2002A&A...395..915A

We present 26 ISO-SWS spectra taken from a sample of 13 M-type Semiregular, Lb and Mira variables and covering the wavelength range between 2.36 and 5µm at a medium resolution. All of the studied objects show intense water bands producing a deep absorption dip around 2.5µm. Features of CO, OH, SiO and CO₂ are also visible. Using the new H₂O linelist published in the first paper of this series and available opacity data for the other important molecules, we calculated a grid of hydrostatic MARCS atmospheres and the corresponding synthetic ISO-SWS spectra. Based on the comparison with these theoretical results the ISO observations can be divided into four classes. The first two groups include the spectra of the Semiregular (SRb) and Lb variables in our sample. For all of them the region between 2.36 and 4.2µm can be quite well reproduced by our hydrostatic models. Only the predicted SiO bands above 4µm are in some cases too strong which is due to known dynamical effects. Depending on the temperature (above or below 3000K) of the atmosphere, which mainly determines the intensity of the water depression at 2.5µm, the spectra of the Semiregular and Lb variables fall into the first or second class. The third group consists of observations of Mira stars obtained around maximum light where the range between 2.36 and 4.2µm can be fitted with our MARCS models except for a strong emission bump appearing in the ISO-SWS data in the region of the SiO features and the slope very close to the short wavelength border. Finally, the last type of spectra corresponds to Mira variables during the phases around the minimum of their visual light curve. For this class the observed water absorption at 2.5µm is much more intense than in any hydrostatic atmosphere with a realistic choice of effective temperature and surface gravity. Thus, we conclude that dynamical models are needed to explain the ISO-SWS data of Mira stars. For all of the cooler objects from our sample the predicted CO₂ bands between 4.2 and 4.6µm are too weak which may be due to the opacity data.