Comparison of Atmospheric River Detection Algorithms in East Asia

This study compares the three detection algorithms of East Asian summer atmospheric rivers (ARs). The algorithms developed by Guan and Waliser (GW15), Park et al. (P21), and Tian et al. (T23) are particularly compared in terms of the AR frequency, the number of AR events, and the AR duration for the period of 2016-2020. All three algorithms show similar spatio-temporal distributions of AR frequency, centered along the edge of the North Pacific high. The maximum AR frequency gradually shifts northward in early summer as the edge of the North Pacific High expands, and retreats in late summer. However, the detailed pattern and the maximum value differ among the algorithms. When the AR frequency is decomposed into the number of AR events and the AR duration, the AR frequencies detected by GW15 and P21 are equally explained by both factors. However, the number of AR events primarily determine the AR frequency in T23. This difference occurs as T23 utilizes the machine learning algorithm applied to moisture field while GW15 and P21 apply the threshold value to moisture transport field. When evaluating AR-related precipitation, the ARs detected by P21 show the closest relationship with total precipitation in East Asia by up to 60%. These results indicate that AR detection in the East Asian summer is sensitive to the choice of the detection algorithm and can be optimized for the target region.