The effects of dry air intrusion on landfalling hurricanes are investigated using eight numerical simulations. The simulations differ in the initial amount of moisture in the storm core and its horizontal extent from the storm center. The storms evolve very differently during the 36-h simulation. Storms with a small radial extent of moisture develop minimal rainbands, intensify rapidly in the first 3 h, and weaken as dry air from the 800–850-hPa layer wraps cyclonically and inward around the storm core. As the air approaches the core, it sinks (possibly by eyewall downdrafts or as a result of evaporative cooling), reaches the storm’s inflow layer, and entrains into the eyewall updrafts. Storms with large radial extent of moisture develop into larger storms with large rainbands, having smaller intensification rates initially, but continue to intensify for a longer period of time. Rainband downdrafts release low equivalent potential temperature air into the moat region. Low-level convergence into the rainbands reduces the magnitude of eyewall inflow. Both factors reduce storm intensification initially. Simultaneously, the rainbands act as a barrier between the moist core and the dry environment, preventing dry air from penetrating the storm core. As land is approached, inflowing air is no longer replenished with heat and moisture. Eventually, rainband convection erodes and dry air approaches the storm core from the landward side causing the storms to weaken. Without the presence of land, a hurricane can sustain itself in a dry environment, provided its moist envelope is large enough.
Several recent hurricanes (Opal, 1995; Georges, 1998; Lili, 2002; Ivan, 2004) weakened just before making landfall on the U.S. Gulf of Mexico coast. This weakening spared Gulf Coast residents from even more extensive damage than was done. In each of these cases, dry air was present in the vicinity of the storms and may have contributed to their weakening. Opal has been studied extensively, and its weakening phase just before landfall occurred in high vertical wind shear and lowered sea surface temperatures (SSTs) as the system moved away from an eddy of warm water (Rodgers et al. 1998). However, an area of dry air existed west of the axis of a midlatitude trough that was also responsible for the high shear values. The dry air intruded cyclonically around the western and southern regions of Opal and came within 222 km of Opal’s center in its southwestern quadrant at the same time the storm began to weaken. Dry air was observed to the east and northwest of Hurricane Georges as it made landfall on the Mississippi coast (Curtis 2004). Hurricane Lili (2002) made landfall on the Louisiana coast as a category 1 hurricane after a period of rapid decay during which its eyewall collapsed (Pasch et al. 2004). Satellite images of Lili before landfall indicate a region of very dry air to the north of the storm. When Ivan approached the Alabama coast on 15 September 2004, dry air existed in its western half and an erosion of the southwestern eyewall was observed in the Mobile, Alabama, Regional Airport (KMOB) Weather Surveillance Radar-1988 Doppler radar imagery and reported by a National Oceanic and Atmospheric Administration research aircraft.
Hurricane Alicia (1983), on the other hand, continued to intensify during landfall in spite of the presence of dry air. Dry air was present beyond about 350 km from the storm center, but there was no evidence of dry air intrusion according to Curtis (2004). On the other hand, Powell (1987) observed that Alicia’s eyewall was open to the south before and during landfall; however, Alicia continued to intensify in spite of half of the storm’s circulation being located over land. Powell (1987) noted that the storm’s rainbands may have served as a boundary between the moist core and the dry environment.
Previous modeling studies of hurricane landfall have focused on the sensitivity of the hurricane to surface boundary conditions and not dry air intrusion (e.g., Tuleya and Kurihara 1978; Tuleya et al. 1984; Tuleya 1994). Bender et al. (1987) modeled the effects of island terrain on tropical cyclones and observed how dry air intrusion from above mountaintops caused the model storms to weaken. Chan and Liang (2003) noted that reduction of the surface moisture flux over land in their idealized numerical study would lead to advection of relatively dry air into the storm center as it approached land. They found that such an intrusion during landfall did not always lead to reduced convective activity. The drier air was advected cyclonically and upward from the landward side, decreasing column static instability on the offshore flow side just off the coast, but increasing it on the onshore flow side over land as the dry air moved over low-level moist air.
When a hurricane approaches a dry environment, forecasters are faced with a difficult decision as to the effect of the dry air on the storm’s intensity. Furthermore, moisture is not well assimilated into hurricane forecast models because of a lack of observations over the tropical oceans, especially within the dangerous environment of a hurricane core. An additional concern about dry air is that its presence and subsequent intrusion at midlevels has been linked to tornado outbreaks during hurricane landfall (e.g., Curtis 2004).
This study will use a numerical modeling approach to explore dry air intrusion in landfalling hurricanes. Specific questions to be addressed are 1) how and where does dry air enter the storm, 2) can a larger moist envelope protect a storm from dry air intrusion, and 3) to what extent does the proximity to land enhance the adverse effects of dry air intrusion on hurricane intensity? The design of the numerical experiments and model configuration are described in the next section. In section 3, the results are presented; followed by a discussion and a list of conclusions in section 4.
2. Method and numerical model configuration
a. Design of the numerical experiments
Eight experiments are compared, each with the same initial hurricane vortex and large-scale analysis (see section 2b), but with a different moist envelope surrounding the storm. The moist envelope is constructed by adding a Gaussian mixing ratio perturbation to the center of the vortex:
where qb(z) is a dry environmental sounding, A(z) is the amplitude of the perturbation (which decreases with increasing height), B is the e-folding radius, and r is the radial distance from the center of the vortex. The dry environment is loosely based on the shape of a sounding from the Geophysical Fluid Dynamics Laboratory (GFDL) analysis at 1200 UTC 18 July 1997 at 36°N and 83°W. Roughly a third of the values observed in that location are taken as qb(z). The moist envelope is defined as the area where q exceeds qb. The magnitudes of A and B are varied to obtain different initial moist envelopes. Table 1 lists the eight different combinations of values for A and B (at the lowest model level) along with the case name of each experiment. The plain solid line in Fig. 1 represents the dry environmental sounding; the other four soundings are taken at 0, 225, 450, and 675 km away from the center of the vortex. In the driest case (A15B250), the sounding at r = 450 km coincides with that of the dry, unperturbed environment, indicating the edge of the moist envelope. In the other extreme (A19B600), the sounding at r = 675 km is still more moist than the original unperturbed environment, indicating a moist envelope with a radial extent larger than 675 km. Figure 2 is a water vapor satellite image of Hurricane Ivan at 1915 UTC on 15 September 2004. Very dry air can be seen around 500 km from the center of the vortex to the southwest and around 1000 km to the northeast of the storm center. Therefore, the values for B are within a realistic range.
In Fig. 3, mixing ratio soundings for each experiment at r = 100 km from the vortex center are compared with observations from Slidell, Louisiana, at 1200 UTC on 18 July 1997, which was, at that time, located 100 km from the center of Hurricane Danny (1997). Danny was a category 1 storm at the time, as are the initial model vortices in this study. A comparison between the observed and model soundings reveals that below about 800 hPa the observed sounding lies between those of the cases with B ≥ 400 km and the smaller B cases. The observed 800–600-hPa layer is about as moist as that of the B ≥ 400 km cases. Above 600 hPa, the observations fluctuate between being about 1 g kg−1 more moist than the wettest experiment and 1.5 g kg−1 drier than the driest experiment. Overall, the observations agree reasonably well with the experimental soundings, providing confidence that the Gaussian perturbations are realistic.
The model vortices are embedded in the temperature and wind fields that surround Hurricane Danny (1997) at 1200 UTC 18 July 1997, which is when the simulation begins. The storm is located over water to the south of the coast of Mississippi at that time and is steered slowly northeastward toward Mobile Bay. Danny enters Mobile Bay around 0900 UTC 19 July. Steering currents weaken further and the storm remains in the bay until it makes landfall near Mullet Point, Alabama, around midday local time on 19 July (Rappaport 1999). The simulation ends at 0000 UTC on 20 July. The Danny environment is chosen because of the absence of strong steering currents and vertical wind shear. In this manner it becomes possible to investigate the effects of just dry air intrusion upon the simulated storms as they approach land.
b. The numerical model
The fifth-generation Pennsylvania State University–National Center for Atmospheric Research (PSU–NCAR) Mesoscale Model (MM5) is a nonhydrostatic, primitive equation model for a fully compressible atmosphere (Grell et al. 1994). In this study, the MM5 is initialized using GFDL atmospheric analysis fields and U.S. Navy Fleet Numerical Meteorology and Oceanography Center SST fields from 1200 UTC 18 July to 0000 UTC 20 July 1997. This time period captures Hurricane Danny (1997) moving from a location southeast of New Orleans to Mobile Bay and eventually making landfall in Alabama. The GFDL analysis includes a bogus vortex to represent Hurricane Danny, which is located about 10 km to the north of the observed center of Danny at 1200 UTC on 18 July 1997. This misplacement is most likely a result of interpolating the coarser-resolution GFDL fields (approximately 18 km) to the finer-resolution MM5 model fields used here. The GFDL vortex is also much larger than the observed Danny because the coarse resolution of the operational GFDL model at that time could not resolve an initial vortex with a small (45 km) radius of maximum winds (RMW). For these reasons, the GFDL vortex is removed and replaced by a new artificial vortex spun up by MM5, following the method described by Kimball and Evans (2002). A vortex constructed in the above manner has a consistent internal structure (moisture, wind, temperature fields, etc. will be in balance with one another), it will resemble the corresponding real storm, and it will be compatible with the numerical model physics, computational schemes, and grid resolution (Kurihara et al. 1993).
The MM5 simulations make use of two two-way nested domains with horizontal resolutions of 9 and 3 km, respectively, and 24 vertical levels, 7 of which are located in the first 1.5 km above the surface of the model. Convection is modeled explicitly on both meshes. Microphysics is modeled using the Reisner graupel scheme (Reisner et al. 1993, 1998) and includes snow, supercooled water, graupel, and ice number prediction equations. Time inflow–outflow-dependent boundaries are used (Grell et al. 1994). The high-resolution Blackadar planetary boundary layer parameterization scheme (Blackadar 1979; Zhang and Anthes 1982) is used on both grids. The surface flux parameterization makes use of a form of the bulk aerodynamic formulations. The modest intensity of the modeled hurricanes in this study makes it unlikely that the fluxes are overestimated because of extrapolation of the surface exchange coefficients into high-wind regimes. The magnitude of the wind speed above the boundary layer beyond which this becomes a problem is 40–50 m s−1 (Franklin et al. 2003); the hurricanes in this study only cross that threshold by a small amount and for a short period of time.
As low-level winds over the warm tropical ocean converge toward the hurricane center, surface sensible and latent heat fluxes supply the low-level inflow with moist entropy. This moist entropy is often measured in terms of equivalent potential temperature θe because of its conservational properties and because the mean θe of the eyewall column has been related to tropical cyclone (TC) intensity (Malkus and Riehl 1960; Betts and Simpson 1987). Furthermore, θe combines temperature and moisture content in one variable. To make an impact on hurricane intensity, a rising parcel needs both a high temperature and high moisture content. The former will allow a parcel to rise in the first place, the latter will allow it to continue to rise and will supply fuel to the storm in the form of latent heat release. For these reasons, dry air intrusion will be mostly discussed in terms of θe.
a. Storm evolution
The evolution of the storms’ intensity and wind radii are presented in Figs. 4 –6. An initial adjustment period is seen in the first 3–6 h of the simulation. During this time, the larger-scale environment of the model adjusts to the inserted bogus vortex (e.g., Liu et al. 1997). The use of nested domains for prediction and a single domain for initialization may contribute to the adjustment period (Xiao et al. 2000). Storm intensity is measured in terms of minimum sea level pressure (PSMIN; Fig. 4) and it is readily seen that hurricanes with different initial moisture contents evolve very differently. Increasing the low-level initial moisture content in the core of a TC, as well as its radial extent, leads to an increased low-level θe under the eyewall. At 300 hPa, the initial moisture content of all cases is almost equal in value (Fig. 1); hence, higher boundary layer θe implies higher convective available potential energy (CAPE). As high-θe air ascends in the eyewall, latent heat release occurs, enhancing parcel buoyancy. The strong ascent in the eyewall is compensated for by low-level radial inflow toward the eyewall and subsidence in the eye (Willoughby 1998). This subsidence warming leads to a hydrostatic surface pressure drop in the eye, a subsequent increase in the low-level radial pressure gradient, and an increase in low-level winds via gradient wind balance. Hence, it would seem reasonable to expect an initially wetter (in terms of larger A and/or B) storm to become more intense. If B is held constant and the initial mixing ratio amplitude A is increased, this expectation is indeed realized. For the four storms with B = 250 km (dotted lines in Fig. 4), the storms become more intense as A increases. The same holds true for the two cases with B = 400 km (dashed lines in Fig. 4). However, a larger horizontal extent (B) of initial moisture, while A is kept constant, does not necessarily lead to a more intense storm. For example, A19B400 is a more intense storm than the initially wetter A19B500 throughout most of the 36 h of the simulation. Intensification rates (Fig. 4) differ throughout the simulation and are not necessarily greater for wetter storms. In fact, the B = 250 km cases (dotted lines) show a large intensification rate during the first 3 h of the simulation, followed by mostly weakening. Larger B cases intensify more slowly initially but continue to intensify for a longer period of time.
The RMW and 17 m s−1 winds (or size) are calculated at the lowest model level (40 m from the surface) and over water only, because increased friction over land causes an abrupt change in wind speed and smaller radii over land. This eliminates changes in wind radii as a result of different landfall times, but also causes the RMW of some storms (Fig. 5) to increase as the storms track inland. As the storm center moves away from the coastline, the distance between storm center and maximum winds (i.e., the RMW) over water increases. After around t = 24 h, the RMWs of the two largest B cases (A19B500 and A19B600) suddenly increase to very large sizes (close to 45 km at t = 36 h). Both these storms track farther from the coastline after landfall than the other cases (see track discussion below) and, hence, their centers move farther away from their maximum winds over water. Before moving inland, the RMW of each case contracts as the simulation progresses, even if the storm is weakening. Usually, contracting RMWs are associated with intensifying tropical cyclones (e.g., Willoughby 1998), but cases where the opposite occurs have been observed in real hurricanes (e.g., Kimball and Mulekar 2004). The two cases with the largest e-folding radii (A19B500 and A19B600) have the largest RMW for most of the simulation time and are significantly larger in size (Fig. 6). The B = 250 km cases have smaller RMWs during most of the simulation time and are smaller in size.
Figure 7 shows that the tracks, landfall locations, and landfall times of the storms differ. This is to be expected given the differences in size and intensity evolution. Past research (e.g., Elsberry 1995) has shown that storms of different sizes and intensities are steered by different levels and shallower or deeper layers of atmospheric steering flows. The impact of the β effect also depends on storm structure. Landfall times (i.e., the time the storm’s center crosses the coastline) and the time of the onset of storm weakening are listed in Table 2. All of the storms begin to weaken before they make landfall; column 3 in Table 2 indicates how many hours before landfall this occurs. This number ranges from 9 to 21 h. The smallest A cases (A = 15 and 16 g kg−1) begin to weaken early (t = 3 h) but also make an early landfall because of their faster motion during the first 9 h. This explains why their onset of weakening precedes their moment of landfall by a relatively small amount. Considering just the large A cases (A = 18 and 19 g kg−1) suggests that the smaller the value of B, the larger the amount of time by which the onset of weakening precedes landfall. This may indicate that dry air intrusion plays a stronger role in the small B cases than in storms with an initially larger moist envelope. The correlation is not statistically significant, however, because of the small sample size.
All storms begin to weaken before their center crosses the coastline (Fig. 4, Table 2); therefore, it is possible that land effects begin to play a role before landfall, in addition to or instead of, dry air intrusion. However, this moment is difficult to measure. When hurricanes encounter land, friction increases and surface fluxes reduce because of the reduced heat and moisture content of the land surface compared with the sea surface. Figure 8 shows the low-level wind speed, surface fluxes, and surface temperature for A19B600 at t = 18 h when part of the storm’s eyewall has crossed the coastline. The white dashed line marks the 33 m s−1 wind contour at 40 m from the surface and roughly marks the storm’s eyewall. Temperatures over land (Fig. 8d) are significantly lower than water temperatures (because of overcast conditions and the presence of vegetation). This leads to significantly lower sensible heat fluxes over land than over water for the same low-level wind speed (Figs. 8b and 8d). The same applies to the surface moisture content (not shown) and latent heat fluxes (Fig. 8c). Over water the surface fluxes fall off as the low-level wind decreases; the largest fluxes occur near the storm’s RMW. However, the model latent and sensible heat fluxes depend on low-level wind speed, as well as the moisture q and temperature T difference, respectively, between the surface and lowest model level (Blackadar 1979; Zhang and Anthes 1982). Therefore, the surface fluxes change in response to low-level T and q changes that may occur in dry air intrusion events. This can be seen in Fig. 8, where the surface fluxes are larger under the western eyewall than the eastern eyewall in spite of equal values in wind speed (Fig. 8a), surface temperature (Fig. 8d), and surface moisture content (both are over water). Furthermore, if a storm weakens, its low-level winds and, hence, the surface fluxes decline. This makes it difficult to decide whether reduced fluxes cause storm weakening or vice versa. Therefore, a change in surface flux magnitude is not a good indicator of when land effects begins to play a role.
Figure 8 illustrates that surface friction immediately reduces a storm’s strongest low-level winds and that the fluxes directly under the eyewall decrease by a large amount when the RMW crosses the coastline. Table 3 lists the time the RMW crosses the coastline for each case. Also shown is the amount of time by which storm weakening (Table 2) precedes this moment. Again disregarding the two smallest A cases, it is clear that the B = 250 km cases begin to weaken a significantly longer time before RMW landfall than do the B ≥ 400 km cases. This would indicate that these cases probably begin to weaken as a result of dry air intrusion alone; that is, well before land begins to play a role. However, all cases begin to weaken before the RMW crosses the coastline. Therefore, dry air intrusion probably plays a significant role in all of these simulated hurricanes. The low-level energy supplied to the storms seems insufficient to counteract the negative effects from dry air entrainment on storm intensification, even when storms have an initially large moist envelope. Some real hurricanes have been observed to continue intensifying (e.g., Alicia, 1983) or to remain in steady state (e.g., Frederic, 1979) until the time when their center crossed the coast in spite of reduced access to surface energy and an increase in land-induced friction beneath portions of their eyewalls. In these cases, the reduced supply of energy from the surface was apparently still sufficient to sustain the storm.
b. Dry air intrusion
1) A19B250 and A19B600
These two cases have the same initial moisture amplitude (A) and make landfall at the same time, but their initial moist envelopes (B) differ substantially in radial extent. Their evolution in terms of intensity and structure also differs dramatically. Both storms approach the coastline at roughly the same pace and their centers reach the coastline at t = 21 h, within 10 km of one another (Fig. 7, Table 2). Hence, different land effects (reduced surface moisture and heat fluxes and enhanced friction) do not explain the differences in storm evolution.
Case A19B250 has an initially small moist envelope and remains a small storm (Fig. 6). The storm intensifies at a greater rate than any other in the first 3 h (Fig. 4), but this is followed by steady weakening before rapid weakening occurs after landfall at t = 21 h. In contrast, A19B600, with a large moist envelope, remains a large storm throughout the simulation. This case intensifies slowly in the first 3 h followed by continued intensification until t = 12 h. Between t = 6 and 12 h, this storm intensifies more rapidly than any other case.
In the remainder of this section, the above two cases will be closely compared with the aid of Figs. 9 –14. Figure 9 compares the θe at 950 hPa of A19B250 and A19B600 at t = 6 h. Figure 10 shows a vertical cross section of θe at t = 6 h for case A19B250 at the 90° azimuth. Figure 11 compares the reflectivity of the two cases at t = 6 h, while Fig. 12 compares the radial component of the wind at that same time. Figures 13 and 14 display the vertical wind speed at t = 6 h and the θe at t = 12 h, respectively, for case A19B600.
In the small radius case (Fig. 9a), a central moist region of θe ≥ 330 K exists, beyond which θe drops off rapidly to less than 316 K. About 30 km to the north of the storm an arc of very low θe can be seen embedded in warmer and moister air. The arc is isolated from the dry environment surrounding the storm at this level, but between 900 and 850 hPa it is connected to the dry environment. In the southeastern half of the storm another dry arc is seen, this one connected to the dry environment. At higher levels, between 900 and 800 hPa, this arc is still present and wraps farther around the storm center to reach the western side. Both arcs spiral cyclonically and inward toward the storm center. The northern end of the southeastern arc wraps inward and approaches the storm center very closely. Radial–height cross sections of θe at 15° azimuth intervals through this arc (i.e., between the 135° and 75° azimuths, using meteorological coordinates) reveal that low-θe air sinks and moves cyclonically toward the eyewall of the storm. At the 135° azimuth, the dry air is located 65 km away from the storm center at ∼800 hPa, whereas by 75° it has approached the surface and the storm center within a distance of 35 km, which coincides with the outer periphery of the storm eyewall. Figure 10 shows a vertical cross section of θe at t = 6 h at the 90° azimuth. The eyewall is marked by strong updrafts between r = 15 and 30 km, while weak subsidence occurs in the eye. Near the surface, high values of θe are seen in the eye because of a combination of low pressure and large moisture content that typically exists at low levels in hurricane eyes (Willoughby 1998). The intruding dry air can be seen at r = 40 km accompanied by sinking and radially inward flow. As the dry air enters the eyewall, it lowers eyewall CAPE and spirals cyclonically and upward within eyewall updrafts, possibly explaining the eroded western eyewall seen in Fig. 11a.
Although the dry air surrounding A19B250 is seen to extend upward to almost 600 hPa (Fig. 10), it seems to have little impact on the storm above about 800 hPa because of the lack of inward motion at those levels. The dry air circles around the storm center until it comes into contact with a pocket of sinking motion, which brings it down to the inflow layer of the storm. This sinking motion may be an eyewall convective downdraft. Alternatively, the dry air may have been cooled by evaporating hydrometeors causing an increase in density, allowing it to sink spontaneously. Eyewall entrainment of dry air like that described above continues throughout the simulation of A18B250.
The dry arc to the north of the storm center does not manage to approach the storm eyewall by t = 6 h. Figure 12a shows the radial component of the wind at t = 6 h and 40 m above the surface. Inflow toward the storm center is substantially weaker in the northwestern half of the storm than the southeastern half. This is most likely caused by a combination of 1) outflowing air from strong convection in the northern eyewall and 2) air flowing toward outer convective cells to the north and west of the storm core. This can be seen by comparing Figs. 11a and 12a, which show a good correlation between converging air and convective activity. To the southeast of the storm, a pattern of weaker and stronger inflowing air (i.e., convergence and divergence) coincides with the small convective bands in that quadrant. In spite of the reduced inflow to the north, strong tangential winds allow the moisture content of the dry, subsiding air in the northern arc to be replenished via latent and sensible heat fluxes over the warm sea waters to the west of the storm, before it enters the storm eyewall.
In spite of being initialized with more moisture in terms of e-folding radius, A19B600 intensifies significantly more slowly during the first 3 h of the simulation than A19B250. The most striking difference between the two cases is the larger size of A19B600 (Fig. 6) and the presence of longer and broader rainbands (Fig. 11). Apparently the presence of a larger initial moist envelope and, hence, a larger area of unstable air surrounding the storm center, allows the formation of large rainbands. As a result, the low-level θe field of this case takes on an entirely different picture (Fig. 9). Collocated with the convective cells in the rainband are pockets of low value θe air embedded in a much larger region of θe > 330 K than A19B250. To the southeast, very low θe air attempts to penetrate this large moist envelope.
A map of the low-level divergence (not shown) reveals a band of strong convergence coincident with the rainband and eyewall. Low-level convergence into the convective cells of the rainband causes air beyond the band to flow toward the band, in a direction toward the storm center, or negative radial wind ur < 0 in Fig. 12b. This is especially prominent to the southwest of the storm over water. At the same time, air in the moat region between the eyewall and rainband also flows toward the band, but away from the storm center, or ur > 0 in Fig. 12b. Powell (1987) observed a similar pattern of radial flow between rainband and eyewall as Hurricane Alicia (1983) was making landfall. As a result, the overall inflow into the storm eyewall is reduced compared with a case without rainbands (Fig. 12a), which may explain the slow intensification rate of A19B600 during the first 6 h of the simulation. The vertical motion field (Fig. 13) reveals that the rainbands consist of strong up- and downdrafts. High-θe (or unstable) air ascends in the updrafts, while low-θe air descends in convective downdrafts (Powell 1990). This explains the pockets of low-θe air seen in Fig. 9b. Over land, low-θe air in the moat region is even more prominent because of the reduced surface fluxes over land. A pool of lower-θe air over land, between the core and eyewall, was also observed in Hurricane Alicia (Powell 1987). The inflow of lower-θe air from the moat further reduces the strength of the eyewall convection and may, therefore, contribute to the slow intensification rate of A19B600 in the first 6 h. Besides low-θe air, rainband downdrafts may also bring down stronger winds from aloft (Powell 1982), contributing to the storm’s larger size (Fig. 6). In addition to vertical advection of air with higher tangential momentum, such air may also be radially advected from the core of the storm toward the rainbands by the outward-flowing air (ur > 0) that was observed in parts of the moat region (Fig. 12b), further explaining the larger size of the storm.
From the above, it appears that the rainbands may initially act to reduce eyewall convection, resulting in low initial intensification rates of A19B600. At the same time, dry air intrusions like those observed in A19B250 were not observed in the large B case (Fig. 9) during the initial 12 h. In other words, the rainband seems to act as a thermodynamic boundary between the dry environmental air at midlevels and the storm core. The rainbands functioning as a boundary between the hurricane core and its environment was also suggested by Willoughby et al. (1984) and observed in Hurricane Alicia (1983) by Powell (1987). This blocking role of the rainbands is what may allow A19B600 to intensify for a longer period of time than A19B250. Between t = 6 and 12 h, rainband convection gradually reduces, especially over land (not shown). However, low-level winds at large radii remain stronger than in A19B250, maintaining strong surface fluxes at those radii. Both the demise of the rainbands and the persistence of strong surface fluxes at large radii allow more energy to be fed into the storm’s eyewall, increasing the convective activity and allowing rapid intensification of A19B600 between t = 6 and 12 h.
By t = 12 h, the low-θe air seen to the southeast of the storm at t = 6 h has rotated cyclonically and inward around the TC center to reach to the northeastern side of the TC (Fig. 14). Three hours later (not shown), dry air is observed to entrain into the eyewall at low levels (900 hPa and below) on the landward side of the storm where protective rainbands have been eroded away by the impinging dry air and reduced surface fluxes over land.
Limited rainband activity in A19B250 allows early and continuous penetration of dry air into the storm eyewall. The onset of A19B250’s weakening occurs while surface fluxes remain high (not shown) since the RMW and strongest surface fluxes do not reach the coastline until 12 h later. Vertical shear over the system is below the threshold value of 8.5 m s−1 (Fitzpatrick 1996). All three points suggest that dry air intrusion was the main contributor to the onset of weakening in the B = 250 km case.
In both cases, dry air intrusion into the eyewall is seen at low levels (mostly below 900 hPa) because inflow at higher levels is weak. This lowers CAPE and reduces both the latent heat release in the eyewall and compensating subsidence in the eye. Both contribute to a rise in central surface pressure and storm weakening. In observed storms, dry air intrusion was seen at 500 and 700 hPa (e.g., Curtis 2004) and Special Sensor Microwave Imager images of total precipitable water around Hurricane Opal (1995) indicate a dry air intrusion to within 222 km of the core (Rodgers et al. 1998). These observations, however, were at too coarse a resolution to precisely observe at which levels and radii the dry air intruded. Chan and Liang (2003) observed a cyclonic and upward rotation of dry air from the landward side of an idealized landfalling TC. Their land surface was flat with a straight coastline and the model storm approached land at a perpendicular angle. In this study there is no evidence of upward advection of dry air, possibly because of the more complex configuration of the experiments and the fact that once dry air entrains into the eyewall it is modified by microphysical processes.
This case behaves similarly to A19B600 discussed above. The storm is slightly smaller (Fig. 6) than its B = 600 km counterpart and follows a similar track for most of the simulation (Fig. 7), but makes landfall about 20 km farther east. Deepening during the first 9 h is slightly larger, but filling commences 3 h earlier. The area of θe > 332 K is slightly smaller as is to be expected given the smaller B. Rainbands form, but convection is somewhat weaker and dissipates earlier than in A19B600. Hence, the rainband’s initial dampening effect on eyewall convection is reduced, but so is its protective role against environmental dry air intrusion as compared with the larger B case. This likely explains the larger initial intensification rate and the earlier start of the filling process of A19B500. Similarly to A19B600, dry air wraps cyclonically around the storm from the southeastern quadrant. After the rainbands over land have eroded, dry air is observed to intrude into the eyewall on the landward side of the storm at t = 12 h, 3 h after the onset of weakening. However, because of the 3-hourly model output times, the exact onset of dry air intrusion and weakening cannot be determined and may have occurred anywhere between t = 9 and 12 h.
3) A18B400 and A18B250
Like small-radius case A19B250, A18B250 displays minimal rainband development and dry air intrusion begins at t = 6 h. In a similar manner to A18B250, dry air from the environment spirals cyclonically and inward toward the TC center where it eventually entrains into eyewall updrafts. The storm begins to weaken at t = 6 h or 13 h before the RMW reaches the coastline. This suggests weakening is caused by dry air intrusion and not by the effects from land. The storm’s large B counterpart, A18B400, has an initially smaller deepening rate but the deepening lasts longer: until t = 12 h or 3 h before the RMW crosses the coastline. A18B400 is a larger storm with more rainband activity than A18B250. A large area of θe > 330 K exists with pockets of lower values of θe in the moat region, especially over land. Weaker inflow into the eyewall occurs to the northwest of the storm center and inflow into the rainband occurs on both sides of the band. This points to a similar dual role of the rainband to that discussed in the previous sections. Initially the band reduces storm intensification by 1) reducing the magnitude and 2) lowering the energy level of the eyewall inflow. Simultaneously, the band forms a barrier against the dry environmental air allowing the storm to continue intensifying for a longer period of time. As the storm approaches land, dry environmental air wraps around the storm from the southeast to reach the landward side of the storm. Over land, reduced surface fluxes cannot replenish the air with heat and moisture and the dry air eventually penetrates the storm eyewall from the landward side.
4) A19B400 and NO-LAND
To investigate the role of land in the weakening process of the B > 250 km cases, a simulation identical to A19B400 but without land (NO-LAND) is performed. If land is not present, the storm may be able to continue to protect itself again dry air intrusion. The minimum central pressure and size evolution of NO-LAND is compared with that of A19B400 in Fig. 15. There is a small difference (<2 hPa) during the initial 12 h of the simulation during which period both storms intensify. At t = 12 h the PSMIN differences become larger as A19B400 begins to weaken while NO-LAND continues to intensify. At t = 24 h, NO-LAND appears to reach a steady state with an intensity of about 974 hPa. For the first 27 h, the two storms remain similar in size (Fig. 15b). At t = 27 h, A19B400 suddenly drops in size as land approaches (Fig. 7). Reflectivity images reveal that over water both storms possess rainbands of similar horizontal extent and intensity. After t = 12 h, rainband convection ceases over land in A19B400, but the storm remains large in size. Since size is calculated over water only, this is in agreement with the observed rainband activity.
Figure 16 compares the 950-hPa θe field at t = 12 h of both cases and shows a tongue of dry air to the southeast of both storms. However, the presence of rainbands and their convective activity and accompanying high surface winds creates a zone of high-θe air around the storm center that prevents this dry air from reaching the core. In A19B400 this zone is considerably less moist over land because of the reduced surface fluxes. As a result, drier air is observed to spiral cyclonically and inward from the landward side of the TC to its southwestern side. This continues through the remainder of the simulation and the drier land air eventually connects with the extremely dry environmental air to the southeast. In NO-LAND, low-θe pockets exist in the moat area as a result of rainband outflow, but the moat air continues to be replenished by strong surface fluxes over water. The tongue of dry air to the southeast surges back and forth and manages to reach the northern side of the TC by t = 21 h; however, by t = 33 h it has receded back to the southeast corner as it cannot overcome strong surface fluxes.
The azimuthally averaged sum of the latent and sensible heat fluxes are displayed in Figs. 17a and 17b as a function of radius and time. Also shown are the lowest-level wind speeds for both cases (Figs. 17c and 17d). As was discussed in section 3a, the surface fluxes are strongest in the eyewall region where the strongest winds occur. Initially, the surface fluxes are stronger in this area in the NO-LAND case consistent with the stronger low-level wind speeds of that case, but between t = 11 and 18 h, the reverse is true despite weaker wind speeds in A19B400. In both cases, the eyewall is located over water during this time, making the surface values of T and q identical in both cases. Therefore, the larger surface fluxes must be caused by a larger vertical gradient in T and q between the surface and the lowest model level (Blackadar 1979; Zhang and Anthes 1982) in A19B400. This means that T and q at the lowest model level are smaller in A19B400, which is most likely caused by the low-θe air that was seen rotating around the storm core from the landward side (Fig. 16a). By t = 15 h, the region of maximum surface fluxes has crossed land (black line in Fig. 17a) and the fluxes reduce while the storm begins to weaken. Surface fluxes at radii beyond the eyewall are consistently larger in the NO-LAND case compared with A19B400. This is because of the presence of land to the north of the storm center in the latter case.
4. Discussion and conclusions
It seems that a storm with an initially large moist envelope can survive in a dry environment without significant wind shear. Such a storm develops rainbands whose convection brings strong wind speeds to low levels at large radii and mixes high-θe air from the surface into the low-level hurricane inflow layer. This prevents the entrainment of dry environmental air into the eyewall. When land is present, reduced surface fluxes break up the protective barrier on the landward side of the storm. Dry air reaches the storm core before the storm center makes landfall but for a while surface fluxes in the core continue to replenish the air with energy before it ascends in the eyewall. As the storm approaches land, however, the core surface fluxes decrease and at some point lose their battle against increasing dry air intrusion.
Real Gulf Coast landfalling hurricanes have been observed to start weakening before their centers crossed the coastline, for example, Opal (1995) and Ivan (2004). In both cases the convection was eroded away in part of the eyewall as a result of dry air intrusion. Both these storms, however, also interacted with a midlatitude trough as they were approaching the coast, the same feature that introduced the dry air. Trough interaction can be unfavorable to hurricane intensification because of large vertical wind shear, but also favorable (e.g., Molinari and Vollaro 1989; Sadler 1976). The results from this study indicate that the already complex problem and difficult predictability of hurricane–trough interaction may be further complicated by the presence of dry air.
On examining the intensity and structure evolution of eight model storms, initialized with different moisture profiles, several important conclusions regarding the role of environmental dry air can be drawn:
Storms with initially small moist envelopes cannot protect themselves against the intrusion of dry air from the environment and weaken well before increased friction and reduced surface fluxes from land begin to play a role.
The scenario for dry air intrusion is as follows. Dry environmental air above approximately 850 hPa rotates cyclonically and inward around the storm center. When it encounters an eyewall downdraft or increases its density because of evaporational cooling, the dry air sinks to lower levels, reaches strong low-level inflow, and is entrained into eyewall updrafts. This reduces eyewall CAPE and, hence, latent heat release in the eyewall as well as compensating subsidence in the eye. Both effects cause weakening of the warm core and a rise in central surface pressure.
Over water, full access to surface energy may allow a hurricane to fend off negative effects from dry air intrusion. This is likely a function of the size of the initial moist envelope, which should be further investigated in future work.
When the proximity to land causes a reduction in surface fluxes and an increase in surface friction, the surface energy supply to the hurricane may be insufficient to counteract the negative effects from enhanced dry air entrainment. The outcome will depend on the size of the initial moist envelop surrounding the storm and the degree of dryness of the environment.
Different initial moisture distributions impact TC intensity, structure, and motion. This is because steering depends on TC intensity and structure. Hence, TC landfall location and damage potential are sensitive to initial moisture content. Landfall locations differ by at most a 20-km distance (Fig. 7). While this may seem small, this lies within the range of the values of the RMWs at landfall (Fig. 5). Therefore, this could mean the difference between hurricane force winds or barely tropical storm force winds at a given location. Storm sizes at landfall range from about 90 to 145 km (Fig. 6); this means that the area of tropical storm force or higher winds ranges from 25 447 km2 to 66 052 km2. This is a substantial increase in the surface area receiving damaging winds.
Moisture is not well measured operationally and, therefore, not directly assimilated in operational hurricane models, yet this study shows that TC models are highly sensitive to the amount and distribution of initial moisture. If moisture is not captured correctly in the analyses or shifted to different scales because of model resolution, the track, intensity, and structure of the TC varies substantially. To obtain more accurate TC model predictions, it is crucial that accurate initial moisture measurements are incorporated both in the core and around the TC. This is supported by a study done by Kamineni et al. (2003) who assimilated Lidar Atmospheric Sensing Experiment moisture profile data into the forecasts of three hurricanes. In each case, a significant improvement in track and intensity forecasts was obtained.
More moisture in the core of the storm, with everything else kept equal, leads to a more intense storm, as expected. More low-level moisture in the core increases CAPE and convective activity in the eyewall. This leads to stronger eye subsidence, a warmer core, and hence a lower PSMIN.
Increasing the radial extent of the initial moisture content enhances the formation of rainbands. Rainband activity causes wind speeds to increase at larger radii and, hence, causes the storm to be larger in size. This, in addition to larger areal coverage of heavy rainfall, makes the storm potentially more destructive.
Rainbands can slow down the intensification of a TC because low-θe air from rainband downdrafts becomes trapped between the eyewall and the rainband. This lower energy air is then entrained into the eyewall. This effect is enhanced when land is present between the eyewall and rainband, because lower surface fluxes over land prevent the air from being replenished with high values of θe.
Rainbands can also prolong the intensification of TCs, by acting as a thermodynamic boundary between the high-θe core and dry environmental air.
While the above rainband conclusions seem to contradict, the extent to which either plays a role depends on the structure of the individual storm as well as its surrounding environment. The protective role of the rainbands will likely dominate in a storm in a dry environment; whereas in moist surroundings, rainbands may prevent storms from reaching their maximum potential intensity. In the latter case, strong rainbands feeding off the moist environment simultaneously discharge dry, cold downdraft air toward the storm core.
A fine balance exists between dry air intrusion and eyewall CAPE production. Over warm water, eyewall CAPE production is sufficient (because of strong surface fluxes) to overcome small amounts of dry air intrusion caused by (i) the proximity to land, (ii) surrounding dry environmental air, and/or (iii) rainband outflow. The TC decays, however, if dry air intrusion increases above a certain threshold and/or local eyewall CAPE production falls below a certain threshold—because of landfall or a decrease in SST. The value of these thresholds depends on both the nature of the environment and the structure of the hurricane. The problem can be further complicated by the presence of a midlatitude trough, often accompanied by dry air in addition to strong vertical wind shear, as well as mechanisms that possibly enhance hurricane intensification. Therefore, predicting if, when, and how much storm weakening occurs before landfall poses a challenging problem.
The author thanks GFDL for supplying the atmospheric analysis and the U.S. Navy for providing the OTIS SST field. Much gratitude is due to Keith Blackwell for in-depth discussions and suggestions. Holly Allen, Christopher Dyke, and Robert Barbre generated many of the graphics to analyze model output. All simulations and analyses were performed on Sun Microsystems Inc. machines that were acquired through a SUN Mircosystems Inc. equipment grant.
Corresponding author address: Sytske Kimball, Dept. of Earth Sciences, LSCB 136, University of South Alabama, Mobile, AL 36688. Email: email@example.com