Continuous thermodynamic profiles can be retrieved from ground-based passive infrared spectrometers (ASSIST, AERI) allowing the investigation of boundary layer evolution and structure and pre-convective storm conditions. The optimal estimation physical retrieval TROPoe (Turner and Löhnert, 2014) determines the optimal state vector consisting of thermodynamic profiles and cloud properties in an iterative process. Starting with the prior as a first guess, the LBLRTM forward model is used to compute pseudo-observations, which are then compared to the actual observations. If the computed and observed values do not agree within the uncertainty of the measurements, the state vector is modified iteratively. The uncertainty of forward model is assumed to be included in the uncertainty of the observations (for computational efficiency). We found that TROPoe struggles to find a valid solution when (i) the uncertainty of the observations is not sufficient to compensate for the missing uncertainty of the forward model leading to overfitting and (ii) the information content of the retrieval decreases as precipitable water vapor (PWV) increases because of saturation in spectral bands traditionally used in retrieval. We were able to drastically increase the solution availability and accuracy compared to radiosondes by implementing a noise floor for infrared radiance uncertainty. Adding an additional spectral band at 793-804 cm-1 for high-moisture environments and by using the new LBLRTM v12.17 where the strength of the absorption by the water vapor continuum in the 800-1300 cm-1 band was decreased (Mlawer et al. 2024) further improved performance.