The case for upgrading Hurricane Andrew to Category 5
Jimmy Franklin
Junior Hurricane Specialist
National Hurricane Center
Miami, FL

This document proposes a set of revisions to the "best track" of 1992's Hurricane Andrew, which is currently classified as a Category 4 hurricane at its landfalls at Eleuthera Island and in south Florida. The revisions would make Andrew a Category 5 hurricane at both landfalls, with a Florida landfall intensity of 145 kt. The peak intensity of Andrew is assessed to be 150 kt at 1800 UTC 23 August. The proposed new wind history is given in Fig. 1 and Table 1; no changes are proposed to the pressure best track (Fig. 2). The original wind best track is shown in Fig. 3.

The case for the revision is outlined below:

1) Franklin et al. (2002, Weather and Forecasting, submitted) examined several hundred GPS dropwindsonde profiles in the hurricane eyewall and have shown that the mean ratio of surface to 700 mb winds was about 91.5% in the eyewall region. However, because there was an effort to preferentially seek out the surface wind maximum at the expense of the flight-level maximum, this ratio may not be quite appropriate for estimating the peak surface wind from the peak flight-level wind. Therefore, a separate analysis was conducted at the flight-level RMW. At the flight-level RMW (FL-RMW), the mean surface to 700 mb wind speed ratio was in fact a bit lower, 88%. Since sondes dropped at the flight-level RMW encounter the peak flight-level wind but not the peak surface wind, the 88% ratio should represent a lower bound on the mean ratio of peak surface winds to peak 700 mb winds, if azimuthal variation is ignored.

Now in fact there is some variability of the mean adjustment ratio with azimuth, but this variability doesn't change the recommended adjustment factors. The strongest FL winds are generally found on the right-hand side of the storm. The eyewall sondes show that the adjustment factor in the right quadrant is a little lower than in the left quadrant (0.891 vs. 0.928, for all eyewall sondes). Assuming this difference is applicable at the FL-RMW, a more accurate lower bound mean adjustment factor for the right quadrant FL-RMW would be just above 86%. The upper bound on the mean adjustment factor for the right quadrant would be just under 90%.

Consequently, unless there is other information to the contrary, current operational practice is to assess the intensity of a hurricane at about 90% of the peak wind observed at the 700 mb level. This factor is at the high end of the range established by the sondes (86%-90%), in part due to a simple rounding of the midpoint of this range, but also to account for the likelihood that the highest 700 mb wind speed was not sampled.

2) From a statistical sense, it is important to distinguish between the following two questions: 1) given a wind observation taken at 700 mb of 100 kt somewhere in the hurricane eyewall, what is the underlying concurrent wind at the surface; and 2) given that a storm's maximum sustained wind at 700 mb is 100 kt, what is the storm's maximum sustained wind at the surface. Individual eyewall profiles have a high variability about the 90% mean relationship (the standard deviation is about 19%). This variance is directly applicable only to the first question posed above, and strictly speaking applies only to instantaneous, rather than 1 min mean winds. This variance is not appropriate for estimating the variability in the ratio of the storm's peak surface to peak 700 mb winds. The variance of the latter quantity is not known, but should be much less than 19%, as illustrated by the following example.

Imagine a steady-state storm in which the maximum sustained wind at the surface is 90 kt, while the maximum sustained wind at 700 mb is 100 kt. Over time, the precise location of the maxima at both the surface and 700 mb may vary, but the peak values present at each level remain constant. By assumption, the variance of peak sfc to peak 700 mb wind is zero. Even so, if the eyewall of this storm were to be sampled by GPS dropsondes, the dropsonde sfc to 700 mb wind ratio will still have a large variance, because (a) the wind distributions are not necessarily the same at the two levels, so a sonde that samples the maximum at 700 mb will not sample the surface maximum, and vice versa, and (b) the sondes don't measure a 1 min sustained wind, but measure whatever turbulence they happen to fall through. These factors contribute the high variance in adjustment factor ratios computed from individual sondes, but it has very little to do with the relationship between peak winds from level to level. It is worth noting that a similar argument applies to adjustment factors computed from SFMR data, through point (a) above.

It has been suggested that applying the 90% rule to the flight-level data is inconsistent with the Fowey Rocks ob. Because of the turbulent and transient nature of hurricane wind field, no conclusions can reasonably be drawn from a storm-relative comparison of adjusted flight-level winds to the couple of observations available for Andrew (neither of which was a sustained surface wind, in any event). It would take a systematic discrepancy over many observation points to warrant a departure from the mean relationship.

3) The question is sometimes raised as to whether the dropsonde-based relationships for adjusting winds from one level to another apply to 1 min sustained winds. The answer is that the relationships are valid for any averaging time. The dropsonde mean profile is an average of several hundred nearly instantaneous wind measurements (the sampling time of the GPS circuitry is on the order of milliseconds, but there is a 5 s filter applied to the profile data). To see that the profile applies regardless of the averaging time, consider our steady state storm again. At 700 mb, the wind speed at a given location at any time, v700, can be expressed as the sum of v700m, a mean value only a function of location, and v700'(t), the time-varying departure from the mean. The wind at the surface can be expressed similarly, as vsfc = vsfcm + vsfc'(t). Now, imagine that we take several hundred paired measurements of v700 and vsfc at this location, and calculate the ratio vsfc/v700. Given a sufficiently large sample, the mean value of vsfc/v700 that we get from this experiment will be vsfcm/v700m, since the long-term average of each the two primed variables is 0. We will get this answer regardless of whether our measuring device measures a 10-s average wind or a 100-min average wind, since in either case the time-varying component of the wind at each level will average to 0.

What this means is that, if you have a wind that you believe is representative at 700 mb, you can use the dropsonde relationship to get a similarly-averaged representative wind at the surface. However, you cannot necessarily use the dropsonde relationship to derive, say, the maximum 1-min wind observed over a 10 min period at the surface from the maximum 1-min wind observed over a 10 min period at 700 mb. You can't do this because the dropsonde relationships only apply to the mean winds, and the relationship between 10 and 1 min winds at 700 mb may be different from the relationship at the surface.

4) About an hour prior to landfall, at 0810Z, reconnaissance aircraft measured a 10-s wind at 700 mb of 162 kt, 90% of which is 146 kt. Ten-second flight-level winds are customarily assumed to represent 1-min sustained winds. One-minute averaging of the flight-level winds would tend to underestimate the true maximum 1-min wind because the aircraft doesn't remain in the peak gradient region that long on a radial flight track. Peak 10-s flight level winds supporting category 5 at the surface (using an adjustment of 90%) were found in three consecutive minobs (0809, 0810, and 0811Z). The 1-min mean flight-level wind at 0810 was 159 kt, or 143 kt when adjusted to the surface using 90%. Neither were these the only aircraft obs that support category 5 winds; a later 10-s report (over land, at 0918Z) yields 138 kt using the same reduction. The 1-min mean aircraft wind at 0918Z was 149 kt, which reduces to 134 kt using 90%. Lacking compelling contradictory data, the flight-level observations would be accepted as representative and a hurricane presenting these obs today would almost surely be assigned an operational intensity of about 145 kt.

5) After the time of the 162 kt aircraft report, Andrew's central pressure continued to fall up to and just after landfall, perhaps by as much as 10 mb. Chances are that flight-level winds at the time of landfall were even higher than what was observed.

6) Assuming that the 162 kt aircraft wind is representative of the peak winds present at 700 mb, a surface adjustment factor of 83% or less is required to keep Andrew a category 4 hurricane. Of all the storms so far examined, only Bonnie, a weakly convective storm with a large eyewall, has been observed to have a mean ratio that low. If the surface winds are kept at 125 kt, this implies an adjustment factor of 77%. No storm has yet been observed to have a mean adjustment factor this low. Furthermore, although sample sizes tend to get small when the dropsonde sample gets stratified, there does appear to be enhanced relative surface winds when the winds are very high, and when vertical motions are particularly vigorous. Andrew at landfall would have satisfied both of these conditions. In other words, there is nothing that we have seen that would lead us to think that Andrew had a lower than normal adjustment factor.

7) Low-altitude radar cell tracking suggests surface winds similar to those implied by the flight-level data. Peter Dodge has generated some vectors based on cell tracking from the Miami WSR-57 radar in and inside of Andrew's eyewall just prior to landfall in south Florida. The two highest cell motions found were 176 kt at 400 m and 180 kt at 1100 m (Fig.4). If we assume that the maximum wind in the eyewall near the boundary layer top (BLT) was 178 kt, we can adjust this wind to the surface using mean dropsonde profiles. If we use the overall eyewall mean profile to go from the BLT to the surface (75%), we get a surface-adjusted wind of 134 kt, just shy of category 5. However, as mentioned above, there is an apparent relationship between boundary layer adjustment factors and wind speed, with adjustment factors increasing with wind speed once the BLT speed exceeds about 105 kt (Fig. 12 of Franklin et al.). Franklin et al suggest that the surface to BLT adjustment factor increases to 82% for BLT wind speeds of 135-155 kt (this is their highest category). Using this adjustment, the BLT wind of 178 kt converts to 146 kt at the surface. Remembering that cell track motions represent mean winds over several minutes, the corresponding maximum 1 min sustained speed would be even higher.

8) Andrew's pressure of 922 mb corresponds to a wind of 139 kt (Cat 5), using the operational pressure-wind relationship. Small storms will tend to have higher winds for the same central pressure than large storms, and so given Andrew's size, its 922 mb pressure could easily have supported winds of 145 kt. The proposed revisions to Andrew's best track fall within the range of existing scatter in the operational pressure-wind relationship (Fig 5). The current best track pressure-wind pair at landfall (922 mb-125 kt) also falls within the observed scatter around the operational pressure wind curve; however, it is near the edge on the light-wind (left) side of the curve. A storm of Andrew's size and structure should not be this far to the left of the pressure-wind curve.

9) The proposed peak wind for the S. Fla. landfall is 20 kt above the Dvorak estimates. However, the proposed points fall within the envelope of previously observed scatter, as 10% of all Dvorak intensity estimates are in error by 20 kt or more (Brown and Franklin, 2002). More importantly, Dvorak estimates of Andrew's intensity were consistently too weak, as can be seen in Fig 2. Beginning at 6Z on the 23rd, measured recon pressures were running 10-15 mb lower than the Dvorak estimates. Dvorak pressure estimates remained consistently too weak until after the Florida landfall. Given this, as well as the likelihood of Andrew's winds exceeding the Dvorak pressure-wind expectation because of Andrew's size, the Dvorak wind estimates for Andrew cannot reasonably be expected to yield an accurate intensity estimate.

10) Chris Velden retrospectively ran the objective Dvorak algorithm on Andrew. The ODT shows an intensity spike (T6.7) at 0900Z (Table 2), in good agreement with the pressure tendency after the last recon pressure. Chris cautions that due to the look angle associated with single GOES operations in 1992, that the ODT is probably an underestimate.

11) In a run of SLOSH recently conducted by Brian Jarvinen, the model gives a storm tide of 14.7 ft using a 1 min sustained wind of 145 kt, near where a tide of 16.9 ft was observed. Thus, the peak observed surge is consistent with cat 5 winds. SLOSH overestimates outside of the peak area are probably due to a pressure/wind field that is not sufficiently concentrated in the eyewall. In a 1999 paper by Houston, Shaffer, Powell, and Chen, the authors describe another SLOSH run for Andrew that also produced less surge than what was observed. Peak SLOSH 10-min winds in this run were 63.1 m/s. The SLOSH model uses a factor of 1.15 to convert its 10-min winds to 1 min winds; this yields a peak 1-min wind of 141 kt. Even using an 11% adjustment to go from 10 min to peak 1 min yields cat 5 winds of 136 kt.

12) The purpose of the Best Track Committee is to ensure the most accurate historical hurricane record possible, consistent with contemporary science. It has been suggested that the record in the case of Andrew should not be changed, in part because of the uncertainty surrounding the maximum wind, and an analogy has been made to the rules of instant replay in the NFL, where conclusive evidence is required to overturn an official's call. Such a standard seems inappropriate for this Committee's deliberations, though, and is inconsistent with the Committee's mission statement. No storm's intensity can be determined with precision; the observations are almost never sufficiently comprehensive, and indirect measures must always be used. Uncertainty in a wind speed estimate should not be an obstacle to revising an earlier estimate that is inconsistent with the observations, interpreted using current understanding.