Abstract. The measurement of low-level, liquid-only and mixed-phase clouds in the polar regions is a necessary building block to understand the regional surface energy and mass budgets over ice sheets. The unambiguous retrieval of cloud phase from polarimetric lidar observations is dependent on the assumption that only cloud scattering processes alter the transmitted polarization. However, due to clouds varying in range, optical depth, and scatterer size and shape, most atmospheric lidar systems must observe high dynamic ranges in scattered signal strengths. Depending on the polarization component measured, these signals can far exceed the linear range of a detection system. Thus, due to the high optical thickness and predominately low-lying nature of liquid-only and mixed-phase clouds in the polar regions, relative to ice only clouds, a systematic overestimate of the traditional depolarization ratio, which uses the co-polarized and cross-polarized signals, can occur due to the large dynamic range signals. For both liquid-only and mixed-phase clouds, this results in a misidentification of liquid water in clouds as ice, which has broad implications on evaluating surface energy budgets. The Clouds Aerosol Polarization and Backscatter Lidar (CAPABL) at Summit, Greenland employs multiple planes of linear polarization, and photon counting and analog detection schemes, to self evaluate, correct, and optimize signal combinations to improve cloud classification. For example, an examination of observations of liquid-only and mixed-phase clouds at Summit shows as much as a 2 kilometer offset in median cloud height of those identified as liquid due to a systematic bias in photon counting signals. At a constant altitude, more than 94 % of the liquid pixels identified with analog signals can be misidentified as ice with photon counting signals. This results in a possible error of fractional occurrence of cloud liquid of approximately 30 %. It is shown that by observing polarization planes that are non-orthogonal, the dynamic range of observed signals is reduced, the coverage of the expected signal dynamic range is increased, and more linear response can be captured. Using non-orthogonal polarization observations is shown to enhance measurement sensitivity increasing the effective sampling range for CAPABL by as much as 18 % or approximately 1.5 km.