15 JANUARY 2012
CECIL AND BLANKENSHIP
687
Toward a Global Climatology of Severe Hailstorms as Estimated by Satellite Passive Microwave Imagers DANIEL J. CECIL Earth System Science Center, University of Alabama in Huntsville, Huntsville, Alabama
CLAY B. BLANKENSHIP Universities Space Research Association, Huntsville, Alabama (Manuscript received 1 March 2011, in final form 12 July 2011) ABSTRACT An 8-yr climatology of storms producing large hail is estimated from satellite measurements using Advanced Microwave Scanning Radiometer for Earth Observing System (AMSR-E). This allows a unique, consistent comparison between regions that cannot be consistently compared using ground-based records because of varying data collection standards. Severe hailstorms are indicated most often in a broad region of northern Argentina and southern Paraguay and a smaller region in Bangladesh and eastern India. Numerous hailstorms are also estimated in the central and southeastern United States, northern Pakistan and northwestern India, central and western Africa, and southeastern Africa (and adjacent waters). Fewer hailstorms are estimated for other regions over land and scattered across subtropical oceans. Very few are estimated in the deep tropics other than in Africa. Most continental regions show seasonality with hailstorms peaking in late spring or summer. The South Asian monsoon alters the hailstorm climatology around the Indian subcontinent. About 75% of the hailstorms on the eastern side (around Bangladesh) occur from April through June, generally before monsoon onset. Activity shifts northwest to northern India in late June and July. An arc along the foothills in northern Pakistan becomes particularly active from mid-June through mid-August. The AMSR-E measurements are limited to early afternoon and late night. Tropical Rainfall Measuring Mission (TRMM) measurements are used to investigate diurnal variability in the tropics and subtropics. All of the prominent regions have hailstorm peaks in late afternoon and early evening. The United States and central Africa have the fewest overnight and early morning storms, while subtropical South America and Bangladesh have the most.
1. Introduction Surface-based climatologies of hailstorms are limited by inconsistencies in observational networks and reporting practices. These generally vary from country to country and also vary with population density within individual countries. Several previous studies have examined climatologies from individual countries or groups of countries. Frisby and Sansom (1967), Williams (1973), and Barnes (2001) provide reviews of hailstorm climatologies from many countries, but the disparities in reporting between those countries are obvious. Figure 1 (from Williams
Corresponding author address: Dr. Daniel J. Cecil, University of Alabama in Huntsville, 320 Sparkman Dr. NW, Huntsville, AL 35805. E-mail:
[email protected] DOI: 10.1175/JCLI-D-11-00130.1 Ó 2012 American Meteorological Society
1973) presents a composite global map based on earlier studies; however, it highlights mountainous areas that are more prone to graupel or small hail showers instead of the large hail we are interested in here. Knight and Knight (2001) discuss this in the context of the lack of global hailstorm climatologies. Studies using data from standard surface weather stations or hailpads may be dominated by cases with hailstones ;1 cm in diameter or smaller. While such small hail can be damaging to certain crops, we are interested here in hail large enough to satisfy the National Weather Service’s criteria for a severe thunderstorm (1-in. diameter, or ;2.5 cm). Of course, there is little ‘‘ground’’ truth from the oceans. A search of Storm Data (severe weather reports compiled by the National Climatic Data Center) reveals very few storms producing hail from 0.75 in. (;2 cm) to 1 in. (;2.5 cm) in diameter on the Hawaiian Islands. Russell
688
JOURNAL OF CLIMATE
VOLUME 25
FIG. 1. Global map of annual hail days, from Williams (1973), based largely on Frisby and Sansom (1967).
(1893) compiled many descriptions of hailstorms from around the world, including one that damaged a ship’s compass about 300 mi (;500 km) south of the Cape of Good Hope. In response to the dearth of oceanic entries in Russell’s book, Harries (1895) examined 14 ship logs and concluded that hail is rather common over the ocean. The ensuing discussion between Russell and Harries [published with the Harries (1895) article] demonstrates some of the issues involved in compiling hail climatologies. Russell pointed out that most of Harries’s examples of hailstorms were likely graupel, sleet, or small hail from cool-season storms (unlike the large hail with which Russell’s book had been concerned). The examples that mentioned lightning, thunder, and hail sizes ;2 cm or larger were almost all within ;500 km of the coasts (mostly the Cape of Good Hope). Harries countered that large quantities of small hailstones can be damaging, and that shipping routes left the midocean regions sparsely sampled. Even with a robust system for reporting severe thunderstorms in the United States, population density plays a large role in the geographic distribution of hailstorm reports. Besides large metropolitan areas (such as Dallas– Fort Worth, Texas; Oklahoma City, Oklahoma; Kansas City, Kansas and Missouri; and Atlanta, Georgia), many
smaller cities (such as Lubbock, Texas, and Dodge City, Kansas) have a factor of 2–3 more hailstorm reports than adjacent rural areas (even after filtering out multiple reports of the same storm). Severe thunderstorm warning verification studies (e.g., Dobur 2005) have shown that more urban counties tend to have more severe weather reports per unit area, while more rural counties have greater false-alarm rates (warnings issued but no subsequent severe weather reports received). The lack of a report does not necessarily mean hail did not fall to the surface. Instead, it means that no one reported it if it did fall, or reports were not continuous throughout the hail swath. This paper uses passive microwave satellite data to objectively estimate severe hailstorm climatologies. Large ice hydrometeors scatter upwelling microwave radiation away from a satellite’s field of view, causing brightness temperature (TB) depressions far below the thermodynamic temperature in the atmosphere. Cecil (2009) compared reports of large (at least 0.75-in. diameter, or ;2 cm) hail in the United States to brightness temperatures from the Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI). The probability of large hail being reported increased with decreasing brightness temperature. The largest hail size categories were associated with
15 JANUARY 2012
CECIL AND BLANKENSHIP
FIG. 2. Fraction of TRMM precipitation features in south-central and southeast United States with Storm Data reports of large (1 in. diameter/2.5 cm, or greater) hail, as a function of precipitation feature’s minimum 85-GHz PCT (gray) or 37-GHz PCT (black). Methodology as in Cecil (2009).
progressively lower brightness temperatures, but that signal was not robust enough to predict the sizes. The 37-GHz channel seemed most useful for identifying severe hailstorms. The 85-GHz channel sometimes had extremely low brightness temperatures without reports of large hail, presumably because a deep column of large graupel or small hail is sufficient to scatter the shorter wavelength radiation. With the 19-GHz channel, the footprint size (;20 3 30 km) is too large, and many cases do not produce strong signatures. Cecil’s (2009) approach using TRMM precipitation features (Nesbitt et al. 2000) is used here, with the hail probabilities rederived for the 1.00 in. (2.5 cm) diameter now used by the National Weather Service as a criterion for severe thunderstorms (Fig. 2). We apply those probabilities to global measurements of brightness temperatures from TMI and from the Advanced Microwave Scanning Radiometer for Earth Observing System (EOS) (AMSR-E). Twelve years (1998–2009) of TMI data provides characterization of the global tropics and subtropics (up to 6388 latitude), including the diurnal cycle there. Eight years of AMSR-E (2003–10) (Ashcroft and Wentz 2006) data extends our results through the higher latitudes, with sampling limited to early afternoon and late night. We present maps using the AMSR-E data by itself, and use the TMI data for a diurnal context. The upwelling brightness temperature represents column-integrated effects, more responsive to middle and upper layers of the atmosphere in these storms (not directly seeing hail near the surface). A given brightness temperature does not uniquely map to a particular vertical profile of hydrometeors having particular distributions of particle type, size, and density. There are infinite combinations that can yield the same brightness temperature.
689
If there are systematic differences in the vertical profiles or hydrometeor distributions between the intense storms in the United States and those in another region, then the empirical relationships from the United States are not directly applicable to that region. To address this, Cecil (2011) analyzed radar-derived ice water content and vertical profiles of radar reflectivity for storms in different regions having the same upwelling 37-GHz polarizationcorrected brightness temperature (PCT). Profiles could generally be grouped together after identifying the region as tropical ocean, tropical land, subtropical ocean, or subtropical land. Based on the linear fits between 37-GHz PCT and mixed-phase-region ice water content, we scale the brightness temperatures toward the values that would be expected from storms over subtropical land (e.g., the United States, where the hail probabilities were derived). The scaling coefficients are taken from Cecil (2011), with Fig. 3 demonstrating their effect. The comparisons with radar data suggest that a given brightness temperature from a subtropical land region corresponds to vertical profiles that produce lower brightness temperatures in other regions. Another source of uncertainty could come from melting hailstones before they fall to the surface. However, large hailstones have large terminal velocities (tens of m s21). Roos (1972) estimated a 47 m s21 fall speed for the giant 1970 Coffeyville, Kansas, hailstone. Matson and Huggins (1980) measured 15–20 m s21 fall speeds for 1-cm-diameter hailstones, which are smaller than the threshold for large hail in this study. Even if there is some melting during the fall to the surface, a storm with large hail at middle levels (leading to the very low brightness temperatures used in this study) is still likely to have large hail reaching the surface. The training dataset from Cecil (2009) did include hot summer environments in the United States, where large hail reached the surface. It is important to emphasize that our measurements do not unambiguously identify the presence of large hail, but instead identify radiometric signatures consistent with observed hailstorms. Spencer et al. (1987) showed that similar measurements correlate with severe weather in general (strong wind, large hail, and/or tornadoes). We focus here on large hail, since the measurements are most directly related to large precipitation particles. Some storm types (e.g., bow echoes, downbursts, low-topped supercells) are more likely to produce damaging winds than large hail, and they would not necessarily be expected to have strong signatures in our analyses. The data sources and approaches are described in section 2. In section 3 we estimate global distributions of the frequency of hailstorm occurrence and variability through the year. These distributions appear reasonable, although they are limited to an 8-yr period. A longer record with more complete
690
JOURNAL OF CLIMATE
VOLUME 25
FIG. 3. (a) Conversion of measured 37-GHz PCT from tropical ocean, tropical land, or subtropical ocean values to equivalent subtropical land values, using Table 1 in Cecil (2011). (b) Geographic regions used in the conversion. The southern tip of India and Sri Lanka are treated the same as tropical Africa. The Maritime Continent, tropical South America, and adjacent waters are treated as ‘‘tropical mixed.’’
sampling (or merely a different set of years) would probably shift the locations and magnitudes of some local maxima (or minima) in the distributions. The general regions of high activity should remain unchanged. We briefly examine diurnal variability, although additional satellites should be used to more adequately sample the diurnal cycle. Some of the major differences between our results and prior studies of hail occurrence are discussed in section 4. Conclusions are summarized in section 5.
2. Data and methods AMSR-E 36.5- and 89.0-GHz PCT from January 2003 to December 2010 are the primary data used in this paper. The footprint sizes are about 14 km 3 8 km and 6 km 3 4 km, respectively. AMSR-E is on the Aqua satellite in a sun-synchronous orbit. Aqua’s equator crossing times are near 0130 and 1330 local solar time (LST); near the edges of the 1450-km-wide AMSR-E swath, the LST is earlier or later. The observation times are generally between approximately 0000 and 0300 and between 1200 and 1500 LST (varying with latitude and distance from the satellite subtrack). Convective storms are identified in the AMSR-E data as a pixel or set of contiguous pixels with 89-GHz PCT at or below 200 K. This is somewhat analogous to the approach used by Mohr et al. (1999) for mesoscale convective systems (MCSs) and other rain systems seen by passive microwave radiometers, which helped motivate Nesbitt et al. (2000) and subsequent TRMM precipitation feature definitions. We use a lower threshold PCT value than those in other studies, because hailstorms generally produce much lower PCT values. Artifacts with low brightness temperature due to surface snow cover are removed following Grody (1991), and by requiring at least
one pixel with 89-GHz PCT , 130 K. We impose no size limit on the AMSR-E identified storms. A single ‘‘storm’’ may extend for hundreds of kilometers and contain several discrete hail cores. This complicates the quantitative interpretation of storm counts but makes the data processing more tractable. TMI 37.0-GHz and 85.5-GHz PCT are used for the empirical comparison with ground-based hail reports in Fig. 2. The footprint sizes were 16 km 3 9.7 km and 6.7 km 3 4.1 km, respectively, prior to the TRMM orbit boost in August 2001. TMI footprint sizes increased by about 10% after the boost. TRMM precesses through the diurnal cycle, so TMI is also used to assess diurnal variability in the tropics and subtropics in section 3d. Microwave brightness temperatures are dependent on 1) gaseous absorption and emission, primarily from water vapor and oxygen; 2) scattering from hydrometeors; and 3) surface emissivity. Changing the observation wavelength will change the observed upwelling TB via changes in these parameters. For two frequencies with similar values of absorption and scattering coefficients, observed TB will be similar and highly correlated. Since the frequency shifts between AMSR-E and TMI for the 36.5- and 37.0-GHz channels, respectively, are small, and not located near regions of high variability (e.g., absorption lines), we can reasonably calculate an expected TMI PCT from an AMSR-E PCT. If there are biases due to calibration, then this method will correct for that. The TB distribution will also be affected by differing footprint sizes, but we expect these differences to be small. We derived a TMI-equivalent PCT for the AMSR-E 36.5-GHz channel by generating histograms of observed PCT for both instruments for a 7-yr period. It was assumed that an AMSR-E 36.5-GHz PCT was equivalent to the TMI 37.0-GHz PCT having the same frequency of
15 JANUARY 2012
CECIL AND BLANKENSHIP
FIG. 4. Conversion of AMSR-E 36.5-GHz PCT to equivalent TMI 37.0-GHz PCT, using cumulative histogram matching method.
occurrence. To reduce diurnal, seasonal, and geographic sampling biases, observation counts from each instrument were scaled to identical values for each 58 3 58 box and time of day (2-h bins). The derived adjustment from AMSR-E to TMI is shown in Fig. 4. AMSR-E 36.5-GHz PCT is adjusted higher, with the magnitude of adjustment increasing for the lowest brightness temperatures. For example, AMSR-E values of 150 K are adjusted up to about 160 K, and measured AMSR-E values of 100 K are adjusted to almost 120 K. The PCT values are further scaled using Fig. 3, as mentioned in section 1. The regions used are shown in Fig. 3b. These scalings (Cecil 2011) were derived from TMI and TRMM precipitation radar (PR). As such, they do not incorporate any data from poleward of ;368. For the lack of a better approach, we extend the subtropical land and subtropical ocean scalings to high latitudes. We acknowledge that factors such as the lower surface temperature, lower freezing-level height, and shallower troposphere could lead to substantially different relationships between microwave TB and hail occurrence at high latitudes. As a sanity check, we compare our scaled global map of estimated hailstorm frequency to an unscaled map in section 3b. The empirical relationship between TMI PCT and large hail follows the approach of Cecil (2009), except using Storm Data reports of hail at least 1 in. (2.5 cm) in diameter in the southeast and south-central United States between 1998 and 2006. For each 10-K interval of PCT, both the number of precipitation features associated with hail and the total number of precipitation
691
features were counted. The hail probabilities were smoothed by summing those counts over 30-K intervals before dividing and then recording the smoothed probability in 10-K increments. These probabilities were extrapolated to 1-K increments and then further smoothed using a moving 8-K boxcar average. This smoothing reduces the appearance of noise in the relationship and makes the function monotonic (thin lines in Fig. 2) but artificially reduces the hail probabilities for the strongest storms and increases them for weaker storms. Each AMSR-E-derived storm was assigned a probability of having hail, based on its scaled minimum 37-GHz PCT and Fig. 2. Storms with minimum 37-GHz PCT greater than 200 K or 89-GHz PCT greater than 130 K were omitted. Their derived hail probabilities are artificially high because of the smoothing and limitations in the assignment of hail reports to precipitation features [see Cecil (2009) for details on the procedure]. The 130-K threshold at 89 GHz is also a useful secondary screen for surface snow cover—allowing higher 89-GHz PCT introduces artifacts over Antarctica and Greenland. The number of storms was counted for each 2.58 3 2.58 grid box, with individual storms being weighted between 0 (if 37-GHz PCT . 200 K) and 1 (if 37-GHz PCT , 105 K). The number of opportunities to see storms in each grid box was also recorded, by summing the fractional coverage of the grid box by AMSR-E in each Aqua overpass. A particular location is not necessarily sampled every day from Aqua’s low-Earth orbit, with high latitudes being sampled more often than low latitudes. We scale the storm counts as if every location is observed 1461 times (4 times per day for a year).
3. Satellite-derived severe hail climatology a. Comparison with ground reports in the United States As a first test of our approach, Figs. 5a,b compares the AMSR-E-derived hailstorm climatology in the United States to ground truth reports from Storm Data during a similar period. The region with hailstorms estimated by AMSR-E is very similar to the region with at least 100 Storm Data large hail reports per 500 kilometers squared per year—that is, from east of the Rockies to about the Great Lakes and mid-Atlantic states, and throughout the South. AMSR-E places the maximum from eastern Oklahoma to southern Iowa, while Storm Data reports place the maximum in Kansas. These similarities give confidence in the satellite-based approach. Doswell et al. (2005) show a maximum in south-central Oklahoma for the years 1980–84. They show a broader maximum from western Kansas to southeastern Oklahoma for 1995–99,
692
JOURNAL OF CLIMATE
VOLUME 25
FIG. 5. (a) Estimated number of hailstorms per 500 kilometers squared per year from AMSR-E from 2003 to 2010. (b) Number of Storm Data reports of 1 in. or larger hail per 500 kilometers squared per year from 2003 to 2009. Note that a single storm may have several hail reports. (c) Percentage of Storm Data large hail reports per hour (LST) for western United States (west of 1058), central United States, and eastern United States (east of 908). (d) As in (b), but for 1970–79; 2.58 grid boxes are used in (a),(b), and (d).
with a secondary maximum near Atlanta. Figure 5d shows that the 1970s had a maximum in Nebraska and eastern Kansas, with another maximum in northwest Texas. This, together with Fig. 5b, demonstrates that a record covering only a few years does not produce a complete picture of a longer-term hailstorm climatology. Short-term climate variability probably drives the shifting spatial patterns in the ground-based reports, but changes in local reporting practices (Doswell et al. 2005) could also contribute. Recent emphasis on severe thunderstorm warning verification has driven up the number of large hail reports (Doswell et al. 2005) by an order of magnitude over recent decades (compare Figs. 5b,d). The early afternoon and late night sampling of AMSR-E introduces a potential bias, since hailstorms more often occur in late afternoon and rarely in the morning. Differences in the phase or magnitude of the hailstorm diurnal cycle cause spatial variability for this bias. Using Storm Data reports to evaluate this diurnal cycle in the United States (Fig. 5c), both the eastern United States (generally east of the Mississippi River) and western United States (generally from the Rocky Mountains westward) have peaks between 1500 and 1700 LST. Hail reports in the central United States peak later (1700–1800 LST),
with a greater fraction occurring in the evening. A much smaller fraction occurs in early afternoon than in the rest of the country. This suggests that the AMSR-E estimates from the North American plains are biased low, by as much as ;50%, because of diurnal sampling. The rest of the United States would have a much smaller negative bias from this analysis.
b. Global climatology Applying this method to AMSR-E observations globally yields the estimated hailstorm frequencies in Fig. 6a. Some of the areas highlighted in prior studies are also shown here, but the satellite observations bring a uniformity of measurements that is lacking in ground-based climatologies. Northern Argentina and Paraguay have the most estimated hailstorms in the AMSR-E climatology, with other active regions including the central United States, Bangladesh, Pakistan, central and West Africa, far southeast Africa, and East Asia. Most of these regions have several days per year with environmental conditions favorable for severe thunderstorms, according to Brooks et al. (2003). There are some notable differences between our map of satellite-derived hailstorm locations and Brooks et al.’s map of severe weather environments.
15 JANUARY 2012
CECIL AND BLANKENSHIP
693
FIG. 6. (a) Hailstorm frequency of occurrence estimated from AMSR-E 36-GHz PCT, 2003–10. Units are storms per 500 kilometers squared per year, using 2.58 grid spacing and bilinear interpolation. (b) As in (a), but without scaling the tropical ocean, tropical land, and subtropical ocean brightness temperatures following Cecil (2011).
Brooks et al. have many favorable environments in Tibet and near the Gulf of Aden, where AMSR-E suggests few severe hailstorms. A climatology of Chinese hail observations does show a strong maximum on the Tibetan Plateau (Zhang et al. 2008), but that dataset does not distinguish hail size and is likely dominated by hail smaller than 2 cm in diameter (Xie et al. 2010). While the Gulf of Aden region has large thermodynamic instability, convective initiation is somewhat limited there. The local maximum near Bangladesh is more prominent in our analysis, and the local maxima in the subtropical Americas are located a bit farther west by Brooks et al. Some differences may result from using
different periods (1997–99 for Brooks et al.; 2003–10 in our study). Without applying the regional scalings to the brightness temperatures (from Cecil 2011), the estimated number of hailstorms in central and West Africa increases dramatically (Fig. 6b). Considering that there are not many ground-based reports of large hail from this region in the literature, we suspect that the scaled version (Fig. 6a) is more accurate. Frisby and Sansom (1967) indicate that many countries in this region do have some reports of hail reaching the ground but not often at most locations. The unscaled version (Fig. 6b) also includes a strong maximum over Colombia and Venezuela,
694
JOURNAL OF CLIMATE
VOLUME 25
FIG. 7. Bimonthly AMSR-E hail climatology for (a) January–February, (b) March–April, (c) May–June, (d) July–August, (e) September–October, and (f) November–December.
15 JANUARY 2012
CECIL AND BLANKENSHIP
FIG. 7. (Continued)
695
696
JOURNAL OF CLIMATE
and occasional hailstorms over broad regions of the oceans. Lightning measurements do confirm a thunderstorm maximum over northwestern South America, but we are unaware of evidence of many hailstorms in that region. The vast majority of potential hailstorms over the oceans are located within a few hundred kilometers of continents. A few are scattered across the open oceans, mostly between 6158 and 358 latitude in the Pacific and North Atlantic. Others are found over regions extending several hundred kilometers east from continents in the 6158–408 latitude belts (offshore from the United States, Brazil, Uruguay, Argentina, South Africa, Mozambique, Australia, and China). Very few hailstorms are estimated south of 408S, but several are scattered across high latitudes in Canada and Russia. The farthest poleward storm in our dataset is near Igarka, Russia, near 67.38N, 908E on the afternoon of 28 July 2002. Another Russian storm near 718N, 1198E produced a 223 K, 37-GHz PCT, failing to reach our 200-K threshold. The northernmost hailstorm we are aware of in the literature occurred near 68.58N in Finland (Tuovinen et al. 2009), with 4-cm-diameter hail denting a car in northern Lapland. We do not imply that locations with no hailstorms estimated by AMSR-E (e.g., Finland) are truly free of large hailstorms. Using a low-Earth-orbiting satellite with sampling in late night and early afternoon ensures a low probability of detection for hailstorms, so places with a low frequency of hailstorm occurrence are likely to go undetected. This should also be recalled when we describe storms as being rare at certain times of year in the next section.
c. Annual cycle Figure 7 shows the global distribution of satelliteinferred hailstorms on a 2.58 grid in bimonthly intervals. Figures 8–10 show annual and spring/summer bimonthly distributions for the active regions of South Asia, southeastern South America, and North America. As expected, the satellite-inferred hailstorms are most common in late spring and summer in particular locations. Tropical Africa has evidence of some hailstorms throughout the year, with the location of the maximum moving north and south following the noontime sun. This leads to seasonal peaks in Northern Hemisphere summer in West Africa and the Sahel (Fig. 7d), and peaks near the equinoxes in equatorial Africa (Figs. 7b,e). The seasonal peak for southeastern Africa is less pronounced (Figs. 7a,b,f), with storms over the offshore waters during much of the year. European activity generally peaks in summer (Fig. 7d), but storm counts are greater in autumn for the Mediterranean (Fig. 7e). This probably has to do with the Mediterranean Sea remaining warm while cooler air advances from Europe.
VOLUME 25
The Australian hailstorms inferred by AMSR-E are almost entirely during late spring and summer (Figs. 7f,a). The few occurring in early spring tend to be near the east coast (roughly from Brisbane to Sydney), consistent with the result of Schuster et al. (2005), that the hail season extends from October to February in New South Wales and peaks in November–December. The AMSR-Ederived storms in summer cover a broader area, with a peak in the northwest. Such details may be beyond the limits of our sample’s robustness. We do estimate that hailstorms occur over most of the continent, including the interior deserts. In East Asia, we estimate that springtime hailstorms are concentrated from Thailand to Beijing, with peak activity between Hanoi and Hong Kong (Figs. 7b,c). Frisby and Sansom (1967) list several stations in Thailand with hail reports, primarily in March and April. They describe hail observations as rare in Vietnam and Hong Kong. In summer, the AMSR-E-derived storms extend through all of eastern and northeastern China, and into far southeastern Russia. Peak concentration is in the lower Yellow River basin (Fig. 7d). Very few storms meet our criteria in this region after August. Xie et al. (2010) show a strong summer peak in large hail occurrence for northern parts of China, with a spring peak in southwestern China (Guizhou Province). Their data included hail sizes for only four regions in China, with only 6% of hail observations reaching the 2-cm diameter threshold they used to define ‘‘severe’’ hail. The percentage was about twice as large in Guizhou Province as in the northern regions, making it difficult to infer a climatology of large hail from the more general climatology of hail observations in China (Zhang et al. 2008). North of 408 across Europe and Asia, the hailstorms are inferred almost entirely during the summer months (Figs. 7c,d), except for the aforementioned autumn storms near the Mediterranean. The summer peak in hailstorm frequency at high latitudes is expected and consistent with ground-based studies (e.g., Webb et al. 2001 for Britain; Tuovinen et al. 2009 for Finland). Not much seasonality is seen from the few storms scattered across open ocean regions, well removed from land. There may be a slight preference for autumn (South Pacific in Figs. 7b,c; North Pacific in Fig. 7f) and winter (North Pacific in Fig. 7a). When vigorous extratropical troughs penetrate the subtropics and tropics in these seasons, they occasionally provide a favorable combination of dynamic support and thermodynamic instability over the relatively warm waters. The South Asian monsoon plays an important role in the distribution of hailstorms from Pakistan to east India. This region has two centers of peak activity (Fig. 8a), with the greatest concentrations in Bangladesh and
15 JANUARY 2012
CECIL AND BLANKENSHIP
697
FIG. 8. AMSR-E hail climatology near India for (a) annual, (b) March–April, (c) May–June, and (d) July–August. Light contours are elevation, contoured at 1-km intervals.
northern Pakistan. During March and April, hailstorm locations are tightly concentrated over Bangladesh and east India (Fig. 8b). In May and June (Fig. 8c), this region expands to include the northern Bay of Bengal and more of eastern India, including the coasts of Orissa and Andhra Pradesh. Hailstorms begin to occur near the foothills of the Himalayas in northern India and northern Pakistan in late June. By July and August (Fig. 8d), northern Pakistan has a large number of hailstorms. Occasional hailstorms are spread across southeastern Pakistan
and the northern half of India. Romatschke et al. (2010) showed a similar shift of deep convective cores from east in the premonsoon season to west during the monsoon. In most individual years in our dataset, hailstorm occurrence in and near Bangladesh begins very near 1 April. The season peaks around 20 May, but it varies by a few weeks from year to year. Storm counts dramatically decrease after about 20 June, presumably with the onset of the monsoon (we have not incorporated ancillary
698
JOURNAL OF CLIMATE
VOLUME 25
FIG. 9. AMSR-E hail climatology for southeastern South America for (a) annual, (b) September–October, (c) November–December, and (d) January–February. Light contours are elevation, contoured at 1-km intervals.
data to determine yearly monsoon onset). About 75% of the retrieved hailstorms in and near Bangladesh occur between 1 April and 20 June. Most years have a few hailstorms derived from AMSR-E data in September, but they are rare during July and August. Chaudhury and Banerjee (1983) show a sharp peak in April but with
very few hailstorms after May. Frisby and Sansom (1967) list about 90% of the hail days in this region in 1951–60 as occurring during March–May, with a few more in June, and a few more in August–October. Yamane et al. (2010) show a similar annual cycle for severe convective storms (not restricted to hailstorms) as ours, with 84% in the
15 JANUARY 2012
CECIL AND BLANKENSHIP
699
FIG. 10. As in Fig. 8, but for North America.
premonsoon season and a slight increase in September– October. With the shift of storm locations from east to west, the region near the foothills in northern India has more storms during June and July. Some years have a concentrated period of 1–3 weeks with several storms. For the entire dataset, about one-third of hailstorms in the region occur between 20 June and 9 July. Almost half occur between 20 June and 29 July. In the global, annual hailstorm climatology (Fig. 6a), an area in northern Pakistan (308–358N, 708–758E) has a higher concentration of storms than any place other than Bangladesh, the central United States, and southeastern South America. Storm locations show a preference for
the foothills, in an arc from southwest to north to southeast. A large number of hailstorms are restricted to a short period, usually from late June to mid-August (plus or minus a couple weeks from year to year). Only 1 out of 79 storms in the AMSR-E database for this region occurred earlier than 9 June. About 75% occurred between 20 June and 18 August. Houze et al. (2007) and Romatschke et al. (2010) describe the meteorological conditions as favoring deep, intense convection in this region: moist southwesterly flow from the Arabian Sea warms while crossing the desert and builds instability under a cap of warm, dry continental air advected from the Afghan plateau. It is lifted by the terrain, and the concave indentation of the mountain barrier may add low-level convergence.
700
JOURNAL OF CLIMATE
Southeastern South America has a broader region of frequent hailstorms, centered on northeastern Argentina and southern Paraguay (Fig. 9a). The peak in the concentration from AMSR-E is in the Chaco state of northern Argentina. Matsudo and Salio (2010) place the hail maximum farther southwest, based on reports at conventional surface weather stations in Argentina from 2000 to 2005. In early spring (Fig. 9b), the common storm locations according to AMSR-E center on southeastern Paraguay and extend into adjoining parts of southern Brazil and northeastern Argentina. Late spring and early summer (Fig. 9c) have the most hailstorms, and the most active region extends from Bolivia and southern Brazil southward across Paraguay, Uruguay, and northern Argentina to about Santa Rosa in the state of La Pampa. No particular location has as high a concentration of storms as premonsoon Bangladesh, but the South American active region covers a much broader area. Further into summer (Fig. 9d), the center of activity shifts a bit southwestward and fewer storms are found in Brazil. Storm counts decrease in autumn and winter, with locations shifting toward the eastern side of the La Plata basin (Uruguay to eastern Paraguay) by midwinter (Fig. 7d). In North America, winter (Fig. 7a) and early spring (Fig. 10b) hailstorms are most common over the Gulf of Mexico and Gulf Coast states. By late spring (Fig. 10c), the hailstorm region extends into southern Canada and covers most of the United States east of the Rocky Mountains and south or west of New England. The highest concentrations in our AMSR-E sample are from Oklahoma to Iowa. A broad maximum centered on Iowa covers the midwestern United States by summer (Fig. 10d). Storm counts decrease in autumn (Fig. 7e) and are mostly limited to the southeastern United States for late autumn and winter (Figs. 7f,a).
d. Diurnal cycle The TRMM satellite precesses through the diurnal cycle, so we use it for a cursory examination of the inferred hail diurnal cycle in the tropics and subtropics. We divide the TMI-derived hailstorm counts into 3-hourly bins (LST) in Fig. 11 for five geographical regions. Sample sizes require that we use rather large grid boxes for this. All five regions have minima between sunrise and 1200 LST and have maxima between 1500 and 2100 LST. The southeastern South American (broadly defined in this subsection as 188–388S, 728–528W) grouping actually has a broad peak between 1500 and 2400 LST, with a little more than half its storms during those 9 h. The other regions have a rapid decrease in storm counts after 9:00 p.m. A disproportionate share of the southeastern South American storms are horizontally extensive, mature MCSs, lasting well into the night.
VOLUME 25
FIG. 11. Percentage of TMI-estimated hail cases in 3-h increments of LST for south central/southeastern United States, southeast South America, central Africa, Pakistan, and Bangladesh.
Both Southeastern South America and Bangladesh (188–348N, 848–968E) have a greater fraction of their storms overnight than the other regions, leading to a better chance that AMSR-E’s late night overpass will catch a storm there. Bangladesh also has a greater fraction during very early afternoon than the other regions, so the AMSR-E estimates around 0130 and 1330 for Bangladesh could be biased high compared to the rest of the world. Ground-based hail reports from Chaudhury and Banerjee (1983) show a sharper afternoon peak for this region, between 1400 and 1800 LST. The diurnal cycle of Bangladesh’s severe thunderstorms from Yamane et al. (2010) is more consistent with ours, with a peak occurrence between 2000 and 2100 and several still occurring after 2400 LST. The diurnal curves for the United States (308–388N, 1088– 808W), central and West Africa (108S–188N, 208W–408E), and Pakistan (228–388N, 648–768E) are very similar to each other, with more than 50% of storms occurring between 1500 and 2100 LST and less than 10% between 0600 and 1200 LST. Pakistan has a smaller fraction during late afternoon than the United States and central and West Africa, and more between midnight and sunrise. To put the AMSR-E observations into a diurnal context, we also gridded and mapped the percentage of TMI-derived hailstorms occurring between 0000 and 0300 and between 1200 and 1500 LST (not shown). This analysis, along with Fig. 11, suggests that AMSR-E likely overestimates the number of hailstorms in the Bangladesh, Pakistan, and southeastern South American regions compared to the United States and central and West Africa. The TMI data and the ground-based storm reports in Fig. 5c suggest that AMSR-E’s diurnal sampling may underestimate the number of severe hailstorms in the
15 JANUARY 2012
CECIL AND BLANKENSHIP
central United States by ;20%–50%. That being said, a global (tropics and subtropics) climatology derived from TMI instead of AMSR-E also suggests Bangladesh and southeastern South America have more hailstorms per year than the other regions (equatorward of 388).
4. Discussion Considering how these results compare to prior knowledge of hail climatologies, two points deserve emphasis: 1) hailstone size distributions are dominated by small sizes, and many prior studies were not limited to the large (2.5-cm diameter or larger) sizes addressed in this study; and 2) the satellite-based measurements used here are ambiguous and do not directly measure hail reaching the surface. The Andes and other mountainous regions are prominent in the ground-based hail climatology in Fig. 1 but not in the satellite-based severe hail climatology in Fig. 6a. The small hail and graupel that fall in these elevated regions would melt before reaching the surface at lower elevations. Northeast Colorado is well known for hailstorms, with the National Hail Research Experiment (NHRE) conducted there in the 1970s (Foote and Knight 1979). Of the 28 days with NHRE hailpads recording hail in 1972–74, only 4 days had hailstone sizes 2.55 cm or larger (Crow et al. 1979). Besides storms with small hail that might melt before reaching the surface at lower elevations, this region certainly does have some storms with very large hail. The AMSR-E observations around 0200 and 1400 LST may have a more negative bias for the western high plains than for some other locations where late afternoon storms more often evolve into the night. South Africa is also prominent in the ground-based hail climatology (Fig. 1). Its Highveld region had a dense, long-lasting hail observing network. Small hail is very common there and a threat to crops, with hail observed on 38% of days during a 6-month hail season (Smith et al. 1998). Only 5% of those hail days reach the 31-mmdiameter criteria used by Smith et al. to define severe hail days. A large, long-lasting network also existed in central Alberta, but only 10% of the hail days featured severe hail (Smith et al. 1998). The satellite-based climatology (Fig. 6a) does indicate severe hailstorms in both these regions, but they are not among the most active regions. The large number of severe hailstorms indicated by satellite in tropical Africa is noteworthy, since very few reports of large hail there are found in the literature. It is plausible that large hail goes mostly undetected or unreported, with limited infrastructure for compiling and reporting such events. It is also plausible that the hail
701
signatures seen by satellite are produced by very large numbers of small hailstones (and large graupel) that melt before reaching the surface. The scaling in Fig. 3 is an attempt to account for such effects, as it reduces the hail counts in the tropics and over the oceans (compare Figs. 6a,b). We suspect the truth involves some combination of both of these plausible explanations. Elsewhere in the deep tropics and open oceans, similar arguments could be made as those for tropical Africa. The scaling from Fig. 3 is even more restrictive for these regions, making it almost impossible for a tropical oceanic storm to register as a severe hailstorm in this analysis. We suspect these estimates are a bit too conservative. The precise locations of the scattered storms over the oceans, deep tropics, and high latitudes should not be taken too seriously, as the grid spacing is too fine for regions with only a few events. Maps from TRMM data [not shown; see Fig. 3 from Zipser et al. (2006) for maps without the weightings applied in section 2] with more complete sampling between 6388 show storms scattered across the subtropical oceans (particularly the South Pacific convergence zone) but also add a few storms across the Amazon basin and Maritime Continent.
5. Conclusions The AMSR-E 36-GHz channel is used here to develop a global climatology of thunderstorms likely to produce large (at least 2.5 cm) hail. Studies relying on ground-based hail reports from individual regions cannot be merged into a cohesive global climatology because of differences in reporting standards, definitions, station density, spotter density, and record keeping. Our satellitebased approach also has uncertainties—it relies on empirical relationships between brightness temperatures and hail reports in the United States, which may not be directly applicable in other places; it uses a satellite that does not observe locations during the midafternoon to evening diurnal maximum for hail; and our record only covers 8 yr. The strength of this approach is that it applies a uniform measurement strategy globally, allowing quantitative comparison of storm counts from diverse regions. We believe this is a major step forward in characterizing the global distribution of severe hailstorms. The places with the most hailstorms estimated in our database (Fig. 6) are a broad region of northeastern Argentina and southern Paraguay and a small region in Bangladesh and eastern India. These are followed by the central United States, northern Pakistan and northwestern India, central Africa, West Africa, and southeastern Africa (and adjacent waters). All of these regions have been mentioned in prior hail studies but without much
702
JOURNAL OF CLIMATE
ability to quantitatively compare the regions. Compared to the locations listed above, we estimate fewer hailstorms in East Asia, Australia, Europe, tropical South America, Russia, Canada, the Middle East, and occasionally over the oceans between about 158 and 408 latitude. In most places, hailstorm occurrence peaks in late spring (Fig. 7). It extends to higher latitudes (e.g., Canada, Russia, northeastern China) in summer. The storms are more common near the equinox in equatorial Africa and in autumn near the Mediterranean. The South Asian monsoon plays a major role in the distribution of inferred hailstorms across the Indian subcontinent. Hailstorms typically begin to affect Bangladesh in the east around 1 April and become common in May. Their numbers in the east dramatically decrease in mid-June with monsoon onset. The northern India foothills of the Himalayas typically have a short period with hailstorms during a few weeks near the end of June and beginning of July. Farther northwest, the hailstorm season around northern Pakistan abruptly begins in late June and peaks in early July. About 75% of the Bangladesh hailstorms occur between 1 April and 20 June; about 75% of the Pakistan hailstorms occur between 20 June and 18 August. We use TRMM to evaluate the diurnal cycle in its tropical and subtropical domain. Further evaluation is warranted using multiple satellites to sample high latitudes at different times of day. All of the regions examined with TRMM have minimum hailstorm occurrence between sunrise and 1200 LST, followed by a sharp increase in the afternoon. The peak is between 1500 and 2100 LST in all regions, except that the peak extends a few hours later in southeastern South America. Southeastern South America and Bangladesh have a greater fraction of their hailstorms overnight than the other regions. The diurnal signals in TRMM data suggest that Bangladesh, Pakistan, and southeastern South America may be a bit oversampled by AMSR-E relative to the United States and central and West Africa. Data from both AMSR-E and TRMM agree on southeastern South America and Bangladesh having more hailstorms than any other location. Acknowledgments. This work was supported by NASA precipitation science program Grants NNX07AD73G and NNX10AG78G. AMSR-E data were downloaded from the National Snow and Ice Data Center. Hail reports were downloaded from the Storm Prediction Center. TRMM data in precipitation feature format were downloaded from the TRMM Science Data and Information System and from the University of Utah. We appreciate the comments of the four anonymous reviewers, whose suggestions improved the original manuscript.
VOLUME 25 REFERENCES
Ashcroft, P., and F. Wentz, 2006: AMSR-E/Aqua L2A global swath spatially-resampled brightness temperatures V002. National Snow and Ice Data Center, Boulder, CO, digital media. [Available online at http://nsidc.org/data/ae_l2a.html.] Barnes, G., 2001: Severe local storms in the tropics. Severe Convective Storms, Meteor. Monogr., No. 50, Amer. Meteor. Soc., 359–432. Brooks, H. E., J. W. Lee, and J. P. Craven, 2003: The spatial distribution of severe thunderstorm and tornado environments from global reanalysis data. Atmos. Res., 67 & 68, 73–94. Cecil, D. J., 2009: Passive microwave brightness temperatures as proxies for hailstorms. J. Appl. Meteor. Climatol., 48, 1281–1286. ——, 2011: Relating passive 37-GHz scattering to radar profiles in strong convection. J. Appl. Meteor. Climatol., 50, 233–240. Chaudhury, A., and A. K. Banerjee, 1983: A study of hailstorms over northeast India. Vayu Mandal, 13, 91–95. Crow, E. L., A. B. Long, J. E. Dye, and C. W. Ulbrich, 1979: Results of a randomized hail suppression experiment in northeast Colorado. Part III: Analysis of hailstone size distributions for seeding and yearly effects. J. Appl. Meteor., 18, 1559–1568. Dobur, J. C., 2005: A comparison of severe thunderstorm warning verification statistics and population density within the NWS Atlanta county warning area. Preprints, Fourth Annual Severe Storms Symp., Starkville, MS, East Mississippi Chapter National Weather Association and Amer. Meteor. Soc., D2-6. Doswell, C. A., III, H. E. Brooks, and M. P. Kay, 2005: Climatological estimates of daily nontornadic severe thunderstorm probability for the United States. Wea. Forecasting, 20, 577– 595. Foote, G. B., and C. A. Knight, 1979: Results of a randomized hail suppression experiment in northeast Colorado. Part I: Design and conduct of the experiment. J. Appl. Meteor., 18, 1526– 1537. Frisby, E. M., and H. W. Sansom, 1967: Hail incidence in the tropics. J. Appl. Meteor., 6, 339–354. Grody, N. C., 1991: Classification of snow cover and precipitation using the Special Sensor Microwave Imager. J. Geophys. Res., 96, 7423–7435. Harries, H., 1895: The frequency, size, and distribution of hail at sea. Quart. J. Roy. Meteor. Soc., 21, 230–244. Houze, R. A., Jr., D. C. Wilton, and B. F. Smull, 2007: Monsoon convection in the Himalayan region as seen by the TRMM precipitation radar. Quart. J. Roy. Meteor. Soc., 133, 1389–1411. Knight, C. A., and N. C. Knight, 2001: Hailstorms. Severe Convective Storms, Meteor. Monogr., No. 50, Amer. Meteor. Soc., 223–248. Matson, R. J., and A. W. Huggins, 1980: The direct measurement of the sizes, shapes, and kinematics of falling hailstones. J. Atmos. Sci., 37, 1107–1125. Matsudo, C. M., and P. V. Salio, 2010: Severe weather reports and proximity to deep convection over northern Argentina. Atmos. Res., 100, 523–537, doi:10.1016/j.atmosres.2010.11.004. Mohr, K. I., J. S. Famiglietti, and E. J. Zipser, 1999: The contribution to tropical rainfall with respect to convective system type, size, and intensity estimated from the 85-GHz ice-scattering signature. J. Appl. Meteor., 38, 596–606. Nesbitt, S. W., E. J. Zipser, and D. J. Cecil, 2000: A census of precipitation features in the tropics using TRMM: Radar, ice scattering, and lightning observations. J. Climate, 13, 4087–4106. Romatschke, U., S. Medina, and R. A. Houze, 2010: Regional, seasonal, and diurnal variations of extreme convection in the South Asian region. J. Climate, 23, 419–439.
15 JANUARY 2012
CECIL AND BLANKENSHIP
Roos, D. S., 1972: A giant hailstone from Kansas in free fall. J. Appl. Meteor., 11, 1008–1011. Russell, F. A. R., 1893: On Hail. Edward Stanford, 224 pp. Schuster, S. S., R. J. Blong, and M. S. Speer, 2005: A hail climatology of the greater Sydney area and New South Wales, Australia. Int. J. Climatol., 25, 1633–1650. Smith, S. B., G. W. Reuter, and M. K. Yau, 1998: The episodic occurrence of hail in central Alberta and the Highveld of South Africa. Atmos.–Ocean, 36, 169–178. Spencer, R. W., M. R. Howland, and D. A. Santek, 1987: Severe storm identification with satellite microwave radiometry: An initial investigation with Nimbus-7 SMMR data. J. Climate Appl. Meteor., 26, 749–754. Tuovinen, J.-P., A.-J. Punkka, J. Rauhala, and H. Hohti, 2009: Climatology of severe hail in Finland: 1930–2006. Mon. Wea. Rev., 137, 2238–2249.
703
Webb, J., D. M. Elsom, and D. J. Reynolds, 2001: Climatology of severe hailstorms in Great Britain. Atmos. Res., 56, 291–308. Williams, L., 1973: Hail and its distribution. Studies of the Army Aviation (V/STOL Environment), Army Engineer Topographic Laboratories Rep. 8, ETL-SR-73-3, 27 pp. Xie, B., Q. Zhang, and Y. Wang, 2010: Observed characteristics of hail size in four regions in China during 1980–2005. J. Climate, 23, 4973–4982. Yamane, Y., T. Hayashi, A. M. Dewan, and F. Akter, 2010: Severe local convective storms in Bangladesh: Part I. Climatology. Atmos. Res., 95, 400–406. Zhang, C., Q. Zhang, and Y. Wang, 2008: Climatology of hail in China: 1961–2005. J. Appl. Meteor. Climatol., 47, 795–804. Zipser, E. J., D. J. Cecil, C. Liu, S. W. Nesbitt, and D. P. Yorty, 2006: Where are the most intense thunderstorms on Earth? Bull. Amer. Meteor. Soc., 87, 1057–1071.
