At the specified time t, within the IVT background field, all grids where IVT exceeds the corresponding IVT threshold are identified and marked. Starting from any unvisited marked grid (i.e., a grid that has been marked but not yet incorporated into any AR candidate), the algorithm recursively searches the eight neighboring grids for unvisited marked grids. This process continues iteratively until no new unvisited marked grids are found in the neighborhood, and resulting visited marked grids collectively form a continuous region as an AR candidate. Continuous regions that span the circular longitudinal boundary (connecting 360° back to 0°) artificially imposed in the data will be considered as a single AR candidate. Subsequent extraction of AR candidates starts from new unvisited marked grids and repeats the above procedure. When all marked grids on the IVT background field have been visited, the AR identification for this time step is deemed complete.

The identification algorithm employs the spatiotemporal quantile method proposed by Guan and Waliser (2024) to construct the IVT threshold field. Specifically, for the IVT threshold at time t, a 5-month window centered around the calendar month of t is defined, with a time interval of 6 hours (00:00, 06:00, 12:00, 18:00 UTC). The 85th percentile IVT value during this window, calculated over the period from 1980 to 2019, is used as the climatological base threshold for each grid globally. This 85th percentile criterion is selected empirically, based on the statistical results of ARs in existing studies (Guan and Waliser, 2015; Lavers et al., 2012). However, given that the thresholds are generally low in regions with climatologically low moisture content, which makes it difficult to effectively extract AR candidates, the algorithm also introduces hemispheric threshold lower limits: for each calendar month, the base threshold of grids where the area-weighted cumulative values reach the 5th percentile in the Northern and Southern Hemisphere are respectively adopted as the lower threshold limits for all grids within each hemisphere. This 5th percentile approximates the areal proportion of polar and other low-moisture regions, representing the base threshold at the margins of these regions and thereby enhancing the extraction of AR candidates within them. Compared to conventional fixed threshold lower limits, this hemispheric lower limit considers inter-hemispheric and seasonal variations in IVT, and its values are more consistent with thresholds used in polar-specific AR identification algorithms (Francis et al., 2020). Further, because the hemispheric lower limit dynamically adjusts in response to the base threshold, their combination forms a universally applicable global threshold-setting method suitable for various detection variables. Figure 1 illustrates the resulting IVT fields after merging the 85th percentile base threshold with the hemispheric threshold lower limit for each calendar month. This set of threshold fields, also denoted as the Level-0 threshold fields, will be described in detail in Step 4.

Accurately and robustly quantifying the key characteristics of AR candidates remains one of the central challenges in current AR identification. Existing algorithms typically construct a single characteristic axis along the core region of a candidate to capture its dominant trajectory, which is subsequently used to quantify its key characteristics. This study proposes to define two distinct characteristic axes for each AR candidate using two complementary methods with different emphases, thereby enabling a more comprehensive assessment of its relevant characteristics. One way to define AR characteristic axis is to connect local maxima in IVT. This approach is computationally efficient but sensitive to small-scale change. The other way is to connect the centroids of IVT. This approach ensures a smoother and more stable AR track but often neglect small-scale change. In this study, two characteristic axes, derived from the two different approaches following Brands et al. (2017) and Pan and Lu (2019), are referred to as Axis₁ and Axis₂, respectively. Overviews of the two axes are provided below.

Axis1. The axis generation begins at the grid with the maximum IVT value within the AR candidate. For the axis point, the surrounding 360° is divided into eight 45° sectors corresponding to the cardinal directions (N, NE, E, SE, S, SW, W, NW), each encompassing the adjacent grid in the direction. Within the sector aligned with the IVT direction at the axis point and its two neighboring sectors, the grid with the maximum IVT value among the three adjacent grids is selected as the next axis point. This iterative process continues downstream until the axis reaches the boundary of the AR candidate. The upstream extension of Axis₁ follows the same procedure, except that the sectors are determined based on the opposite IVT direction. For both extensions, a maximum of 500 axis points is imposed to avoid closed-loop iterations in vortex-like moisture structures.

On the one hand, the elongated and narrow morphology of ARs serves as a key distinguishing feature from other moisture-rich weather systems. On the other hand, as critical pathways for global meridional water vapor transport, ARs are generally associated with organized poleward moisture transport structures that should be accounted for in their identification. Therefore, this study focuses on two aspects of AR candidates: morphological characteristics and poleward moisture transport, with most diagnostic metrics quantified based on their characteristic axes. Figure 2 illustrates an example of an AR candidate, together with its Axis₁ and Axis₂. For both axes, the metrics are calculated separately, yielding two sets of axis-based metrics that are jointly used in the subsequent diagnostic assessment of the corresponding characteristics of the AR candidate.

For the morphological aspect, the AR candidate is characterized using its length, length–width ratio, and two curvature-related metrics. For each axis, the length of the AR candidate is defined as the sum of the geographical distances between consecutive axis points. The corresponding equivalent width is then calculated as the ratio of the geographic area of the candidate to this axis-based length, and the length–width ratio is derived as the ratio of the length to the equivalent width. In addition to these length- and width-related metrics, two metrics are used to describe the curvature characteristics of the candidate: the sum of turning angles along the axis (Pan and Lu, 2019) and the straightness ratio of the axis. The sum of turning angles is computed as follows: each pair of adjacent axis points forms a vector pointing downstream; starting from the first vector at the upstream end, the turning angle between the current vector and the next adjacent vector is calculated sequentially (Fig. 2) and cumulatively summed, with counterclockwise angles assigned positive values. The straightness ratio is defined as the ratio of the great-circle distance between the axis endpoints to the total axis length.

For the poleward moisture transport aspect, the algorithm computes the total length of axis segments satisfying the poleward moisture flux condition (Guan and Waliser, 2024). Each segment consists of consecutive axis points where the ratio of the poleward component of IVT to the IVT magnitude exceeds 0.25, as marked with stars in Fig. 2. This metric therefore provides an axis-based quantitative measure of the spatial extent of organized poleward moisture transport embedded within the AR candidate. The design of this metric makes it particularly well suited for characterizing curved or directionally heterogeneous moisture transport features, in which transport directions may vary among different portions of the same candidate.

In the diagnostic assessment based on morphological features, the AR candidate will be excluded if it fails to satisfy all of the following criteria: (1) At least one axis has a length exceeding 2000 km; (2) At least one axis exhibits a length–width ratio greater than 2; (3) For each axis, the absolute value of the sum of turning angles is less than 300°-i×30°, and the standard deviation of the two numbers (i.e., sums of turning angles for Axis₁ and Axis₂, respectively) is less than 60°, where i∈{0,1,2,3,4} and denotes the current threshold-level index within the threshold adjustment step, as detailed in the next step; 4) For each axis, the straightness ratio exceeds 0.85-l/10000, where l denotes the axis length in kilometers. The first two criteria are geometric constraints based on statistical studies of ARs, and are designed to ensure the AR candidate exhibits elongated and narrow morphology. The latter two criteria are curvature constraints derived from extensive visual inspection and sensitivity tests, and are designed to effectively filter out highly curved moisture transport systems, particularly tropical cyclones. Specifically, the third criterion is based on the sum of turning angles along the axis, which can more directly quantify and diagnose the curvature characteristics of candidates at different scales, offering broad applicability and stability. The fourth criterion serves as a supplementary filter, primarily targeting a few high-curvature AR candidates with incomplete structures extracted by IVT threshold (Fig. 3a). Although such cases are rare, they are difficult to completely avoid in all threshold-based identification algorithms. When structural incompleteness prevents such candidates from being filtered out by the third criterion, the fourth criterion detects their low straightness ratios of the axes and accordingly excludes them. Notably, this criterion incorporates scale dependence, enhancing its sensitivity to small-scale candidates, which are the most common in such special cases.

It is worth emphasizing that the dual-axis collaborative representation proposed in this algorithm exhibits unique advantages in characterizing the morphological features of AR candidates and related diagnostic assessments. Each of the two axes (Axis₁ versus Axis₂) has its own strengths and limitations when independently representing the morphological characteristics of AR candidates. Axis₁, whose trajectory is determined by dynamically selecting the maximum IVT value immediately around each grid, offers high computational efficiency and sensitivity to local turbulent disturbances. This enables it to respond effectively to small-scale changes but may also lead to premature termination of the axis finding process due to preset boundary conditions (Fig. 3b). In contrast, Axis₂ is the connection line of the centroids of each part of the AR candidate constructed based on the K-nearest neighbors and local weighted regression, which ensures a smoother and more stable axis but may overlook some smaller vortex systems (Fig. 3c). To reduce the influence of the respective limitations of the two methods, our algorithm integrates both methods and refines the diagnostic logic at the level of axis-based metrics. Specifically, each quantitative metric is treated as an independent diagnostic item, and the two axes are cross-checked using metric-specific decision rules, with scale- and elongation-related criteria requiring at least one axis to satisfy the corresponding requirement and curvature-related criteria requiring both axes to satisfy it. In terms of axis representation, Axis₁ can more accurately capture small-scale disturbances than Axis₂, whereas Axis₂ can correct unrealistic path diversions that may arise in Axis₁ due to local vortices. Conceptually, the axis definitions in Guan and Waliser (2015, 2024) have elements similar to Axis₁. A dual-method axis identification strategy here is expected to enhance the robustness of AR identification in complex weather conditions, particularly to have a better representation of AR systems with multi-core structures or irregular shapes.

The poleward water vapor transport characteristics of AR candidates can reflect their potential dynamical linkages with weather systems across multiple scales. The diagnostic assessment of these characteristics aims to exclude candidates exhibiting pronounced zonal features, such as tropical moisture filaments and water vapor transport belts in the mid- and high latitudes that are almost entirely controlled by prevailing winds. Drawing upon established screening criteria from prior algorithms (Pan and Lu, 2020; Guan and Waliser, 2024), the AR candidate will be excluded if it fails to satisfy all of the following criteria: (1) The fraction of grids within the 20° S–20° N latitude band must not exceed 95 %; additionally, if this fraction exceeds 50 %, then the proportion of grids where the angle between the IVT direction and the latitude circle is less than 15° must not exceed 50 %; (2) There must be at least one axis with a total length of segments exhibiting poleward water vapor transport characteristics that exceeds 800 km, or at least one axis whose length exceeds 3500-Rp×5000 km, where Rp is required to be greater than 0 and is defined as the ratio of the area-weighted mean of the poleward IVT component to the area-weighted average of IVT magnitude within the candidate. The first criterion can effectively eliminate zonally-oriented tropical moisture filaments (Fig. 3d), while retaining water vapor transport corridors in the tropics that exhibit characteristics consistent with ARs, such as the strong water vapor transport belts embedded in the East Asian summer monsoon circulation (Kamae et al., 2017; Yang et al., 2018; Fig. 3e) and the tropical moisture exports over the North Pacific (Knippertz and Wernli, 2010). The second criterion is applied globally but is particularly useful for mid- and high-latitude candidates, where moisture transport structures are often affected by complex synoptic conditions. This criterion emphasizes the presence of poleward-directed moisture transport structures within individual candidates, rather than requiring all grids or local portions of a candidate to exhibit poleward transport. This flexible design helps avoid the erroneous exclusion of some ARs that include zonally oriented extensions (Fig. 2) or locally equatorward portions. As a supplement, the second half of this criterion accounts for a few large-scale candidates whose poleward transport along the axes may be locally weak (Fig. 3f); this supplementary condition helps retain such AR cases when they still exhibit a positive object-scale poleward IVT tendency and a sufficiently large spatial scale. As with the morphological criteria, the specific parameter thresholds in these criteria are based on prior research, combined with extensive visual inspection and sensitivity tests.

When the AR candidate satisfies all the above criteria, the algorithm identifies it as an AR and archives the associated data. Subsequently, a new AR candidate is extracted from the IVT background field in Step 1, and its main characteristics are quantified and diagnosed following the same procedure. If the candidate is excluded due to failing to meet the criteria, the algorithm performs an additional verification of the corresponding region through the threshold adjustment step in the next step.

For an initially excluded AR candidate, the algorithm employs a threshold adjustment step to re-check if water vapor transport structure is consistent with AR characteristics in the interior of the corresponding region, as proposed by Guan et al. (2018). Specifically, based on the Level-0 IVT threshold fields at the 85th percentile combined with the hemispheric threshold lower limit (see Step 1), the algorithm additionally establishes four sets of ascending threshold fields, corresponding to the combinations of the 87.5th, 90th, 92.5th, and 95th percentile base thresholds, denoted as Level-1 to Level-4 in the threshold adjustment step. For an AR candidate initially excluded, the IVT values of each grid within the region are compared against the corresponding thresholds in the Level-1 threshold field. Grids with IVT values exceeding this threshold are re-registered as unvisited marked grids in the IVT background field, and retained for extraction as the potential new AR candidate in Step 1. Concurrently, the algorithm records the threshold-level index of the threshold field used to extract the current candidate. If the candidate fails to satisfy the diagnostic criteria, the procedure is repeated using the threshold field of the next level. If a candidate extracted from the Level-4 threshold field still does not meet the criteria, the verification process for that region is concluded.

The threshold adjustment step modifies the IVT threshold to regulate the spatial extent of AR candidates, with two primary functions. First, for some overly broad AR candidates that fail to meet the length–width ratio criterion, their core regions obtained after spatial contraction may satisfy the criterion, consistent with the sensitivity of AR identification to IVT thresholds and geometric constraints discussed by Reid et al. (2020). This scenario predominantly occurs over moisture-rich oceanic regions, where broad candidates with loose peripheral structures are more likely to be extracted. Second, certain AR candidates may consist of multiple water vapor transport systems. When merged candidates exhibit clear discontinuities in water vapor transport, the threshold adjustment step enables their separation. For example, Fig. 4a illustrates an AR candidate initially extracted based on the Level-0 IVT threshold field, which would be excluded due to its failure to meet the diagnostic criterion based on the curvature metric derived from the sum of turning angles. Through the threshold adjustment step, the constituent tropical cyclone and AR are separated, thereby preventing the erroneous removal of the latter. Furthermore, this step also alters the curvature metric of the tropical cyclone component due to changes in the region (Fig. 4b). If the original criterion was still applied, it might be incorrectly retained. Therefore, the algorithm dynamically adjusts the relevant criterion based on the current IVT threshold-level index. In fact, the aforementioned two scenarios are inherent limitations of using threshold to extract AR candidates, but these can be effectively improved by incorporating a threshold adjustment step.

This study employed the IVT dataset derived from ERA5 reanalysis data as input, with a remapped horizontal resolution of 1° × 1° at a time interval of 6 h (00:00, 06:00, 12:00, 18:00 UTC). Applying the aforementioned AR identification algorithm, a global AR database spanning from 1940 to 2024 with a 6 h time interval is constructed. The database is constructed as a three-dimensional spatiotemporal array that records AR occurrences on a two-dimensional grid for each timestep. At every timestep, individual ARs are sequentially assigned an integer label starting from 1, which are mapped onto the corresponding grids within their occurrence regions, while non-AR regions are assigned a value of 0. The combination of timestamp and integer label provides a unique identifier for each AR, facilitating the direct extraction of data for individual ARs in practical applications, as well as further tracking of their complete life cycles by analyzing spatial overlaps between consecutive timesteps (Zhou et al., 2018).

To evaluate the performance of the algorithm and the AR database (hereinafter referred to as DB1) it constructed, this study conducted a comparative analysis between DB1 and two existing global AR databases (hereinafter referred to as DB2 and DB3). DB2 was developed by Guan and Waliser (2024) using the Tracking Atmospheric Rivers Globally as Elongated Targets (tARget) algorithm with ERA5-derived IVT data as input, and this algorithm has been comprehensively evaluated in previous studies. DB3 was generated by Wang et al. (2024) through a data fusion algorithm integrating 12 independently developed AR datasets from both global and regional identification efforts, thereby providing a broadly representative synthesis of AR identification results. Methodologically, DB2 is closer to DB1 than is DB3, because both DB1 and DB2 are generated by object-based AR identification algorithms that first extract connected regions of enhanced IVT and then apply a series of diagnostic criteria to assess the extracted candidates. The algorithm used to construct DB1 generally follows the IVT threshold strategy used for DB2, and the two algorithms share the same requirements for candidate length and length–width ratio. Their main differences include characteristic axis construction, candidate quantification and diagnostic screening. In particular, DB1 employs a dual-axis test, whereas DB2 uses a single-axis representation. DB2 also uses the area-weighted standard deviation of IVT direction to exclude candidates with strongly inconsistent transport directions. By comparison, DB1 characterizes the curvature-related properties of candidates using two axis-based metrics and evaluates them against the corresponding diagnostic criteria. DB1 further incorporates a tropical-specific criterion for assessing poleward moisture transport in AR candidates. By contrast, DB3 is a fusion product of multiple AR datasets rather than the output of a single identification algorithm; consequently, the specific settings used, including IVT thresholds, methods for quantifying candidate characteristics, and diagnostic requirements, are not uniform.

A common period from 1980 to 2016 is selected for analysis, during which the total number of ARs and their latitudinal distribution are calculated for all three databases. The number of ARs refers to the cumulative AR snap counts across all timesteps (00:00, 06:00, 12:00, 18:00 UTC), and the location of each AR is determined based on its area distribution in different latitude bands. This study further quantifies the number of ARs identified consistently in DB1 and DB2 (DB3). Specifically, for each AR in DB1, the longer axis (Axis₁ versus Axis₂) is mapped onto the grid of the same timestep in DB2 (DB3). If more than 1000 km of the axis lies within an AR region in the target database, these two ARs are considered consistent. This criterion is designed to assess spatial overlap in core structural regions, ensuring both computational efficiency and robustness. As shown in Table 1, all three databases exhibit consistent spatial patterns, indicating that ARs predominantly occur in the mid-latitudes, with higher frequency observed in the Southern Hemisphere than in the Northern Hemisphere. However, discrepancies exist in the absolute counts across different databases. Overall, the counts in DB1 and DB2 are comparable and exceed those in DB3. In terms of identification agreement, DB1 captures the majority of ARs present in DB2 and DB3, achieving the highest consistency rate in the mid-latitudes despite the largest sample size there.

To further evaluate the potential differences in ARs among DB1, DB2, and DB3, as well as the characteristics of AR frequency distribution in DB1, this study separately calculated the occurrence frequencies of ARs in all three databases during the period from 1980 to 2016 for the northern wintertime (November to March), northern summertime (May to September), and the entire year. The occurrence frequency at each grid is determined by dividing the number of time steps identified with ARs by the total number of time steps. Figure 5a–c shows the spatial distribution of AR frequency in DB1 for wintertime and summertime, respectively. ARs exhibit the highest occurrence frequency in mid-latitude ocean basins, possibly due to frequent extratropical storms and relatively abundant moistures (Eiras-Barca et al., 2018; Guan and Waliser, 2019). Further, the initial activity positions are usually located closer to the equator in the wintertime hemisphere and farther poleward in the summertime hemisphere. This spatial pattern and seasonal evolution of AR frequency distribution is highly consistent with storm tracks (Bengtsson et al., 2006; Priestley and Catto, 2022). Specifically, ARs exhibit a southwest-northeast direction under the background of the Northern Hemisphere westerly circulation. Therefore, ARs have a more direct and intuitive impact on the west coasts of the continents especially in the northern winter (Lavers and Villarini, 2015; Mahoney et al., 2016). Notably, seasonal expansion of the Western North Pacific Subtropical High (WNPSH) builds a moisture transport pathway over East Asia (Kwon et al., 2024; Park et al., 2021). Figure 5d–i illustrates the differences in AR occurrence frequency between DB1 and DB2, and between DB1 and DB3. The discrepancies between DB1 and DB2 are mainly evident in tropical and mid-latitude regions: DB1 shows slightly lower frequency in the tropics but slightly higher frequency in the mid-latitudes compared to DB2. Relative to DB3, DB1 exhibits higher frequency over continents and high-latitude regions, which are typically characterized by low moisture content, while showing close agreement in the tropics and in mid-latitude ocean basins where ARs are most active.

The discrepancies in AR identification across the three databases arise from multiple factors. A critical determinant is the threshold applied to the magnitude of the detection variable in the underlying identification methods, particularly its lower limit, which governs the number of ARs identified in lower-moisture regions. In low-moisture continental and high-latitude regions, the DB3 fusion is based primarily on constituent datasets with global coverage. Because these datasets generally adopt fixed IVT threshold lower limits that are higher than the hemispheric lower limits used in DB1 and DB2, DB3 identifies significantly fewer ARs in these regions. In contrast, over mid-latitude ocean basins where moisture is abundant and AR activity is frequent, the threshold values used by the corresponding methods are comparable, resulting in minimal differences in AR counts among the databases. The statistical discrepancies in AR counts between DB1 and DB2 mainly stem from differing diagnostic criteria for the poleward water vapor transport characteristics of AR candidates. The algorithm used in this study applies stricter filtering of tropical moisture filaments, leading to fewer identified ARs in tropical regions in DB1 than DB2. Conversely, in the mid-latitudes, the algorithm adopts less strict criteria for zonally-oriented ARs, yielding a slightly higher AR count in DB1. Furthermore, variations exist across algorithms in the quantification methods and diagnostic criteria for the morphological characteristics of candidates, as well as in the spatial resolution of input data. These technical differences all affect the specific AR identification results; for instance, DB1 excludes the AR candidates shown in Fig. 3a, c and d while retaining the candidate in Fig. 3f, compared to other databases.

To assess the feasibility of DB1 in tracking the complete life cycle of ARs and to further understand the dynamic evolution characteristics, this study selects two weather extreme events with broad impacts for case analysis. The first case is the East Asian Meiyu rainfalls in late June 2018, which successively affected eastern China, the Korean Peninsula, and Japan, with extreme precipitation triggering severe disasters in wide areas of East Asia (Tsuguti et al., 2019). The second case is the Australian forest wildfires in late January 2020 (Black Summer), when the wildfires were gradually brought under control and were followed by persistent precipitation in early February, leading to multiple cascading hazards (Kemter et al., 2021).

The Meiyu is a seasonal rainy period that occurs as the East Asian summer monsoon advances northward into the subtropical region of East Asia, typically during June and July (Ding et al., 2020). The Meiyu season is characterized by persistent precipitation, and Wang et al. (2021) showed that AR is an important component of Meiyu system, exerting significant impacts on Meiyu rainfalls. In late June 2018, the Meiyu rainfalls were influenced by interactions of multi-scale weather systems, including the WNPSH, extratropical cyclone, and Typhoon Prapiroon, exhibiting a long duration and high precipitation intensity. Figure 6 illustrates the evolution of the AR during this period. Initially, the AR was embedded within the monsoon flow and extended along the WNPSH edge in a northeast-southwest orientation, continuously transporting the warm, moist air northward from the Bay of Bengal, the South China Sea, and western Pacific (Fig. 6a). As the extratropical cyclone located northwest of the AR downstream slowly moved eastward, its trailing cold air mass advanced southward and converged with the warm, moist airflow in the central region of the AR, generating a low-level cyclonic shear. This interaction intensified moisture convergence and strengthened moisture transport in the midsection of the AR; as the cyclone continued moving eastward, the water vapor transport was enhanced and shifted downstream (Fig. 6b, c). Concurrently, Typhoon Prapiroon in the south developed and moved poleward, while the WNPSH gradually shifted northward, guiding the AR to reorient into an east-west direction. Subsequently, as the extratropical cyclone moved offshore, its influence on the AR diminished. The AR stabilized along the northwestern periphery of the WNPSH (Fig. 6d, e), maintaining a quasi-stationary state until the mature Typhoon Prapiroon approached, disrupting the original circulation and terminating the AR (Fig. 6f). This process, combined with the subsequent typhoon landfall, contributed to the “Heavy Rain Event of July 2018” in Japan (Sayama et al., 2020). In summary, this AR was primarily governed by the WNPSH and modulated by both the extratropical cyclone and the typhoon, exhibiting pronounced quasi-stationary feature that distinguishes it from typical ARs. Figure 6 also presents the AR identification results from DB2. While generally consistent with DB1, discrepancies are evident in Fig. 6a, f. Specifically, no AR is identified in DB2 in Fig. 6a. In Fig. 6f, the AR identified in DB2 includes the southern Typhoon Prapiroon, whereas the result in DB1 excludes it; this distinction arises because the Axis₁ used in DB1 effectively captures the internal vortex structure, enabling its separation from the AR through the threshold adjustment step.

Australia is a typical monsoon-controlled region, and Asian and Australian monsoon is usually viewed as a unified system (Cheng et al., 2024). The weather and climate in the Asian and Australian monsoon-controlled regions are influenced by large-scale climate variability across various time scale (Lim et al., 2019; Risbey et al., 2009). During the 2019–2020 austral Black Summer, a series of uncontrollable forest wildfires broke out across Australia. The extreme drought occurred under the strong positive Indian Ocean Dipole (IOD) and the stratospheric disturbance was possibly an important contributing factor (Rao et al., 2020). In late January 2020, a well-developed AR traversed Australia, delivering substantial moist air that effectively suppressed the wildfires and creating conditions for extinguishing the wildfires. Figure 7 shows the evolution of the ARs in January 2020 for several snaps at the end of the Australian wildfire. Initially, a narrow, zonal AR was present over the southwestern seas of Australia, moving eastward at a relatively low speed, while a tropical cyclone was moving inland from northern Australia (Fig. 7a). Once landing, the tropical cyclone gradually weakened, and its residual circulation merged with the upstream portion of the AR, causing changes in the upstream circulation and forming a trough-ridge dipole system on both sides of the AR (Fig. 7b). Subsequently, a substantial amount of tropical moisture was transported along the AR, traversing Australia in a meridional direction (Fig. 7c, d). Concurrently, cold air from the southwest converged with poleward-flowing warm, moist tropical air on the western flank of the AR, where the cyclic shear zone developed, which in turn fed the upstream AR. Thereafter, the entire system continued to move eastward and left Australia; under the influence of a northern anticyclone, the AR reoriented into an east-west direction and persisted along its southern periphery (Fig. 7e, f). Overall, the most prominent feature of this AR is the direct influence from the tropical cyclone residual circulation, which led to deformation of the upstream circulation and the AR orientation. As a consequence, abundant moisture was transported across Australia, helping to alleviate the severe wildfire activity that had lasted for months. Similar to Fig. 6, the AR identification results in Fig. 7 are largely consistent between DB1 and DB2, with the exception of Fig. 7a. In Fig. 7a, an AR is identified in DB2 within the northern tropical cyclone, whereas it is excluded in DB1; this exclusion is attributed to the diagnostic criterion for the axis straightness ratio employed in our algorithm, which effectively filters out such small-scale AR candidates with incomplete vortex structures.