Experimental GOES Sounder Products for the Assessment of Downburst Potential (2024)

a. Wind index

The melting-level height (H_M) is an important factor in the generation of strong surface wind gusts because, within convective downdrafts, it is the altitude where cooling due to phase change occurs. While the latent heat lost due to melting is relatively small, the cooling process occurs rapidly, thus providing a strong impetus in downdraft generation. That process continues via evaporative cooling in an unsaturated subcloud layer, resulting in significant negative buoyancy, and strong downdraft winds that may reach the surface.

The WINDEX is a quantitative parameter that employs some simplifying assumptions, such as a linear dependence between outflow speed and vertical lapse rate, and other approximations. The multiplicand 5 in the WINDEX equation converts the wind gust values to knots. The R_Q factor adjusts the WI for overestimation within dry low-level (i.e., dry microburst) environments (McCann 1994). This correction accounts for situations where all of the precipitation evaporates, and the descending downdraft is very weak, perhaps not reaching the surface at all. The subtractand value 30 is approximately 5.5², which is the lowest lapse rate capable of producing microbursts in numerical model simulations, according to Srivastava (1985). When the quantity (Γ² − 30 + Q_L − 2Q_M) is less than or equal to 0, WI is set to zero. The maximum wind gust values are conditional upon the occurrence of convection and, thus, will not be observed in many areas. WINDEX may be used in any type of microburst environment (wet, dry, or hybrid).

b. Dry microburst index

An important consideration in deciding whether a dry microburst may occur is the strength of the convective updrafts. There is evidence that weak updrafts (characterized by slightly negative LI values and small CAPE) result in small rimed snowflakes that evaporate very quickly [as proposed by Brown et al. (1982) and verified by subcloud measurements by Rodi et al. (1983)]. Although the DMI does not yet consider CAPE as a limiting factor, CAPE is a GOES-derived Sounder product that is easily accessible via the Internet from NESDIS¹ and will soon become operationally available to National Weather Service (NWS) forecasters. The NWS Storm Prediction Center currently uses a CAPE value of 50 J kg⁻¹ as a minimum threshold in assessing the likelihood of “dry” thunderstorms that may trigger wildfires. Use of DMI and CAPE together should be helpful in evaluating the potential risk of dry microbursts on a given day.

c. Maximum vertical theta-e (θ_e) differential

The equivalent potential temperature (θ_e or theta-e) is a measure of the total static energy (sensible heat, latent heat, and geopotential) in an atmospheric column. Due to its strong dependence on moisture, theta-e typically decreases rapidly with height above the boundary layer, reaching a minimum in the middle troposphere, then increases again into the upper troposphere and stratosphere due to the increase in potential temperature (with minimal amounts of moisture) at those levels. In situations described earlier where the environment is more favorable for wet microbursts, the maximum vertical theta-e differential (hereafter referred to as TED) from the boundary layer to the middle troposphere can be very large, sometimes exceeding 30 K (Atkins and Wakimoto 1991). Large values of TED have been noted to precede severe convection (Darkow 1968) and were sometimes associated with tornadoes. More recently, large values of TED have been associated with the occurrence of wet microbursts (Ellrod 1989; Atkins and Wakimoto 1991; Wheeler 1996). It should be noted that there are many situations where TED may be large and, yet, convection does not occur due to the presence of a capping inversion above the boundary layer.

At Cape Canaveral Air Station (CCAS), Florida, the maximum TED value from the surface to midtroposphere is used to forecast wet microburst activity (Wheeler and Roeder 1996). CCAS is responsible for launch weather support for all United States space operations at Cape Canaveral and Kennedy Space Center. Division of TED by the value 30 K results in what is referred to as the microburst day potential index (MDPI). The 30-K threshold was based on local empirical tuning. An MDPI of ≥1 indicates a high probability of microbursts if convection were to occur. MDPI is only used operationally when the probability of thunderstorm activity is forecast to be ≥60%. Other weather forecast offices in the southeast United States are using a form of TED or MDPI, but sometimes with different threshold values.

a. Southern Arizona windstorms: 14–15 August 1996

On the evening of 14–15 August 1996, severe thunderstorms struck portions of southern Arizona from the northern suburbs of Phoenix (PHX) westward to Yuma. Winds exceeded 100 kt (51 m s⁻¹) at Deer Valley just north of Phoenix, and there were widespread reports of 50–60 kt (26–31 m s⁻¹) winds. Heavy rains ranging up to a maximum of 1.63 in. accompanied the high winds. It was the strongest wind ever recorded in the state of Arizona, and damage exceeded $100 million (Haro and Green 1996). Conditions in the two days preceding the event were characterized by extreme heat and steadily rising dewpoint temperatures that were >5°C above the climatological mean for mid-August. The evolution of storms on the afternoon of 14 August is shown by a four-panel GOES IR image (Fig. 4). By 2130 UTC, convection was developing rapidly over higher terrain to the north and northeast of Phoenix. Around 0030 UTC, an outflow boundary intersection from two separate cells occurred north of Phoenix, resulting in a strong new cell that produced the highest winds and most severe damage (lower-left panel of Fig. 4). The storms swept across southern Arizona from east to west, reaching Yuma by around midnight, Pacific standard time.

Downburst products and soundings were generated retrospectively for this event using data from GOES-9 beginning at 1803 UTC. A color-coded WINDEX product valid at 2103 UTC is shown in Fig. 5. Maximum potential wind gusts of >60 kt (31 m s⁻¹) are shown for the Phoenix area, with values >70 kt (37 m s⁻¹) possible in the desert just to the southwest. It should be noted that after a couple of summers observing WINDEX data for the desert Southwest, it is apparent that values of this magnitude are quite common. However, coupled with the extreme instability and resultant likelihood of storms on this day, WINDEX could have been useful in predicting wind gusts for storms in this area. Although WINDEX values significantly underestimated the 100-kt maximum observed winds, they were representative of most of the other wind reports on this evening. An enlarged plot of GOES-derived WINDEX at 2103 UTC, along with maximum wind gusts from airports and storm spotters that evening (Fig. 6), show generally good agreement, except for the small area just north of Phoenix.

Sounder data from 2103 UTC indicated that there was a strong gradient of DMI across the state of Arizona (Fig. 7). High DMI values in northeast Arizona diminished rapidly to near zero in the southern part of the state, indicating a better likelihood of wet (or hybrid) microburst conditions in the Phoenix area.

Figure 8 is a GOES sounding at 0003 UTC 15 August 1996, located near Gila Bend, about 100 km southwest of PHX, the closest available retrieval at this time. Overlain are radiosonde data from Tucson (TUS) at 0000 UTC and a special sounding at Luke Air Force Base (LUF), 30 km northwest of Phoenix, taken as part of the SouthWest America Monsoon Project (Maddox et al. 1996). The LUF sounding was obtained from a highly accurate Cross-Chain Loran Sounding System (CLASS) instrument. There is excellent correspondence between the temperature profiles from the three systems, indicating that the strong lapse rates (8–9 K km⁻¹) shown by the GOES Sounder data were reliable. The moisture appears to be underestimated by GOES from approximately the surface to 800 hPa, with an overestimate in the middle/upper troposphere from approximately 400 to 200 hPa. These moisture differences resulted in somewhat higher WINDEX values derived from the two radiosonde soundings, compared with GOES. For example, WINDEX calculated from the LUF sounding was 81 kt (42 m s⁻¹), and from the TUS sounding 73 kt (38 m s⁻¹), versus only 66 kt (34 m s⁻¹) from GOES. Since the three soundings were not collocated, some differences are to be expected. A summary of all microburst parameters around 0000 UTC is shown by Table 2. In summary, the downburst parameters derived from GOES on 14–15 August 1996 indicated the potential for very strong winds (60–65 kt), but not of the extreme magnitude observed, and that the storms would most likely result in wet or hybrid-type downbursts.

b. Western Kansas downbursts: 24 June 1998

During the evening of 24 June 1998, there were more than 20 reports of downbursts in western and central Kansas that eventually spread into the eastern part of the state. Maximum winds were estimated to be as high as 100 kt (51 m s⁻¹) with a dry microburst at Ulysses in the far southwest corner of the state, but most were in the 50–70-kt (26–36 m s⁻¹) range in west-central Kansas. There was some minimal damage reported (∼$30 000) and no injuries or fatalities (National Climatic Data Center 1998). GOES downburst products on this day demonstrated their ability to detect changes in mesoscale conditions in the lower and middle troposphere that contributed to these high wind events.

A surface trough and dryline were present across western Kansas during the day, as a weak surface low developed in the northwest corner of the state. There was moderate [30–50 kt (15–26 m s⁻¹)] southwest flow aloft. By late afternoon (2046 UTC), GOES DMI showed a narrow band of high values (yellow = ≥16 units) associated with some midlevel moisture observed in the GOES 6.7-μm water vapor channel (left panel, Fig. 9), and a deep layer of nearly dry-adiabatic lapse rates. The 2246 UTC GOES DMI still showed this maximum somewhat, although there were fewer retrievals in the moist band due to the formation of high-based, convective clouds (right panel, Fig. 9).

Figure 10 shows the locations of surface wind gust reports relative to GOES WINDEX values at 2200 UTC, about 2 h prior to the onset of most wind gust events. The approximate location of the dryline is shown, based on surface data at 0000 UTC 25 June 1998. A strong gradient of WINDEX can be seen, with values in the middle to high 60s (kt) along and southeast of the dryline (Fig. 10) decreasing to the 40s west of the dryline. Surface temperatures on this afternoon were around 100°F (38°C), and dewpoints were in the low to middle 40s (°F; 6°–8°C) west of the dry line. Although no precipitation data were available, the earliest high wind reports (from 2230 to 0100 UTC) located to the west of the dryline were most likely dry microbursts. After 0000 UTC, deep convection with heavy rainfall developed in the moist, high-θ_e air east of the dryline. Figure 10 shows that the GOES WINDEX values were representative of the observed maximum wind reports in Kansas on this day, except for the isolated dry microburst at Ulysses. Based on operational experience, the limiting value for WINDEX can be expected to be around 80 kt (41 m s⁻¹) so it is unlikely that an event of this magnitude could have been anticipated from the GOES sounder data.

A possible mechanism for the production of the dry microbursts in western Kansas is one that commonly occurs on the lee side of the Rocky Mountains when convective clouds that develop over elevated terrain drift eastward over the high plains into an environment that is highly unstable, but dry below cloud base. Referred to as “cloud base detrainment instability” (CBDI), this process is described in detail by Emmanuel (1981). CBDI theory accounts for the generation of penetrative, unsaturated downdrafts from small- to medium-sized cumulus clouds that in some cases can reach the ground through evaporative cooling in the subcloud layer.

A GOES sounding at 2246 UTC, located about 20 km southeast of Dodge City (DDC) is compared with the 0000 UTC DDC radiosonde profile in Fig. 11. Based on comparison with the moisture profile from an earlier (1946 UTC) GOES sounding (solid trace), the most significant increase in moisture occurred from just above the surface (800 hPa) up to about 600 hPa. Thus, there was a deepening in the boundary layer moisture, indicating that conditions near DDC were being modified from a dry downburst environment to one more suitable for wet (or hybrid) downbursts. These soundings appeared to be located just to the east of the midlevel moist band depicted in Fig. 9.

Figure 11 shows excellent agreement between the GOES and DDC temperature profiles. GOES dewpoints were slightly higher than DDC in the 600–750-hPa layer, and slightly lower than DDC below 750 hPa, but were mostly representative of existing conditions. A comparison of all three microburst parameters from GOES versus the DDC radiosonde (Table 3) also compares favorably. Standard convective indices such as LI and CAPE (not shown) were unspectacular (around −1°C and 300–700 J kg⁻¹, respectively), which is typical of dry microburst conditions (Wakimoto 1985), but had destabilized throughout the afternoon. In summary, GOES microburst products, including vertical soundings, would have been valuable in this case to 1) highlight the potential for dry microbursts in western Kansas, 2) estimate the maximum expected wind gusts, and 3) show the steady increase in low- and midlevel moisture, signaling a transition from dry to wet microburst conditions.

a. Wind index

Values of GOES WINDEX were compared with adjacent reports of straight-line wind damage or wind gusts >50 kt (26 m s⁻¹) reported in the NWS Storm Prediction Center’s (SPC) preliminary storm data. GOES data used in the verification were obtained for retrieval times 1–3 h prior to the observed surface wind gust or damage report so that their predictive value could be assessed. Care was taken to obtain WINDEX away from the influence of any low-level outflow boundaries that would result in values unrepresentative of the inflow air mass feeding the storm. Based on more than 300 reports of wind damage during the summer of 1997, most (92%) of the corresponding WINDEX values were >40 kt, while a majority (60%) exceeded 50 kt (26 m s⁻¹). The distribution of GOES WINDEX for storm damage events is shown in Fig. 12. McCann (1994) obtained a WINDEX distribution derived from surface and radiosonde data for 207 cases that is similar to Fig. 12 with a peak frequency in the 56–60-kt (29–31 m s⁻¹) range and a sample mean of 58 kt (30 m s⁻¹). It should be noted that there are many microburst events with wind gusts less than 50 kt (26 m s⁻¹) that are operationally significant for aviation. For example, using a 4-yr data sample from the Cape Canaveral mesonet, it was found that 90% of peak wind speeds for 282 microbursts fell between 25 and 44 kt (13–23 m s⁻¹) (Sanger 1999).

As previously mentioned, WINDEX typically has an upper limit of near 80 kt (41 m s⁻¹), supported by Fig. 12 and results from McCann (1994). Since downburst wind gusts have been measured as high as 130 kt (67 m s⁻¹) (e.g., Fujita 1983), WINDEX will not be able to account for these extreme events. WINDEX actually estimates the maximum downdraft velocity due to negative buoyancy from a stationary storm. Thus, WINDEX does not include several effects that could help account for these underestimates of extreme events: 1) horizontal acceleration on the ground from hydraulic pressure as cold dense air continues to accumulate behind the downburst leading edge, 2) conversion of rotational energy into straight-line kinetic energy when the rotor vortex collapses, 3) translational speed of the storm, 4) downward advection of momentum, and 5) venturi effects as local land features and structures focus the outflow.

Mean absolute values of GOES WINDEX were also compared with observed surface wind gust reports obtained from SPC storm data, as well as WINDEX calculated from the Eta Model FG. Estimated wind gusts (often reported as 52 or 61 kt, for example) were removed from the database. A tabulation of these data based on 82 comparisons is shown in Table 4. Mean GOES WINDEX differed from mean SPC surface wind gusts by <2 kt (1 m s⁻¹) with a slight negative bias. Results for the GOES data were slightly better than from the Eta FG (by 0.4 kt), but this difference is not statistically significant. In the 26 cases where GOES differed from the FG by ≥3 kt, GOES improved upon the FG in 12 of those cases (46%), while the FG was superior in the remaining 14 cases. In their evaluation of temperature and moisture from GOES-8 retrievals, Rao and Fuelberg (1998) also found that while there was improvement over the FG (which in their study was the Nested Grid Model) the majority of the time, there was also occasional degradation. For example, 620-hPa temperatures derived from GOES-VAS were worse than the FG in 11% of the cases, and 620-hPa dewpoints were worse 34% of the time.

The correlation between GOES WINDEX and SPC data for all 82 events was quite poor (0.25), but by eliminating the 7 nighttime [later than 2000 local standard time (LST) and before 1000 LST] events, the correlation improved slightly to 0.37. The correlations for both GOES retrievals and Eta FG are summarized in Table 5. The mean GOES WINDEX error for the nighttime events soared to −22 kt (−11 m s⁻¹). The diminished boundary layer lapse rate beginning around sunset is undoubtedly an important factor in these poor results, suggesting that a different approach must be used at night.

A recent study (Dickerson 2000) that compared WINDEX derived from the 1500 UTC (1000 EST) Cape Canaveral radiosonde versus maximum wind gusts reported by the CCAS mesonetwork between 1500 and 2200 UTC on 114 days showed a correlation of 0.24, a lower value than that obtained from the GOES WINDEX daytime data sample. These results suggest that the accuracy of GOES WINDEX is comparable to, or perhaps better than that which can be obtained using radiosonde data. The poor correlation of these data is likely due to factors such as the high time and space variability of the microburst event, storm motion, and the inability of WINDEX to accurately simulate physical processes in some instances. McCann (1994) found a good qualitative correlation between WINDEX and a small sample of observed microburst events. In a postanalysis of the Dallas microburst storm of 2 August 1985, McCann obtained a WINDEX value of 70 kt (36 m s⁻¹), which was exactly what was observed. Analysis of GOES-VAS sounder data at 2218 UTC for the Dallas case (the time period used to produce Fig. 1) determined a maximum WINDEX of 68 kt (35 m s⁻¹).

The sensitivity of the GOES retrievals to their FG temperature and moisture profiles is illustrated by a recent in-house evaluation at NESDIS. In this study, information in the Eta FG below 700 hPa was found to be less accurate than corresponding Aviation Model (AVN) FG data. Corresponding PW and stability indices were found to be significantly improved when the Aviation Model was used as an FG instead of the Eta. The differences were greatest in moist atmospheric conditions when convection is most likely. It is possible that using these superior AVN FG data in the initial retrieval processing could have resulted in improvement upon the FG in more than 46% of the cases. As a result, the AVN model was installed as the FG for GOES retrievals at the NESDIS Office of Research and Applications beginning in early October 1999. Evaluation of the effects of this change on WINDEX and other parameters is planned.

WINDEX typically exhibits a highly diurnal variability due to its dependence on the square of the lapse rate. Figure 13 shows how WINDEX changed through the course of a day at six inland locations. WINDEX values increased rapidly through the morning hours, normally reached a maximum by approximately 1100–1300 LST, and remained rather steady until late afternoon, when a more gradual decline began. Since the development of cumulus and cumulonimbus clouds may prevent the determination of WINDEX in some areas during the course of an afternoon, these results suggest that an earlier WINDEX would still be valid. Further analysis of these data may allow prediction of maximum possible winds later in the day by extrapolation of midmorning values or other statistical techniques. A major assumption would be that no significant changes in airmass characteristics are expected to occur.

b. DMI

DMI from GOES was compared with the same index derived from radiosonde data. A plot of data from the two sources (Fig. 14) shows considerable scatter, with a correspondingly low correlation (r = 0.505). Rao and Fuelberg (1998) also found significant differences in satellite-retrieved versus radiosonde-measured moisture at 500 hPa. Most of the scatter can be attributed to 1) poor vertical moisture resolution of the GOES Sounder, 2) errors in radiosonde measurements, 3) differences in the two types of measurements (point vs bulk layer), and most notably 4) the large variability of moisture in time and space. Typical radiosonde errors could be on the order of 0.5 K for temperature, and 10% for relative humidity (Schmidlin 1988; Wade 1994). These factors are compounded in the DMI by the use of temperature–dewpoint differences at two levels.

Qualitative evaluation has shown, however, that in areas where DMI is high, atmospheric temperature and humidity profiles obtained from the GOES-9 Sounder often exhibit the classic hourglass profiles typical for dry microburst conditions. A good example of this is the Dodge City profile shown in Fig. 11. Since the DMI uses data from 500 to 700 hPa, it is best suited as a diagnostic tool for higher terrain west of approximately 100°W.

c. Maximum theta-e difference (surface–300 hPa)

The maximum TED from the surface to 300 hPa was validated by calculating similar values using Eta Model FG data. A scatterplot (Fig. 15) for 100 data points indicates good agreement, with a correlation coefficient of 0.77. Additional validation of TED versus radiosonde data is planned. As with WINDEX, the distribution of TED for a large number of wet microbursts in the central and eastern United States was obtained during the summer of 1997 (Fig. 16). More than two-thirds of the events (67%) occurred with TED ≥ 20 K, which has been specified previously as a minimum threshold for wet microburst activity (Atkins and Wakimoto 1991). The large number of cases less than this threshold is an indication that this value must be used with caution, however.

Maximum TED divided by 30 K (MDPI) derived from radiosonde data has been validated at CCAS since 1995 (Wheeler 1996). For this east-central Florida environment, MDPI had a POD of 97%, with a FAR of 28%, a CSI (i.e., a threat score) of 70%, and a Heidke Skill Score of 71% (skill relative to random forecasting) based on 176 events in 1995 and 1997 (Wheeler 1997). These values were sustained even during the summer of 1997, which was considered to be an abnormal year in that many of the significant convective events were synoptically forced (Wheeler 1997), which speaks well for the robustness of the technique. In an independent study conducted by Florida Institute of Technology based on data collected on 26 days during August 1998, the MDPI (once again based on radiosonde data) was evaluated for its utility as a yes–no forecast parameter. Using TED ≥ 20 (MDPI = 0.67) as a predictor, the POD was determined to be 100%, but there was a fairly high FAR of 42%. Thus, further tuning of the threshold, or inclusion of other predictors, will be required to reduce the FAR to acceptable levels.

Table 1.

GOES Sounder channel characteristics

Table 2.

Microburst parameters at 0000 UTC 25 Aug 1996

Table 3.

Microburst parameters at 0000 UTC 25 Jun 1998

Table 4.

WINDEX (GOES retrievals/Eta first guess) vs storm data (N = 82)

Table 5.

Correlation of WINDEX to storm data