To help understand possible impacts of anthropogenic greenhouse warming on hurricane activity, we assess model-projected changes in large-scale environmental factors tied to variations in hurricane statistics. This study focuses on vertical wind shear (Vs) over the tropical Atlantic during hurricane season, the increase of which has been historically associated with diminished hurricane activity and intensity. A suite of state-of-the-art global climate model experiments is used to project changes in Vs over the 21st century. Substantial increases in tropical Atlantic and East Pacific shear are robust features of these experiments, and are shown to be connected to the model-projected decrease in the Pacific Walker circulation. The relative changes in shear are found to be comparable to those of other large-scale environmental parameters associated with Atlantic hurricane activity. The influence of these Vs changes should be incorporated into projections of long-term hurricane activity.
 We explore 21st Century projected changes in Vs over the tropical Atlantic and its ties to the Pacific Walker circulation, using a suite of coupled ocean-atmosphere models forced by emissions Scenario A1B (atmospheric CO2 stabilization at 720 ppm by year 2100) for the Intergovernmental Panel on Climate Change 4th Assessment Report (IPCC-AR4). Changes are computed between two 20-year periods: 2001–2020 and 2081–2100 (use of linear trends or other averaging periods does not alter the character of the results presented here). Our index of the strength of the Pacific Walker circulation is the difference of SLP averaged over the eastern (160°W–80°W, 5°S–5°N) and western (80°E–160°E, 5°S–5°N) equatorial Pacific Ocean [Vecchi et al., 2006; Vecchi and Soden, 2007, hereinafter referred to as VS07]. We define Vs as the magnitude of the vector difference between monthly-mean winds at 850 hPa and 200 hPa (Vs = ∣u850 − u200∣) following a typical Vs definition in the literature [e.g., Goldenberg et al., 2001; Zhang and Delworth, 2006]. For models where daily data was available we found little difference in the 21st Century Vs changes computed using daily winds and monthly winds over the global tropics. See Auxiliary Material Text S1 for a list of models used. We restrict our attention to changes in Vs during the northern Atlantic hurricane season (Jun.–Nov.), though the results hold for other subsets of boreal summer/fall months.
Figure 1a shows the 18-model ensemble-mean projected change in Vs (normalized per °C global warming) over the 21st Century; for reference, contours show the background Vs. There is a prominent increase in Vs over the tropical Atlantic and East Pacific (10°N–25°N) (Figure 1a), which is distinct from a tendency for weakened Vs across much of the northern hemisphere tropics (see below). The amplitude of the projected Vs increase is considerable, given the 1.5–3.5°C global-mean surface air temperature increase in these models by the end of the 21st Century [Held and Soden, 2006; VS07]. These Vs changes are robust across the multi-model suite, with all but a handful of models projecting an increase in the 21st Century (Figure 1b). We define the tropical Atlantic region in which there is large increase of Vs in the ensemble mean (90°W–40°W, 13°N–25°N) as the “Shear Enhancement Region” or SER (see Figure 2a). The Scenario A1B 21st Century Vs changes in the SER are between −2% and 30% of the mean shear.
 On interannual timescales, changes in the Pacific Walker circulation associated with El Niño have been connected to enhanced shear over the tropical Atlantic, via atmospheric teleconnections from the related eastward shift of equatorial Pacific atmospheric convection [e.g., Pielke and Landsea, 1999; Camargo et al., 2007]. Here we explore the extent to which the model-projected increase in Vs is related to the model projections of a weakened Pacific Walker circulation over the 21st Century [e.g., Held and Soden, 2006; VS07]. Figure 2a shows the inter-model correlation between the change in the Pacific Walker circulation index and the change in Vs at each location; warm colors in Figure 2a indicate regions where a decrease in the Pacific Walker circulation is associated with increased shear. Notice that the region of strongest correlation corresponds to the SER. That is, inter-model differences in the region of largest ensemble-mean shear increase are correlated to the deceleration of the Walker circulation in each model. The connection between decreased Pacific Walker circulation and increased shear in these models is further highlighted in Figure 2b. The models with larger Walker circulation weakening tend to show larger Vs increase over the SER region (the correlation coefficient across models is 0.71; p < 0.05).
 We note that the SER is displaced to the north of the region of most frequent cyclogenesis over the period 1981–2005, which we shall refer to as the “Main Development Region” or MDR (60°W–20°W, 8°N–15°N; see Figure 1). We chose to define Vs as ∣u850 − u200∣ because there is substantial literature indicating some relationship between Vs defined in this manner and hurricanes. Over the SER this definition captures the principal wind features that contribute to vertical shear (Figure 3a). However, over the MDR, both the model background and ensemble-mean change of tropospheric vertical wind shear are better captured by the difference between 700 hPa and 150 hPa winds (Figure 3b). The IPCC-AR4 models show a statistically significant (p < 0.05) increase in MDR shear between 700 hPa and 150 hPa (Figure 3b). To the extent that the effect of an increase of 700 hPa to 150 hPa wind shear of equal relevance to that of 850 hPa to 200 hPa wind shear, the multi-model ensemble also projects an increase in shear over the MDR. If one adopts an alternative definition for vertical shear as the vertical standard deviation of wind over the model free troposphere (850 hPa–150 hPa), rather than the magnitude of the vector difference at two pressure levels, the models project a substantial increase of shear over both the MDR and SER (not shown).
 So far we have focused on the June–November tropical North Atlantic shear, though there are robust Vs changes evident globally, in other seasons (e.g., auxiliary material) and in the annual mean. For example, between 20°–40° latitude in the southern hemisphere (and both hemispheres in the annual-mean) there is a zonally-symmetric Vs increase (e.g., Figure 4a). Within 5° of the Equator there is a noticeable weakening of Vs over all three oceanic basins (Figure 4a), which is present in all seasons. In these models the near-equatorial Vs weakening appears related to their robust weakening of near-equatorial zonal overturning [e.g., Vecchi et al., 2006; VS07], resulting from global thermodynamic constraints [Held and Soden, 2006].
3. Changes in Other Hurricane-Related Indices
 Increases in lower tropospheric absolute vorticity (η850), mid-tropospheric relative humidity (rh700) and Emanuel's  hurricane maximum potential intensity for velocity (MPIv) have been linked to increased hurricane activity. Emanuel and Nolan  have developed a “Cyclone Genesis Potential Index” – or GPI – which looks at the combined effect of all four parameters on storm genesis. As is shown in the auxiliary material, changes in the various terms would have comparable effects on GPI if their fractional changes are similar. In Figure 4 we compare the fractional changes in the parameters relevant to GPI.
 The changes in η850 are an order of magnitude smaller than those of the other parameters and therefore not shown. The tropical Atlantic rh700 changes are dominated by drying over the Caribbean Sea (Figure 4b). Tropical-mean rh700 shows very little change, consistent with the largely Clausius-Clapeyron driven increase in specific humidity of these models [Held and Soden, 2006]. Many of the regional rh700 changes appear connected to the local changes in 500 hPa pressure velocity (ω500, contours in Figure 4b), with regions of anomalous descent (ascent) showing relative drying (moistening)-a relationship consistent with anomalous advection of drier (moister) air from above (below).
 While June–November MPIv increases over most of the northern hemisphere tropics, there is a large region in the northern tropical Atlantic where the ensemble-mean MPIv actually decreases (Figure 4c). This region of MPIv decrease is associated with a relative minimum in the sea surface temperature (SST) warming (contours in Figure 4c). MPIv changes around the globe track the structure of SST changes very tightly – with regions that warm more (less) than the tropical mean showing an MPI increase (decrease). Since changes in upper tropospheric temperatures are determined by changes in the tropical-mean SST, rather than changes in local SST [e.g., Sobel et al., 2002], a local minimum (maximum) in surface warming results in an anomalous increase (decrease) in static stability. This relationship between MPIv and local SST changes (relative to the tropical mean SST change) holds not only for the ensemble mean, but also for each of the models. A similar mechanism has been suggested to be important in the El Niño response of tropical Atlantic hurricane activity [Tang and Neelin, 2004]. Understanding the processes that control both regional and global tropical SST changes [e.g., Knutson et al., 2006; Santer et al., 2006] is essential for projecting regional MPIv changes. The SST warming minimum in the tropical Atlantic is also present in the ensemble-mean of IPCC-AR4 climate model runs with a mixed-layer ocean forced with a doubling of CO2 (not shown), suggesting that the minimum in surface warming may result primarily from changes in atmospheric forcing, rather than from ocean dynamics.
 The multi-model ensemble-mean change in GPI is shown in Figure 4d. Model-projected GPI increases substantially in the western and central Pacific, but the changes in the tropical Atlantic and East Pacific are more modest – showing both regions of increase and decrease – due in part to the local increase in wind shear (e.g., Auxiliary Figure S1). In the multi-model ensemble, the North Atlantic and East Pacific contribution of Vs to the fractional change in GPI is comparable to that of each of the other three terms (Auxiliary Figure S1), although the region of largest percentage Atlantic GPI changes caused by shear is a region of relatively modest GPI.
4. Summary and Discussion
 Global climate model projections for the 21st Century indicate a robust increase in June–November vertical wind shear in the tropical Atlantic and East Pacific Oceans. Over the Caribbean Sea, the northern tropical Atlantic (the SER) and the eastern tropical Pacific, the multi-model ensemble-mean shear increases by 0.5-1 ms−1 per °C global warming (Figures 1 and 3). The Atlantic shear changes result largely from changes to upper tropospheric zonal winds (Figure 3). Aspects of the projected shear increase in the SER are strongly related to a reduction in Pacific Walker circulation, with the inter-model variability in Walker circulation changes explaining ∼50% of the inter-model variability in SER shear change (Figure 2). The relative amplitude of the shear increase in these models is comparable to or larger than model-projected changes in other large-scale parameters related to tropical cyclone activity (Figure 4), indicating that these shear changes should be considered in projections of future changes in tropical cyclone activity. Based on published connections between large-scale environmental parameters and hurricane activity [e.g., Emanuel and Nolan, 2004], the changes shown here alone would not suggest a strong anthropogenic increase in tropical Atlantic or East Pacific hurricane activity during the 21st Century; although other regions (e.g., Indian and western/central Pacific oceans) show consistent changes towards more hurricane-favorable conditions (Figure 4).
 In addition to impacting cyclogenesis, the increase in SER shear could act to inhibit the intensification of tropical cyclones as they traverse from the MDR to the Caribbean and North America (e.g., Auxiliary Material Figure S2). Although the response of the frequency and intensity of tropical storms to the shear changes documented here remains to be fully understood, the robustness of the shear changes across models, their impact on GPI (Figure 4d and Auxiliary Material Figure S1), and the potential influence of shear on cyclone intensity underscore their importance in projections of future Atlantic hurricane activity.
 The detailed mechanisms behind the modeled Tropical Atlantic Vs changes should be comprehensively explored, in order to fully understand the robustness and limitations of the model Vs projections. For example, the extent to which El Niño serves as a useful analogue for the mechanisms behind the projected shear changes should be further examined: although the sign of the relationship in Figure 2 is the same as during El Niño, the structure of the Vs changes differs from that associated with El Niño. It is also important to keep in mind that the Pacific Walker circulation can exhibit energetic variability – even on decadal timescales – independently of external forcing [e.g., Vecchi et al., 2006], and that Atlantic shear is influenced by a variety of factors besides the Pacific Walker circulation. For example, both the meridional temperature gradient in the tropical Atlantic [e.g., Zhang and Delworth, 2006] and the extent of the Atlantic Warm Pool [e.g., Wang et al., 2006] have been connected to changes in Vs. A full understanding of the projected and historical patterns of tropical Atlantic shears must take into consideration the full set of factors that influence shear, including those resulting from internal climate variability as well as forced climate change.
 We acknowledge the various modeling groups for providing their data, and PCMDI and the IPCC Data Archive at Lawrence Livermore National Laboratory for collecting, archiving and making the data readily available. We thank the two anonymous reviewers, T. Delworth, K. Dixon, S. Garner, D.E. Harrison, I. Held, A. Johansson, T. Knutson, J. Lu, T. Marchok, J. Sirutis, R. Stouffer, A. Wittenberg, Sebastian Ilcane, and Autumn Laperra for helpful discussion. This work partially supported by NASA and NOAA-OGP. Code to compute the MPI is available at http://wind.mit.edu/∼emanuel/home.html.