1. Tropical cyclone translation speed and stalls
Tropical cyclones (TCs) or their remnants have produced the seven costliest billion-dollar disasters in the United States as of mid-2022 (National Climatic Data Center 2022). The immense damage produced by TCs arises from a variety of hazards. The most damaging aspect of Hurricane Katrina (2005) was storm surge and associated levee breaches/failures (Link 2010), while the damage produced by Hurricane Andrew (1992) resulted from strong winds (Willoughby and Black 1996). Other damages result from the extended duration of a storm as it stalls or remains nearly stationary over a region. For example, Hurricane Harvey (2017) in southeast Texas was associated with unremitting rainfall caused, in part, by the storm’s unusually slow motion (Emanuel 2017; Risser and Wehner 2017; van Oldenborgh et al. 2017). Five-day rainfall totals exceeded 1270 mm (50 in.) at several locations (Blake and Zelinsky 2018)—an amount nearly equal to the climatological annual average for the region. Perhaps even more remarkable than the historic point precipitation totals was the geographical extent of the rainfall. Recent estimates show that in 120 h, Hurricane Harvey produced roughly 572 mm (22.5 in.) of rainfall across 90 650 km2 (35 000 mi2), an area exceeding the size of South Carolina (Applied Weather Associates, LLC 2019).
Hurricane Harvey was not the first time that the slow motion/recurvature of a TC greatly increased its destructive potential. In September 1950, Hurricane Easy reached category 3 intensity on the Saffir–Simpson hurricane wind scale as it approached northwestern Florida (National Weather Service 2023). The TC completed two loops over north and central Florida and produced a 72-h rainfall total of 1148 mm (45.2 in.) for over 25.9 km2 (10 mi2) (Schreiner and Riedel 1978). Rainfall totals reached 983 mm (38.7 in.) in Yankeetown, Florida, in 24 h, breaking the then-current 24-h rainfall record for the United States (Norton 1951).
Recently, North Atlantic basin stalling TCs have been studied (Hall and Kossin 2019), but many fundamental aspects of stalling storms, such as their geographical and seasonal distribution, are unknown. Additionally, risks from slow-moving TCs are likely changing as communities grow along the U.S. Gulf of Mexico and Atlantic coastal zones; the coastal population has increased by 7.7 million since 2000 to 59.6 million by 2016 (Cohen 2018). The Gulf Coast is the fastest-growing coastal region in the United States. Harris County, Texas, including Harvey-ravaged Houston, observed a population increase of 1.2 million people from 2000 to 2016, the largest gain of any U.S. county. In Latin America and the Caribbean, the population is expected to grow past the middle of the century, potentially placing more people at risk of stalling TCs (Ham-Chande and Nava-Bolaños 2019). Additionally, TCs are expected to produce more intense precipitation because of increased atmospheric water vapor as a response to climate warming and stronger upward motion in storms (Emanuel 2017; Risser and Wehner 2017; van Oldenborgh et al. 2017), and TC tracks that recurve or loop over a region can cause multiple landfalls and increase rainfall totals (Patricola 2018). A global analysis found a 10% reduction in translation speed from 1949 to 2016, with an even greater decrease (17%) in the North Atlantic (Vecchi and Soden 2007; Kossin 2018; Hall and Kossin 2019). This contributed to a statistically significant increase in coastal annual-mean rainfall in the second half of the twentieth century and into the twenty-first century (Hall and Kossin 2019).
Studies claim decreasing translation speed can be attributed to poor low-latitude presatellite observations, with average translation speed affected by greater postsatellite TC counts in low latitudes where translation speed is slower (Chan 2019; Lanzante 2019; Moon et al. 2019). Approximately 30% of the global slowdown mentioned above was associated with variations in TC frequency across basins, and a long-term decreasing trend in translation speed over the North Atlantic (−0.14 m s−1 decade−1) exists, where satellite inhomogeneities are less problematic (Kossin 2019). In coastal China, translation speed decreased by 0.11 m s−1 decade−1 from 1961 to 2017 (Lai et al. 2020), and in South China, a decrease of 0.3 m s−1 decade−1 has been reported (Wu et al. 2022). In the western North Pacific (WNP), a historical slowdown in midlatitude translation speeds in September has been associated with climate change and multidecadal variability (Yamaguchi and Maeda 2022). The decline in high-latitude autumn TCs in the WNP was attributed to natural variability (Wang et al. 2021), and an enhancement of low-latitude translation speed and steering flow has been observed over the past couple of decades (Gong et al. 2022).
Whether trends in translation speeds are a manifestation of climate change or data inhomogeneities or some combination of both, the threat of a stalling TC at or near coastal populations remains. Potential risk will increase with continued sea level rise (Sweet et al. 2022), population migration (Cohen 2018), and increasing sea surface temperatures in a warmer climate (Johnson and Lyman 2020). The impact of stalled TCs could be further exacerbated by an increase in the frequency of intense TCs, as suggested by maximum potential intensity theory and recent modeling studies (Emanuel 1987; Knutson and Tuleya 2004; Hill and Lackman 2011; Walsh et al. 2015; Patricola 2018; Trenberth et al. 2018). A specific quantitative metric to evaluate stalled TCs (i.e., what constitutes a stall) is provided by Hall and Kossin (2019). They examined trends in stall frequency (and associated precipitation) and found an annual increase of 0.7–1.2 yr−1 from 1948 to 2017; however, reasons for the observed changes and spatial patterns were not addressed. Here, we provide a complete spatiotemporal climatology of a specific set of stalling storms. An assessment of temporal trends in stall frequency is also provided while considering whether the storms are related to the month of occurrence or varying intensity metrics. We consider multistalling storms and those close to coastlines to aid emergency preparedness.
2. Data
TC data are taken from the National Hurricane Center’s “best track” hurricane database (HURDAT2) (Landsea and Franklin 2013), dated 29 May 2021 (available at https://www.nhc.noaa.gov/data/hurdat/hurdat2-1851-2020-052921.txt). HURDAT2 includes 6-h data on the location and maximum sustained winds of all known TCs in the North Atlantic from 1851 to 2020. Initially, all TCs listed in HURDAT2 were used to calculate corrals (i.e., circles) where TC points are confined (i.e., encapsulated); however, implausibly stationary TC tracks were present in HURDAT2 before 1900. For example, the smallest calculated corral occurred in 1862 with an estimated 3-day radius of 16 km. The next smallest stall had a 61-km radius and occurred with Hurricane Gilda in 1973. Thus, this analysis is restricted to 1900–2020 because of the known limitations with the earliest period in HURDAT2 [e.g., undercounting weaker storms before 1953 (Hagen and Landsea 2012)]. To test the sensitivity of the results pertaining to the introduction and improvement of storm detection and tracking, we subset the analysis to data post-1966 and compare results to the longer period of 1900–2020.
3. Analyzing stalls
a. Stall definition
Here, a corral is defined as the smallest possible circle (i.e., geographical area) containing all storm center locations (six-hourly position points) over 72 h. Note that a tropical system traveling in a straight line at an average translation speed of 20 km h−1 would travel 1440 km in 72 h and have a corral radius of 720 km. Stalls, as defined here, are TCs with a minimum corral radius of ≤200 km based on their 6-hourly track locations for 72 h. No minimum wind speed threshold was applied to the corrals (e.g., tropical depressions and extratropical storms are included). While some analyses here consider the minimum stall per event, we also consider the number of multiple stalls along one TC track. A multistall event must fit two criteria: The additional stall cannot overlap with the first stall zone, and the storm must have had an intervening corral radius greater than 200 km.
Hurricane Harvey’s track is depicted in Fig. 1. The pink buffer shows Harvey’s smallest circular stall region, or corral (123 km radius), which occurred between 0000 UTC 26 August and 0000 UTC 29 August. The dotted line (circle) is the defined maximum radius for a stall (i.e., 200 km).
Hurricane Harvey’s track and corral method. The pink buffer over Houston, Texas, is Harvey’s observed minimum stall zone with a radius of 123 km. The larger dashed circle denotes the 200-km radius used as the corral size definition in this study.
Citation: Journal of Applied Meteorology and Climatology 63, 11; 10.1175/JAMC-D-23-0229.1
Hall and Kossin (2019) use circles to predefine coastal impact regions and consider storm residence time within those circles. Here, circles are calculated from the tracks of individual TCs focused on residence time in any area and then applied to coastal zones separately. We explore the results using different stall definitions in Table 1. The 72-h, 200-km radial definition leads to 191 individual TCs (15% of all TCs in HURDAT2). The Hall and Kossin (2019) definition (i.e., 48 h, 200 km) leads to 491 TCs if applied throughout the North Atlantic basin rather than just along coasts. The distribution of minimum corral sizes at 72 h for all TCs is shown in Fig. 2a, with the 200-km line marked in red. The 200-km size was chosen because it yielded a large enough sample size for statistical estimation and because a radius > 200 km begins to lose applicability for emergency preparedness purposes for a given region.
Stall duration and TC counts. TC counts for varying stall time periods and corral sizes from 1900 to 2020.
Corral distribution. (a) Distribution of minimum 72-h corral sizes for all TCs from 1900 to 2020. Bins are made for every 50-km increment. A red vertical line is placed at 200 km to denote the stall definition. (b) Density plot of minimum corral sizes for all TCs 1900–2020.
Citation: Journal of Applied Meteorology and Climatology 63, 11; 10.1175/JAMC-D-23-0229.1
b. Statistical estimation
Poisson regression is used to determine whether trends exist in the annual frequency of stalling TCs in the North Atlantic basin (Elsner and Jagger 2013). Two periods are considered for analysis: post-1900 and post-1966, with the latter period when TC instances and tracks are more reliably estimated by satellite (Landsea 2007). There is also a bias due to weak and short-lived storms within the record. Landsea et al. (2010) found that the occurrence of short-lived storms (≤2 days) increased after the turn of the twenty-first century, likely related to our increased capacity to observe these storms. Given the imposed duration requirement, our 72-h definition is less sensitive to short-lived storms.
Clusters were identified using the hierarchical density-based spatial clustering of applications with noise (HDBSCAN) algorithm, an unsupervised machine learning algorithm introduced by Campello et al. (2013) as an improvement on the density-based spatial clustering of applications with noise (DBSCAN) algorithm (Ester et al. 2013). Density-based clustering is a nonparametric approach that identifies groups or clusters and is suitable for spatial data (Kriegel et al. 2011). HDBSCAN has been used in meteorological research concerning extreme precipitation (Zhang et al. 2019) and lightning-induced wildfires (Coogan et al. 2022), and it has also been implemented in archeology, criminology, and aviation research (Butt et al. 2021; Golden et al. 2021; Kumar et al. 2021). For a detailed review of the algorithm, please see McInnes et al. (2017) and Stewart and Al-Khassaweneh (2022). The algorithm creates a density surface based on the “mutual reachability distance,” obtained from the distances between points and their kth-nearest neighbor. The created density surface generates a hierarchy to identify clusters. Points not contained within a cluster are labeled as noise. Finally, clusters are analyzed for stability, with the highest levels of stability belonging to the most prominent clusters.
4. Climatology and trends of stalling tropical cyclones
a. Spatiotemporal distribution
From 1900 to 2020, 1274 North Atlantic TCs had tracks that lasted at least 72 h, according to the HURDAT2 database, of which 191 TCs (15%) met the 200-km stall definition. The 191 event tracks are shown in Fig. 3 and color coded by intensity for hourly segments using spline interpolation to estimate hourly position (Elsner and Jagger 2013). Tracks of stalling TCs span the whole of the North Atlantic basin.
Stalling TC tracks. The track history for stalling TCs with colors based on wind intensity (kt) according to the Saffir–Simpson hurricane wind scale and coded at each hourly segment (1900–2020).
Citation: Journal of Applied Meteorology and Climatology 63, 11; 10.1175/JAMC-D-23-0229.1
Minimum stall event types include tropical depressions (n = 34; almost all occurring since 1960), tropical storms (n = 87), hurricanes (n = 47), extratropical storms (n = 9), and other types (e.g., subtropical storms; n = 14). The minimum corral radius for all TCs since 1900 occurred with Tropical Storm Gilda in 1973 at 61 km, followed by Hurricane Bonnie at 70 km in 1992. Gilda brought heavy rainfall to Jamaica for 3 days, leading to landslides across eastern Jamaica (Evans 1975). The slowdown resulted from nearly balanced upper-level and lower-level steering currents (Hebert and Frank 1974). After passing near Jamaica and over Cuba, Gilda stalled again and crossed over the Bahamas, moving only 6 mi (9.7 km) in 24 h. Gilda remained nearly stationary for another 24 h before moving northward and transitioning into a subtropical cyclone (the first to do this in the known record) (Hebert and Frank 1974). Two days later, the storm began moving northeastward under the influence of an upper-level trough moving east of the United States (Hebert and Frank 1974). Hurricane Bonnie stalled in the middle of the central Atlantic because it was embedded in a larger-scale deep cyclonic circulation at higher latitudes. The storm slowly tracked toward the Azores without damaging the islands (Mayfield et al. 1994).
Ten of the 191 TCs experienced multiple stalling periods. Table 2 shows the 10 storms that experienced more than one stall with minimum and additional stall zones listed. Hurricane Ginger in 1971 experienced four separate stall periods as the second-longest-lived Atlantic hurricane in the known record.
Multistalling storms. List of 10 multistalling TC storms.
A spatiotemporal snapshot of all stalls per TC from 1900 to 2020 is shown in Fig. 4a, with the stall color coded by month and average intensity during the stall by circle size (graduated symbol). Panels include early-season stalls (May–August; Fig. 4b), midseason stalls (September; Fig. 4c), and late-season stalls (October–December; Fig. 4d). Stalls in June (n = 11) and July (n = 14) are scattered and infrequent throughout the basin. They also exhibit generally weak intensities during the stalls (i.e., smaller circles). All stalls in August (n = 24) occurred poleward of 24.5°N latitude. Stalls are more frequent in September (n = 71) and October (n = 55). Stalled storms occur throughout the entire basin in September and October, but October has more stalled storms in the Caribbean Sea than any other month. The intensity of some October stalls reached major hurricane status (i.e., 50 m s−1), further exacerbating the risk of severe damage during a stall. For example, Hurricane Mitch in 1998 (corral size: 121 km) was a devastating, slow-moving October storm that stalled 17.2 km away from the Honduran coast. Mitch was the second deadliest hurricane on record in the North Atlantic basin (11 374 deaths), second only to the Great Hurricane of 1780 that moved through the Lesser Antilles islands in October (the exact track is unknown) (Britannica 2021). Stall frequency decreases dramatically by November (n = 18), with most stalls located in the Caribbean. Historically, storms have not stalled near Hispaniola, Puerto Rico, the Virgin Islands, or along the immediate coastal region from Miami, Florida, to Charleston, South Carolina.
Spatiotemporal map of stalling TCs. All TC stall locations by month (color) and average intensity during the stall (m s−1) (graduated symbol) from 1900 to 2020. The stalls are separated into (a) all stalls and three seasonal periods: (b) early season, (c) midseason, and (d) late season.
Citation: Journal of Applied Meteorology and Climatology 63, 11; 10.1175/JAMC-D-23-0229.1
Table 3 compares the monthly frequency distribution of stalling TCs to the full HURDAT2 population starting in 1900. For direct comparison, the counts of all HURDAT2 storms include the month the storm experienced its smallest corral over 72 h. The monthly distribution is similar between the two sets. In the HURDAT2 record, there are more TCs in August and fewer stalls (n = 294; HURDAT2 = 270; stalls = 24) than in October (n = 289; HURDAT2 = 239; stalls = 50), and the proportion of TCs that stall in October is 17.3% compared to the 8.2% that stall in August. This further highlights the increased likelihood of a stalling event as the season progresses. The highest proportion of storms that stall occurred in December (n = 15; HURDAT2 = 11; stalls = 4; 33%).
Monthly distribution. Months with the number of storms from HURDAT2 without stalling storms and months with stalling TCs (1900–2020). Stalling storms indicate the number of individual storms, and stalls include any multistalling counts.
To compare the approach offered by Hall and Kossin (2019) (i.e., ≤200 km radial movement in 48 h) with the approach here (i.e., ≤200 km radial movement in 72 h), we provide Figs. 5a and 5b, respectively. These panels show the stall location across three separate time bins and their proximity to the coast. Stalls within 200 km of a coastal zone are given an “x” symbol. Hall and Kossin (2019) considered the coasts of the continental United States, Mexico, and Belize; here, we add the remainder of the Caribbean coasts of central and South America, and the coasts of Cuba, Hispaniola, Puerto Rico, and the arc of the windward and leeward islands, collectively “major coasts.” The three time bins represent the presatellite era (1990–66), the early satellite era (1967–94), and the most recent satellite period (1995–2020). The more conservative, 72-h definition showcases similar patterns as the definition used in Hall and Kossin (2019), with one crucial exception: Their definition has stalls north of 40°, including those in the southern margin of the Sargasso Sea, whereas ours does not. We also provide unique insight into the coastal areas more likely to experience TCs that linger for longer. Based on the periods, more storms stalled near the coasts in the most recent satellite period, which could be a product of observational bias; however, in the first period (1900–66), stalls occurred in the central Gulf of Mexico and Caribbean Sea away from the coast, suggesting there may be a slight trend toward stalls occurring closer to the coast. This manuscript does not explore the potential causes, but further insight into coastal stalls is provided below.
Comparison of two approaches. (a) Location of stalls defined as 48-h periods with a radius of 200 km according to Hall and Kossin (2019) and (b) stalls defined as 72-h periods with a radius of 200 km (including multistalls). Colors are based on periods and given x symbols if they are within 200 km of a coast. Symbol size is based on intensity (m s−1).
Citation: Journal of Applied Meteorology and Climatology 63, 11; 10.1175/JAMC-D-23-0229.1
From 2000 to 2020, 60 TCs stalled in the North Atlantic; one of the most devastating was Hurricane Harvey (2017). The mean total life cycle duration for all stalling storms was nearly 11 days (i.e., 264 h or 43.9 observations every 6 h). Of storms that met the stall definition, Hurricane Otto in 2016 had the longest duration stall period (7.5 days) and started as a tropical depression and finished its stall as a 46.3 m s−1 hurricane. The stall occurred near 10.9°N, making it the second most southerly stall in the record. Three other storms had stalls lasting 7 days: Roxanne (1995) in the Bay of Campeche and two others in the open Atlantic (1915 and 1965).
b. Coastal zones
Stalling TCs close to a coastline pose more immediate threats to people and infrastructure. The coastal zone is defined here as a 200-km buffer around either side of the coastline because average precipitation starts increasing substantially within 200 km of the storm center (Lonfat et al. 2004).
Potentially most devastating are those stalls within 200 km of the coast, but even more so, those that stall over or partially over land. Of the 191 stalling storms with 204 total stall periods, 17 occurred within 200 km of the coast over land. Alice (1953) and Ginger (1971) stalled over land, but these were not their minimum stall points. Table 4 shows the 17 storms and their inland distance calculated from the nearest proximity to a coastal zone. Tropical Storm Allison in 2001 stalled at the farthest inland distance near Lufkin, Texas, before heading south into the Gulf of Mexico and making a second landfall near Morgan City, Louisiana.
Inland stalling storms. List of inland stalling TCs within 200 km of a coastal zone and over land, including name, year, month, intensity (m s−1), stall radius (km), and inland distance (km) from the nearest coast.
Figure 6 shows all TC stalls and the proximity to coastal zones. Colors denote proximity to the coast, and the size of the points represents wind intensity according to the Saffir–Simpson hurricane wind scale. Those with the highest intensity and closest coastal stalls are within the western Caribbean Sea. Figure 7a shows the total number of storms with corral centers within 200 km of any given location. Figure 7b shows the same information for those storms whose corral radius is less than 200 km. It is a measure of overall storm motion rather than stalls since stalls are defined relative to the minimum corral size. This graphic highlights areas most at risk for storms that move slowly for several days, suggesting they will likely experience extended exposure to storm characteristics (e.g., rainfall, winds, and storm surge). Persistent, slowly moving storms are rare in the main development region because, even though many TCs occur here, they tend to move quickly. A noticeable enhancement in frequency around 30°N aligns with the 700 mb (1 mb = 1 hPa) wind climatology during the peak of the TC season, possibly as a marker of the transition zone between the easterlies and the westerlies. Interestingly, the winds are not similarly calm over the Caribbean later in the season, though they weaken. This may account for the fewer stalls at higher latitudes in the later months of hurricane season. The slowest-moving TCs included in the study occur near the central Gulf of Mexico coast, the area surrounding the Carolinas, and the western Caribbean Sea. Figure 8 shows the 10th-percentile corral size for all storms within 200 km of any given location. The average corral radius is used for storms with more than one six-hourly HURDAT2 location within that distance. The geographical pattern of frequency of slow-moving storms (Fig. 7) is similar to that of the 10th-percentile corrals, but Fig. 8 also shows that more than 10% of all storms passing through the southwestern Gulf of Mexico and southwestern Caribbean will stall there.
Stalling coastal TCs. Intensity and location of stalls, color coded by distance from the coasts of North and central America and the islands defining the margin of the Caribbean Sea. Symbol sizes vary continuously with intensity. Sizes shown in the legend are examples.
Citation: Journal of Applied Meteorology and Climatology 63, 11; 10.1175/JAMC-D-23-0229.1
Corral centers. (a) The total number of storms with corral centers within 200 km of any given location, 1900–2020. (b) The total number of storms with corral centers within 200 km of any given location, 1900–2020, whose corral radius is smaller than 200 km.
Citation: Journal of Applied Meteorology and Climatology 63, 11; 10.1175/JAMC-D-23-0229.1
10th-percentile corral size. The 10th-percentile corral size for all storms within 200 km of any given location, 1900–2020. If a storm had multiple corral centers within 200 km, the average corral radius is used.
Citation: Journal of Applied Meteorology and Climatology 63, 11; 10.1175/JAMC-D-23-0229.1
Figure 9 shows the cumulative count (Fig. 9a) and fraction (Fig. 9b) of all stalls in relation to the distance from a major coastline. To investigate temporal trends in TC behavior, the focus is only on those storms considered a tropical depression (TD), tropical storm (TS), or hurricane (i.e., all extratropical and subtropical systems were removed). The TCs are separated into 30-yr groups representing varying climatological periods. They are further shown through all wind strengths combined and those that are at least 17 m s−1 or TS strength. The increase in stalls from 1931–60 to 1961–90 is fully attributable to the increased tracking of non-TS-strength storms in HURDAT2 (i.e., when comparing the solid and dashed lines in Fig. 9a). In Fig. 9b, the fraction of nearshore stalls has stayed constant, and the increase in nearshore stalls in Fig. 9a is fully matched by an increase in offshore stalls. The fraction of stalls from 300 to 500 km offshore is large during the presatellite era (Fig. 9b), possibly as storms could be detected onshore by clouds and increased winds, but exact distances to shore were unknown and, likely, too far away for the position to be accurately triangulated using varying coastal wind directions. The fraction of stalls over 500 km offshore has increased, possibly due to better satellite data allowing clearer insight into storm tracks far from sustained in situ observations. The solid, black line in Fig. 9b is included to show that nearshore storms post-1960 are more likely to stall than storms farther offshore. This is important for emergency managers to realize, as those stalling near the coast will have a higher potential to inflict more damage to a population.
Proximity of stalls to major coasts. (a) The cumulative count of TD, TS, and hurricane stalls and their proximity to a major coastline. Different colors separate 30-yr climatology across the record from 1900 to 2020. Solid lines represent all wind strengths. Dashed lines represent storms with greater than 17 m s−1 or TS strength. (b) The cumulative fraction of all stalls and their proximity to a major coastline. The solid black line shows all corrals (stalls and nonstalls) since 1960. The dashed black line shows those at minimum TS strength. Positive values are offshore storms. Negative values are over land.
Citation: Journal of Applied Meteorology and Climatology 63, 11; 10.1175/JAMC-D-23-0229.1
c. Trends and spatial clusters
The annual frequency distribution of all HURDAT2 tracked storms, the annual number of storms, and the smoothed number of stalls as a percentage of tracked storms are displayed in Fig. 10. Ten storms stalled more than once along their tracks, including one triple and one quad-stalling event (Hurricane Ginger in 1971). There are 31 years of data with no stalls recorded, with the five most recent years being 2006, 1991, 1988, 1983, and 1981. One year (2005) had eight individual storm stalls.
HURDAT2 storms and stalls over time. (a) Annual number of tracked storms, annual number of stalls, and smoothed number of stalls as a percentage of tracked storms. (b) Cumulative number of tracked storms and stalls.
Citation: Journal of Applied Meteorology and Climatology 63, 11; 10.1175/JAMC-D-23-0229.1
Poisson regression of annual counts of stalling storms on year using the minimum stall count only is performed for 1900–2020 and 1966–2020. The distribution of the annual number of TC stalls throughout the record closely aligns with a Poisson distribution. Poisson regression is used to estimate trends in North Atlantic TC counts based on the distribution of landfalling storms. Both periods show a positive statistically significant trend. From 1900 to 2020, the regression trend is 1.1% yr−1 increase in the number of stalling storms (p value = 0.04). The regression trend is 1.5% yr−1 if data are limited to the 1966–2020 satellite era (p value = 0.1). A chi-square test on the residual deviance of the satellite-era model yielded significance at the 90% confidence level (p value = 0.1). This result, paired with Fig. 10, suggests that the frequency of stalling storms has modestly increased. This is similar to the results in Hall and Kossin (2019), who also noted an increase in stalling storms over time, particularly near the coast.
HDBSCAN identified five zones for stalling TCs (Fig. 12). We excluded all storms east of 60°W longitude because those storms are unlikely to affect human populations. We determined stall point locations based on the region of the storm center when the stall occurred. The “Find Point Cluster” tool in ArcGIS Pro [Environmental Systems Research Institute (ESRI) 2022] was used to execute the HDBSCAN algorithm. This tool requires two inputs: a point layer and a minimum number of features that constitute a cluster, so TC characteristics such as intensity, size, rainfall rate, or longevity are not considered in determining a cluster. The smallest corral radius location per storm was used. We determined the optimal number of minimum features using the density-based clustering validation (DBCV) index (Moulavi et al. 2014). The DBCV was chosen because it accounts for noise and uses density measures rather than distance measures, which is appropriate for a density-based method. Using 1000 simulations, the optimum number of minimum features per cluster was 8, with a mean DBCV value of 0.24 (median = 0.27; Fig. 11). While DBCV values greater than 0.45 are considered ideal, values greater than 0.20 are considered acceptable (Corporal-Lodangco et al. 2014).
Density-based clustering simulation boxplot. Boxplot of DBCV scores by minimum number of features used in the HDBSCAN algorithm, after 1000 seed iterations.
Citation: Journal of Applied Meteorology and Climatology 63, 11; 10.1175/JAMC-D-23-0229.1
Points are separated into clusters or considered noise. Points are treated as noise if the algorithm determines that a point does not have a significant probability of being a cluster member. This can occur when there are fewer nearby points than the minimum number of features or a minimum number of samples. This prevents it from being a cluster member entirely, or the point is identified as not being a part of a cluster more often than being part of a cluster. For context, the minimum number of features is defined by Environmental Systems Research Institute (ESRI) as the number of features neighboring each point (including the point) that will be considered when estimating density. It is also the minimum cluster size allowed when extracting clusters. The tool’s output also includes measures related to the probability that a given point is in the correct cluster, the probability that the point is an outlier in its respective cluster, the stability of the point (i.e., how long the point remains in its cluster during the algorithm’s sorting), and whether the point is the most representative of its cluster (referred to as an exemplar).
In Fig. 12, exemplar points are denoted by diamonds and are the “core” of the cluster (i.e., the probability that these points are in a given cluster is 1). For instance, every point in cluster 5 was considered an exemplar point, meaning the probability of every stall being in that particular cluster was 1. The remaining points in each cluster are color coded based on the probability they are in the correct cluster, with lighter-shaded points indicating a lower probability of being in the correct cluster. Stalls with lower probabilities of being in the correct cluster have higher probabilities of being outliers in their given cluster, so points in Fig. 12 that are a lighter shade have a greater chance of belonging to a different cluster or being classified as noise. Black points are not considered in any cluster and are considered noise. Longitude 60°W is the first line to the legend’s left, and no points are included east of that line.
Stalling TC clusters. Point clusters based on HDBSCAN approach.
Citation: Journal of Applied Meteorology and Climatology 63, 11; 10.1175/JAMC-D-23-0229.1
Cluster 1 consists of storms near Texas and Louisiana with a mean wind speed within the cluster of 48.6 kt (25 m s−1). Cluster 2 consists of those located in the Bay of Campeche (53.4 kt; 27.5 m s−1), cluster 3 consists of those in the western Caribbean Sea (47.2 kt; 24.3 m s−1), cluster 4 consists of stalling storms from eastern Florida through the North Carolina coast (46.0 kt; 23.7 m s−1), and cluster 5 consists of those near the coast of western Florida (45.4 kt; 23.4 m s−1). Cluster 2 has the highest mean wind speed expected within a cluster zone.
d. Wind intensity and rapid intensification
It is interesting to consider the wind intensity of stalling TC compared with those that do not stall in the known record. Figure 13 compares the average annual number of corrals separated by intensity ranges to the storms that meet our stall definition. Figure 13a shows the total number of overlapping 72-h periods within all storm tracks and is directly comparable to Fig. 13b, which includes only average stalling storm counts.
Intensity comparison. (a) Ten-year running average annual number of corrals over time separated by average wind intensities (m s−1) during the corral periods of all TC tracks in HURDAT2 and (b) average annual number of stalls separated by wind intensity for all stalling TCs.
Citation: Journal of Applied Meteorology and Climatology 63, 11; 10.1175/JAMC-D-23-0229.1
During a stall, wind intensity is most often less than category 1 hurricane strength (≤64 kt; 33 m s−1). Four systems average category 3 wind strength (≥50 m s−1) during their stalls. The most recent of these was Hurricane Betsy in 1965, which stalled as a category 3 system near the Bahamas, bringing hurricane-force winds for >20 h in Abaco Island and ultimately causing $14 million (1966 dollars) in damage (Sugg 1966). Hurricane Dorian in 2019 nearly stalled in a similar location at category 5 strength, devastating the islands, but is not included in the data assessed here as the stall radius was measured at 222 km. Spearman’s rank correlation test and ordinary least squares regression show a statistically significant negative trend over time in the mean wind intensity during minimum stalls (1900–2020), suggesting stalling TCs are weaker in wind intensity during their stalls today than in the past. This is likely a product of storms being tracked at weaker intensities in more recent years, leading to data inhomogeneity from better tracking practices rather than a product of actual wind weakening.
Differences in wind intensity are considered between each observation of per-event minimum stall periods. The 12-h differences are averaged and plotted in Fig. 14. Values above zero indicate periods of intensification, but the downward trend through time implies the intensification rate is slowing. Around the center of the stall (hour 36), storms are not changing in wind intensity. After hour 42, the intensity continues to decay over time, with the decay rate fluctuating toward the end of the stall, trending toward no change.
Wind intensity change within stall zone. Average differences in wind intensity within the TC stall zone. The temporal center of the stall zone is shown with a red line. A 95% confidence band is shown in gray.
Citation: Journal of Applied Meteorology and Climatology 63, 11; 10.1175/JAMC-D-23-0229.1
A chi-square test was used to determine whether TCs that stall at any point along their track reach category 3+ strength more often than TCs that do not stall, similar to the conclusion reached by Sun et al. (2021). Two periods were tested (1900–2020 and 1966–2020) to address potential observational biases in the HURDAT2 database. For both periods, storms that stalled had a higher frequency of major hurricane status than storms that did not stall (1900–2020, p < 0.001; 1966–2020, p < 0.01). This could demonstrate that the background meteorological environment that produces stalling storms is also favorable for the continued strengthening of TCs; that is, the environment that produces a stall also supports TC intensification, at least initially. However, the result may also be a product of the longer lifespan of the stalled storms (i.e., at least 72 h) and may not be directly related to the actual tendency for stalled storms to reach major hurricane status. This is explored further in the discussion below. Results of the chi-square tests are shown in Table 5.
Chi-square test results. Listed values are observed frequency, expected frequency, percent, row percent, column percent, mean, and median. Values are separated by time periods (a) 1900–2020, (b) 1966–2020, and (c) 1991–2020.
Wind speeds do not necessarily need hurricane status for a storm to cause damage. Hurricane Harvey may have been a category 4 storm when it made landfall at peak intensity on San Jose Island, Texas, but during the stall, the TC exhibited wind speeds of only 27.1 m s−1 (52.7 kt). The most damaging characteristic was the estimated 1270 mm (50 in.) of rain produced over 5 days across 2590 km2 (1000 mi2) (Nielsen-Gammon and Larson 2022).
The biggest differential in duration was found in weaker (<category 3) TC storms (i.e., weak stalling storms persisted longer than weak nonstalling storms). From 1900 to 2020, weak (<category 3) TCs that stalled (n = 142) at some point during their life cycle had a median of 35 observations (
TCs that stall reach category 3+ strength more frequently than TCs that do not stall because they tend to have longer life cycles. When considering all moments of a TC in HURDAT2, storms that reached major hurricane status and stalled at some point during their life cycle (n = 49) persisted for 324 (median) hours or 13.5 days (
Rapidly intensifying TCs exhibit a wind increase of at least 30 kt (1 kt ≈ 0.51 m s−1) over 24 h (Kaplan et al. 2015). Benedetto and Trepanier (2020) suggest these storms have become more frequent and are occurring closer to the coast in recent years. TC wind intensity and position are interpolated to the hour using splines from 1900 to 2020 to identify those that underwent rapid intensification. Of the 191 minimum stalls since 1900, 77 rapidly intensified (∼40%). Only minimum stall corrals per event are used to maintain storm independence. Since 1966, 47 storms have stalled and rapidly intensified at some point in their life cycle. Of those, 11 began rapid intensification within 6 h of the stall start time; thus, they were still undergoing rapid intensification during the stall period (i.e., Martha in 1969, an Unnamed Storm in 1972, Alberto in 1982, Charley in 1992, Keith and Michael in 2000, Wilma and Beta in 2005, Hanna in 2008, and Katia and Lee in 2017). Seven rapidly intensified after stalling, and the remaining 30 (∼64%) rapidly intensified before stalling. If a TC is already a hurricane (winds ≥ 33 m s−1) when rapid intensification begins, it will become at least a category 3 intensity on the Saffir–Simpson hurricane wind scale (Benedetto and Trepanier 2020), suggesting the 30 storms that rapidly intensified before stalling were major hurricanes at least before or during their stalls. This increases the risk of major hurricane threats for a population being influenced by a stalling TC. Stalling storms that rapidly intensify are more dangerous than nonstalling storms that rapidly intensify because this leads to nearly stationary storms at higher and possibly unexpected intensities, extending exposure times and making evacuations more difficult. Eight of the 11 storms that simultaneously stalled and underwent rapid intensification began the process, on average, 32 h after the initial stall start time, with a range of 1–64 h after stall start. Five of those 11 TCs did this within 200 km of a coastal zone. These distances (±10 km) from the coast include Martha (91.6 km), Alberto (174.6 km), Keith (89.1 km), Beta (158.0 km), and Katia (182.8 km). The average intensity of these storms during the stall period is 32.2 m s−1, with a maximum of 41.9 m s−1 in Hurricane Keith. Hurricane Keith stalled just southeast of the Yucatán Peninsula at the same time that rapid intensification began, causing severe flooding and damage in Belize due to the combined high winds and near coastal stall.
Table 6 shows the number of stalled storms that also rapidly intensified during different decades. The increase suggests a temporal trend related to bias introduced early in the record due to underreporting. However, nearly 30% of the rapidly intensifying and stalling TCs occurred since 2000. This may be related to better observational capabilities and a more conducive environment for rapidly intensifying storms. However, ∼31% of the stalling TCs in the data occurred since 2000, and the increased number of stalls is likely the reason for the increased number of rapid intensification events in the most recent decades.
Rapidly intensifying stalls. Number of stalled TCs that also rapidly intensified.
5. Final discussion
TC motion is regulated by many factors, with large-scale atmospheric circulation often being the primary control (Holland 1983; Kossin 2018). TCs stall when weak upper-level steering currents prevail (Mayfield et al. 1994) or when lower-level and upper-level winds work against one another (Hebert and Frank 1974). A warming environment is linked to weakening tropical summertime circulation and decreasing TC translation speed (Kossin 2018). Some results indicate stalling is related to arctic amplification and the reduction in meridional temperature gradients, allowing for less steering flow for TCs and more persistent blocking patterns (Coumou et al. 2015).
Stalling TCs can inflict many hours of devastating consequences, including high winds and heavy rains, in a small geographic area. This research offers an approach for identifying stalling storms based on the minimum geographical area they travel within 72 h. It provides a spatiotemporal analysis of stalling TCs and temporal trends. There is a statistically significant increase in the satellite-era stalling TC rate of 1.5% yr−1 (p value = 0.1). Stalling TCs tend to cluster near the Louisiana/Texas coast, the Bay of Campeche, the western Caribbean Sea, and around eastern and western Florida into the Carolinas. Of the 191 stalling TCs, 10 stalled multiple times, including a maximum of four times in Hurricane Ginger in 1971.
Storms that stall tend to have a longer average duration than those that do not stall in all categories. Category 3+ storms persist longer than weaker TCs due to favorable conditions, providing more stall opportunities. TCs decay and dissipate when they move into unfavorable environments, including over land, higher-latitude, high-shear environments, or when low-latitude environments become less favorable. Based on its definition, a stalling TC will take longer to move over land or to higher latitudes and will eventually move to a point of decay. Stalls are, thus, more likely to lead to longer life cycles. However, stalling storms draw heat from the ocean, and upwelling and high amounts of rainfall cool the ocean’s surface, potentially making the local environment less favorable. These storms can and do rapidly intensify; however, stalling storms may not influence their environments enough to stop rapid intensification from happening. Of the 47 storms that rapidly intensified and experienced the minimum stall point, 23% did so concurrently, and 45% of those did so within 200 km of a coastal zone (n = 5).
A limitation of this study is that stalls, or corrals, only include the center location of the TC event. The areal extent of influencing winds, storm surge, and rain is not considered beyond the center location, and the impacted areas on the ground are much larger than what is represented by the 200-km or less stall region. This should be considered in emergency preparations. Future research will investigate the synoptic setting of stalled TCs to determine which characteristics cause a TC to stall and reach a higher intensity than nonstalling storms. Additionally, exploring the coastal stalls versus open water stalls may provide insight into causal factors for temporal changes. Future research will also use the stall data in combination with available rainfall to estimate recurrence intervals for the precipitation produced from stalled storms within the stall area and for the TC areal extent to estimate flood risks similar to Zhang et al. (2023). Finally, exploratory analyses will be conducted to identify any similar environmental conditions in the storms occurring near the stall belt around 30°N or in those that rapidly intensify after or during stalling periods to provide insight into the environment leading to rapidly intensifying storms.
If stalling storms become more frequent throughout the basin, the risk of devastating conditions in areas impacted by TCs will increase. Before Hurricane Harvey, the previous largest 5-day, 10 000-mi2 total accumulated rainfall record in Texas occurred with a TC in 1899, followed by Hurricane Beulah in 1967 (Nielsen-Gammon 2019). Researchers are attempting to understand the future risk of rainfall throughout the region, some suggesting slowing translation speed and increased rainfall (Nogueira and Keim 2011; Zhu et al. 2021), while others find decreased rainfall due to an increased likelihood of fast-moving landfalling TCs in Texas (Hassanzadeh et al. 2020). Research must continue to attempt to further explain these discrepancies and geographic differences.
Acknowledgments.
The authors acknowledge support from NOAA Grant NA21OAR4310306 and NOAA Contract 1332KP21FNEEN0021. The data and analysis software for this manuscript are available upon request from the corresponding author.
Data availability statement.
The data are available through the corresponding author by request, and an active URL is provided in the manuscript with the data, which are publicly available through the National Oceanic and Atmospheric Administration’s National Hurricane Center.
REFERENCES
Applied Weather Associates, LLC, 2019: Regional probable maximum precipitation study for Oklahoma, Arkansas, Louisiana, and Mississippi final report. State of Oklahoma Tech. Rep., 126 pp., https://oklahoma.gov/content/dam/ok/en/owrb/documents/dam-safety/2019RegionalPMPStudy.pdf.
Benedetto, K. M., and J. C. Trepanier, 2020: Climatology and spatiotemporal analysis of North Atlantic rapidly intensifying hurricanes (1851–2017). Atmosphere, 11, 291, https://doi.org/10.3390/atmos11030291.
Blake, E. S., and D. A. Zelinsky, 2018: Tropical cyclone report: Hurricane Harvey (AL092017), 17 August–1 September 2017. NHC Tech. Rep., 77 pp., https://www.nhc.noaa.gov/data/tcr/AL092017_Harvey.pdf.
Britannica, 2021: Hurricane Mitch. Encyclopedia Britannica, https://www.britannica.com/event/Hurricane-Mitch.
Butt, U. M., and Coauthors, 2021: Spatio-temporal crime predictions by leveraging artificial intelligence for citizens security in smart cities. IEEE Access, 9, 47 516–47 529, https://doi.org/10.1109/ACCESS.2021.3068306.
Campello, R. J. G. B., D. Moulavi, and J. Sander, 2013: Density-based clustering based on hierarchical density estimates. Advances in Knowledge Discovery and Data Mining. PAKDD 2013, J. Pei et al., Eds., Lecture Notes in Computer Science, Vol. 7819, Springer, 160–172, https://doi.org/10.1007/978-3-642-37456-2_14.
Chan, K. T. F., 2019: Are global tropical cyclones moving slower in a warming climate? Environ. Res. Lett., 14, 104015, https://doi.org/10.1088/1748-9326/ab4031.
Cohen, D. T., 2018: All American Counts Stories: 60 million live in the path of hurricanes. United States Census Bureau, 6 August, https://www.census.gov/library/stories/2018/08/coastal-county-population-rises.html.
Coogan, S. C. P., O. Aftergood, and M. D. Flannigan, 2022: Human- and lightning-caused wildland fire ignition clusters in British Columbia, Canada. Int. J. Wildland Fire, 31, 1043–1055, https://doi.org/10.1071/WF21177.
Corporal-Lodangco, I. L., M. B. Richman, L. M. Leslie, and P. J. Lamb, 2014: Cluster analysis of North Atlantic tropical cyclones. Procedia Comput. Sci., 36, 293–300, https://doi.org/10.1016/j.procs.2014.09.096.
Coumou, D., L. Lehmann, and J. Beckmann, 2015: The weakening summer circulation in the Northern Hemisphere mid-latitudes. Science, 348, 324–327, https://doi.org/10.1126/science.1261768.
Elsner, J. B., and T. H. Jagger, 2013: Hurricane Climatology: A Modern Statistical Guide Using R. Oxford University Press, 430 pp.
Emanuel, K. A., 1987: The dependence of hurricane intensity on climate. Nature, 326, 483–485, https://doi.org/10.1038/326483a0.
Emanuel, K. A., 2017: Assessing the present and future probability of Hurricane Harvey’s rainfall. Proc. Natl. Acad. Sci. USA, 114, 12 681–12 684, https://doi.org/10.1073/pnas.1716222114.
Environmental Systems Research Institute, 2022: Find point clusters (GeoAnalytics Desktop). ArcGIS Pro, https://pro.arcgis.com/en/pro-app/latest/tool-reference/geoanalytics-desktop/find-point-clusters.htm.
Ester, M., H.-P. Kriegel, J. Sander, and X. Xu, 2013: A density-based algorithm for discovering clusters in large spatial databases with noise. Proc. Second Int. Conf. on Knowledge Discovery and Data Mining, Menlo Park, CA, AAAI, 226–231, https://file.biolab.si/papers/1996-DBSCAN-KDD.pdf.
Evans, C. J., 1975: Heavy rainfall in Jamaica associated with Hurricane Flora 1963 and Tropical Storm Gilda 1973. Weather, 30, 157–161, https://doi.org/10.1002/j.1477-8696.1975.tb03360.x.
Golden, C., and Coauthors, 2021: Airborne lidar survey, density-based clustering, and Ancient Maya settlement in the Upper Usumacinta River Region of Mexico and Guatemala. Remote Sens., 13, 4109, https://doi.org/10.3390/rs13204109.
Gong, D., X. Tang, J. C. L. Chan, and Q. Wang, 2022: Trends of tropical cyclone translation speed over the Western North Pacific during 1980–2018. Atmosphere, 13, 896, https://doi.org/10.3390/atmos13060896.
Hagen, A. B., and C. W. Landsea, 2012: On the classification of extreme Atlantic hurricanes utilizing mid-twentieth-century monitoring capabilities. J. Climate, 25, 4461–4475, https://doi.org/10.1175/JCLI-D-11-00420.1.
Hall, T. M., and J. P. Kossin, 2019: Hurricane stalling along the North American coast and implications for rainfall. Climate Atmos. Sci., 2, 17, https://doi.org/10.1038/s41612-019-0074-8.
Ham-Chande, R., and I. Nava-Bolaños, 2019: Convergence toward demographic aging in Latin America and the Caribbean. Annu. Rev. Sociol., 45, 607–623, https://doi.org/10.1146/annurev-soc-073018-022532.
Hassanzadeh, P., C.-Y. Lee, E. Nabizadeh, S. J. Camargo, D. Ma, and L. Y. Yeung, 2020: Effects of climate change on the movement of future landfalling Texas tropical cyclones. Nat. Commun., 11, 3319, https://doi.org/10.1038/s41467-020-17130-7.
Hebert, P. J., and N. L. Frank, 1974: Atlantic hurricane season of 1973. Mon. Wea. Rev., 102, 280–289, https://doi.org/10.1175/1520-0493(1974)102<0280:AHSO>2.0.CO;2.
Hill, K. A., and G. M. Lackman, 2011: The impact of future climate change on TC intensity and structure: A downscaling approach. J. Climate, 24, 4644–4661, https://doi.org/10.1175/2011JCLI3761.1.
Holland, G. J., 1983: Tropical cyclone motion: Environmental interaction plus a beta effect. J. Atmos. Sci., 40, 328–342, https://doi.org/10.1175/1520-0469(1983)040<0328:TCMEIP>2.0.CO;2.
Johnson, G. C., and J. M. Lyman, 2020: Warming trends increasingly dominate global ocean. Nat. Climate Change, 10, 757–761, https://doi.org/10.1038/s41558-020-0822-0.
Kaplan, J., and Coauthors, 2015: Evaluating environmental impacts on tropical cyclone rapid intensification predictability utilizing statistical models. J. Climate, 30, 1374–1396, https://doi.org/10.1175/WAF-D-15-0032.1.
Knutson, T. R., and R. E. Tuleya, 2004: Impact of CO2-induced warming on simulated hurricane intensity and precipitation: Sensitivity to the choice of climate model and convective parameterization. J. Climate, 17, 3477–3495, https://doi.org/10.1175/1520-0442(2004)017<3477:IOCWOS>2.0.CO;2.
Kossin, J. P., 2018: A global slowdown of tropical-cyclone translation speed. Nature, 558, 104–107, https://doi.org/10.1038/s41586-018-0158-3.
Kossin, J. P., 2019: Reply to: Moon, I.-J. et al.; Lanzante, J. R. Nature, 570, E16–E22, https://doi.org/10.1038/s41586-019-1224-1.
Kriegel, H.-P., P. Kroger, J. Sander, and A. Zimek, 2011: Density-based clustering. WIREs Data Min. Knowl. Discovery, 1, 231–240, https://doi.org/10.1002/widm.30.
Kumar, S. G., S. J. Corrado, T. G. Puranik, and D. N. Mavris, 2021: Classification and analysis of go-arounds in commercial aviation using ADS-B Data. Aerospace, 8, 291, https://doi.org/10.3390/aerospace8100291.
Lai, Y., and Coauthors, 2020: Greater flood risks in response to slowdown of tropical cyclones over the coast of China. Proc. Natl. Acad. Sci. USA, 117, 14 751–14 755, https://doi.org/10.1073/pnas.1918987117.
Landsea, C. W., 2007: Counting Atlantic tropical cyclones back to 1900. EOS Trans., 88, 197–202, https://doi.org/10.1029/2007EO180001.
Landsea, C. W., and J. L. Franklin, 2013: Atlantic hurricane database uncertainty and presentation of a new database format. Mon. Wea. Rev., 141, 3576–3592, https://doi.org/10.1175/MWR-D-12-00254.1.
Landsea, C. W., G. A. Vecchi, L. Bengtsson, and T. R. Knutson, 2010: Impact of duration thresholds on Atlantic tropical cyclone counts. J. Climate, 23, 2508–2519, https://doi.org/10.1175/2009JCLI3034.1.
Lanzante, J. R., 2019: Uncertainties in tropical-cyclone translation speed. Nature, 570, E6–E15, https://doi.org/10.1038/s41586-019-1223-2.
Link, L. E., 2010: The anatomy of a disaster, an overview of Hurricane Katrina and New Orleans. Ocean Eng., 37, 4–12, https://doi.org/10.1016/j.oceaneng.2009.09.002.
Lonfat, M., F. D. Marks Jr., and S. S. Chen, 2004: Precipitation distribution in tropical cyclones using the Tropical Rainfall Measuring Mission (TRMM) microwave imager: A global perspective. Mon. Wea. Rev., 132, 1645–1660, https://doi.org/10.1175/1520-0493(2004)132<1645:PDITCU>2.0.CO;2.
Mayfield, M., L. Avila, and E. N. Rappaport, 1994: Atlantic hurricane season of 1992. NHC Tech. Rep., 22 pp., https://www.aoml.noaa.gov/general/lib/lib1/nhclib/mwreviews/1992.pdf.
McInnes, L., J. Healy, and S. Astels, 2017: hdbscan: Hierarchical density based clustering. J. Open Source Software, 2, 205, https://doi.org/10.21105/joss.00205.
Moon, I.-J., S.-H. Kim, and J. C. L. Chan, 2019: Climate change and tropical cyclone trend. Nature, 570, E3–E5, https://doi.org/10.1038/s41586-019-1222-3.
Moulavi, D., P. A. Jaskowiak, R. J. G. B. Campello, A. Zimek, and J. Sander, 2014: Density-based clustering validation. Proc. 2014 SIAM Int. Conf. on Data Mining, Philadelphia, PA, Society for Industrial and Applied Mathematics, 839–847, https://doi.org/10.1137/1.9781611973440.96.
National Climatic Data Center, 2022: Billion dollar weather and climate disasters. https://www.ncdc.noaa.gov/billions/events/US/2021.
National Weather Service, 2023: Saffir-Simpson Hurricane wind scale. National Oceanic and Atmospheric Administration, https://www.weather.gov/mfl/saffirsimpson.
Nielsen-Gammon, J. W., 2019: Stalling tropical cyclones. State Climatology Meeting, Nebraska City, NE, American Association of State Climatologist.
Nielsen-Gammon, J. W., and C. J. Larson, 2022: Hurricane Harvey’s maximum rainfall: A forensic analysis. 36th Conf. on Hydrology/102nd Annual Meeting of the American Meteorological Society, Online, Amer. Meteor. Soc., 22, https://ams.confex.com/ams/102ANNUAL/meetingapp.cgi/Paper/397650.
Nogueira, R. C., and B. D. Keim, 2011: Contributions of Atlantic tropical cyclones to monthly and seasonal rainfall in the eastern United States 1960–2007. Theor. Appl. Climatol., 103, 213–227, https://doi.org/10.1007/s00704-010-0292-9.
Norton, G., 1951: Hurricanes of the 1950 season. Mon. Wea. Rev., 79, 8–15, https://doi.org/10.1175/1520-0493-79.1.8.
Patricola, C. M., 2018: Tropical cyclones are becoming sluggish. Nature, 558, 36–37, https://doi.org/10.1038/d41586-018-05303-w.
Risser, M. D., and M. F. Wehner, 2017: Attributable human-induced changes in the likelihood and magnitude of the observed extreme precipitation during Hurricane Harvey. Geophys. Res. Lett., 44, 12 457–12 464, https://doi.org/10.1002/2017GL075888.
Schreiner, L. C., and J. T. Riedel, 1978: Probable maximum precipitation estimates, United States East of the 105th Meridian. Hydrometeorological Rep. 51, 100 pp., https://repository.library.noaa.gov/view/noaa/6499.
Stewart, G., and M. Al-Khassaweneh, 2022: An implementation of the HDBSCAN* clustering algorithm. Appl. Sci., 12, 2045, https://doi.org/10.3390/app12052405.
Sugg, A. L., 1966: Hurricane season of 1965. Mon. Wea. Rev., 94, 183–191.
Sun, Y., Z. Zhong, T. Li, L. Yi, and Y. Shen, 2021: The slowdown tends to be greater for stronger tropical cyclones. J. Climate, 34, 5741–5751, https://doi.org/10.1175/JCLI-D-20-0449.1.
Sweet, W. V., and Coauthors, 2022: Global and regional sea level rise scenarios for the United States: Updated mean projections and extreme water level probabilities along U.S. coastlines. NOAA Tech. Rep. NOS 01, 111 pp., https://aambpublicoceanservice.blob.core.windows.net/oceanserviceprod/hazards/sealevelrise/noaa-nos-techrpt01-global-regional-SLR-scenarios-US.pdf.
Trenberth, K. E., L. Cheng, P. Jacobs, Y. Zhang, and J. Fasullo, 2018: Hurricane Harvey links to ocean heat content and climate change adaptation. Earth’s Future, 6, 730–744, https://doi.org/10.1029/2018EF000825.
van Oldenborgh, G. J., and Coauthors, 2017: Attribution of extreme rainfall from Hurricane Harvey, August 2017. Environ. Res. Lett., 12, 124009, https://doi.org/10.1088/1748-9326/aa9ef2.
Vecchi, G. A., and B. J. Soden, 2007: Global warming and the weakening of the tropical circulation. J. Climate, 20, 4316–4340, https://doi.org/10.1175/JCLI4258.1.
Walsh, K. J. E., and Coauthors, 2015: Hurricanes and climate: The U.S. CLIVAR working group on hurricanes. Bull. Amer. Meteor. Soc., 96, 997–1017, https://doi.org/10.1175/BAMS-D-13-00242.1.
Wang, C., L. Wu, H. Zhao, Q. Liu, and J. Wang, 2021: An abrupt slowdown of late season tropical cyclone over the western North Pacific in the early 1980s. J. Meteor. Soc. Japan, 99, 1413–1422, https://doi.org/10.2151/jmsj.2021-068.
Willoughby, H. E., and P. G. Black, 1996: Hurricane Andrew in Florida: Dynamics of a disaster. Bull. Amer. Meteor. Soc., 77, 543–550, https://doi.org/10.1175/1520-0477(1996)077<0543:HAIFDO>2.0.CO;2.
Wu, K., C. Wang, L. Wu, H. Zhao, and J. Cao, 2022: Slowdown in landfalling tropical cyclone motion in South China. Proc. Natl. Acad. Sci. USA, 49, e2022GL100428, https://doi.org/10.1029/2022GL100428.
Yamaguchi, M., and S. Maeda, 2022: Slowdown of typhoon translation speeds in mid-latitudes in September influenced by the Pacific Decadal Oscillation and global warming. J. Meteor. Soc. Japan, 98, 1321–1334, https://doi.org/10.2151/jmsj.2020-068.
Zhang, L., T. F. Cheng, M. Lu, R. Xiong, and J. Gan, 2023: Tropical cyclone stalling shifts northward and brings increasing flood risks to East Asian Coast. Geophys. Res. Lett., 50, e2022GL102509, https://doi.org/10.1029/2022GL102509.
Zhang, Y., W. Huang, and D. Zhong, 2019: Major moisture pathways and their importance to rainy season precipitation over the Sanjiangyuan Region of the Tibetan Plateau. J. Climate, 32, 6837–6857, https://doi.org/10.1175/JCLI-D-19-0196.1.
Zhu, L., K. Emanuel, and S. M. Quiring, 2021: Elevated risk of tropical cyclone precipitation and pluvial flood in Houston under global warming. Environ. Res. Lett., 16, 094030, https://doi.org/10.1088/1748-9326/ac1e3d.
