Evaluation of simulated cloud liquid water in low clouds over the Beaufort Sea in the Arctic System Reanalysis using ARISE airborne in situ observations

Arctic low clouds and the water they contain influence the evolution of the Arctic system through their effects on radiative fluxes, boundary layer mixing, stability, turbulence, humidity, and precipitation. Atmospheric models struggle to accurately simulate the occurrence and properties of Arctic low clouds, stemming from errors in both the simulated atmospheric state and the dependence of cloud properties on the atmospheric state. Knowledge of the contributions from these two factors to the model errors allows for the isolation of the process contributions to the model–observation differences. We analyze the differences between the Arctic System Reanalysis version 2 (ASR) and data taken during the September 2014 Arctic Radiation–IceBridge Sea and Ice Experiment (ARISE) airborne campaign conducted over the Beaufort Sea. The results show that ASR produces less total and liquid cloud water than observed along the flight track and is unable to simulate observed large in-cloud water content. Contributing to this bias, ASR is warmer by nearly 1.5 K and drier by 0.06 g kg−1 (relative humidity 4.3 % lower) than observed. Moreover, ASR produces cloud water over a much narrower range of thermodynamic conditions than shown in ARISE observations. Analyzing the ARISE–ASR differences by thermodynamic conditions, our results indicate that the differences are primarily attributed to disagreements in the cloud–thermodynamic relationships and secondarily (but importantly) to differences in the occurrence frequency of thermodynamic regimes. The ratio of the factors is about 2/3 to 1/3. Substantial sampling uncertainties are found within low-likelihood atmospheric regimes; sampling noise cannot be ruled out as a cause of observation–model differences, despite large differences. Thus, an important lesson from this analysis is that when comparing in situ airborne data and model output, one should not restrict the comparison to flight-track-only model output.