Abstract

Aims. We present J, H, K spectrally dispersed interferometry with a spectral resolution of 35 for the Mira variable S Orionis. We aim at measuring the diameter variation as a function of wavelength that is expected due to molecular layers lying above the continuum-forming photosphere. Our final goal is a better understanding of the pulsating atmosphere and its role in the mass-loss process. Methods. Visibility data of S Ori were obtained at phase 0.78 with the VLTI/ AMBER instrument using the fringe tracker FINITO at 29 spectral channels between 1.29 mu m and 2.32 mu m. Apparent uniform disk (UD) diameters were computed for each spectral channel. In addition, the visibility data were directly compared to predictions by recent self-excited dynamic model atmospheres. Results. S Ori shows significant variations in the visibility values as a function of spectral channel that can only be described by a clear variation in the apparent angular size with wavelength. The closure phase values are close to zero at all spectral channels, indicating the absence of asymmetric intensity features. The apparent UD angular diameter is smallest at about 1.3 mu m and 1.7 mu m and increases by a factor of similar to 1.4 around 2.0 mu m. The minimum UD angular diameter at near-continuum wavelengths is Theta(UD) = 8.1 +/- 0.5mas, corresponding to R similar to 420 R-circle dot. The S Ori visibility data and the apparent UD variations can be explained reasonably well by a dynamic atmosphere model that includes molecular layers, particularly water vapor and CO. The best-fitting photospheric angular diameter of the model atmosphere is Theta(Phot) = 8.3 +/- 0.2mas, consistent with the UD diameter measured at near-continuum wavelengths. Conclusions. The measured visibility and UD diameter variations with wavelength resemble and generally confirm the predictions by recent dynamic model atmospheres. These size variations with wavelength can be understood as the effects from water vapor and CO layers lying above the continuum-forming photosphere. The major remaining differences between observations and model prediction are very likely due to an imperfect match of the phase and cycle combination between observation and available models.