Quantifying the possible existence of undocumented Atlantic warm-core cyclones in NOAA/CIRES 20th Century Reanalysis data



[1] Prior to 1966, the lack of satellite imagery led to an unknown number of tropical cyclones (TCs) in the Atlantic Basin remaining undetected by traditional surface observational networks. Previous research has shown that this historical undersampling has led to difficulties in interpreting long-term trends in TC activity. In this study, advances in global reanalysis methods are used to further guide existing TC Best-Track revision efforts, specifically for the pre-satellite era. First, a series of atmospheric proxies of TC passage in the NOAA/CIRES 20th Century Reanalysis dataset are constructed for the satellite-era. Next, similar magnitude signatures (not corresponding to known Best-Track TCs) are identified in the pre-satellite era and subjected to further synoptic verification. Surface observations are used to categorize candidate events into three confidence bins. A small but significant number of the candidate events are cases that warrant further examination by ongoing reanalysis efforts. As this method shows promise for use in other TC basins around the world, preliminary results suggest that this technique may potentially lead to a more complete climatological record of global TC incidence and an improved understanding of long-term trends in activity.

1. Introduction

[2] The tremendous strides in observational technology over roughly the latter half of the 20th and first decade of the 21st century have revolutionized the forecasting and understanding of tropical cyclones (TCs). After decades of relying solely on surface-based observations, an impressive array of new tools has become available in the past 65 years. Major advances, including the advent of aircraft reconnaissance in 1944, near real-time geostationary satellite imagery in 1966, and remotely sensed microwave data and surface windfield estimates in the past decade [Landsea, 2007] have resulted in improved theories of the fundamental physical processes that sustain TCs and superior operational TC coverage of the Atlantic Basin.

[3] However, evolution of observational platforms has led to large discontinuities in the climatological record, as the percentage of TCs that were detected as they occurred increases with time. For example, Vecchi and Knutson [2008] find that an adjustment to annual TC count is necessary to correct for the relative scarcity of ship reports in the open Atlantic prior to 1965. Additionally, Chang and Guo [2007] show that the percentage of open ocean cyclones in the pre-satellite era (1904–1965) is 12.8% of TC count, increasing to 21.2% between 1966 and 2005. As Walsh et al. [2007] find this disagreement is unlikely to have arisen through chance or spatial variation in TC development preferences, it is well-founded that the climatological record of TCs in the Atlantic Basin is incomplete prior to the advent of satellites.

[4] With the “true” number of historical TCs unknowable, improving the quality and reliability of the climatological database has been a focus of research within the past decade. One such major effort is the ongoing HURDAT re-evaluation, in which Landsea et al. [2008] employ a careful methodology to both systematically revise the track and intensity of existing Best-Track (BT) cyclones and add “new” or remove “existing” TCs to/from the historical record when strongly supported by newly discovered observations or when prior observations are reconsidered in the context of new science. In order to be considered for inclusion in BT, candidate events must satisfy the National Hurricane Center's criteria for adding a new TC to HURDAT by presenting a closed surface circulation, two wind or pressure observations supporting tropical storm intensity, and a non-frontal thermodynamic structure [Landsea et al., 2008]. While the HURDAT re-analysis gathers the widest possible set of available surface data, recent advances in global reanalysis techniques that extend reliable and high-resolution gridded state data back to 1871 may offer a new pathway to objectively identify candidate TC events. In this study, we endeavor to quantify a mean tropospheric signature of Atlantic Basin TCs in the NOAA/CIRES Twentieth Century Reanalysis [Compo et al., 2011] in order to determine whether ongoing BT reanalysis efforts can be enhanced by seeking such TC signatures in the pre-satellite era.

2. Data and Methodology

2.1. The Twentieth Century Reanalysis

[5] In general, a reanalysis can be defined as a hindcasting numerical weather prediction scheme that assimilates historical observations and returns the most likely atmospheric state for a given time [Thorne and Vose, 2010]. However, all of the high-quality global re-analyses released prior to 2009 are dependent on assimilating upper-level radiosonde observations to resolve the vertical structure of the atmosphere; thus, none of the reanalyses produce fields prior to 1948 when such observations are rare. This limits the usefulness of traditional reanalyses as tools for improving pre-satellite-era TC climatology.

[6] The Twentieth Century Reanalysis (20CR) is the first product to make global reanalysis data available prior to the advent of systematic radiosonde data. This is accomplished using a technique first described by Whitaker et al. [2004], in which an ensemble Kalman filter [Burgers et al., 1998] is applied to assimilated sea level pressure (SLP) observations to yield a best guess of the vertical structure of the atmosphere, along with uncertainty of the estimate given the large number of degrees of freedom possible. Two datasets were produced using this technique: the Twentieth Century Reanalysis version 1 (20CRv1) which spans the years 1908 through 1958 [Compo et al., 2008], and the newly-released version 2 (20CRv2) which encompasses 1871 through 2008 [Compo et al., 2011].

[7] Briefly, the 20CR has a spatial resolution of two degrees of longitude and latitude globally and seventeen vertical levels at or below 200 hPa. Gridded output fields and uncertainties are produced four times a day from a 56-member filtered ensemble. The SLP observations assimilated into the model are taken from the International Surface Pressure Databank version 2 [Yin et al., 2008]. In the 20CRv2, estimated minimum central pressure “observations” for TCs from the International Best Track Archive for Climate Stewardship (IBTrACS) [Knapp et al., 2010] are assimilated into the reanalysis model; in 20CRv1 they are not. Using the 20CRv2 ensemble mean as an initial condition, 24-hour forecasts of observed SLP demonstrate significant skill against persistence in the northern hemisphere. For documentation of the technical specifications and performance of the 20CRv2, refer to Compo et al. [2011].

2.2. Reanalysis Climatology

[8] The first step in determining whether the 20CR is a useful tool for aiding studies of TC climatology is to assess the mean response of the reanalyzed historical atmosphere to known TCs. To this end, at each gridpoint within the northern Atlantic TC Basin, defined as 100°–20°W, 0°–60°N, the mean and variance of geopotential height at eight pressure levels are calculated for each of the 1460 six-hourly output times in the annual cycle (removing leap days). All 20CR datapoints within five degrees of an IBTrACS TC for three days before and after TC passage were not included in the calculations of local climatology. This step is an attempt to remove as much of the influence of TCs on the height climatology as possible, consistent with the atmospheric temporal footprint found by Hart et al. [2007].

[9] This process was performed for the full 20CRv1 and an 1891 to 2008 subset of the 20CRv2, yielding the model's assessment of normal heights for a particular location and synoptic time. In a matrix comprised of three lower tropospheric levels (950, 900, and 850 hPa) and three upper tropospheric levels (300, 250, and 200 hPa), the 300–850 hPa thickness yielded the highest average normalized response at TC passage in the 20CRv1 (2.84σ). Therefore, the thickness of the 300 to 850hPa layer is found to be a useful proxy for warm-core cyclones (consistent with Hart [2003]) and serves as the experimental metric in this study.

[10] This database of thickness climatology is used to identify the temporal and spatial thermodynamic signature of TC passage using a modified version of Hart et al.'s [2007] methodology. For thirty days before and after IBTrACS-indicated passage, the normalized thickness anomaly ΔZ in 300–850hPa layer was calculated every six hours using the following:

equation image

Where L is the nearest reanalysis gridpoint to the IBTrACS TC location at passage time, t is the appropriate six-hour timestep in the annual cycle, M is the thickness in the 20CR, and μ and σ are the mean and variance of layer thickness from the climatology of the reanalysis, respectively. This normalization process was shown to reliably identify significant synoptic-scale events by Hart and Grumm [2001]. The procedure is repeated in the 20CRv1 and 20CRv2 datasets for each six-hourly TC position in IBTrACS, excluding tropical depressions and extratropical cyclones. The process yields a set of 9,195 sixty-day TC passage response sequences for the 20CRv1 and 23,570 sequences for the 20CRv2 dataset.

3. Results

3.1. TC Passage Response in 20CR

[11] The composite-mean normalized anomaly series, broken out by BT TC maximum sustained wind at passage, are shown in Figure 1. Positive values of thickness anomaly represent a deeper 300–850hPa layer than the reanalysis climatological value. As layer thickness is a proxy for layer-integrated virtual temperature, warm-core cyclones can be expected to yield positive thickness anomalies. Additionally, the normalization of the thickness anomaly by the local variance further highlights areas of warm-core structure, overcoming limitations in absolute representation due to the lower-resolution nature of the grids. Outside of five days centered on TC passage, there is little to no observed thickness anomaly in the both 20CRv1 and 20CRv2 datasets. Peak response also scales well with intensity in both versions. Mean thickness anomaly at passage was roughly +1.1, +1.7, and +2.7 σ in 20CRv1 and +1.8, +3.9, and +5.7 σ in 20CRv2 for tropical storms, minor hurricanes, and major hurricanes, respectively.

Figure 1.

(top) Normalized 300–850 hPa thickness anomalies for sixty days centered on TC passage in version 1 of the Twentieth Century Reanalysis for the 1908 to 1958 Atlantic Basin hurricane seasons. (bottom) Normalized 300–850 hPa thickness anomalies for sixty days centered on TC passage in version 2 of the Twentieth Century Reanalysis for the 1891 to 2008 Atlantic Basin hurricane seasons. In both plots, results are broken out by the maximum sustained wind at the time of TC passage as tropical storms (17 to 32 ms−1), minor hurricanes (33 to 49 ms−1), and major hurricanes (≥50 ms−1), with N the number of six-hourly TC position points in Best Track belonging to each bin.

[12] This result shows that the reanalysis is capable of capturing to first order the magnitude of the TC's signature. Another significant result is that assimilating estimated minimum MSLPs from IBTrACS in 20CRv2 results in a roughly doubled peak response compared to 20CRv1, in which there is no assimilation of IBTrACS data. Even in version 1, the normalized thickness anomaly at passage is significantly different from zero at a 95% confidence level for both minor and major hurricane intensities. Therefore, even when the 20CR model was only fed with historical surface observations from ships, buoys, and land-based stations for the 1908–1958 period, 20CRv1 is able to meaningfully evaluate a warm-core low in the vicinity of the IBTrACS position without the “help” of directly assimilating a minimum central pressure estimate at that TC location. This result may illustrate the benefit to reanalysis TC representation of assimilating nearby observations over a long time period (e.g., on a tropical wave's track after departing Africa), even when the number of observations at a given time is quite low. A separate test shows that the average thickness anomaly for all named TCs is nearly constant at just under +3σ in both the satellite and pre-satellite eras. Thus, the 20CR analyzes TCs equally well (with respect to the appropriate climatology) in either regime.

3.2. Locating Candidate Events

[13] Given TCs are shown to have a particular passage signature in the 20CR, searching for similar cases that do not correspond to an IBTrACS cyclone position is possible. The criteria applied to locate candidate events identified cases in 20CRv2 six-hourly output for which the given event 1) was located over the Atlantic between 0°–60°N latitude, 2) did not correspond to an IBTrACS TC, and 3) had a normalized 300–850 hPa thickness exceeding the 95th percentile (ΔZ = 1.65σ). To further refine these results and filter out larger-scale warm thickness anomalies caused by interannual variability (e.g., ENSO) and low- and mid-level ridging inconsistent with cyclonic signatures, three-panel synoptic maps were created. These maps illustrate 300–850 hPa thickness anomaly, 850 hPa streamlines, and 850 hPa relative vorticity and SLP in Figures 2a, 2b (left), and 2b (right), respectively. If an IBTrACS TC exists at the synoptic time, its location and identifying number are also plotted in Figures 2a and 2b. Figures 2a and 2b were generated at 12 h time steps for the TC season (1 June to 30 November) during the years 1951 through 1958, which are selected because they had not yet been subjected to the HURDAT re-analysis process at the time of this study.

Figure 2a.

Normalized 300–850 hPa thickness anomaly in version 2 of the Twentieth Century Reanalysis for the North Atlantic Basin, in sigma, for 0000 UTC 18 October 1958.

Figure 2b.

(left) Streamlines at 850 hPa in version 2 of the Twentieth Century Reanalysis for the North Atlantic Basin at 0000 UTC 18 October 1958. (right) Relative vorticity at 850 hPa, shaded in units of 10−5 s−1, and SLP contoured for each 2 hPa in version 2 of the Twentieth Century Reanalysis at 0000 UTC 18 October 1958. Note that while several local thickness anomaly maxima exceeding 2 sigma are observed in Figure 2a, inspection of streamline, vorticity, and SLP data show that only the anomalies in the central Atlantic and western Gulf of Mexico are associated with potential surface cyclones and are considered candidate events.

[14] The resulting 2912 plots are then individually scrutinized for evidence of TCs or TC-like events, which will be referred to as candidate events (CEs). In general, in order to be defined as a CE, an area of interest needs to show persistence by maintaining significant thickness anomalies for two consecutive twelve-hourly periods. Additionally, the anomalies need to be corroborated in low-level fields by demonstrating a compact and symmetric presentation broadly consistent with a warm-core cyclone. Increased consideration is also given to events that possess a closed SLP isobar of 1010 hPa or lower, positive 850 hPa relative vorticity exceeding 2.0 × 10−5s−1, or a closed circulation in the analyzed low-level streamlines.

[15] Applying this search heuristic to the 1951–1958 hurricane seasons in the 20CRv2 yields a list of 68 CEs meriting closer evaluation outside of the model. In each of the eight seasons, there are at least seven 20CR candidate cases where a “missing TC” or track extension is possible, with a maximum of 15 in the 1955 hurricane season. In general, the technique is a hybrid of objective and subjective methods and utilizes the discretion of the synoptician to determine which events are possible missing TCs.

3.3. Evidence for Candidate Events

[16] To demonstrate that this technique is useful, it must be shown that the 20CR can identify events that further analysis shows likely merit inclusion in BT. In order to do this, the 68 CEs identified for the 1951 through 1958 hurricane seasons in Section 3.2 are assessed using historical databases of ship- and land-based observation records. Maritime records were taken from ICOADS Release 2.5 [Woodruff et al., 2011]. Records for land-based station observations were provided by the National Climatic Data Center's Integrated Surface Database Version 2 (ISD) [Lott, 2004].

[17] Over the lifetime of each CE, storm-centered maps are made twice daily, overlaying normalized thickness anomalies and estimated SLP fields from the 20CRv2 with all available historical wind speed, wind direction, and pressure observations at the synoptic time. Each of these maps are subsequently analyzed in accordance with the criteria for adding new TCs to the Atlantic hurricane database outlined by Landsea et al. [2008], with the synoptician specifically looking for a closed circulation, wind reports exceeding 34 knots, and SLP lower than 1005 hPa.

[18] The CEs are subsequently classified into one of three broad confidence categories depending on the level of observational support:

[19] 1. Type 1 events are those for which the available surface observations are sufficient to demonstrate conclusively that there is no real-world TC associated with the reanalysis thickness anomaly.

[20] 2. Type 2 events are those for which there are too few surface observations in the vicinity to come to a meaningful judgment. This category includes cases for which there is observational support for a closed circulation but the thermodynamic structure of the cyclone is ambiguous.

[21] 3. Type 3 events reflect those for which surface observations generally support the existence of a warm core, closed circulation, and tropical storm force winds at some point in the CE window, meaning the event is potentially a “missing” TC in accordance with Landsea et al. [2008] classification criteria.

[22] It should be stressed that this methodology is only designed to act as a probabilistic filter for CEs, and that any event identified would still require the more rigorous HURDAT re-analysis process, only after which would revisions to BT be considered.

[23] Observational verification of the 68 CEs between June of 1951 and November of 1958 yields 34 Type 1 (50%), 20 Type 2 (29%), and 14 Type 3 (21%) events, with an average thickness anomaly over the life of the CE of 1.54, 1.79, and 1.95σ, for Type 1, Type 2, and Type 3 cases, respectively. Just under half (32) of the events occur in the climatological peak months of Atlantic TC season of August and September, as shown in Table 1. Geographically, 53% (47%) of events are primarily east (west) of 60°W. The type distribution of the CEs is distinct between the two regions, with 15 (19) Type 1, 16 (4) Type 2, and 5 (9) Type 3 events in the eastern (western) half. Figure 3 plots the track in the 20CR and classification type of all 54 “missing TC” CEs during the study period. This figure shows that over 80% of Type 2 events are registered in the eastern half of the Atlantic Basin, where observational density remained quite poor in the middle part of the 20th century [Vecchi and Knutson, 2008]; many of these cases involve tropical waves in the Cape Verde region with strong signatures in the reanalysis. While the available reports were not inconsistent with the existence of a TC near the 20CR location, they were simply too few observations to make a case for Type 3 classification. For this reason, it is not surprising that a majority of the Type 3 cases identified are in the Gulf (3 CEs), Caribbean (2 CEs), or southwestern North Atlantic (4 CEs). In these regions, ship traffic and land observations are denser and more reliable, and the time integration of prior observations increases representation quality. Qualitative descriptions of each of the 68 CEs along with quantitative verification data are provided as auxiliary material.

Figure 3.

Track map for all 1951–1958 Atlantic hurricane season “missing TC” CEs identified in 20CRv2, by classification type. Position data is determined every six hours from 20CR SLP fields.

Table 1. Candidate Event Counts for the 1951–1958 North Atlantic Hurricane Seasons by Month and Classification
Type 158783334
Type 232753020
Type 312145114

4. Case Study

[24] In order to demonstrate the process by which individual CEs are evaluated and classified, a complete case study is presented here. The selected event is candidate event #4 of the 1958 TC season, active between 0600UTC on 16 October 1958 and 0000UTC on 19 October. For each six-hour period in this timeframe, a plot was made of the normalized 300–850 hPa thickness anomaly and estimated SLP fields from the 20CR along with all available ship- and land-based wind, pressure, and surface temperature observations from ICOADS and the IDS (e.g., Figure 4). These synoptic maps were analyzed for evidence of a closed circulation and wind or pressure observations supporting TC status. Using Hart's [2003] methodology, plots of cyclone phase space were also made for the life of the CE within 20CR in order to further assess its thermodynamic and dynamic structure.

Figure 4.

(a) Normalized 300–850 hPa thickness anomalies (shaded, in sigma) and MSLP (contoured at an interval of 2 hPa) in version 2 of the Twentieth Century Reanalysis, overlayed with wind direction, wind speed, and surface pressure and temperature observations from all available ICOADS ship reports [Woodruff et al., 2011] at 0000UTC on 18 October 1958. Synoptic surface observations show the existence of a vigorous surface vortex in the vicinity of 35N 60W, while thickness anomalies and surface temperature observations indicate that at least a shallow warm-core is present. On the basis of these observations, this candidate event is designated a Type 3, meaning that the feature potentially represents a “missing TC” and future addition to the Atlantic BT database pursuant to further verification.

[25] Candidate event #4 first appears as an elongated area of 1σ warm anomalies and a sub-1010 hPa low near 24°N, 68°W at 0600UTC on 16 October. There are three wind observations at 1200UTC exceeding 30 kt (16 ms−1), including a 38 kt (20 ms−1) sustained east-northeasterly wind near 28°N, 68°W. Cyclone phase space shows a shallow, symmetric warm core structure to the low on the 16th. Therefore, it is possible that a tropical or subtropical cyclone developed as early as 1200UTC on 16 October 1958, though the surface observations are not unambiguous due to a possible trough axis southwest of the proposed surface center.

[26] As the cyclone continued northeast early on the 17th, surface observations show a vigorous circulation with a large area of winds exceeding 30 kt (16 ms−1) elongated southwest to northeast. While maximum normalized thickness anomalies of over 1.5σ by 1200UTC on the 17th are co-located with a 34 kt (18 ms−1) sustained wind out of the south near 30°N, 62°W, phase space diagrams show increasing thermal asymmetry, which possibly indicates a somewhat subtropical structure to the cyclone – although the resolution of the reanalysis likely introduces a bias into the depth of the warm-core structure within the phase space [Manning and Hart, 2007]. Nevertheless, intensification continues, and at 0000UTC on the 18th, a SLP observation of 999.9hPa is associated with a 58 kt (30 ms−1) sustained wind report out of the southwest near 35°N, 60°W, indicating that the low-level circulation center of a vigorous tropical or subtropical cyclone is likely nearby. Early on the 18th, gale force winds expand outward from the surface center as the cyclone begins to interact with an extratropical low over eastern Canada. Surface temperature observations indicate that the CE completes extratropical transition (ET) by 1200UTC on the 18th, which is supported by the decoupling of warm thicknesses from the surface circulation.

[27] Overall, the case for the addition of 1958 candidate event #4 to the BT database is among the strongest of the 68 candidate events. With roughly one dozen observations of sustained winds exceeding 34 kt (18ms−1) as well as several sub-1000 hPa SLP reports, there is strong evidence to support the existence of a closed surface circulation on the 17th and early 18th. While the thermodynamic evolution of the low remains uncertain, thickness anomaly data suggest the CE was at least partially tropical and in the context of the 20CR resolution as well as prior research, may have been consistent with a purely tropical cyclone that eventually undergoes ET.

5. Conclusion

[28] An investigation of TC signatures in the 20th Century Reanalysis dataset reveals significant tropospheric thickness anomalies versus climatology in the pre-satellite era. Cases of warm anomalies in the deep tropics were filtered by thermodynamic structure, revealing 68 candidate cases in the Atlantic Basin between 1951 and 1958 for which persistent warm-core anomalies do not correspond to an existing BT cyclone. Refining the confidence of the CEs using synoptic observations further narrows the number of potential additions to Best Track in this time frame to roughly one dozen, or approximately 1.5 per year. Results thus far in potentially locating “missing TCs” in the Atlantic Basin suggest that newer reanalysis datasets may be useful in aiding ongoing efforts to improve the climatological database of historical TCs, and may by extension enhance the interpretation of long-term trends in TC activity.


[29] The work was supported by the US National Science Foundation (ATM-0842618) and by the Risk Prediction Initiative of the Bermuda Institute for Ocean Studies. The research has benefitted from discussions and feedback with Andrew Hagen of University of Miami and Chris Landsea of the National Hurricane Center. The authors are grateful to NOAA/CIRES for the availability of the 20th Century Reanalysis. Support for the Twentieth Century Reanalysis Project dataset is provided by the U.S. Department of Energy, Office of Science Innovative and Novel Computational Impact on Theory and Experiment (DOE INCITE) program, and Office of Biological and Environmental Research (BER), and by the National Oceanic and Atmospheric Administration Climate Program Office.

[30] The Editor would like to thank the two anonymous reviewers for their assistance in evaluating this paper.