The longest cirrus time series are ground-based, visual observations captured by human observers [synoptic observations (SYNOP)]. However, their reliability is impacted by an unfavorable viewing geometry (cloud overlap) and misclassification due to low cloud optical thickness, especially at night. For the very first time, this study assigns a quantitative value to uncertainty. We validate 15 years of SYNOP observations (2006–20) against data from the cloud lidar flown on board the Cloud–Aerosol Lidar and Infrared Pathfinder (CALIPSO) spacecraft. We develop a dedicated method to match SYNOP reports (with a hemispherical field of view) with lidar samples (along-track profiles). Our evaluation of the human eye’s sensitivity to cirrus revealed that it is moderate, at best. In perfect conditions (daytime with no mid/low-level clouds) the probability of correct detection was 44%–83% (Cohen’s kappa coefficient < 0.6), and this fell to 24%–42% (kappa < 0.3) at night. Lunar illumination improved detection, but only when the moon’s phase exceeded 50%. Cirrus optical depth had a clear impact on detection. When clouds at all levels were considered (i.e., real-life conditions), the reliability of the visual method was moderate to poor: it detected 47%–71% of cirrus (kappa < 0.45) during the day and 28%–43% (kappa < 0.2) at night and decreased with an increasing low/midlevel cloud fraction. These kappa coefficients suggest that agreement with CALIPSO data was close to random. Our findings can be directly applied to estimations of cirrus frequency/trends. Our reported probabilities of detection can serve as a benchmark for other ground-based cirrus detection methods.
Kotarba, A.Z, Nguyen Huu, Ż. (2022) Accuracy of cirrus detection by surface-based human observers