will benefit from large number of galaxies with multiple abundance measurements as included in our DESI DR2 dataset. The derivation of robust strong line metallicity calibrations that account for secondary dependencies also requires reliable measurements of metallicities of large samples of galaxies ( Nakajima et al. 2022 , see also e.g.,) . As shown by Scholte et al. 2025 , once secondary dependencies are accounted for, these strong line metallicity calibrations will be applicable at any redshift. The \sim 2 orders of magnitude step change in the number of individual galaxies for which we can measure auroral emission lines is due to the increased survey depth and number of galaxies observed by DESI. As DESI continues its 8-year survey we may double the size of this sample once again. Therefore, DESI will remain the largest statistical sample of galaxy abundances for the foreseeable future and be a core resource for the study of the chemical evolution of galaxies. It is already providing a valuable local comparison for chemical evolution studies at high-redshift using JWST ( Scholte et al. 2025 ; Laseter et al. 2025 , see e.g.,) and will provide a core reference for chemical evolution studies in upcoming surveys such as MOONS (Multi-Object Optical and Near-infrared Spectrograph), PFS (Prime Focus Spectrograph), WEAVE (WHT Enhanced Area Velocity Explorer) and 4MOST (4-metre Multi-Object Spectrograph Telescope). Data Availability The data from DESI Data Release 2 is not yet publicly available. However, the data from the DESI Survey Validation period and Data Release 1 is publicly available at https://data.desi.lbl.gov/doc/ . This includes spectra and derived data such as emission line flux measurements from FastSpecFit ( Moustakas et al. 2023a ) . The Legacy Survey imaging is available at https://www.legacysurvey.org/ . After the public release of DESI Data Release 2, the derived electron temperature and abundance measurements will be made available. A link to the data will also be shared at https://dirkscholte.github.io/data . A summary of our data model is already available in appendix A . Data shown in figures will be available on Zenodo upon acceptance ( https://doi.org/10.5281/zenodo.17839238 ). Appendix A Data model The measurements derived in this work comprise a catalogue that will be made available after the public release of DESI DR2. The data model for this catalogue is provided in Table 5 . Table 5: The data model of the measurement catalogue presented in this work. Quantity Units Description DESINAME — The official DESI name of the galaxy TARGETID — Unique target ID SURVEY — Survey name PROGRAM — Programme name HEALPIX — Healpixel number RA deg Right ascension DEC deg Declination Z — Redshift LOGMSTAR \rm M_{\odot} The stellar mass computed using FastSpecFit assuming h=0.6766 . LOGSFR_p[ n ] \rm M_{\odot}~yr^{-1} The [ n=16,50,84 ] {}^{\rm th}~ percentile value of star formation rate based on dust corrected H \alpha or H \beta line emission. AV_p[ n ] mag The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the dust attenuation measurement. NE_p[ n ] cm -3 The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the electron density. TE_OIII_p[ n ] K The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm O^{++} electron temperature. TE_SIII_p[ n ] K The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm S^{++} electron temperature. TE_OII_p[ n ] K The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm O^{+} electron temperature. TE_NII_p[ n ] K The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm N^{+} electron temperature. TE_SII_p[ n ] K The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm S^{+} electron temperature. N+_p[ n ] dex The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm log(N^{+}/H^{+}) abundance. NH_p[ n ] dex The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm log(N/H) abundance. O+_p[ n ] dex The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm log(O^{+}/H^{+}) abundance. O++_p[ n ] dex The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm log(O^{++}/H^{+}) abundance. OH_p[ n ] dex The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm log(O/H) abundance. Ne++_p[ n ] dex The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm log(Ne^{++}/H^{+}) abundance. NeH_p[ n ] dex The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm log(Ne/H) abundance. S+_p[ n ] dex The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm log(S^{+}/H^{+}) abundance. S++_p[ n ] dex The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm log(S^{++}/H^{+}) abundance. SH_p[ n ] dex The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm log(S/H) abundance. Ar++_p[ n ] dex The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm log(Ar^{++}/H^{+}) abundance. Ar+++_p[ n ] dex The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm log(Ar^{3+}/H^{+}) abundance. ArH_p[ n ] dex The [ n=16,50,84 ] {}^{\rm th}~ percentile value of the \rm log(Ar/H) abundance. Appendix B Author affiliations 1 Institute for Astronomy, University of Edinburgh, Royal Observatory, Blackford Hill, Edinburgh EH9 3HJ, UK 2 Department of Physics and Astronomy, Siena College, 515 Loudon Road, Loudonville, NY 12110, USA 3 National Astronomical Observatories, Chinese Academy of Sciences, A20 Datun Road, Chaoyang District, Beijing 100012, China 4 Department of Physics & Astronomy, University College London, Gower Street, London, WC1E 6BT, UK 5 Max-Planck Institute for Radio Astronomy, Auf dem Hugel ¨ 69, D-53121 Bonn, Germany 6 Department of Physics and Astronomy and PITT PACC, University of Pittsburgh, Pittsburgh, PA 15260, USA 7 Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, CA 94720, USA 8 Department of Physics, Boston University, 590 Commonwealth Avenue, Boston, MA 02215 USA 9 Dipartimento di Fisica “Aldo Pontremoli”, Università degli Studi di Milano, Via Celoria 16, I-20133 Milano, Italy 10 INAF-Osservatorio Astronomico di Brera, Via Brera 28, 20122 Milano, Italy 11 Institut d’Estudis Espacials de Catalunya (IEEC), c/ Esteve Terradas 1, Edifici RDIT, Campus PMT-UPC, 08860 Castelldefels, Spain 12 Institute of Space Sciences, ICE-CSIC, Campus UAB, Carrer de Can Magrans s/n, 08913 Bellaterra, Barcelona, Spain 13 Kapteyn Astronomical Institute, University of Groningen, Landleven 12 (Kapteynborg, 5419), 9747 AD Groningen, The Netherlands 14 Instituto de Física, Universidad Nacional Autónoma de México, Circuito de la Investigación Científica, Ciudad Universitaria, Cd. de México C. P. 04510, México 15 Department of Astronomy & Astrophysics, University of Toronto, Toronto, ON M5S 3H4, Canada 16 Department of Physics & Astronomy, University of Rochester, 206 Bausch and Lomb Hall, P.O. Box 270171, Rochester, NY 14627-0171, USA 17 University of California, Berkeley, 110 Sproul Hall #5800 Berkeley, CA 94720, USA 18 Departamento de Física, Universidad de los Andes, Cra. 1 No. 18A-10, Edificio Ip, CP 111711, Bogotá, Colombia 19 Observatorio Astronómico, Universidad de los Andes, Cra. 1 No. 18A-10, Edificio H, CP 111711 Bogotá, Colombia 20 Institute of Cosmology and Gravitation, University of Portsmouth, Dennis Sciama Building, Portsmouth, PO1 3FX, UK 21 University of Virginia, Department of Astronomy, Charlottesville, VA 22904, USA 22 Fermi National Accelerator Laboratory, PO Box 500, Batavia, IL 60510, USA 23 NSF NOIRLab, 950 N. Cherry Ave., Tucson, AZ 85719, USA 24 Sorbonne Université, CNRS/IN2P3, Laboratoire de Physique Nucléaire et de Hautes Energies (LPNHE), FR-75005 Paris, France 25 Center for Cosmology and AstroParticle Physics, The Ohio State University, 191 West Woodruff Avenue, Columbus, OH 43210, USA 26 Department of Astronomy, The Ohio State University, 4055 McPherson Laboratory, 140 W 18th Avenue, Columbus, OH 43210, USA 27 The Ohio State University, Columbus, 43210 OH, USA 28 Institució Catalana de Recerca i Estudis Avançats, Passeig de Lluís Companys, 23, 08010 Barcelona, Spain 29 Institut de Física d’Altes Energies (IFAE), The Barcelona Institute of Science and Technology, Edifici Cn, Campus UAB, 08193, Bellaterra (Barcelona), Spain 30 Department of Physics and Astronomy, University of Waterloo, 200 University Ave W, Waterloo, ON N2L 3G1, Canada 31 Perimeter Institute for Theoretical Physics, 31 Caroline St. North, Waterloo, ON N2L 2Y5, Canada 32 Waterloo Centre for Astrophysics, University of Waterloo, 200 University Ave W, Waterloo, ON N2L 3G1, Canada 33 Space Sciences Laboratory, University of California, Berkeley, 7 Gauss Way, Berkeley, CA 94720, USA 34 Instituto de Astrofísica de Andalucía (CSIC), Glorieta de la Astronomía, s/n, E-18008 Granada, Spain 35 Departament de Física, EEBE, Universitat Politècnica de Catalunya, c/Eduard Maristany 10, 08930 Barcelona, Spain 36 Department of Physics and Astronomy, Sejong University, 209 Neungdong-ro, Gwangjin-gu, Seoul 05006, Republic of Korea 37 CIEMAT, Avenida Complutense 40, E-28040 Madrid, Spain 38 Department of Astronomy, School of Physics and Astronomy, Shanghai Jiao Tong University, Shanghai 200240, China 39 University of Michigan, 500 S. State Street, Ann Arbor, MI 48109, USA