Title: On the accuracy of HI observations in molecular clouds -- More cold HI than thought?

Abstract: We present a study of the cold atomic hydrogen (HI) content of molecular clouds simulated within the SILCC-Zoom project. We produce synthetic observations of HI at 21 cm including HI self-absorption (HISA) and observational effects. We find that HI column densities, $N_\textrm{HI}$, of $\gtrsim$10$^{22}$ cm$^{-2}$ are frequently reached in molecular clouds with HI temperatures as low as $\sim$10 K. Hence, HISA observations assuming a fixed HI temperature tend to underestimate the amount of cold HI in molecular clouds by a factor of 3 - 10 and produce an artificial upper limit of $N_\textrm{HI}$ around 10$^{21}$ cm$^{-2}$. We thus argue that the cold HI mass in clouds could be a factor of a few higher than previously estimated. Also $N_\textrm{HI}$-PDFs obtained from HISA observations might be subject to observational biases and should be considered with caution. The underestimation of cold HI in HISA observations is due to both the large HI temperature variations and the effect of noise in regions of high optical depth. We find optical depths of cold HI around 1 - 10 making optical depth corrections essential. We show that the high HI column densities ($\gtrsim$10$^{22}$ cm$^{-2}$) can in parts be attributed to the occurrence of up to 10 individual HI-H$_2$ transitions along the line of sight. This is also reflected in the spectra, necessitating Gaussian decomposition algorithms for their analysis. However, also for a single HI-H$_2$ transition, $N_\textrm{HI}$ frequently exceeds 10$^{21}$ cm$^{-2}$, challenging 1D, semi-analytical models. This is due to non-equilibrium chemistry effects and the fact that HI-H$_2$ transition regions usually do not possess a 1-dimensional geometry. Finally, we show that the HI gas is moderately supersonic with Mach numbers of a few. The corresponding non-thermal velocity dispersion can be determined via HISA observations within a factor of $\sim$2.