Reliability of NH3 as the temperature probe of cold cloud cores

Context. The temperature is a central parameter affecting the chemical and physical properties of dense cores of interstellar clouds and their potential evolution towards star formation. The chemistry and the dust properties are temperature dependent and, therefore, interpretation of any observation requires the knowledge of the temperature and its variations. Direct measurement of the gas kinetic temperature is possible with molecular line spectroscopy, the ammonia molecule, NH3, being the most commonly used tracer. Aims. We want to determine the accuracy of the temperature estimates derived from ammonia spectra. The normal interpretation of NH3 observations assumes that all the hyperfine line components are tracing the same volume of gas. However, in the case of strong temperature gradients they may be sensitive to different layers and this could cause errors in the optical depth and gas temperature estimates. Methods. We examine a series of spherically symmetric cloud models, 1.0 and 0.5 M-circle dot Bonnor-Ebert spheres, with different radial temperature profiles. We calculate synthetic NH3 spectra and compare the derived column densities and temperatures to the actual values in the models. Results. For high signal-to-noise observations, the estimated gas kinetic temperatures are within similar to 0.3K of the real mass averaged temperature and the column densities are correct to within similar to 10%. When the S/N ratio of the (2, 2) spectrum decreases below 10, the temperature errors are of the order of 1K but without a significant bias. Only when the density of the models is increased by a factor of a few, the results begin to show significant bias because of the saturation of the (1, 1) main group. Conclusions. The ammonia spectra are found to be a reliable tracer of the real mass averaged gas temperature. Because the radial temperature profiles of the cores are not well constrained, the central temperature could still be different from this value. If the cores are optically very thick, there are no longer guarantees of the accuracy of the estimates.