List of Changes in SSPP Since DR8 - SSPP

[Back to main SSPP page]

The SSPP has been refined and modified with each new SDSS data release in order to provide more accurate estimates of the stellar atmospheric parameters. Here, we describe major changes and improvements made to the SSPP since Data Release 8 (DR8).

Metallicity determined from Auto-correlation Function (ACF), Ca II triplet line index (CaIIT) method, and ANNSR is no longer reported and does not contribute to the best metallicity estimate for each star.
Gravity estimate from (MgH) and (CaI2) line index methods and k24 is not reported any more and does not contribute to the best gravity estimate for each star.
NGS1 and CaIIK1 make use of a new grid of synthetic spectra generated with α-enhanced Kurucz model atmospheres and with variable micro-turbulence velocity with surface gravities. The adopted relation is V_t=-0.345×log g+2.225, and this relation is derived from a sample of high-resolution spectra.
NGS2 utilizes a new grid of synthetic spectra created with α-enhanced Kurucz model atmospheres with micro-turbulence velocity fixed at 1 km/s.
The grid of synthetic spectra used in the NGS1, CaIIK1, and estimates is extended to [Fe/H]=-4.5 and [Fe/H]=+1.0 in order to improve the estimates at low metallicity and above solar metallicity.
Instead of linear interpolation, cubic spline interpolation is implemented for k24, ki13, NGS1, NGS2, and CaIIK1 during χ² minimization. That is, a new trial synthetic spectrum to calculate χ² value is generated by interpolating fluxes from existing grid of synthetic spectra with a cubic spline function.
If more than 5 % of a region of a spectrum used by a particular parameter estimation method has the inverse variance of the spectrum flux array set to zero, no parameter estimate is reported for that particular method. This was done to improve the reliability of the parameter estimates, especially at very low metallicity.
A new set of g-r and S/N limits is introduced for individual methods. These changes were made after evaluating the accuracy of the individual estimators in the SSPP estimates for a set of synthetic spectra with realistic noise properties covering the range of g-r and S/N in the SEGUE data, as well as the high resolution and cluster spectra.
New neural networks trained on new grid of synthetic spectra and the DR8 parameters are implemented for ANNSR and ANNRR, respectively.
The color-based temperature estimates by T_K, T_G, and T_I are not reported and do not contribute to the best temperature estimate.
Effective temperature based on the InfraRed Flux Method (IRFM) is estimated, using g-i, log g, and [Fe/H] from NGS1.
A comparison with the DR8 stellar parameters indicates that the DR9 T_eff is, on average, higher by ~60 K, the DR9 log g is lower by ~0.2 dex, and the metallicity does not change significantly (less than 0.02 index). The temperature shift occurs because we adopt the IRFM temperature scale to calibrate each temperature estimate in the SSPP. The reason for the gravity shift can be partly explained by the new log g values from the re-analysis of the high-resolution spectra, which are 0.13 dex lower than the previous high-resolution values. Another contribution to the shift in log g is the new grid of synthetic spectra for NGS1, NGS2, and CaI. However, the biggest causes of this shift are the retirement of the gravity estimates by MgH and CaI2 line index methods, which consistently deliver much higher (and less accurate) log g estimates that the other estimators in the DR8 SSPP.