Cloud Observation and Cloud Cover Calculation at Nighttime Using the Automatic Cloud Observation System (ACOS) Package

Abstract

An Automatic Cloud Observation System (ACOS) and cloud cover calculation algorithm were developed to calculate the cloud cover at night, and the calculation results were compared with the cloud cover data of a manned observatory (Daejeon Regional Office of Meteorology, DROM) that records human observations. Annual and seasonal analyses were conducted using the 1900–0600 local standard time (LST) hourly data from January to December 2019. Prior to calculating the cloud cover of ACOS, pre-processing was performed by removing surrounding obstacles and correcting the distortion caused by the fish-eye lens. In addition, the red–blue ratio (RBR) threshold was determined, according to the image characteristics (RBR and luminance) using the red, green, and blue (RGB) brightness value of the area in which the solar zenith angle (SZA) was less than 80°, to calculate the cloud cover. The calculated cloud cover exhibited a bias of −0.28 tenths, root mean square error (RMSE) of 1.78 tenths, and a correlation coefficient of 0.91 for DROM across all cases. The frequency of the cases that exhibited differences less than 1 tenth between the observed and calculated cloud cover was 46.82%, while the frequency of cases that exhibited differences less than 2 tenths was 87.79%.

On the Generalization Ability of Data-Driven Models in the Problem of Total Cloud Cover Retrieval

“…within the approaches of Maximum Likelihood Estimator or Maximum a Posteriori Probability Estimator, similar to regression statistical models. However, the "less or equal than one-okta error accuracy" ("Leq1A" hereafter) is frequently considered as additional quality measure in the problem of TCC retrieval [43,49]. In our understanding, this metric is still not valid and may be biased due to the reasons given above, though we provide its estimates for our results to be comparable with other studies.…”

“…Most of these algorithms are designed by experts introducing their understanding of the physical processes resulting in the all-sky imagery similar to the one presented in 1. Given all-sky imagery acquired, in most simple cases, an index is calculated pixel-wise, e.g., red-to-blue ratio (RBR) in the series of papers of Long et al [32,33,41] or in following studies [42][43][44], or the ratio B−R B+R in [36,37,45], or even a set of indices [46]. Then, an empirical threshold is applied for the classification of pixels dividing them into two classes: "cloudy" and "clear sky" ones.…”

“…Then, these human-engineered features are used in some sort of an algorithm for the computation of an index or multiple indices which then are aggregated over the image to a quantitative measure. The algorithm may be considered simple [32,33,36,38,48], adaptive [39,47] or designed to be complex to some extent in attempt of taking some advanced spatial features into account [34,35,41,43,44] or in attempt of correcting the distortion of imagery or other features of imagery that are not consistent with the initial researcher's assumptions [36,43,44,46]. However, all these methods rely on the aggregation step at some point that is commonly implemented as a variation of thresholding.…”

“…The threshold value(s) is(are) empirical and should be adjusted minimizing the error of TCC estimates or maximizing the quality of the method that is proposed in the corresponding study. This adjustment stage is commonly described vaguely [43] or briefly [33], though in this sense, the above mentioned expert-designed algorithms are essentially data-driven and inherently have optimization nature, thus the optimization is a key part of the development of these schemes.…”

“…However, the definition of a quality measure should never depend on the way the problem is solved. In some studies, the quality measures are used that are common for regression problems, e.g., correlation coefficient [49], MSE or RMSE [43]. In probability theory, these measures usually imply the specific assumptions about the distribution of the target value (TCC), and also the assumptions about the set of all possible outcomes and their type (real values).…”