Predicting Atlantic Hurricanes Using Machine Learning

Despite the observed high correlation between the Atlantic sea surface temperature (SST) and the Atlantic tropical cyclone (TC) activity, interpretation of this relationship remains uncertain. This study suggests that the tropical Atlantic sea surface warming induces a pair of anomalous low-level cyclones on each side of the equator, providing favorable conditions for enhancing TC formation east of 45°W, while the effect of SST warming in the tropical Indian Ocean and Pacific Ocean tends to suppress the TC formation. Over the past 30 years (1978–2007), the TC activity in the Atlantic basin is characterized with significant enhancement of TC formation east of 45°W, where the total TC number increased significantly compared to the period 1948–77. Despite the possible undercount of TCs, this study shows that the recently enhanced TC formation may not be totally accounted for by the poor TC observing network prior to the satellite era. The Atlantic sea surface warming that occurred in recent decades might have allowed more TCs to form, to form earlier, and to take a longer track, while the effect is partially offset by the SST warming in Indian and Pacific Oceans. This study suggests that the close relationship between the Atlantic SST and TC activity over the past 30 years, including basinwide increases in the average lifetime, annual frequency, proportion of intense hurricanes, and annual accumulated power dissipation index (PDI), as reported in previous studies, is mainly a result of the SST warming in the tropical Atlantic exceeding that in the tropical Indian and Pacific Oceans. The results agree with recent argument that the relative Atlantic SST change or the SST difference between the tropical Atlantic and other oceans play an important role in controlling long-term TC activity in the Atlantic basin.

The North American Multi-Model Ensemble (NMME)-Phase II models are evaluated in terms of their retrospective seasonal forecast skill of the North Atlantic (NA) tropical cyclone (TC) activity, with a focus on TC frequency. The TC identification and tracking algorithm is modified to accommodate model data at daily resolution. It is also applied to three reanalysis products at the spatial and temporal resolution of the NMME-Phase II ensemble to allow for a more objective estimation of forecast skill. When used with the reanalysis data, the TC tracking generates realistic climatological distributions of the NA TC formation and tracks, and represents the interannual variability of the NA TC frequency quite well. Forecasts with the multi-model ensemble (MME) when initialized in April and later tend to have skill in predicting the NA seasonal TC counts (and TC days). At longer leads, the skill is low or marginal, although one of the models produces skillful forecasts when initialized as early as January and February. At short lead times, while demonstrating the highest skill levels the MME also tends to significantly outperform the individual models and attain skill comparable to the reanalysis. In addition, the short-lead MME forecasts are quite reliable. At regional scales, the skill is rather limited and mostly present in the western tropical NA and the Caribbean Sea. It is found that the overall MME forecast skill is limited by poor representation of the low-frequency variability in the predicted TC frequency, and large fluctuations in skill on decadal time scales. Addressing these deficiencies is thought to increase the value of the NMME ensemble in providing operational guidance.

The main historical archive of all tropical storms, subtropical storms, and hurricanes in the North Atlantic Ocean, Caribbean Sea, and Gulf of Mexico from 1851 to the present is known as the Atlantic hurricane database (HURDAT), which is the fundamental database for meteorological, engineering, and financial studies of these cyclones. Previous work has demonstrated that a reanalysis of HURDAT is necessary because it contains many random errors and systematic biases. The Atlantic Hurricane Reanalysis Project is an ongoing effort to correct the errors in HURDAT and to make HURDAT as accurate a database as possible with utilization of all available data. For this study, HURDAT is reanalyzed for the period 1944–53, the first decade of the “aircraft reconnaissance era.” The track and intensity of each existing tropical cyclone in HURDAT are reassessed, and previously unrecognized tropical cyclones are discovered, analyzed, and recommended to the HURDAT Best Track Change Committee for inclusion into HURDAT (existing tropical cyclones may be removed from the database as well if analyses indicate evidence that no tropical storm existed). Changes to the number of tropical storms, hurricanes, major hurricanes, accumulated cyclone energy, and U.S. landfalling hurricanes are recommended for most years of the decade. Estimates of uncertainty in the reanalyzed database for the decade are also provided.

The study of past tropical cyclone activity by means of geological proxies and/or historical documentary records is known as paleotempestology. This scientific discipline has become prominent over the course of the last decade partially in response to the recent increase in tropical cyclone count and intensity in the North Atlantic basin witnessed since 1995. The field has also developed due to the socioeconomic impacts of tropical cyclones particularly along vulnerable coastal regions. During the twenty-five years prior to the start of the most recent increase in hurricane activity, major (Category 3, 4, or 5) hurricanes were less frequent than in previous decades. Yet, property losses from the hurricanes that did make landfall in the United States increased during this period due to development in damage prone areas (NOAA Paleoclimatology Program, 2000). Many researchers hence stress the importance of identifying historical tropical cyclones to understand long term trends in tropical cyclone climatology and to determine the influence of anthropogenic global warming on tropical cyclone activity and intensity. The North Atlantic Hurricane Database (HURDAT) has been one of the authoritative sources for examining North Atlantic tropical cyclone activity trends since 1850. However, some of the deadliest known hurricanes and potentially most active seasons in the North Atlantic basin occurred prior to the beginning of the HURDAT record (Table 1), including the Great Hurricane of 1780 that killed an estimated 22,000 people and was one of eight known tropical cyclones during that season. This chapter will provide a brief overview of some paleotempestology techniques and illustrate a methodology for identifying and reconstructing historical North Atlantic tropical cyclone tracks in the pre-HURDAT era employing a Geographic Information System (GIS) and utilizing readily accessible archival data.

In this chapter, we performed a climatological study of the tropical cyclone (TC) precipitation in the satellite observations and the reanalysis from the ECMWF and NCEP/NCAR centers over the North Atlantic (NATL) basin. Using the recently developed best track IBTrACS (Kruk et al. 2009; Knapp et al. 2009) we derived the mean daily TCs rainfall within 10∘ ×10∘ box around the center of the TCs, the fraction of TCs rainfall to total rainfall and the TCs precipitation efficiency (TCPE). These variables were used to assess the ability of the reanalysis to represent the impact of TCs in altering the total rainfall over the North Atlantic basin. The main results show that: The maximum of cyclonic precipitation is located in the Caribbean Sea and the Gulf of Mexico, for the observation and the reanalysis. TCs contribute to a maximum of precipitation (between 15∘ and 25∘N) over the Southern Gulf of Mexico, the Caribbean Sea and the West Atlantic Ocean for the observation and the reanalysis. The most efficient TCs are located in the Gulf of Mexico, the Caribbean Sea and the West Atlantic Ocean for the observation and the reanalysis.

Organized tropical convection, often characterized by overshooting tops, is a distinguishing quality of tropical cyclones (TCs). In this study, the climatology of tropical overshooting tops (TOTs) in North Atlantic Ocean TCs from 2004 to 2015 is examined. Previous studies have investigated the distribution of convection in TCs based on lightning data. The purpose of this study is to examine the distribution of TC convection from geostationary satellites using an objective TOT detection algorithm based on infrared brightness temperatures and empirically dependent thresholds. It will be shown that TOTs can provide an additional metric for identifying the characteristics of TC convection. Based on the 12-yr (2004–15) climatology, a distinct semidiurnal cycle in TOT activity is detected within 500 km of the TC center. In agreement with lightning data from previous studies, a predawn maximum (local to the TC) in TOTs is observed within 300 km of the TC center. A second predusk maximum is associated with TOTs between 300 and 500 km of the TC center. TC intensity and intensity trend along with environmental factors can affect the number and distribution of TOTs. For example, an exponential relationship exists between the number of TOTs and increasing sea surface temperatures. Conversely, increasing vertical wind shear magnitude decreases the density of TOTs, with a higher percentage of TOTs observed downshear of the wind direction. Generally, within 100 km (100–300 km) of the TC center, the preferred quadrant for TOTs is downshear left (downshear right), and increased TOT activity is observed right of TC motion. The findings corroborate previous lightning study results while providing additional insights into TC convection.

An empirical relationship between climatological sea surface temperatures (SST) and the maximum intensities of tropical cyclones over the eastern North Pacific Ocean is developed from a 31-yr sample (1963–93). This relationship is compared with an empirical relationship for tropical cyclones over the Atlantic Ocean and with theoretical results. Over the period of study, the storms over the eastern North Pacific Ocean reached a lower percentage of their empirical maximum potential intensity (MPI) than tropical cyclones over the Atlantic Ocean. At the time of their maximum intensity, only 11% of eastern North Pacific storms reach 80% of their MPI, while 19% of the Atlantic tropical cyclones reach that proportion of their MPIs. Poleward recurvature of Atlantic storms over cooler waters appears to be a major factor in the difference between the two regions. The storms were stratified by latitude, longitude, the phase of the quasi-biennial oscillation (QBO), and the status of the El Nino phenomenon. T...

Integrated Kinetic Energy (IKE) is a recently developed metric that measures the destructive potential of tropical cyclones (TCs) by integrating the square of the surface winds across these powerful storms. In this chapter, the previous literature is reviewed to provide insights on the factors that make IKE a desirable metric. IKE complements existing scales and metrics by considering a TC’s entire wind field, in lieu of just focusing on the maximum intensity of a storm. Using a dataset of six-hourly IKE estimates for two decades of North Atlantic TC activity, the climatology of IKE in individual storms is explored, with emphasis on seasonal and spatial variability. The driving mechanisms for IKE variability during the lifetime of a TC are also reviewed to determine which environmental and storm-scale features promote IKE growth. The historical record of IKE can also be aggregated to a seasonal metric, called Track Integrated Kinetic Energy (TIKE), which is shown to offer a comprehensive overview of seasonal TC activity and can be used to explore interannual TC variability over the last two to three decades.

A significant increase of tropical cyclone (TC) frequency is observed over the North Atlantic (NATL) basin during the recent decades (1995–2014). In this study, the changes in large-scale controls of the NATL TC activity are compared between two periods, one before and one since 1995, when a regime change is observed. The results herein suggest that the significantly enhanced NATL TC frequency is related mainly to the combined effect of changes in the magnitudes of large-scale atmospheric and oceanic factors and their association with TC frequency. Interdecadal changes in the role of vertical wind shear and local sea surface temperatures (SSTs) over the NATL appear to be two important contributors to the recent increase of NATL TC frequency. Low-level vorticity plays a relatively weak role in the recent increase of TC frequency. These changes in the role of large-scale factors largely depend on interdecadal changes of tropical SST anomalies (SSTAs). Enhanced low-level westerlies to the east of the positive SSTAs have been observed over the tropical Atlantic since 1995, with a pattern nearly opposite to that seen before 1995. Moreover, the large-scale contributors to the NATL TC frequency increase since 1995 are likely related to both local and remote SSTAs. Quantification of the impacts of local and remote SSTAs on the increase of TC frequency over the NATL basin and the physical mechanisms require numerical simulations and further observational analyses.

El Nino-Southern Oscillation (ENSO) is known to significantly modulate Atlantic tropical cyclone (TC) activity. In this paper, we separate El Nino events into El Nino resurgence (ELR) and La Nina-transition-El Nino (LAT) events according to the warm or cold ENSO phase in the preceding winter. Different influences of the two types of El Nino events on North Atlantic TC activity are explored during the hurricane season of June–November. The average number of TCs is more, with shorter average duration and slightly larger intensity during ELR events than during LAT events. There are more TCs passing through and making landfall at the Gulf of Mexico, Caribbean Sea during ELR events, indicating that TCs may have a greater impact on coastal cities of Gulf of Mexico and Caribbean Sea during ELR events than LAT events. Composite analysis reveals that thermodynamic factors such as relative humidity (RH), sea surface temperature and tropical cyclone heat potential are mainly responsible for the TC activity difference between ELR and LAT events. Genesis potential index diagnosis shows that the RH and potential intensity are the dominant factors that influence TC genesis in the MDR during ELR events. During LAT events, the relative vorticity, vertical wind shear and RH over Cape Verde are favorable for TC genesis. In summary, the preceding winter ENSO phases associated with ELR and LAT events lead to opposite changes of environmental factors, and thus induce different TC activities in ELR and LAT events.

The aim of this work is to provide a classification of tropical cyclones (TC) according to their location, month of genesis and their lifecycles and to study the role of African Easterly Waves (AEWs) in North Atlantic cyclogenesis. Between 1980 and 2004, 269 tropical cyclones (TCs) were formed over the North Atlantic, 77% of which occurred during the August-October period and 95% of major hurricanes (TCs in which the maximum sustained wind was greater or equal to 50 m/s) were generated in the same period. The high activity of the August-October period was due to the favourable thermodynamic and dynamic conditions that simultaneously exist over the main development region (MDR). A classification of TCs according to their months of genesis and lifecycles showed that TCs generated before August and after October have no preferential latitude and longitude of genesis. TCs generated at the East of 60°W and over the southern part of the MDR have long lifecycles. Major hurricanes have long lifecycles and are generally initiated over the southern part of the MDR and East of 60°W. The spatial representation of the cyclogenesis area shows that cyclones are generated along two main axes; one is situated around 12.5°N (southern axis) and the second around 27.5°N (northern axis). 56% (44%) of TCs are generated over the southern axis (northern axis). 92% of TCs initiated along the southern axis originate from African Easterly Waves. While 64% of those generated over the northern axis are linked to mid-latitude baroclinic systems and upper level cold lows. This study contributes to a better understanding of North Atlantic cyclogenesis characteristics as well as the role of AEWs. Key words: North Atlantic tropical cyclones, major hurricanes, African easterly waves

Coastal flooding poses the greatest threat to human life and is often the most common source of damage from coastal storms. From 1980 to 2020, the top 6, and 17 of the top 25, costliest natural disasters in the U.S. were caused by coastal storms, most of these tropical systems. The Delaware and Chesapeake Bays, two of the largest and most densely populated estuaries in the U.S. located in the Mid-Atlantic coastal region, have been significantly impacted by strong tropical cyclones in recent decades, notably Hurricanes Isabel (2003), Irene (2011), and Sandy (2012). Current scenarios of future climate project an increase in major hurricanes and the continued rise of sea levels, amplifying coastal flooding threat. We look at all North Atlantic tropical cyclones (TC) in the International Best Track Archive for Climate Stewardship (IBTrACS) database that came within 750 km of the Delmarva Peninsula from 1980 to 2019. For each TC, skew surge and storm tide are computed at 12 NOAA tide gauges throughout the two bays. Spatial variability of the detrended and normalized skew surge is investigated through cross-correlations, regional storm rankings, and comparison to storm tracks. We find Hurricanes Sandy (2012) and Isabel (2003) had the largest surge impact on the Delaware and Chesapeake Bay, respectively. Surge response to TCs in upper and lower bay regions are more similar across bays than to the opposing region in their own bay. TCs that impacted lower bay more than upper bay regions tended to stay offshore east of Delmarva, whereas TCs that impacted upper bay regions tended to stay to the west of Delmarva. Although tropical cyclones are multi-hazard weather events, there continues to be a need to improve storm surge forecasting and implement strategies to minimize the damage of coastal flooding. Results from this analysis can provide insight on the potential regional impacts of coastal flooding from tropical cyclones in the Mid-Atlantic.

The variability of intense (or major) hurricanes in the Atlantic basin is investigated on both intraseasonal and interannual time scales. Differences are highlighted in characteristics between intense hurricanes and the weaker minor hurricanes and tropical storms. Intense hurricanes show a much more peaked annual cycle than do weaker tropical cyclones. Ninety-five percent of all intense hurricane activity occurs during August to October. In addition, over 80% of all intense hurricanes originate from African easterly waves, a much higher proportion than is observed for weaker cyclones. Of all classes of Atlantic basin tropical cyclones, the intense hurricanes display the greatest year-to-year variability. The incidence of intense hurricanes also has decreased during the last two decades. A small portion of this decreased activity appears to be due to an overestimation of hurricane intensity during the period spanning the 1940s through the 1960s. After adjusting for this bias, however, a substantial downward trend in intense hurricane activity during recent years is still apparent. Given that intense hurricanes are responsible for more than 70% of all destruction caused by tropical cyclones in the United States, an understanding is needed of the physical mechanisms for these observed variations of intense hurricane activity.

A method for early detection of the systems that become tropical cyclones (TCs) in the Atlantic hurricane basin is developed using the SeaWinds scatterometer aboard the QuikSCAT satellite. The method is based on finding positive vorticity signals exceeding a threshold magnitude and horizontal extent within the swath of vector wind observations. The thresholds applied herein are subjectively derived from the TCs of the 1999 Atlantic hurricane season. The thresholds are applied to two sets of data for the 2000 season: research-quality data and near-real-time (< 3-h delay) data (available starting 18 August 2000). For the 2000 research-quality data, 7 of 18 TCs had signals that were detected an average of 27 h before the National Hurricane Center (NHC) classified them as tropical depressions. For the near-real-time data, 3 of 12 TCs had signals that were detected an average of 20 h before NHC classification. The SeaWinds scatterometer is a powerful new tool that, in addition to other conventional products (e.g., satellite images that determine if convection is organized and persistent), could help the NHC detect potential TCs earlier.

&amp;lt;p&amp;gt;We will first examine the skill of probabilistic tropical cyclone (TC) occurrence and intensity (ACE - accumulated cyclone energy) predictions in the Subseasonal to Seasonal (S2S) dataset. We show that some of the models in the S2S dataset have skill in predicting TC occurrence 4 weeks in advance. In contrast, only one of the models (ECMWF) has skill in predicting the anomaly of TC occurrence from the seasonal climatology beyond week 1. For models with significant mean biases, calibrating the forecast can improve the models&amp;amp;#8217; prediction skill. In contrast, for models with small mean biases, calibration does not guarantee an improvement in model skill as measured by the Brier Skill Score.&amp;amp;#160;&amp;lt;/p&amp;gt;&amp;lt;p&amp;gt;We then focus only on the ECMWF model and using cluster analysis examine the sensitivity of the North Atlantic TC tracks biases to various factors, such as model resolution, lead time, and tracking. We also explore how well the ECMWF North Atlantic TC model tracks in each cluster simulate the known response to climate modes, such as ENSO and MJO. By applying simple bias corrections to each cluster of Atlantic TC tracks, we examine if we can improve the model skill in landfall prediction in the US and Caribbean.&amp;lt;/p&amp;gt;