Guest Commentary from Urs Neu
To understand the influence of climate change on tropical cyclone and hurricane activity, it is crucial to know how this activity has varied in the past. There have been a number of interesting new studies of Atlantic tropical cyclones (TCs) and hurricanes (tropical cyclones with maximum sustained winds exceeding 74 miles per hour) since my review of the topic a couple years ago (see here and here). These newer studies underscore that, while our knowledge continues to improve in this area, key uncertainties persist. In particular, it remains very difficult to confidently estimate trends in the past.
In assessing past trends, one must distinguish between two distinct time intervals: 1) the period of historical observations (mainly after 1850), and 2) the earlier period for which TC activity can only be reconstructed using proxy data. Furthermore, we have to distinguish between trends in tropical cyclone (TC) number and TC intensity–the latter measure is particularly important from the standpoint of impacts. There is no a priori reason to expect these quantities to vary in concert, either in the past, or in the future. Unfortunately, uncertainties are much greater for intensity than for counts.
In this article, I will review our current understanding of Atlantic TC and hurricane trends with respect to: A) the historical record of basin-wide TC numbers; B) the historical record of hurricanes and TC intensity; C) distant past proxy estimates of TC (primarily, hurricane only) counts; and D) distant past proxy measures of TC/hurricane intensity. I will conclude with a discussion of current methods for forecasting Atlantic hurricane activity.
The historical record of Atlantic tropical cyclones of the U.S. National Hurricane Center (HURDAT) goes back to 1850. However, only since the start of the satellite area in the 1970s has an area-wide observation system been available. Before that, the density of observations increased with time, either gradually (e.g. density of ship tracks or settlements on coasts), or stepwise (e.g. the introduction of reconnaissance flights in 1944 and launch of GEOS satellites in 1975). There was also subjectivity in various analysis methods, be they the interpretation of local observational data from stations and ships, or of satellite pictures (Dvorak method). Some improvements are obtained by the reanalysis of the whole data set with the same method (e.g. Kossin et al. 2007) or by the analysis of additional data (past meteorological observations not yet included in the analysis, e.g. Landsea et al. 2008).
However, the inhomogeneity caused by a changing and incomplete areal coverage of measurements before the start of the satellite era in 1975 will never be completely eliminated. The only way to correct for this is to estimate the number of TCs ‘missed’ by the observations in earlier times, using indirect methods. One way to do this is to estimate average numbers of ‘missed’ TCs for certain periods using the relationship of general parameters which are known in the past to the TC numbers in the satellite era. While such estimates won’t reveal the ‘right’ TC numbers for individual years, the average ‘missings’ (“undercount bias”) will improve the analysis of long-term trends.
In recent years there have been a number of attempts to estimate the undercount bias. A first attempt of Landsea (2007) using the percentage of landfalling TCs as a basis has been argued to be implausible, as the percentage of landfalling TCs has multidecadal variations and is thus not constant as assumed (Holland 2007). A second approach, by Mann et al (2007), used the statistical relationships of seasonal TC counts to climate variables (NAO, El Niño, and Main Development Region SST). A third method analyzed past ship tracks (Chang and Guo 2007, Vecchi and Knutson 2008 (“VK08″)). These estimates used a new reanalysis of the ICOADS ship track data. Reanalysis of the ICOADS data for 1911-1925 has led to the detection of about one additional TC per year (Landsea et al. 2008), which suggests that the same reanalyis of pre-1911 data might also lead to the detection of additional TCs.
A recent analysis (Landsea et al. 2009, “L09″) looked at trends of TCs of different life times and estimated the undercount bias of medium to long living TCs (more than two days) using the ship track method of VK08. They find that most of the positive trend in total TC numbers over the record originates from a strong positive trend of short lived TCs (two days or shorter) and, accordingly, that there has been no significant long-term trend in moderate to long-lived TCs. Moreover, the observed long-term decrease in average TC life-time is also explained by the increase in short-lived TCs. The authors discuss the possible reasons of the increase in short-lived TCs and suggest that the trend is primarily due to improved observations rather than a real phenomenon.
The approach of looking at short-lived TCs seems basically reasonable, since the probability of missing short-lived cyclones with local observations (ships etc.) seems higher than for longer-lived TCs. The strong trend in short-lived TCs and absence of the same for longer-lived TCs is an interesting result. However, the nature of the observed strong trend of ‘shorties’ is very unclear: If we assume that the whole trend in short-lived TCs is due to observational changes, the undercount bias before 1920 would be about two TCs per year on average compared to the 1975-2000 period (see Fig. 2 in L09; about 0.5 vs. 2.5 TCs per year). However, the undercount bias of short-lived TCs revealed by the ship track (VK08) method is only between 0.5 and 1 per year for the period 1880-1920 (see Fig. 4 in L09). Thus there is a clear discrepancy between the bias suggested by the ship track reconstruction and a bias resulting from the hypothesis of no trend in short-lived TCs. At first sight the result of the ICOADS reanalysis (Landsea 2008) for 1911-1925 suggesting the possible detection of one additional TC per year before 1910 would be a nice explanation for the discrepancy of about 1 to 1.5 TCs per year. However, a quick look at the newly detected TCs for 1911-1925 reveals that most of them have a life-time of more than two days and thus do not add to the trend of short-lived TCs. What else could explain the discrepancy? The VK08 method assumes no trend in TC number. Thus the method only allows to test if it is possible to reject the hypothesis that there is no change in TC frequency. If the analysis had revealed a significant trend, there would be an inconsistency with the method (because less or more cyclones in the early period would alter the bias correction).
Another new paper (Emanuel 2010, currently in discussion at JAMES) provides additional information: A downscaling method (described in Emanuel et al. 2008) using a reanalysis data set driven by surface temperature, pressure, and sea ice for the period 1908-1958 reveals an increasing trend of short-lived TCs (as defined in L09), which is not much smaller than the trend in the HURDAT record. This indicates that the observed positive trend in short-lived TCs, which is particularly pronounced in this period, might well be real and only partly due to an observational bias. On the other hand, the analysis of all TCs in the North Atlantic reveals a trend for 1908-1958 that is significantly lower than the HURDAT record, which is in line with the existence of a general undercount bias.
Another explanation that has been offered for the strong increase in short-lived TCs during the last decade is that due to new analysis methods, like e.g. Quikscat, additional short and weak cyclones have been detected. While Landsea (2007) estimated the corresponding ‘overcount bias’ at about one TC per year, Landsea et al. 2009 suggest now that this bias might be much larger. To explain the recent increase it would require to be about two TCs per year. However, this hypothesis is mainly based on one year with an exceptional high fraction of short-lived and relatively weak TCs (2007 with 9 out of 15). For the other recent years these fractions (2006: 1 out of 9; 2008: 4 out of 16; and 2009: 3 out of 9) were not far from the average before 2000 (about 2.5 out of 9). With regard to the high variability of the number of short-lived cyclones as well as their fraction of the total number, it is hard to see any evidence for an observational bias that is higher than suggested by Landsea (2007) which is based on those storms actually detected through reanalysis of Quikscat data (less than one per year).
In summary, the new analyses reveal that most of the long-term trend in TCs over the hurricane record is due to an increase in short-lived TCs and that there seems to be no significant trend of medium to long-lived TCs. However, it seems unlikely that this trend can be explained solely by an observational bias. Establishing the underlying physical reasons for these trends is a challenge for future research.
If we are interested in damages, of course trends in the number of hurricanes or even major hurricanes is arguably more important than trends in e.g. basin-wide TC counts. Most of the undercount bias discussion has focused on tropical storm number in general. The problem of observational biases in the historical record for hurricanes is more or less the same than for tropical storms in general. On the one hand, the probability for a tropical storm system with hurricane force winds to remain undetected likely is smaller than for a weak tropical storm (because of its higher wind speed and longer life-time). But on the other hand, the probability that hurricane force winds in a tropical storm are missed might be similar to the one for missing a weak tropical storm, because in a hurricane the diameter of the area of hurricane force wind is only about half that of tropical storm force winds. Moreover, it has to be considered that ships tried to avoid strong winds if ever possible. Thus the ship track method to estimate the undercount bias might underestimate the number of missings. However, the analysis of ship records and observations uses extrapolations if there is evidence for stronger winds than are measured, which would compensate at least partly for the ‘avoiding problem’.
Until now the analysis methods described above have rarely been extended to hurricanes. The VK08 ship track method has recently been applied to hurricanes, with the above mentioned caveats (Vecchi and Knutson, submitted). The results show that more or less as many hurricanes per year would have been ‘missed’ as tropical storms (rising from 1 in the mid-20th century to about 3 in the 1880ies). This would turn the long-term trend from positive to slightly negative. Even considering the caveats of the method, this analysis shows that uncertainties in the record are too strong to get any reliable trend information for hurricanes.
Moreover, confident estimates of trends in intensity or related indices (like “ACE” or “PDI”) are even more difficult to achieve, since their estimation requires wide areal data coverage. While the HURDAT data base contains an ACE index calculation, the uncertainties are probably quite a bit larger than for other metrics. A downscaling-based PDI estimate over the period 1908-1950 (Emanuel, 2010) reveals a factor of two difference with HURDAT. It has been argued that wind speed has been overestimated even during the aircraft reconnaissance era (1944-1970), based on shifts in the observed wind speed/pressure relationship (Landsea 1993, Emanuel 2007). There is also evidence for an underestimation of wind speed in ship-based and coastal observations before the change from Beaufort to anemometer wind measurements (e.g. Cardone 1990). This change took place spatially and temporally inhomogeneous.
The problem of reconstructing past basin wide hurricane activity is the lack of spatially comprehensive proxy data because a) most of the area where hurricanes occur is open ocean, b) hurricanes are of geographically limited extent, and c) the number of occurrence at a given location is very restricted. Local proxies of hurricane occurrence only give a very patchy impression of what has happened. There are two options to tackle these problems: 1) to summarize as much individual records as possible (over space and time) to circumvent the high spatial and temporal variability to some extent, and 2) to look for proxies of spatially more homogeneous parameters which are linked to basin wide hurricane activity.
Until now, there have been made reconstructions of hurricane activity for several locations at the American coast (e.g. Donnelly and Woodruff 2007, Elsner et al. 2008). However, these reconstructions can only give a very rough impression due to the high spatial variability of hurricane activity. Donnelly and Woodruff e.g. found some correlations of past hurricane activity to El Niño and West African Monsoon, relations that are well-known from current records.
Until recently there has been only one attempt to reconstruct basin wide past activity: Nyberg et al. (2007) have used proxies which they suggested to represent basin-wide wind shear for such a reconstruction. The problems of this reconstruction – mainly an opposite long-term trend of their two proxy series, problems with calibrations, and discrepancies of reconstructed and observed records – have been discussed by Neu (2008).
Recent work by Mann et al. (2009) has sought to employ and compare both approaches, i.e. integration of as many individual sites as available as well as using a statistical model based on proxies for the key governing climate variables. They present two independent reconstructions of hurricane activity over the last 1500 years. Both reconstructions consistently show a peak in Atlantic hurricane activity during medieval times, although the subsequent decrease is not synchronous. The statistical model allows to explain the medieval peak, which is similar to current levels of activity, by the La-Niña-like climate conditions and a relatively warm tropical Atlantic. This work is an important step forward and provides the most comprehensive information that is possible with existing records. However, as the authors acknowledge, the number of individual records is still rather low, and proxies for potentially important large-scale influences (e.g. African monsoon) are not used. Nevertheless, although the assessment leaves many open questions, there are some features which seem reasonably robust and match the current knowledge on external influences on Atlantic hurricane activity, like ENSO and surface temperature. However, there is still lots to be done.
While the number of tropical cyclones or hurricanes in the distant past is difficult to expolore, changes in tropical cyclone intensity is even more demanding. Most proxies like overwash sediments are based on cut-off effects, i.e. a yes-no information (i.e. if an event lead to an overwash or not). The amount of deposit is very difficult to link to intensity since there are a lot of confounding factors. Proxies linked to basin-wide information on existing pre-conditions influencing tropical cyclone strength might be a solution. However, corresponding uncertainties are so high that there is little chance to get a reasonable signal-to-noise ratio to distinguish any trends, at least with existing proxy records.
Inherent uncertainties in the observation Atlantic TC/hurricane record preclude confident estimates of trends prior to the mid 20th century. This does not mean that was cannot draw some instructive conclusions from the record. For example, it is possible to draw useful conclusions regarding the factors that influence TC and hurricane activity from interannual through interdecadal timescale relationships, even if long-term trends might be compromised by observational biases. These relationships, in turn, might inform our understanding of future climate change impacts on Atlantic tropical cyclone behavior.
It is worth a short remark on last year’s hurricane season: As is expected from the development of El Niño, the hurricane season 2009 was rather quiet (9 tropical storms, 3 hurricanes, 2 major hurricanes). Exceptional was the development of a tropical storm (“Grace”) from an extratropical cyclone as far northeast as never observed before (near the Azores), which only became extratropical some 200 miles southwest of the British Isles. It is difficult to speculate about the meaning of such single events (like also the first tropical storm in the South Atlantic some years ago), e.g. if this is a signal of the extension of the area of occurrence of tropical storms, since these phenomena would likely not have been detected before the satellite period and also within that period not for sure.
Finally, some remarks about the outlook for the current season: Current atmospheric and oceanic conditions seem very conducive for an active hurricane season, including very warm SSTs in the tropical Atlantic and the Caribbean and a medium probability for the development of La Niña which is supportive for the development of tropical storms in the Atlantic. For a more detailed description see the seasonal forecast of NOAA. The forecasts from different authors are almost all in the same range as the one from NOAA, including the one of Klotzbach and Gray, who are applying the third forecast algorithm in the last four years, since the older ones haven’t shown any forecast skill at all (correlations near zero). However, seasonal forecasts seem really tricky, as the attempts of Klotzbach and Gray to find predictive patterns in atmosphere and ocean have shown. While their patterns show very good hindcast skills over about 50 years (correlations of about 0.8), the forecast skill for the following years immediately dropped near zero. It is not clear if they have tested the hindcast skill for different periods (e.g. even vs. odd years) and could have found out earlier. If their new method is more skillful, the years to come should demonstrate this.
A recently developed forecasting approach by Sabbatelli and Mann has shown promising skill, but until now has only been applied for two years (2007/2009), which provides a limited basis for evaluation. For 2010, the method predicts between 23 +/-5 named storms, a number that is somewhat higher than the other forecasts.
The current forecast range for the 2010 season based on all published predictions is:
14-28 tropical storms
3-7 major hurricanes
It will surely be instructive to do a post-mortem on the forecasts when the season is done.
I’d like to thank Gabriel Vecchi for helpful discussions and Mike Mann and Chris Landsea for their comments.
Cardone V. J., J.G. Greenwood, and M. A. Cane, 1990: On trends in historical marine wind data. J. Climate, 3, 113–127.
Chang E.K.M, Y. Guo, 2007: Is the number of North Atlantic tropical cyclones significantly underestimated prior to the availability of satellite observations? Geophys. Res. Lett., 34, L14801.
Donnelly, J. P. and Woodruff, J. D., 2006: Intense hurricane activity over the past 5,000 years controlled by El Niño and the West African monsoon. Nature 447, 465–468
Elsner, J. B., Jagger, T. H. and Liu, K. B., 2008: Comparison of hurricane return levels using historical and geological records. J. Appl. Met. Climatol. 47, 368–374.
Emanuel K., 2007: Environmental factors affecting tropical cyclone power dissipation. J. Climate, 20, 5497-5509.
Emanuel K., R. Sundararajan, and J. Williams, 2008: Hurricanes and global warming: Results from downscaling IPCC AR4 simulations. Bull. Amer. Meteor. Soc. 89, 347‐367.
Emanuel K, 2010: Tropical Cyclone Activity Downscaled from NOAA‐CIRES Reanalysis, 1908‐1958. J. Adv. Modeling Earth Syst., in press.
Holland G.J., 2007: Misuse of Landfall as a Proxy for Atlantic Tropical Cyclone Activity. EOS Trans., 88, 349-356.
Landsea, C. W., 1993: A climatology of intense (or major) Atlantic hurricanes. Mon. Wea. Rev., 121, 1703–1713.
Landsea C.W., 2007: Counting Atlantic Tropical Cyclones Back to 1900. EOS Trans., 88, 197-208.
Landsea C.W., D.A. Glenn, W. Bredemeyer, M. Chenoweth, R. Ellis, J. Gamache, L. Hufstetler, C. Mock, R. Perez, R. Prieto, J. Sanchez-Sesma, D. Thomas, and L. Woolcock, 2008: A Reanalysis of the 1911–20 Atlantic Hurricane Database. J. Climate 21, 2138-2168.
Landsea C.W., G.A. Vecchi, L. Bengtsson, and T.R. Knutson, 2009: Impact of Duration Threshods on Atlantic Tropical Cyclone Counts. J. Climate, DOI: 10.1175/2009JCLI3034.1 (published online)
Mann, M.E., Sabbatelli, T.A., Neu, U., 2007: Evidence for a Modest Undercount Bias in Early Historical Atlantic Tropical Cyclone Counts, Geophys. Res. Lett., 34, L22707, doi:10.1029/2007GL031781.
Mann M.E., J.D. Woodruff, J.P. Donelly, and Z. Zhang, 2009: Atlantic hurricanes and climate over the past 1,500 years. Nature, 460, 880-883.
Neu, U., 2008: Is recent major hurricane activity normal? Nature 451, E5
Nyberg, J. et al., 2007: Low Atlantic hurricane activity in the 1970s and 1980s compared to the past 270 years. Nature 447, 698–702
Sabbatelli T.A, M.E. Mann, 2007: The influence of Climate State Variables on Atlantic Tropical Cyclone Occurrence. J. Geophys. Res., 112, D17114.
Scott, D. B., Collins, E. S., Gayes, P. T. and Wright, E., 2003: Records of prehistoric hurricanes on the South Carolina coast based on micropaleontological and sedimentological evidence, with comparison to other Atlantic Coast records. GSA Bull. 115, 1027–1039.
Vecchi G.A., T.R. Knutson, 2008: On Estimates of Historical North Atlantic Tropical Cyclone Activity. J. Climate, 21, 3580-3600.
Vecchi G.A., T.R. Knutson, 2010: Estimating annual numbers of Atlantic hurricanes missing from the HURDAT2 database (1878-1965) using ship track density (submitted)