Assessment of the Quality of Thunderstorm Data at First-Order Stations (2024)

Audibility problems

It is certain that changes in audibility due to varying atmospheric conditions have an affect on the number of thunderstorms heard at a location during any given month or year. The audibility of thunder has been set at various distances. Fleagle (1949) concluded that thunder seldom could be heard beyond 24 km. Brooks (1925) gave a maximum distance of 16–19 km under “the most favorable atmospheric conditions.” In a massive study of thunderstorms, Byers and Braham (1949) indicated that the typical distance for thunder audibility was 12–18 km. These distances, varying from 8 to 24 km, reveal that varying atmospheric conditions allow the sound of thunder to be heard over widely different distances. Thus, on some days when more distant thunderstorms occurred, their occurrence has undoubtedly been missed. For example, on a given afternoon, atmospheric conditions could limit transmission of the sound of thunder to 8 km, and actual thunderstorms that were 10–12 km away at their closest position to the station would go unreported.

The atmosphere affects the propagation of sound in several ways—included are temperature levels, inversions, sharp horizontal gradients, turbulent eddies, humidity levels, and the density of the atmosphere. The reported audibility differences for thunder result from variations in the atmospheric conditions that vary randomly over both time and space, and thus do not produce a systematic temporal difference at any one place (weather station). However, the relocation of a weather station to a different site with different surrounding structures or physiographic features, or the development of major structural changes at a station site, could alter the “local” audibility. Regardless, the uncertainty over atmospheric audibility means that the historical frequency of thunder days measured anywhere is an underestimate of the true regional frequency at a station. These underestimates have been assessed using lightning data (Changnon 1993).

Various localized conditions and station shifts can act to affect the audibility and thus recording of a thunder day, and there are four primary factors. First are actions that are nonatmospheric and that could affect the audibility of thunder and thus influence the local records of thunder days over time. These include local noise from aircraft operations since the 1940s, which is suspected to limit audibility at stations located near busy airports (Court and Griffiths 1981). Another shift that potentially has reduced audibility includes the advent of air conditioning at weather stations during the 1950s–60s, and the resulting closure of office windows in summer (the prime season for thunderstorms) could have reduced audibility. The relocation of a weather station that changes the physical features near the station, such as nearby hills, an adjacent wall of mountains, or tall buildings, can affect audibility in certain quadrants around a station. Court and Griffiths (1981) state that many western stations located in valleys miss nearby thunder events because the valley walls, or mountain ranges, limit the audibility. Station elevation is a fourth factor and western FOS at high elevations experience more thunderstorms than nearby stations in valleys (Court and Griffiths 1981). They conclude that the average number of storm days increase by 20 for every kilometer in elevation, and they ascribed the greater storm frequencies at higher elevations to measurements at less noisy sites and to less dense atmospheric conditions allowing better audibility.

Urban effects on storm activity

A large city influences the atmosphere and can affect storm frequency. This change will affect thunder-day values at weather stations located in the area where urban effects have systematically changed over time the frequency of thunderstorms. The need to discern urban effects on local storm frequency is just as important in climate change assessments as determing how much an urban heat island affects local temperatures so that the data can either be deleted or the values modified to adjust for local influences (Karl et al. 1988).

At cities with populations of one million or more, such storm changes have been found east, or downwind, of the city (Huff and Changnon 1973), and in larger cities such as New York (Bornstein and Leroy 1990) and Chicago (Changnon 2001), thunderstorm incidences have been altered over the city itself. Figure 1 presents the summer (June–August) average thunder-day pattern for the St. Louis area and shows major increases amounting to 45% over and just east of the city (Changnon 1978).

Shifts of stations into different climatic locales

A third factor that can affect thunder records relates to moves of stations from in-city locales to rural airports. These movements can alter the frequency of storms reported if the areas of the different locations are in a sharp climate gradient of storm activity due to physiographic changes or nearness to large water bodies that affect storm incidence. Figure 2 shows the sizable changes in the average thunder-day pattern for the autumn season downwind of Lakes Superior and Michigan, revealing sharp differences in storm incidences over short distances (Changnon 1968). A station moved 10–20 kilometers inland from a shoreline site would be sampling a different frequency of storm days. Further, cities such as Denver, which is located in the lee of the Rocky Mountains and in an area where storms frequently develop, could have very different incidences of storms depending on the distance of the weather station from the front range. For example, the weather station at Colorado Springs, which is located only 90 km south of Denver but closer to the mountains, averages 21 more thunder days a year than Denver.

Other possible factors

The thunder-day definition is based on an international agreement among national weather services and simply depends on hearing thunder between midnight and midnight. In the central United States, certain atmospheric conditions of the warm season favor the development of nocturnal thunderstorm (Means 1944; Wallace 1975). Such nocturnal storms have the potential for creating a double count in thunder days if a storm begins before midnight and lasts into the early morning, leading to an increase in the number of thunder days not found at FOS with fewer nocturnal storms (Court and Griffiths 1981). This possibility is more of a problem for spatial analysis of storms than for temporal analysis because such events spanning midnight should occur as often in 1900 as in 1990. There is no evidence that the nocturnal storm activity has undergone any systematic change during the past 100 years.

The possibility also exists that certain personnel at some weather stations have been less or more attentive to recording thunder than those at other times. Such possibilities exist but these conditions are not recorded so there is no way of measuring their impact on the actual storm count. Also the use of pilot reports to define a thunder day, a situation that developed with commercial aviation in the 1940s, could also affect local values. Any major vagaries due to record keeping should be detectable using the data evaluation approach described in the following section.

Unusual shifts in thunder-day frequencies

Each station’s historical values, based on 5-yr values, were expressed as percent of their long-term averages, allowing comparisons before and after relocations, plus comparisons with similar values of adjacent stations. These percentages were plotted on graphs and all major moves of each station, as extracted from the station’s history, were plotted by dates on the graphed historical record.

Figure 3 shows the 5-yr thunder-day percentage of average values of Williston, North Dakota, and dates of its station relocations. There were two major moves of the Williston station, one in 1961 and another in 1981. The Williston values are shown along with those at two other nearby stations, Bismarck and Fargo (Fig. 3). The three stations’ 5-yr values matched well from 1901 to 1955, with Williston averaging 27 thunder days per year, but thereafter the Williston values decreased dramatically and remained different, either lower or higher than the other two stations, through 1995. The Williston station’s relocation during 1961 appears to be related to the major shift in the station’s values. The Williston values also became extremely high after the 1981 relocation. The Williston values also failed the hom*ogeneity tests.

Figure 4 presents the 95-yr values for Amarillo, Texas, which had major station moves in 1941 and in 1974, along with values for two nearby stations, Abilene and Oklahoma City. From 1901 to 1940, the Amarillo values were consistently lower than those of the other two stations, averaging 37 thunder days per year, and in 1941–45 the Amarillo percentages increased and became the highest of the three values for the ensuing 20 yr, averaging 52 thunder days annually. The 1941 relocation appears to be related to this major shift. The other move of the Amarillo station in 1974 was also followed by values much higher than those at the other stations.

Figure 5 presents the values at Birmingham, Alabama, a station that had major relocations in 1945 and 1972, and values for nearby stations at Montgomery and Mobile. The Birmingham and Montgomery values reveal they were relatively similar from 1901 through 1945, with Birmingham averaging 70 thunder days per year, but thereafter Birmingham’s values decreased and remained low (averaging 51 thunder days annually) in the succeeding 20 yr. This shift was likely related to the station relocation in 1945. After the relocation in 1972, the Birmingham values for succeeding 5-yr periods (Fig. 5) were in better agreement with those of the other two stations.

The stations found to have such major anomalous shifts in thunder-day frequencies related to station moves or to other unknown influences are listed in Table 1. The station records considered to be questionable because they failed the tests of hom*ogeneity and had extreme shifts in their thunder-day frequencies included Amarillo, Atlanta, Birmingham, Kansas City, Denver, Sheridan, and Williston. Three other stations shown in Table 1 also failed the tests of hom*ogeneity, but had some periods of data considered to be useful but only before or after a questionable frequency shift. For example, the Havre data for 1901–80 are considered to be of good quality, but values after 1980 are poor. The Lander values for 1901–30 are too low, but those after 1930 appear to be representative. The early Grand Island values were questionable, but after 1940 appeared to be satisfactory. However, none of these 10 stations have data considered to be useful for long-term temporal assessments of thunder days.

Urban influences on local thunderstorm activity

Past studies that defined urban influences on storm incidence found that all had occurred at metropolitan areas with populations exceeding one million. Based on this information, a list of stations with the population potential for urban effects was created, as shown in Table 2. Results of studies of urban effects at specific cities (Bornstein and Leroy 1990; Bornstein and Lin 1999; Changnon 1978; Harnack and Landsberg 1975; Huff and Changnon 1973; Westcott 1995) were used to identify cities with known effects on storms, and these are also indicated by asterisks in Table 2.

In all studies of urban influences on storm activity, the altered thunderstorm frequencies were noted to occur in the urban area and/or downwind (east) of the urban area. These locations are important in relation to where the official local weather station was located and when the urban influences developed. For example, a study of eight cities (Huff and Changnon 1973) showed an increase in thunderstorm activity over New Orleans labeled “in city” that began in the 1950s, but the official in-city station site had been shifted to an airport 14 miles west of the city in 1948. Hence, the official New Orleans thunder-day record at the airport did not contain urban influences on recorded storm activity. Hence, New Orleans, as shown in Table 2, is defined as unlikely to have urban influences in its thunder-day records.

Table 3 lists those cities where prior studies found statistically significant increases in thunder days in the city and/or downwind (east) of the city. The eight cities were assessed as to where their thunder-day records had been collected and when the urban influence had begun. Chicago’s influence dates back to the 1930s and clearly has urban influences, given that the local station then and until 1958 was located within the city (Changnon 2001). Kansas City data were collected in the city until the station moved in 1972 to northwest of the city. Hence, portions of the historical record are considered to be urban influenced. The St. Louis in-city observations were moved to the local airport northwest of the urban center in 1948, and the urban influence began in the 1930s (Changnon 1978). This station’s historical thunder-day record contains urban-influenced storm counts. For example, the airport station, as shown in Fig. 1, averaged 24 storms per summer as opposed to 15–18 in areas surrounding the metropolitan area. New York’s record in the city (La Guardia Airport) is clearly urban influenced with storm decreases in certain weather conditions (Bornstein and Leroy 1990). The Dallas–Fort Worth record, with the weather station between the two cities, has also likely been urban influenced. Cleveland’s increase in storms began in the 1950s, but the station had been moved from the city southwest to the airport in 1941. Thus, the Cleveland thunder record is judged to be without urban influences. The Milwaukee record with downwind influences east of the city is also without urban effects, because the station was moved far south of the city in 1941. The Minneapolis–St. Paul effects on storm activity have existed since about 1970, but the in-city station was moved southwest of the city in 1938, and the thunder record likely contains no urban effects. Thus, three of the eight stations with significant urban-related increases in thunder days (Table 3) did not experience effects on their official thunder records, which were taken beyond the area of local change in storm activity.

Other cities were found to have measurable increases in storm activity, but the values were not statistically significant. These included Houston (city and downwind influences), New Orleans (city only), Atlanta (downwind influences), and Washington, District of Columbia (city and downwind influences). Houston and New Orleans, because of timing of airport shifts of observations, are without urban effects in their official records, but Washington, likely has an urban influence on its records (Harnack and Landsberg 1975). Atlanta has experienced urban effects on storm activity (Bornstein and Lin 1999), and the Atlanta records had also been found to contain inhom*ogeneities related to a station relocation (Table 1). Other cities assessed and found to be without local increases in the thunderstorm records included Detroit, Cincinnati, Oklahoma City, Indianapolis, Tulsa, Omaha, and Wichita. In all these cities, the official weather stations had been located south, west, or north of the city and were free from any urban-related shifts in storm activity.

Other major cities with sizes adequate to have possible urban influences included Baltimore, Boston, Los Angeles, Philadelphia, and San Francisco. Comparison of the Baltimore and Philadelphia records with those of nearby Harrisburg did not indicate a relative increase at either major city during the twentieth century. Both cities had station (observation) moves to distant rural airports south of their locations (Baltimore in 1950 and Philadelphia in 1940) that further preclude any marked urban influence. The two California cities did not exhibit any localized storm increases during the twentieth century, relative to other stations. The data for Boston are unclear. The station has been in the city and the adjacent airport since 1900, and may have urban as well as oceanic influences on storm activity.

In conclusion, thunder-day records at eight cities were considered to be urban influenced and are not recommended for use in long-term assessments of regional-scale storm activity. Those stations considered to have thunder-day activity records with urban-influenced values are Atlanta, Boston, Chicago, Dallas–Fort Worth, Kansas City, New York, St. Louis, and Washington.

Inadequate sampling of thunder events

A comparative study using thunder event data at the 130 candidate first-order stations and cloud-to-ground lightning data at varying ranges around each station was made to assess further possible data problems. Lightning data available for the 1986–89 period were employed. This involved documenting all occurrences of nearby lightning that matched conditions of thunder events (i.e., six or more cloud-to-ground strokes evenly distributed over a 1-h period), and occurring when no thunder was recorded. Such cases were defined as “missed” thunder events. The 4-yr frequency of the missed thunder events was compared with the actual number of recorded thunder events at each station, and the two values were summed representing all possible events. The number of actual events recorded was expressed as a percent of all possible thunder events during 1986–89. The resulting percentages for cases when lightning occurred within 15 km of each station are shown in Fig. 6. The values vary from 45% to 60% in the southeast and increase northward and westward, reaching 90% along the Canadian border.

Changery (1981) computed the average number of“thunderstorms,” based on the number of thunder events, for all FOS based on the 1948–77 period. If these station storm values are divided by the average number of events per storm day from Changnon (1993), one can estimate the number of missed thunder days. For example, at Tampa, Florida, the average number of thunder events is 128 and dividing this by the average number of events per day (1.3) reveals an average of 99 thunder days per year. The Tampa records, however, indicate the long-term average is 88 days; thus, it appears that 11 thunder days have been missed annually, on average, representing 10% of the correct total. Underestimates of thunder days elsewhere in the nation ranged from 6% to 15%.

The individual station values of missed thunder events, as defined by the cloud-to-ground lightning frequencies, were assessed through comparison with the values of surrounding stations. The frequency of missed thunder events are the inverse of those shown on Fig. 6. Inspection of the 130 station values revealed that four stations had unrealistically low values, defined as more than 10% less than surrounding values. These stations are listed in Table 4. Data from these four stations, although questionable for only a 4-yr period, are not recommended for use in studies of long-term thunder activity.

APPENDIX  List of First-Order Stations with Quality Thunder Data during the 1896-1995 Period

Table 1.

Stations with major shifts in thunder-day frequencies during 1896–1995 and that failed tests of hom*ogeneity with surrounding stations

Table 2.

Assessment of stations at cities of sufficient size to have potential urban effects on thunderstorms. The official weather station’s location with respect to the city’s center is shown. Assessment of potential effects on local records (column 3) was also based on a comparison of the station locations with where storm activity is typically altered. Changed storm activity in cities assessed as having urban effects has occurred downwind of cities (east quadrant) and over the city in the largest metropolitan areas such as New York and Chicago

Table 3.

Cities with statistically significant increases in thunder days, based on urban thunderstorm studies

Table 4.

Stations assessed as having too many missed thunder events based on comparisons with nearby lightning occurrences