Tropical cyclones, climate change, and scientific uncertainty: what do we know, what does it mean, and what should be done?

Transcription

1 Climatic Change (2011) 108: DOI /s Tropical cyclones, climate change, and scientific uncertainty: what do we know, what does it mean, and what should be done? Iris Grossmann M. Granger Morgan Received: 5 November 2008 / Accepted: 3 December 2010 / Published online: 17 February 2011 Springer Science+Business Media B.V Abstract The question of whether and to what extent global warming may be changing tropical cyclone (TC) activity is of great interest to decision makers. The presence of a possible climate change signal in TC activity is difficult to detect because interannual variability necessitates analysis over longer time periods than available data allow. Projections of future TC activity are hindered by computational limitations and uncertainties about changes in regional climate, large scale patterns, and TC response. This review discusses the state of the field in terms of theory, modeling studies and data. While Atlantic TCs have recently become more intense, evidence for changes in other basins is not persuasive, and changes in the Atlantic cannot be clearly attributed to either natural variability or climate change. However, whatever the actual role of climatic change, these concerns have opened a policy window that, if used appropriately, could lead to improved protection against TCs. 1 Motivation Coastal areas worldwide have become increasingly vulnerable to destruction by tropical cyclones and the associated severe floods. Tropical cyclones (TCs) 1 are 1 The weakest form of a TC is a tropical depression ( tropical low in Australia and tropical disturbance in the South Indian Ocean); from wind speeds of 18 m/s TCs are called tropical storms, from speeds of 33 m/s hurricane in the Atlantic and Northeast Pacific, typhoon in the Northwest Pacific, severe cyclone in the South Pacific and North Indian Ocean, and tropical cyclone in the South Indian Ocean. I. Grossmann (B) M. G. Morgan Department of Engineering and Public Policy, Carnegie Mellon University, Pittsburgh, PA 15213, USA irisg@andrew.cmu.edu

2 544 Climatic Change (2011) 108: non-frontal low-pressure systems with a closed circulation over tropical or subtropical oceans. Damage from intense TCs results from powerful winds and from floods caused by torrential rains and the wind-driven storm surge. During the earlier part of the 20th century, the US experienced a number of extremely destructive TCs, including the 1900 Galveston Hurricane, the Great Miami Hurricane of 1926, and the Long Island Express of During the 1970s and 1980s, remarkably little damage from TCs occurred in the US, despite the rapid growth of coastal populations. However, since the early 1990s, several intense TCs have caused devastation, with estimated total US damages of almost $50 billion in 2004 and 2008, and more than $100 billion in 2005 (Pielke et al. 2008). While a considerable portion of this increase in damages is due to the recent rapid increase in capital and population in high-risk areas (Pielke et al. 2008), there has also been an increase in hurricane activity in the Atlantic (Kossin et al. 2007) and possibly in other basins (Elsner et al. 2008). In 2005, two studies first reported a correlation between the warming of tropical sea surface temperatures (SST) and measures of TC activity: the number and percentage of the most intense hurricanes (Webster et al. 2005) and the power dissipated by North Atlantic and West Pacific TCs (Emanuel 2005a, b). A third study reported that, relative to other factors, long-term trends in SST play a dominant role for the recent increase in intense TCs (Hoyos et al. 2006). The question of whether global warming may be changing TCs has since become a hot topic in climate policy circles and among those dealing with coastal development, insurance, and emergency response. However, to date, no consensus has been reached on whether such changes can be detected (e.g., World Meteorological Organization 2006) or on how large future changes, if any, may be. This paper provides a semi-technical review of the principle schools of thought on the development of TCs in a warmer climate, the underlying theoretical foundations and the associated uncertainties. We begin with a brief discussion of the main reasons why a simple attribution of the recent increase in TC activity to global warming is not possible, and review the extent to which these issues have been addressed in available reviews. The sections that follow investigate the identified uncertainties in greater detail, focusing in particular on four central questions: 1. What is the theoretical basis for a possible increase in intense TCs in a warmer climate? Given the current understanding of TC formation, how might TC frequency and tracks change? 2. How might large scale climate patterns in the Atlantic be influencing TC activity on multi-decadal time scales? 3. What do we know about the extent of inaccuracies and temporal dishomogeneities in TC databases in different ocean basins? Given these, what conclusions can we draw about trends in TC activity? 4. How can the reliability of current modeling studies be assessed, and what conclusions can be drawn from the available results? We conclude with a brief discussion of the implications, for a range of decision makers, of the present and likely continuing uncertainty about future TC frequency and strength.

3 Climatic Change (2011) 108: Global warming and TC activity: trends and attribution 2.1 Why a simple attribution is not possible In 2005, Emanuel (2005a) reported an approximate doubling of the power dissipated annually by Atlantic TCs over the last 30 years and a correlation of r 2 = 0.65 between the smoothed power dissipation index (PDI) and SST in the Atlantic main development region (MDR). A comparable upswing and correlation were reported for Western Pacific PDI and SST. Subsequent studies pointed to problems in the adjustment of TC data including the exclusion of the endpoints in the smoothing process and an overly large downward adjustment of pre-satellite era winds in the Atlantic to obtain a better fit with available TC pressure values (Landsea 2005). After correcting these issues, a high correlation between the SST and PDI timeseries and a significant upswing is maintained, at least over the last three decades (Emanuel2005b). However, without adjustments to the TC data, Atlantic PDI values during the 1950s and 1960s are as high as recent levels (Landsea 2005). A definitive conclusion is precluded by the fact that the exact adjustments needed for earlier data are not known. In addition to a downward adjustment, an upward adjustment of unknown magnitude may be needed due to possible frequent underestimation of the intensities of earlier intense TCs. The lack of a reliable and consistent dataset of TC tracks and intensities over a sufficiently long period of time is a central difficulty in ascertaining the existence of global warming impacts on TCs (World Meteorological Organization 2006; Kossin et al. 2007). This is due to the large changes in observation techniques and methods that have occurred over the period of interest that spans an era of sparsely populated coasts and very limited ship tracks to the current era with synoptic coverage by high resolution satellites (Landsea et al. 2004, 2006).To addressthis problem, studies have sought to either consider data over a shorter more reliable time period (Klotzbach 2006) or to estimate and correct for likely errors. A number of studies find that the significant upward trends in various TC parameters reported in previous work cannot be upheld after applying data error corrections (Landsea 2005, 2007;Klotzbach 2006; Kossin et al. 2007; Vecchi and Knutson 2008). We discuss these issues in more detail in Section 3. Detecting a possible global warming signal in TC data is also complicated by the potential impact of natural variations on multiple time-scales, including the multidecadal time scale in the North Atlantic (Goldenberg et al. 2001) andthewest Pacific (Chan 2006; Swanson 2007). In the North Atlantic basin, which has the most reliable data, TC activity shows a pronounced increase since the 1980s (Kossin et al. 2007). As we will discuss in Section 2.5, this increase has been suggested to match either with anthropogenic warming (Emanuel 2005a; Holland and Webster 2007) or natural multidecadal variability (Goldenberg et al. 2001; Klotzbach and Gray 2008). TC activity in the West Pacific has also been reported to match a multidecadal pattern with similarly high activity levels in the 1950s 1960s and recent years (Chan 2006; Swanson 2007). Finally, both modeling studies and theory indicate that changes in TC activity due to global warming will likely depend on how several thermodynamic and dynamic conditions develop (Vecchi and Soden 2007a, b; Chan2008). Projections of possible

4 546 Climatic Change (2011) 108: changes in these factors have a high degree of uncertainty, especially at regional scales. Further, even if we knew the future of climate with precision, knowledge gaps, in particular with regard to TC formation, would preclude accurate prediction of the response of TCs (Pielke et al. 2005; Emanuel 2008). We discuss these and other uncertainties in current modeling studies in more detail in Section Treatment of these issues in recent reviews A comprehensive consensus statement on climate change impacts on TCs resulted from the World Meteorological Organization s 6th International Workshop on TCs in They argued that a firm conclusion on the existence of an anthropogenic signal in TC activity could not be drawn, but that some increase in TC intensity with global warming is likely, while changes in TC frequency (and the sign of those changes) remain uncertain (World Meteorological Organization 2006). While the statement considered all relevant sources of insight and the major uncertainties, it is relatively brief, and is now several years out of date. Similar conclusions are reached in a review article by Shepherd and Knutson (2007) who discuss the theoretical background for possible changes in TC intensity, the role of natural variability, recent model projections and observational evidence. Three questions are identified as central to future work: (1) TC data quality issues, (2) the relative importance of thermodynamic versus dynamic factors, and (3) the relative role of natural variability and anthropogenic climate change in the Atlantic. Given recent work, a more detailed discussion of item (1), including data correction schemes and implications for trends is now possible. We provide that in Section 3 of this paper. We also present additional material on item (3), Atlantic modes of variability and relevant historic observations in Section 2.4, and on recent modeling studies in Section 4, before combining these different types of evidence into a critical discussion of the current main positions of the debate in Section 5. Working Group I of the IPCC Fourth Assessment Report concludes that an increase in global peak TC intensities is likely according to high-resolution models, while a global decrease in TC numbers and a possible increase in the North Atlantic is expected with medium confidence according to high and low resolution models (Meehl et al. 2007; their Table 11.2). Of the four cited higher resolution studies that investigate TC intensity changes, three project an increase (Knutson and Tuleya 2004; Walsh et al. 2004; Oouchi et al. 2006) while one projects no significant change (Bengtsson et al. 2007). All five studies cited with respect to changes in global TC frequency project a decrease (Sugi et al. 2002; Yoshimura et al. 2006; McDonald et al. 2005; Bengtsson et al. 2007; Oouchi et al. 2006). No discussion is provided of why the mildly divergent results on intensity yield high confidence while the entirely consistent results on frequency yield only medium confidence. Presumably, the interpretation of the model results draws upon theoretical confidence on these projections, but this is not stated. The current disagreements among modeling studies are likely due to the lack of knowledge on the future development of regional conditions influencing TC activity, uncertainties regarding the response of TCs, and model limitations. Aside from technical model limitations these uncertainties are not discussed in the relevant chapter of the IPCC Report; also little effort is made to combine insight derived from models with theoretical insight and observational evidence. Observational

5 Climatic Change (2011) 108: evidence for an increase in the most intense hurricanes is mentioned (p. 783) but the important fact that differing interpretations of data quality issues permit vastly differing conclusions is not discussed. A decrease in TC frequency is given as a less certain possibility that is attributed to a stabilized atmosphere and a weakened tropical hydrological cycle. A more critical discussion and a more careful attribution of the projected TC frequency changes to projected changes in stability would be appropriate, given the remaining large uncertainties on the process of TC formation. More recent work discusses the potential role of an increase in the difference between middle troposphere and boundary layer moist entropy (which should also act to decrease TC frequency, Emanuel et al. 2008) and changes in sensitivity of TC formation to wind shear in a warmer climate (Nolan and Rappin 2008). 2.3 The theoretical background for an increase in TC potential intensity TC genesis and intensification depend on three thermodynamic factors: (1) thermal energy, (2) moist static instability, (3) mid-tropospheric moisture, and three dynamic factors: (4) vertical wind shear (the variation in wind direction or speed with height), (5) cyclonic vorticity and (6) the Coriolis effect (Gray 1968, 1975). The assumed dependency of the highest potential intensity (PI) of a TC on the thermodynamic disequilibrium between the ocean and atmosphere (Emanuel 1986, 1995) provides the theoretical foundation for an expected increase in intense TCs in a warmer climate. In this framework, absent changes in other factors, rising SSTs alone should lead to a higher number of intense TCs than at present. TCs intensify as air at low altitudes is drawn toward the low pressure at the center of the TC while absorbing heat and moisture from the ocean. This heat is subsequently released as the air rises in the eyewall (Gray 1975; Emanuel 1986). The resulting decrease of inner core pressure leads to further convergence and heat release. This process depends both on structural properties of the TC and on environmental properties. Approaches differ regarding the role of different factors in limiting the PI and in certain assumptions on the atmospheric environment. Here we focus on PI approaches that do not rely on convective available potential energy (CAPE) hypothesized to be present in the TC environment. The PI approach and steady-state model of Emanuel (e.g., 1986, 1995, 1999) is the most widely used PI approach today. Intensification depends exclusively on heat transfer from the ocean. A balance between frictional dissipation and energy production is assumed. The PI is the product of the maximum possible latent heat input from the ocean and the thermodynamic efficiency E with which this heat becomes available to the TC. E is proportional to the difference between the enthalpy (heat content) of the ocean surface and of near-surface air. Thus, the PI depends on the thermodynamic ocean-atmosphere disequilibrium. This approach can be idealized as a Carnot heat engine driven by the heat flow arising from the atmosphere-ocean thermodynamic disequilibrium (Emanuel 1986). This framework suggests an approximately 5% increase in PI for every 1 C increase in tropical SST (Emanuel 1987), considerably less than the upward trends reported by Emanuel (2005a) for the Atlantic and West Pacific, and by Sriver and Huber (2006) who find a 60% increase in PDI for a 0.25 C rise in tropical Atlantic SST.

6 548 Climatic Change (2011) 108: Emanuel s PI approach has proven capable of predicting reasonable upper bounds on TC intensity (Emanuel 1999). Possible limitations involve some of the assumptions made (Smith et al. 2008; Camp 1999), inconsistencies with measurements of outflow temperatures and of air temperatures beneath the eyewall, and a lack of attention to eyewall geometry and eye dynamics (Gray 1995). In the framework of Gray (1993, 2003, 1995), TCs develop and intensify by importing more angular momentum than is dissipated through friction. The largest part of the available latent heat goes into expanding and heating rising air. The resulting inner core warming gradually stabilizes lapse rates within the TC. Upward mass flux and intensification are thus limited by thermal stabilization and the decrease of buoyancy within the eyewall due to entrainment. Intensification continues until upward mass flux is no longer able to compete with the effects of friction and upper level wind export. The particular geometry of the TC s core with its outward sloping eyewall makes it possible for upward motion to occur outside the warm center of the eye. This upward motion and intensification of the TC will continue until thermal stabilization of the eyewall is reached. According to Gray s calculation, the eyewall only has to withstand about 35 40% of the warming and stabilization occurring in the eye. This is because a considerable portion of the temperature increase in the eye, e.g. about 60 65% (Gray 1995), is due to subsidence warming which affects primarily the eye and not the eyewall. In this framework, although eyewall convection can maintain itself for a longer time with higher SSTs, inner core thermal stabilization rather than SST is the basic limiting factor to the PI. This model also assumes that the vertical gradient of upper tropospheric equivalent potential temperature e 2 will only change slightly, if at all, in a warmer climate, implying that the intensity at which stabilization is reached should not change. Important limitations of this approach are the lack of an explicit method for calculating the PI and very little discussion of the core assumptions in the published literature. Independent of which PI theory is used as a foundation, how much of the PI is actually attained depends on conditions other than SST such as wind shear or lack of mid-tropospheric moisture (DeMaria 1996; Evans 1993; Chan and Liu 2004; Michaels et al. 2006; Zengetal.2007) and on structural TC properties (Shea and Gray 1973; Willoughby et al. 1982; Gray1995; KaplanandDeMaria2003; Wang and Wu 2004). Wind shear acts to disturb the structure of the TC, thus making the TC more susceptible to other interferences and less efficient at using ocean heat. Saunders and Lea (2008) consider Atlantic TCs between 1965 and 2005 and find that wind shear explains more variance in hurricane activity than SST does. Michaels et al. (2006) investigate 229 Atlantic TCs and the SSTs encountered by them prior to intensification. They find that while SSTs of approximately 28 C may be necessary to reach major TC status, warming beyond 28 C does not act to further increase the intensity reached. They speculate that this could be due to factors other than SST becoming dominant beyond 28 C. 2 e measures the instability or buoyant capacity of an air parcel; it is calculated as the temperature at 1000 mb after releasing the additional heat energy that is available due to the parcel s moisture content.

7 Climatic Change (2011) 108: Using Atlantic and West Pacific TC data from 1882 through 1995, Emanuel (2000) finds a uniform likelihood that a given hurricane will reach a particular intensity across the interval from initial hurricane strength to the PI. In a 31-year sample of Atlantic TCs, DeMaria and Kaplan (1994) find that only 16% of all TCs and 19% of non-landfalling TCs reached 80% or more of their PI, while the average TC reaches 55% of its PI. Wing et al. (2007) find a typical value of 80% of the PI for Atlantic and West Pacific TCs and 65% for hurricanes. With a modified PI scheme that incorporates wind shear, the fraction of TCs that reach at least 80% of the PI is increased by 67% for a 23 year sample of West Pacific TCs (Zeng et al. 2007). In a warmer climate, shear may increase in the Atlantic and East Pacific and decrease in the West Pacific (Vecchi and Soden 2007b). 2.4 TC frequency and tracks in a warmer climate Potential changes in TC frequency in a warmer world are difficult to project due to the lack of a complete genesis theory and the uncertainty of regional changes in factors influencing TC genesis. In today s tropical atmosphere, conditional instability or sufficient warmth and saturation of surface air to allow sustained rising (condition 2 above) requires temperatures of at least 26 C(Palmén1948). This threshold may shift upward as the troposphere warms (Holland 1997; Royer et al. 1998; Bengtsson 2007). If the conditions 1 4 above are met, pre-existing convective disturbances with sufficient cyclonic vorticity (factor 5) and rotation induced by the Coriolis force (factor 6) may become self-sustaining (Gray 1998). Disturbances typically originate from tropical waves or from low level features in the monsoon flow or the Intertropical Convergence Zone (ITCZ). In the Atlantic, about 60% of all TCs and 85% of major hurricanes are spawned by easterly waves (Landsea and Gray 1992) pressure disturbances that arise from instabilities in the African easterly jet (Dunn 1940). Only a fraction of tropical disturbances develop into TCs. The formation of a warm core vortex has remained the most uncertain phase of the TC life cycle. The two main approaches suggest that either a low level cold core forms first and subsequently warms (Emanuel 2003) or that a warm core forms at higher altitudes and subsequently extends downward (Zehr 1992; Gray 1998). The former approach is supported by extensive modeling studies, while the latter is supported by measurements from East Pacific reconnaissance flights (Zehr 1992) and satellite observations (Ritchie et al. 2003). Globally there is no in-situ positive relationship between SST and TC frequency on the inter-annual time scale (e.g., World Meteorological Organization 2006). TC genesis in a warmer climate may depend critically on the development of other factors such as ENSO, shear, atmospheric stability (Chan and Liu 2004; Bengtsson et al. 2007) and the moist entropy difference between the middle troposphere and the boundary layer (Emanuel 2008; Emanuel et al. 2008). We will discuss relevant model projections in Section 4. It is unclear whether or how TC tracks and genesis regions might change with global warming (World Meteorological Organization 2006; Henderson-Sellers et al. 1998). Shifts in genesis regions could occur, for instance, if the relative suitability of different parts of basins in which TC genesis occurs were to change.

8 550 Climatic Change (2011) 108: The potential role of multidecadal natural variability in the North Atlantic Research has shown strong associations between North Atlantic TC activity and atmosphere-ocean variability on different timescales, including multidecadal (Landsea et al. 1999). On the interannual timescale, major landfalling hurricanes increase almost threefold during La Niña events relative to El Niño events (Gray 1984; Boveetal.1998). During El Niño seasons, upper-level winds from the west increase wind shear over the Atlantic MDR (Goldenberg and Shapiro 1996), thus making conditions less favorable for both TC genesis and intensification. Atlantic TC formation, and to a lesser extent intensification, also show a significant negative relationship with the Saharan Air Layer (SAL) on the interannual scale. The SAL is an elevated layer of Saharan air and mineral dust over portions of the North Atlantic that lasts from late spring to early fall. Four possible mechanisms have been discussed: First, the SAL heats the air above its base by absorbing solar radiation. The resulting warm and dry anomalies below 750 hpa raise the lifting condensation level and increase the energetic barrier to convection (Dunion and Velden 2004; Wong and Dessler 2005). Second, the SAL suppresses convection through dry air intrusion into the TC circulation (Dunion and Velden 2004; Sun et al. 2008). Third, the SAL appears to enhance vertical wind shear by strengthening the low to mid-level easterly flow along its southern boundary (Dunion and Velden 2004). Fourth, the dust aerosols cool tropical Atlantic SSTs, thus reducing available thermal energy (Evan et al. 2008). Of particular interest to this study is the potential impact of multidecadal modes of variability on Atlantic TC activity, as this complicates the detection of global warming signals due to the coinciding of the recent increase in Atlantic TC activity with a natural shift in the Atlantic Multidecadal Oscillation (AMO) (Goldenberg et al. 2001; Kossin and Vimont 2007). The AMO is one of three interrelated Atlantic atmosphere-ocean modes with variability in the multidecadal spectrum in addition to shorter-term variations, next to the North Atlantic Oscillation (NAO) and the Atlantic Meridional Mode (AMM) (Marshall et al. 2001; Kossin and Vimont 2007; Grossmann and Klotzbach 2009). Klotzbach and Gray (2008) find that during the positive phase of the AMO, approximately twice as many major hurricanes occur. The AMO has been described both as a separate phenomenon that excites the AMM (Kossin and Vimont 2007) and as a set of atmosphere-ocean variations that include impacts otherwise associated with the AMM (Gray 1990; Gray et al.1997; Zhang and Delworth 2006). The AMO is thought to arise from variations in the strength of the density-dependent Atlantic thermohaline circulation (THC). Several possible causes of systematic long-term changes in density have been discussed and modeled (Bjerknes 1964;Delworth et al.1993; Wohlleben and Weaver1995; Dickson et al. 1988; Grayetal.1997; Dima and Lohmann 2007). We refer the reader to Grossmann and Klotzbach (2009) for a recent summary of these mechanisms and of observations of SST patterns, salinity and sea ice associated with the AMO. Atmosphere-ocean variations associated with the AMO, AMM and NAO have been documented for about a century (e.g., Walker 1924; Smed1943; Namias1963; Bjerknes 1964; Cartonetal.1996; Marshalletal.2001). However, while the NAO is well established as atmosphere-ocean mode, the existence of the AMO and AMM as intrinsic modes has not been definitively confirmed, largely because of the short duration of the available records and the complexity of the causes underlying these

9 Climatic Change (2011) 108: modes. A recent study attributes variations in tropical Atlantic SSTs over the past century that are otherwise associated with the AMO to changes in radiative forcing (Mann and Emanuel 2006). However, aside from SST patterns, the AMO also manifests in distinct ocean subsurface effects(bjerknes 1964; Zhang 2007, 2008) and climate impacts outside the North Atlantic (Gray et al. 1997; Enfield et al. 2001; Sutton and Hodson 2005), which are not as easily explained by changes in radiative forcing. During the positive AMO and AMM phases, the North Atlantic warms relative to the rest of the tropics. Since PI depends on the difference between local SSTs and tropical mean atmospheric temperatures (Vecchi and Soden 2007a; Swanson 2008), this localized warming causes greater PI increases than would be the case for a uniform warming of the tropics. This may explain the large magnitude of the observed recent changes in TC activity relative to the changes projected to occur with global warming (Vecchi and Soden 2007a; Swanson 2008). The warming is heightened in the eastern part of the basin, enabling TCs to form further east so that they can benefit from a longer time over warm waters (Kossin and Vimont 2007). The impacts of the AMO and AMM also differ from projected global warming impacts in that a range of other favorable conditions besides thermal energy are affected. These include (Gray 1990; Shapiro and Goldenberg 1998; Landsea et al. 1998; Kossin and Vimont 2007;VimontandKossin2007; Xie et al. 2005): Reduced wind shear due to a weakening of both the Westerlies and the North Atlantic trades the latter due to an anomalous cross-equatorial flow induced by the warming of the North Atlantic, Increased low-level vorticity, A northward shift of the ITCZ and surface convergence into the MDR, and The positive influence of slow-moving better organized easterly waves. Garner et al. (2009) use a high-resolution dynamical downscaling model (Knutson et al. 2007) to investigate the relative effect of the changes in wind shear and the thermodynamic environment over recent decades that have been associated with the AMO and AMM. They find that changes to the horizontally averaged temperature profile and SST with constant relative humidity explain about 60% of the increase in PDI that is obtained when all fields, including wind shear, are changed to reflect the changes from 1993 to The reliability of TC databases TC databases were originally created by forecasters to assist coastal populations in assessing risk. Over time, major technological improvements have taken place including: (1) the recording of a larger percentage of TCs; (2) more accurate and continuous intensity estimates; and (3) more complete and accurate tracks. While obviously desirable, these improvements have also introduced considerable temporal inhomogeneities into the resulting time series. The North Atlantic is widely regarded as having the most reliable TC data and the longest TC time series (beginning in1851). However, estimates of errors in the pre-satellite era are large enough to allow different analysts to reach very different

10 552 Climatic Change (2011) 108: conclusions about possible trends. In the sections that follow we first discuss data problems in the North Atlantic and then explore the situation more globally. 3.1 TC undercounting in the Atlantic The earliest Atlantic TC records relied on coastal observations and observations reported by ships upon return to port. Beginning around 1905, observations were transmitted from ships via radio. Mann and Emanuel (2006) suggest that the Atlantic TC record can be regarded as reasonably reliable back into the late nineteenth century because ships could not be warned off from approaching TCs. Several arguments cast doubt on the validity of this assumption, however, in particular in the case of weaker or more short-lived TCs and TCs that did not hit land. Until as late as the early 1960s, a number of TCs were only detected at approximately hurricane stage suggesting that some weaker TCs went undetected. A significant increase in weaker, short-lived TCs since the pre-satellite era has also been shown (Landsea et al. 2009). The Atlantic hurricane database (HURDAT) also contains TCs that struck land without previous observations (Landsea et al. 2004). Dunn and Miller (1960) observe that in the late 1950s only about half of all TCs were initially detected by ships (p. 155), leaving a large number of TCs to be detected only once they approached land. Ships may also have changed course to avoid contact with TCs (Vecchi and Knutson 2008), relying on heuristic methods mariners used to determine the possible approach of TCs prior to gale force winds (Bowditch 1841; Government of the United States 1995; Piddington 1860). Some landfalling TCs were likely also missed even after 1900 due to low coastal population density and limited reporting in parts of the US (Landsea 2007). Finally, not all observations of gale force winds were recorded as TCs, since more than one ship observation, evidence of tropical storm (TS) force winds, and evidence of a closed circulation and the TC s non-frontal character were required (Landsea et al. 2004, 2008). Observational capacities improved considerably with the onset of aircraft reconnaissance in 1944 (Jarvinen et al. 1984). However, the first years were limited by the lack of technologies to measure wind speed (Dorst 2007) and the limitation of regular patrols to a triangle west of 55 West between Bermuda, near St. Croix and Miami (Dunn and Miller 1960). Under-detection during this era is expected in particular for TCs that remained east of 55 W. TCs may also have been missed until late into the satellite era due to the difficulties involved in establishing the tropical characteristics and sufficiently high wind speeds of some TCs (Landsea 2007; Edson 2004). Landsea (2007) reports that between , one additional TC per year could be added with a post-seasonal analysis using new technologies (the Quikscat, the Advanced Microwave Sounding Unit and the Cyclone Phase Space analysis tool which have been operational respectively since 2001, 2002 and 2003). Recently, several different methods have been developed to estimate the number of missing TCs. The first group of methods match satellite-era TC tracks with earlier ship tracks and land points (Chang and Guo 2007; Vecchi and Knutson 2008). Another group of methods considers the ratio of TCs with certain characteristics to total annual counts (Solow and Moore 2002; Landsea 2007). A third group predicts TC counts with a statistical model given their assumed dependence on certain climate variables (Mann et al. 2007a; Solow and Beet 2008).

11 Climatic Change (2011) 108: Frequent problems within the first group of approaches arise from difficulties in defining TC detection restrictively enough. For instance, studies have dropped the requirements for two independent ship observations or for evidence of a closed circulation (as in, respectively, Chang and Guo 2007; Vecchi and Knutson 2008) or assumed that coastal densities were generally sufficient for detection. Thus, Bengtsson and Hodges (2008) point out that the increase in TCs during the satellite era found by Chang and Guo (2007) is confined to open ocean or within 300 km of islands. Some undercounting may also have continued during the calibration period, resulting in a possible low bias. A possible high bias could result from changes in TC track properties in recent years. Approaches that use the ratio of TCs with certain characteristics such as landfalling TCs (Landsea 2007) to total counts are problematic if the relevant ratios changed over time. Holland (2007) and Holland and Webster (2007) show a recent eastward shift of genesis and a reduction in landfall associated with the warming of the eastern Atlantic. If this is due to global warming rather than a cyclic natural variation, the results of Landsea (2007) should have a high bias. Nyberg et al. (2007) estimate missed overall TCs from reconstructions of past major hurricanes. In an extension of their results in Table 1, two annual missed major hurricanes between 1900 and 1950 are assumed (scaled from Fig. 3 in Nyberg et al. 2007). 3 Table 1 also lists the annual average number of missed TCs derived from the assumption that the proportion of short-lived storms (up to 30 hours) has remained constant. Solow and Beet (2008) predict TC counts from MDR SST and the Niño3.4 index of the following winter. Mann et al. (2007a) predict TC counts from August-October SSTs in the MDR and the Niño3.4 and NAO indices of the following winter. 4 This use of the Niño3.4 and NAO indices is problematic insofar as a TC-ENSO relationship has only been shown for the August September October El Niño index concurrent with the hurricane season (Pielke and Landsea 1999), while the NAO index is commonly evaluated during the spring or winter preceding the hurricane season (Elsner 2003). Given the possible confounding of SST trends with increasing detection capabilities, another problem may be the choice of two validation periods that both show increasing MDR SSTs ( and ). 3.2 Intensity bias in the Atlantic Peak intensities prior to 1945 were likely only rarely sampled. This is due to limited sampling, limited instrumentation including the common failure of anemometers at high wind speeds, and the fact that ships or properly equipped land stations were unlikely to be present in the radius of maximum winds at the time of peak intensity (Landsea et al. 2008). Unless equipped with barometers, ships estimated TC intensity via an evaluation of the sea state with the Beaufort wind scale (Jarvinen et al. 1984). This scale 3 The lower and upper boundary for the ratio between major hurricanes and total counts is taken, respectively, as the average number of TCs during satellite-era years with two major hurricanes, and years with at least two major hurricanes. 4 Data available at

12 554 Climatic Change (2011) 108: Table 1 Estimates of the numbers of missed annual Atlantic TCs based on a variety of methods as reported by Chang and Guo (2007) (CG-2007), Vecchi and Knutson (2008) (VK-2007), Mann et al. (2007a) (M-2007), Landsea (2007) (L-2007) Source Time periods and estimated missed TCs per year CG N/A N/A SB VK-2007 Mostly <2.0 Mostly < M N/A N/A N/A N N/A N/A Own L For comparison, the recent reanalysis for (which cannot capture TCs that were not observed by land or ship) identified four additional landfalling and nine non-landfalling TCs (Landsea et al. 2008). Note that CG-2007 start their calculation in 1904 SB-2008 own calculation on the basis of Fig. 2 in Solow and Beet (2008), N 2007 own calculation on the basis of Nyberg et al. (2007) (applicable until ca 1950). N/A not available

13 Climatic Change (2011) 108: specifies differences up to category 1 (Landsea et al. 2004, 2008). Wind speeds of 90 kt (i.e., category 2 strength) were assumed if observations were strongly suggestive; major hurricanes were only recorded when corresponding pressure data were available (Landsea et al. 2004, 2008). In 1900, approximately one quarter of ships had barometers; by 1930, most had barometers (C. W. Landsea, pers. comm. 2007). For , Landsea et al. (2008) indicate a likely wind speed error of 20 kt with a bias toward underestimating the true intensity for open ocean TCs and TCs that hit land in sparsely populated areas. The onset of aircraft reconnaissance enabled considerable improvements in intensity recordings. However, limitations in spatial and temporal coverage continued until geostationary satellites were launched in Until the late 1950s when radar altimeters and Doppler radars became available, wind speed was estimated from visual inspection of the sea surface, successive measurements of the plane s position and sometimes with dropsondes (Rappaport and Simpson 2003). Independent of available equipment, stability concerns precluded aircraft from flying into major hurricanes during the 1940s and 1950s (C. W. Landsea, pers. comm. 2007). The recording of intensities higher than category 3 thus required evidence from sources other than aircraft. Consequently, peak intensities prior to landfall were likely missed in a number of major hurricanes. A high likelihood for undersampling of higher intensities in the pre-satellite era is suggested by Chylek and Lesins (2008). They decompose an index of overall annual hurricane activity into the contributions of different TC categories and find an unrealistically high contribution of lower intensity hurricanes and TS in the preaircraft and pre-satellite eras and an unrealistically low contribution of category 4 and 5 hurricanes. 3.3 Conclusions on trends in the Atlantic basin HURDAT appears to be reasonably reliable since about 1980 (Kossin et al. 2007). Kossin et al. (2007) find a pronounced increase in PDI since 1980, which is unchanged after applying a scheme designed to correct for inaccuracies due to changes in satellite technologies. Emanuel (2007) finds that this upward trend is mostly due to increases in TC frequency and to a lesser degree intensity, while Wu et al. (2008) attribute it to increases in average TC lifetime and frequency. Holland and Webster (2007) and Wu and Wang (2008) find an increase in TC tracks starting in the Eastern portion of the basin with the likely consequence of longer TC lifetimes. In an investigation of intensification rates over the period Balling and Cerveny find no detectable trends. Briggs (2008) uses a Bayesian statistical model and finds that an increase in the number of Atlantic TCs since 1966 is unlikely, while an increase since 1975 is probable. He also finds that the rate at which tropical storms become hurricanes has decreased, while the rate at which category 4 or higher is reached has increased. Several studies find that average intensities are probably unchanged (Wu et al. 2008; Briggs 2008). Given cyclic variability in the data, the period since 1980 or 1970 is too short to allow the comparison of phases of similar levels of TC activity (Vecchi and Knutson 2008). If longer time series are evaluated, uncertainties on the reliability of presatellite era data enable quite different conclusions. The unadjusted HURDAT data do not show an increase in PDI since the previous active phase around 1950 (Landsea

14 556 Climatic Change (2011) 108: ). After applying data adjustments and smoothing, a pronounced increase in PDI and annual average peak wind speed since 1950 has been reported (Emanuel 2005a, b). Sriver and Huber (2006) find a 25% increase in Atlantic PDI between and using ERA-40 reanalysis wind data. Saunders and Lea (2008) find a 35% increase in intense hurricanes and a 40% increase in ACE from Pielke et al. (2008) investigate damages from US hurricanes since 1900 and find that after accounting for increases in coastal population and capital, no trend in damages remains. Holland and Webster (2007) define a pattern of three regimes TC1, TC2 and TC3 over the last century with each having about 50% more TCs than the previous regime (TC1 during , TC2 during , and TC3 from 1995 to present). Aberson (2009) reports, however, that these regimes are not significant statistically. Increases in TC frequency are also reported by Saunders and Lea (2008) and by studies applying correction schemes to estimate missed TCs (Mann et al. 2007a; Chang and Guo 2007). In contrast, Vecchi and Knutson (2008) who compare similar regimes of TC activity do not find a significant increase after correcting for missed TCs. Kossin (2008) reports a significant lengthening of the Atlantic TC season over time with the earliest storms forming earlier and the latest storms forming later than previously. For the more reliable periods and , the reported trends are between 5 10 days per decade with a confidence level of 80 90%. 3.4 Joint typhoon warning center TC data Pacific, Indian Ocean and Southern hemisphere TCs are observed by a variety of agencies, each of which maintains its own database (Guard et al. 1992). The Joint Typhoon Warning Center (JTWC) coordinates data collection and publishes best tracks. Discrepancies between agencies individual databases include differences in wind speed, TCs that are only registered in certain databases, and missing track portions. Data are regarded as much less reliable than Atlantic TC data. The official documentation of JTWC data strongly cautions against using these data without bias corrections or error bars (Chu et al. 2002). The recording of West Pacific TCs began in Reconnaissance flights were carried out between the 1950s and late 1980s. Central Pacific observations began in The East Pacific was observed in and then continuously since 1965 (Chu et al. 2002). North Indian Ocean observations began in the 1970s, South Pacific and South Indian observations in Prior to the launch of MeteoSat-7 in 1997, satellite data availability and coverage in the Indian basins was insufficient with a considerable viewing gap in the South Indian Ocean and a frequent oblique viewing angle in both Indian basins (Guard et al. 1992; Knapp and Kossin 2007; Kossin et al. 2007). 3.5 Global satellite era data While to date satellites have not been able to directly measure wind speed, they enable the estimation of central pressures from TC patterns according to the Dvorak technique (Dvorak 1975). The temperature contrast between the TC s warm eye and the colder cloud tops, which is important for this procedure, is affected by the

15 Climatic Change (2011) 108: Table 2 Percentage of 6-hour periods in major ocean basins during which a TC was present but no satellite observations were available due to insufficient coverage, satellite eclipses, and failures (Knapp and Kossin 2007) Decade Indian Ocean South Pacific West Pacific East Pacific North Atlantic 1980s s s available resolution, the viewing angle and the availability of enhanced infrared satellites (IR) (Landsea et al. 2006). Inaccurate intensities may also result from inherent features of the Dvorak technique, such as constraints on allowable intensity changes. For TCs undergoing RI or weakening rapidly this may result, respectively, in a low and high bias (Gray et al. 1991;Veldenetal.2006). Satellites for TC observation first became available in the 1960s. Observations in the early satellite era were restricted by low resolution and insufficient spatial and temporal coverage (Table 2) (Landsea et al. 2006; Knapp and Kossin 2007). For instance, in 1975, only two weather satellites with 9 km resolution were available globally for TC observation (Landsea et al. 2006). Brown and Franklin (2004) find that 50% of Dvorak estimates in the Atlantic between were within 5 kt of reconnaissance intensities, 25% deviated by at least 12 kt, and 10% by at least 18 kt. Some cases in the 1970s and 1980s were in error by kt or more (Gray et al. 1991). Outside the Atlantic, verification of satellite intensities is hindered by the lack of availability of in-situ measurements. The recent reanalysis of Kossin et al. (2007) aimed to correct for changes in resolution, inconsistencies in image interpretation and differences in the application of the Dvorak technique between basins during The large inaccuracies found for the North and South Indian basins, and to a lesser extent the South and West Pacific, are likely due to the late establishment of sufficient coverage over the Indian basins, the lack of reconnaissance observations and lower training level of forecasters outside the Atlantic and East Pacific, and significant changes in West Pacific wind pressure relationships (Knaff and Sampson 2006). Storm-by-storm reanalyses find that some TCs were misestimated by two and in rare cases even three Saffir- Simpson categories (Landsea et al. 2006; Knaff and Sampson 2006; Wuetal.2006; Hoarau et al. 2006). 3.6 Conclusions on global trends Given the serious data problems discussed above, it is not surprising that studies using the original data or reanalyzed datasets have reached very different conclusions on global trends. Pronounced increases in Atlantic and West Pacific PDI (Emanuel 2005a; Sriver and Huber 2006) and in the number and proportion of category 4 and 5 TCs in all basins since 1970 (Webster et al. 2005) have been reported on the basis of original (non-reanalyzed) data. A large contribution to the results of Webster et al. (2005) likely derives from the West Pacific since this basin accounts for almost 40% of all global TCs and more than 40% of category 4 and 5 TCs. Reliable conclusions on trends in the West Pacific are difficult, however, as data are not reliable prior to 1970 (Chu et al. 2002). Even after 1970, discrepancies between the available databases are

16 558 Climatic Change (2011) 108: likely of a similar magnitude as the identified trends (Kamahori et al. 2006;Wuetal. 2006; Kossin et al. 2007; Lander 2008). Wu et al. (2006) report that if the databases of the Royal Specialized Meteorological Centre of Japan (RSMC) and the Hong Kong Observatory (HKO) are employed, a significant decrease in the percentage of category 4 and 5 hurricanes between the periods and is apparent. Lander (2008) shows that the differences in the calculation of maximum wind speeds between the JTWC and the RSMC, i.e. relying on 1-min and 10-min sustained winds, respectively, is not enough to reconcile differences between the databases. To avoid this problem, Klotzbach (2006) evaluates global TC data on a much shorter time scale, i.e. since He finds no significant trend in global Category 4 5 hurricanes and global Accumulated Cyclone Energy. Studies evaluating data on longer time scales report that due to natural multidecadal variability, the 1950s and 1960s were similarly active to recent years (Chan 2006, 2008; Swanson 2007). With the recent shift to colder conditions in the North Pacific, West Pacific TC activity and Northern hemisphere TC activity in general have been much lower (Maue 2009). To address errors in the data, data have been reanalyzed in two ways, i.e. on a storm by storm basis and based on algorithms. In the first category, a reanalysis of West Pacific JTWC TC satellite data between 1966 and 2005 finds a weak increasing trend in category 4 and 5 hurricanes (Knaff and Sampson 2006). Further storm by storm reanalyses are underway (Landsea et al. 2008;Hoarauetal.2006). In the second category, Kossin et al. (2007) report that they cannot corroborate the presence of an upward trend in global PDI or in the number and percentage of global category 4 and 5 hurricanes since A recent study using the same dataset (Elsner et al. 2008) finds significant upward trends for wind speed quantiles above the 70th percentile since Except for the South Pacific, individual basins also show upward trends in the 0.85, 0.9, 0.95 and quantiles, and all basins show an upward trend for the 0.99 quantile. Of the changes in the considered 30 quantiles, 13 are reported statistically significant ( P < 0.05), 13 not significant, while significance levels for the remaining four are not available. Possible concerns with these results are: (a) the shortness of the evaluated time period relative to multidecadal variability in the Atlantic (Goldenberg et al. 2001) and the West Pacific (Chan 2006), (b) the possibility that the reanalysis may not correctly render the most intense TCs, and (c) the use of a cloud top temperature adjustment in the Indian basins prior to 1997 without a corresponding adjustment to TC eye temperatures. The latter is of concern since the known obliquity of the satellite viewing angle prior to 1997 may have prevented an accurate reading of the eye temperature. 4 Modeling studies of future TC activity 4.1 Uncertainties in current modeling studies Modeling studies have improved significantly since early modeling experiments succeeded to simulate low-pressure systems with wind speeds of at least TS force over tropical oceans (Manabe et al. 1970). However, projections of TCs in a warmer climate continue to be much less reliable than projections of measures of global climate. Possible changes in TC activity as the climate warms depend on the develop-