alexa Land Surface Heterogeneity and Tornado Formation: A Comparison of Tornado Alley and Dixie Alley | Open Access Journals
ISSN : 2332-2594
Journal of Climatology & Weather Forecasting
Make the best use of Scientific Research and information from our 700+ peer reviewed, Open Access Journals that operates with the help of 50,000+ Editorial Board Members and esteemed reviewers and 1000+ Scientific associations in Medical, Clinical, Pharmaceutical, Engineering, Technology and Management Fields.
Meet Inspiring Speakers and Experts at our 3000+ Global Conferenceseries Events with over 600+ Conferences, 1200+ Symposiums and 1200+ Workshops on
Medical, Pharma, Engineering, Science, Technology and Business

Land Surface Heterogeneity and Tornado Formation: A Comparison of Tornado Alley and Dixie Alley

Frazier AE1*, Hemingway B1 and Brasher J2

1Department of Geography, Oklahoma State University, Stillwater, OK, 74074, USA

2Department of Geography, University of Tennessee, Knoxville, TN, 37996, USA

*Corresponding Author:
Frazier AE
Oklahoma State University
Stillwater, OK, 74074, USA
Tel: (405) 744-2864
E-mail: [email protected]

Received Date: March 30, 2017; Accepted Date: April 24, 2017; Published Date: May 03, 2017

Citation: Frazier AE, Hemingway B, Brasher J (2017) Land Surface Heterogeneity and Tornado Formation: A Comparison of Tornado Alley and Dixie Alley. J Climatol Weather Forecasting 5:203. doi: 10.4172/2332-2594.1000203

Copyright: © 2017 Frazier AE, et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Visit for more related articles at Journal of Climatology & Weather Forecasting


Tornadoes are among the most destructive hazards to human life and property, and certain areas of the United States are more prone to these events. In particular, scientists use the terms “Tornado Alley” and “Dixie Alley” to refer to two general areas that experience higher incidences of tornadoes. While there is wide recognition that the two regions vary in the number, magnitude, and fatalities caused by tornadoes, more research is needed to better understand the reasons for these differences. The growing recognition that land surface heterogeneity may play a role in tornado formation provides motivation to compare the geographical characteristics of the two regions to determine whether there are significant differences in the landscape characteristics where severe storms form. To investigate the relationship between tornado formation and land surface heterogeneity in these two regions, we first delineate the spatial extent of Tornado Alley and Dixie Alley based on tornadic activity using a statistical test for the detection of significant clusters of spatial association. Next, using severe weather data for tornadoes and storms producing wind/hail (but no tornadoes), we investigate how land surface heterogeneity factors are related to tornado formation of weakly tornadic storms (EF0-EF1) and significantly tornadic storms (EF2-EF5) in each region using binary logistic regression. Lastly, we map probability surfaces for each region to show areas of greater risk. Results show that relationships between land surface heterogeneity and tornado formation vary from region to region. Elevation, slope, and distance to rivers are significant predictors of tornado formation, but the directionality of those relationships varies by region and storm severity. Urban land covers are associated with decreased tornado probability for all storm types in both regions. Spatial trends show an decreasing likelihood for EF0-EF1 from west to east in Tornado Alley but an increasing likelihood in that direction for EF2-EF5 storms.


Severe weather; Environment; Terrain; Tornado touchdown; GIS


Tornadoes are among the most destructive natural hazards to human life and property. In the United States, approximately 1,253 tornadoes occur every year (NCDC) causing an average of 50 fatalities annually and $400 million in yearly economic losses [1-3]. Most tornadoes occur in the central part of the United States in a region known as Tornado Alley [4]. While the exact geographic extent of Tornado Alley is debated in the literature [5], it is often loosely defined as encompassing northern Texas, Oklahoma, Kansas, and Nebraska. This region is not the only tornado prone area in the country though. A second area of high incidence, termed Dixie Alley [5,6], stretches across the Delta region of the United States and is generally considered to encompass Louisiana, Mississippi, and Alabama [7].

While the exact boundaries of Tornado Alley and Dixie Alley have been differentially defined in the literature [5], there are observable differences in the number of violent tornadoes affecting the two regions, both in terms of the seasonality of tornado occurrences as well as the time of day when tornadoes are most likely to occur [5]. For example, Gagan et al. [5] found that a greater number of stronger tornadoes form in Tornado Alley from April through October, while Dixie Alley experiences more stronger tornadoes during the remaining months. Tornado Alley also experiences the majority of its strong tornadoes during the afternoon and early evening hours, whereas Dixie Alley experiences more tornadoes during the late night and early morning hours. These diurnal differences may also affect differences in the number of tornado fatalities, with hazard-related deaths concentrated in the southeastern United States [8,9]. While some research has focused on understanding the differences between the two regions in terms of their distinctive tornadic characteristics [5], more research is needed to investigate the relationships between environmental factors and meteorological processes [10] to understand why these geographic variations occur, with the ultimate goal of improving forecasting and warning.

A growing body of research recognizes the relationship between land surface heterogeneity and the evolution of mesoscale convective systems [11-18]. Land surface heterogeneity refers to the variation in terrain, vegetation, and land cover characteristics that combine to produce different environmental conditions (e.g., soil moisture, urban heat island, etc.), which in turn influence storm formation. For example, in a study comparing tornado frequencies in urban and rural counties, Aguirre et al. [19] suggest that terrain roughness (i.e., land surface heterogeneity measured through land cover) may explain higher frequencies of tornadoes in urban areas. More recently, Pielke et al. [15] document how land cover changes (e.g., conversion of forests to cropland) have altered surface climate fluxes, and Pielke et al. [18] describe how regional weather patterns are a function of land cover. In a study explicitly testing the role of land surface heterogeneity on tornado climatology, Kellner and Niyogi [17] found spatial relationships between tornado formation locations and surface roughness measured through the change in elevation and land cover in Indiana. These initial studies suggest that differences in land surface heterogeneity between Tornado Alley and Dixie Alley may help explain some of the geographical variation in tornado formation. While it remains extremely difficult to predict exactly where a tornado will form [20], a better understanding of the relationship between landscape surface heterogeneity and tornado formation from climate data remains an important factor in reducing risk [4,7].

The objective of this study is to investigate the relationship between tornado formation and land surface heterogeneity in Tornado Alley and Dixie Alley. We first delineate the spatial extent of both Tornado Alley and Dixie Alley using an established statistical test for the detection of significant clusters of spatial association. This extent delineation is not meant to supplant pre-existing conceptualizations but rather provide a spatially explicit area from which to undertake a systematic comparison of the two regions. Next, using severe weather data for tornadoes and storms producing wind/hail (but no tornadoes), we investigate how land surface heterogeneity factors are related to tornado formation in each region through binary logistic regression. Lastly, we map our results from both regions to visually compare the spatial trends of tornado formation probability in each region. Our ultimate goal is to derive useful information that can aid in identifying vulnerable geographic locations and ultimately inform risk assessment, warning dissemination, and infrastructure (e.g., tornado warning systems) development.


Tornado data

The National Weather Service (NWS) Storm Prediction Center maintains a database of severe weather reports for tornado producing storms as well as non-tornado producing wind/hail events, which is available in GIS format. The tornado database contains path geometry for all tornadoes reported since 1950 along with attributes including date, time, and magnitude. Point locations of tornado formation were generated by isolating the starting vertex of each tornado path. The NWS also maintains a database of severe storms that produce wind/ hail (but not tornadoes) dating back to 1955, which includes similar attributes for the date and time of the storm. To ensure there was no duplication between the two datasets, we queried the storm databases according to the date and time and removed any duplicate points in both datasets. Only storms reported during the 15-year period from 2001-2015 were included in the study to align with the land cover variables (discussed below). We divided the database of tornadoes by storm type into weakly tornadic supercells (EF0 and EF1) and significantly tornadic supercells (EF2 or greater) based on other studies using a similar typology [21-24]. For this study, we compare both the weakly tornadic and the significantly tornadic super cells to wind/hail storms that did not produce any tornadic activity.

Land surface heterogeneity data

Topographic variables: Three topographic/terrain variables are analyzed here: elevation, slope, and aspect. Elevation and slope have previously been included in tornado land surface heterogeneity studies [17], but the role of aspect (i.e., slope direction) is not typically considered. Since tornadoes typically have path directions oriented in a southwest to northeast direction [25], we hypothesize that slope direction may be related to tornado formation. Topography variables were derived from a digital elevation model (DEM) obtained from the United States Geologic Survey (USGS) 3D Elevation Program (3DEP) at a spatial resolution of 1 arc second. The DEM was resampled to 90 m spatial resolution for improved processing time. Since there is some uncertainty in the exact location of tornado formation, this resampling is not expected to have a large impact on overall results. Slope and aspect were computed directly from the DEM using the suite of spatial analysis tools in ArcGIS. Slope is computed as the maximum rate of change (i.e., steepest downhill direction) and is reported in degrees. Aspect is the direction of maximum slope measured clockwise from 0 degrees (due north) to 360 degrees (also due north) with flat areas given a value of -1. Aspect data were reclassified into five categories: Flat, North (315 -45 degrees), East (45-135 degrees), South (135-225 degrees), and West (225-315 degrees).

Distance to major rivers: The channelling of wind through valleys may cause a jet of increased wind speed, thereby supporting tornadogenesis [26,27], and we hypothesize that tornado formation probabilities will increase closer to major rivers. Distance to river was computed using the National Hydrography Dataset (NHD), a linear vector shape file, to generate a raster surface using Euclidean distance computations where each cell in the contiguous United States represents the closest distance to a major river. The spatial resolution of the raster was generated at 90 m to match the topographic datasets.

Land cover: Early tornado studies [28] noted the importance of ground cover of tornado paths, and recent studies have continued to link land surface conditions with tornado formation [17,29-31]. Land cover investigations present a challenge for tornado-related research since detailed land cover maps have only become available in recent decades. Datasets produced by the Multi-Resolution Land Cover Consortium (MRLC) prior to 2001 are not compatible for direct comparison with more recent products, but by limiting our study to 2001-2015, we are able to utilize the three most recent, compatible NLCD datasets from 2001, 2006, and 2011 to assign land cover to storm locations. We assigned all storms occurring between 2001-2005 land cover values based on the 2001 version of the NLCD, storms occurring between 2006-2011 values from the 2006 version, and storms spanning 2011-2015 values from the 2011 version.

While it is possible that land covers changed within a single time window, using the multiple NLCD versions is preferred over assuming the current NLCD database represents historical land cover. NLCD land cover categories were reclassified into eight aggregated classes: Water, Developed, Barren, Forest, Shrub land, Grassland, Agriculture, and Wetlands.

Final data preparation

To prepare the data for logistic regression (discussed below) the EF0-EF1 tornadic storms were combined with the wind/hail storms into a single database where tornadic storms were coded 1 and wind/ hail storms were coded 0.

The same process was repeated for the EF2-EF5 storms. All five of the land surface heterogeneity variables (elevation, slope, aspect, distance to river, and land cover) were then extracted to each of the storm points in each set using ArcGIS. The final datasets for analysis included each storm coded as a 1 or 0, depending on whether the storm produced a tornado, and several fields of data indicating the land surface characteristics at the geographical location of the storm.


Delineation of Tornado Alley and Dixie Alley

There are many different criteria and methods that can be used to define Tornado Alley versus Dixie Alley [4], and researchers have debated different boundary delineations following different statistical methods [4,6,32,33]. Our intent here is not to add to existing discourse but rather delineate spatially explicit boundaries for each region that permit systematic statistical comparisons between the two regions. Any definitive areas referenced here should be considered the authors’ own interpretation based on the statistical analyses performed in this paper with recognition that there are several interpretations of the exact spatial extent of Tornado Alley and Dixie Alley.

We use an established statistical test for the detection of clusters of significant spatial association to delineate boundaries for Tornado Alley and Dixie Alley. Cluster detection tests are used to identify “hot spots” of activities [34,35] through local statistics that measure and test for spatial association of a variable (e.g., tornadoes) within a geographic neighbourhood (e.g., counties). Hot spots emerge in areas where neighboring values are unusually high [35], and cold spots occur in areas where values are unusually low compared to surrounding areas. Local spatial statistics are widely used to test hypotheses of clustering to determine whether there is a raised incidence of a phenomenon in an area or location of interest, but to our knowledge they have not yet been applied to tornado events in the United States.

The Getis-Ord generalized (local) Gi* statistic [35,36] allows for analysis of local pockets of increased or decreased incidences and is computed as:

image (1)

Where {wij(d)} is a symmetric one/zero spatial weights matrix with ones for all links defined as being within distance d of a given target region, i. All other links are set to zero. In the standardized version of the statistic used here, the target region i; included in the computation of the statistic. Therefore, Wij ≠ 0. The variables x and s are the sample mean and standard deviation of the observed set of xi, respectively. Gi* will produce high values and a high positive z-score when there is a dominant pattern of high values near other high values and will produce low values when there is clustering of low values [37]. One benefit of using the Gi* statistic over other commonly used measures, such as Moran’s I [38], is that it can find both “hot” and “cold” spots of tornadic activity.

Binary logistic regression

Binary logistic regression is used here to analyze the impact of each land surface heterogeneity variable on tornado formation. Logistic regression uses a nonlinear function to explain the probability of tornado formation where the dependent variable is, whether or not a severe storm formed a tornado (1 or 0 for the database), and this outcome is influenced by a vector of five land surface heterogeneity variables. The dependent variable takes the value 1 if the storm produced a tornado (EF0-EF1 or EF2-EF5, depending on the analysis) and 0 if the storm was severe but produced only wind/hail and did not produce a tornado. In logistic regression, the statistical significance of the coefficients indicates whether the corresponding explanatory variable is significantly related to the dependent variable. However, it must be noted that the coefficients cannot be directly compared to each other as a measure of relative importance to explain probability of demolition. Instead, the odds ratio can be interpreted in the context of the impact of each variable on the probability of tornado formation [39]. When the odds ratio for a particular variable is greater than 1, an increase in the variable of one unit will increase the odds of tornado formation by the amount of the odds ratio. When the odds ratio for a variable is less than one, an increase in the explanatory variable by one unit leads to a decrease in the odds of tornado formation. If the odds ratio is exactly one, the odds of tornado formation do not change as that particular variable changes. The exponent of a coefficient is the odds ratio; therefore the probability of tornado formation can be determined through Eqn. (2):

image (2)

Logistic regression has been used extensively for predicting hazards [40,41] due to its ability to identify the degree of influence of all independent variables. It has also recently been used to investigate the influence of the El Nino/Southern Oscillation on tornado and hail frequency in the U.S. [42]. Here, we use the binary case of logistic regression to investigate the influence of land surface heterogeneity on tornado formation. Since logistic regression investigates the impact of one unit of change in the independent variable on the probability of encountering the dependent variable, we scaled elevation by 100 so that we can assess the impact of every 100 m change in elevation. We scaled the distance to river variable by 1,000 to create similarly meaningful units for interpretation, and we rounded slope to the nearest degree so every degree equals a one unit change. Additionally, the categorical variables (aspect and land cover) must have a reference category selected for comparison. In this case, we compared all land covers to cultivated land and all aspects to east-facing slopes.

Results and Discussion

The hot spot results from the Gi* cluster analysis show a large cluster of highly significant (z-score>1.96) counties stretching from Denver across eastern Colorado, north into Wyoming, through Nebraska, Kansas, Oklahoma, Arkansas, and down into northern Texas. This cluster, which we define as “Tornado Alley”, is visually separate from the clusters of high values in Illinois and Iowa/South Dakota/Minnesota, but is less separated from the cluster of high values along the Gulf Coast, supporting research that has described the area of primary activity between the two regions as connected [6,21].

However, in order to compare “Tornado Alley” to “Dixie Alley”, we sever the two regions at the Mississippi River using the state boundary of Arkansas. While our aim is only to compare Tornado Alley and Dixie Alley, it is interesting to note the locations of clusters of high tornadic activity in Illinois, Iowa, North Dakota/Northwestern Minnesota, Florida, and the Carolina coast. Our analysis did not indicate clusters of high activity in Indiana as other studies have found. It should be noted though that this statistical analysis highlights counties with high tornadic activity that are surrounded by other counties that also have high activity (i.e., clusters of tornadic events). Certain counties with large numbers of tornadoes may not be included in a cluster if they are not surrounded by other counties with high activity (Figures 1 and 2).


Figure 1: Getis-Ord Gi* hotspot results (z-scores) for tornado touchdown locations with Tornado Alley and Dixie Alley delineated based on 95% confidence bounds.


Figure 2: Tornado formation probability surfaces based on logistic regression coefficients for Tornado Alley and Dixie Alley. Green areas indicate relatively lower probabilities while orange and red areas indicate relatively higher probabilities.

With explicit boundaries for Tornado Alley and Dixie Alley, we next analyzed the number and type of storm in each region. There were 100,421 total storms recorded in Tornado Alley during the study period (2001-2015), and 28,906 total storms recorded in Dixie Alley (Table 1). Of those storms, 94.8% were non-tornadic (only wind/hail) in Tornado Alley and 92.1% were non-tornadic in Dixie Alley. While the area of Tornado Alley is about 2.7 times larger than the size of Dixie Alley (based on our delineation procedure [1,058,126 km2 vs. 400,149 km2]), Tornado Alley received approximately 3.5 times more severe storms than Dixie Alley over the study period, so based on area alone, Tornado Alley is receiving a larger number of severe storms. The remaining 5.2% of storms occurring in Tornado Alley and 7.9% in Dixie Alley were tornadic. Most of the tornadic storms were weakly tornadic (90.5% in Tornado Alley and 86.4% in Dixie Alley), while only 9.5% and 13.6% were significantly tornadic (EF2-5) in the two regions, respectively. When comparing the number of significantly tornadic events, the number in Tornado Alley (4,711) was about 2.4 times greater than the number in Dixie Alley (1,975), which roughly corresponds to their proportional sizes. So, while Tornado Alley receives proportionally more total storms that Dixie Alley, the two regions receive roughly equivalent EF2-EF5 storms according to their respective sizes.

  Total No. Storms Wind/Hail (% Total Storms) All Tornadoes (% Total Storms) EF0-EF1 (% of Tornadoes) EF2-EF5 (% of Tornadoes)
Tornado Alley 100,421 95,213 (94.8%) 5,208 (5.2%) 4,711 (90.5%) 497 (9.5%)
Dixie Alley 28,906 26,620 (92.1%) 2,286 (7.9%) 1,975 (86.4%) 311 (13.6%)

Table 1: Observation totals for Tornado Alley and Dixie Alley (2001-2015).

Logistic regression results: weakly tornadic supercells

For the weakly tornadic supercells (EF0-EF1), elevation, distance to rivers, and urban, forest, and wetland land covers were all significant predictors in both regions. In Tornado Alley, a one unit (100 m) increase in elevation was accompanied by an increase in the likelihood of tornado formation (p<0.001), as evidenced by the odds ratio value of 1.018 (Table 2). Recall, odds ratio values above one indicate increases in likelihood of occurrence while values below one indicate decreases in likelihood. In Dixie Alley, a one unit increase in elevation decreased the odds ratio (0.742), and thus indicates a significant decrease in the likelihood of tornado formation (p<0.001). The finding that weakly tornadic storms are more likely to occur in higher elevations in Tornado Alley but lower elevations in Dixie Alley shows a difference in the effects of land surface heterogeneity on tornado formation across regions, supporting the need for local/regional analyses and comparisons.

Variable Tornado Alley Dixie Alley
β Coef. Odds Ratio Std. Error β Coef. Odds Ratio Std. Error
Intercept -2.787 0.062*** 0.044 -1.958 0.141*** 0.073
Elevation 0.018 1.018*** 0.004 -0.299 0.742*** 0.038
Slope 0.016 1.016* 0.008 -0.006 0.994 0.015
Distance to River -0.009 0.991. 0.005 0.035 1.035** 0.013
Flat 0.194 1.214 0.279 0.532 1.703. 0.306
East -0.048 0.953 0.040 -0.075 0.928 0.065
North 0.005 1.005 0.046 -0.006 0.994 0.070
South -0.107 0.898* 0.045 -0.015 0.985 0.065
Land Cover
Water -0.146 0.864 0.162 -0.029 0.971 0.179
Urban -1.491 0.225*** 0.056 -0.790 0.454*** 0.068
Barren 0.295 1.344 0.303 0.238 1.268 0.408
Forest -0.141 0.868* 0.059 -0.222 0.801*** 0.064
Shrubland -0.003 0.994 0.070 -0.197 0.821* 0.098
Grassland -0.045 0.965 0.038 -0.075 0.927 0.164
Wetland -0.121 0.886* 0.058 -0.264 0.768* 0.134

Table 2: Logistic regression results for weakly tornadic supercells (EF0- EF1) in Tornado Alley versus Dixie Alley.

Distance to river was also a significant variable in both regions with the directionality of the relationship again reversed. In Tornado Alley, as the distance from a major river increased (one unit increase equals 1000 m), the probability of tornado formation decreased (p<0.1), which supports our hypothesis that tornadoes are more likely to form in riverbeds where winds can be channelled. However, in Dixie Alley, moving away from a major river increased the likelihood of tornado occurrence (p<0.01).

This disparity can possibly be explained by the topographical differences between the two regions. The rivers in Tornado Alley are located in relatively flatter areas compared to Dixie Alley, and so it is possible the significance of this relationship may indirectly be related to terrain with flatter areas being the primary contributing factor rather than a channelling of winds through river valleys. However, this is an area in need of further investigation.

For land cover, urban land, forests, and wetlands all significantly decreased the likelihood of tornado formation in both regions. The extremely low odds ratios for urban land cover in both regions (0.225 and 0.454, respectively, p<0.001) are significant because many prior studies have suggested that tornado frequencies may be greater around urban areas [17,19]. Urban and developed land covers are known to increase the direct heating of the lower atmosphere [15] and have been linked to enhanced thunderstorm activity [43].

However, our analysis indicated no increase in tornado formation for weakly tornadic supercells in urban areas compared to other land covers. These results may partially be explained by Niyogi et al. [44] who found that storms may diverge, or split, when passing over urban areas, which may lead to fewer relative tornado touchdown events in the actual urban area. Early research on the impact of surface heterogeneity also found that hotter areas over cities suppressed development of tornadoes in certain areas of London [45,46], which may help explain the significantly decreased probabilities of tornado formation in urban areas.

Forested land covers were negatively related to tornado formation in both regions (Tornado Alley: p<0.05; Dixie Alley, p<0.001). Forests have a darker albedo and increased surface roughness compared to other land covers and therefore reflects less electromagnetic radiation. This reduction in reflectance results in a higher atmospheric boundary layer and higher sensible heat flux (lower latent heat flux) over wooded areas compared to cultivated lands [47]. The higher these conditions, the less likely tornadoes are to form, which would result in decreased likelihood of tornado formation over forested areas. Our results corroborate these findings with forested areas decreasing the likelihood of tornado formation in both regions.

Lastly, wetlands were associated with a significant decrease in storms for both regions (p<0.05), and south-facing slopes were associated with a significant decrease in Tornado Alley only.

Overall, the results of the analysis for weakly tornadic supercells (EF0-EF1) highlight several key findings. First, the relationships between land surface heterogeneity and tornado formation varied across the two regions. Specifically, elevation and distance to rivers were significant predictors in both regions, but the directionality of the relationships was reversed in Tornado Alley versus Dixie Alley. Knowledge that such differences exist may help target future forecasting and warning tools according to specific geographic criteria. Second, the low odds ratios for both urban and forest land covers suggest that these areas are not at greater risk for tornado formation compared to cultivated areas.

Logistic regression results: significantly tornadic storms

For the EF2-EF5 storms, elevation remained highly significant in Tornado Alley (p<0.001), but the direction of the relationship changed from that of the EF0-EF1 tornadoes. An increase in elevation decreased the probability of EF2-EF5 tornado formation in Tornado Alley (Table 3). Highest elevations in Tornado Alley are located along the western edge of the region in the Colorado plateau.

Variable Tornado Alley Dixie Alley
β Coef. Odds Ratio Std. Error β Coef. Odds Ratio Std. Error
Intercept -4.548 0.011*** 0.135 -4.070 0.017*** 0.184
Elevation -0.105 0.900*** 0.015 -0.088 0.916 0.085
Slope 0.047 1.048* 0.020 0.069 1.071** 0.027
Distance to River -0.012 0.988 0.025 -0.221 0.802* 0.096
Flat 0.911 2.488 0.717 -12.238 0.000 253.0
East -0.080 0.923 0.122 -0.177 0.838 0.158
North -0.079 0.924 0.139 0.152 1.164 0.160
South -0.017 0.984 0.129 -0.140 0.869 0.161
Land Cover
Water -0.607 0.545 0.556 0.498 1.645 0.405
Urban -1.147 0.318*** 0.157 -0.834 0.434*** 0.178
Barren 0.185 1.203 1.009 -12.408 0.000 345.0
Forest 0.231 1.259 0.149 -0.128 0.880 0.157
Shrubland 0.120 1.127 0.228 0.021 1.021 0.222
Grassland 0.053 1.054 0.125 0.196 1.216 0.359
Wetland 0.395 1.485* 0.168 0.155 1.168 0.308

Table 3: Logistic regression results for significantly tornadic super cells (EF2-EF5).

This area experiences many weakly tornadic super cells and not as many significantly tornadic super cells, which is likely driving this relationship. Elevation was not a significant predictor of EF2-EF5 storms in Dixie Alley.

Increases in slope slightly increased the probability of strong tornadoes (EF2-EF5) in both regions. Logic suggests that flatter areas would be more prone to tornado formation and in their study of tornadoes in Indiana, Kellner and Niyogi [17] found that topographic changes over short distances did not strongly correlate with tornado formation. However, in this study, we found a significant relationship between tornado formation and slope for weaker storms in Tornado Alley (Table 2) and for stronger storms in both Tornado and Dixie Alleys (Table 3). In all cases, an increase in slope signalled an increase in the odds of formation. These regional differences further support the need for targeted investigations in specific geographic areas as relationships between land surface heterogeneity and tornado formation do not appear to be constant across space.

Distance to river remained a significant predictor in Dixie Alley (p<0.05) with the probability of tornado formation decreasing as distance from river increased. In Dixie Alley, rougher terrain supports the formation of valleys through which wind may be channelled, possibly leading to increased probabilities of significant tornado formation closer to rivers. In Tornado Alley, where the terrain is much less variable, distance to river was insignificant. Lastly, urban land cover was again a highly significant predictor of tornado formation in both regions with urban lands signaling a decrease in the probability of tornado formation.

Several relationships that were not significant in the model are nonetheless interesting for understanding differences between Tornado Alley and Dixie Alley. First, our hypothesis that flat areas would experience an increased probability of tornadoes was true in Tornado Alley (odds ratio=2.488) but not in Dixie Alley. The high standard error in Dixie Alley (253.0) suggests that the mean is not reliable, which may be due to a lack of truly flat areas in this region as there were only 55 total instances of storms in flat areas (out of 28,906 total storms). The low numbers issue is also likely causing the high standard error for the Barren land cover in Dixie Alley where there were only 102 instances (Table 3). Water also showed contrasting (but not significant) relationships between the two regions with an odds ratio below one in Tornado Alley (decrease in probability) and above one in Dixie Alley (increase in probability). This dichotomy may be due to the presence of more water in Dixie Alley compared to Tornado Alley.

Probability maps

Tornado probabilities were computed for each pixel using the β coefficients computed from Eqn. 2 and presented in Tables 2 and 3. We mapped the resulting probability surfaces for the two categories of storms (EF0-EF1 and EF2-EF5) across the two study areas (Figure 2). Several interesting patterns are apparent. In Tornado Alley, weakly tornadic storms (Figure 2a) have the highest probabilities of forming in the western part of the region, with probabilities decreasing from west to east.

However, for the stronger storms (Figure 2b), probabilities of formation are highest in the eastern part of the region, notably in Arkansas and southeastern Oklahoma. The reverse patterns for the weak versus strong storms in Tornado Alley appear to be strongly influenced by the elevation gradient. For Dixie Alley, the patterns are not as stark as in Tornado Alley, but there are clear differences in the spatial distribution of probabilities for weaker storms versus stronger storms. The weaker storms (Figure 2c) show a slightly higher probability of forming in the western portion of the region, with several small pockets of very high probabilities along the Gulf Coast (areas in red). The stronger storms (Figure 2d) show a slight gradient of increasing probability from west to east but no strong trends.


This study has several limitations that warrant consideration. First, some uncertainty and error exist in the spatial locations of tornado data provided by the NWS. The 90 m spatial resolution of the four terrain variables (elevation, slope, aspect, and distance to river) minimizes some of this uncertainty, but the land cover datasets from MRLC are generated at a nominal 30 m spatial resolution, and since land cover is categorical, it cannot easily be aggregated to coarser resolutions. However, moderately high spatial and temporal resolution satellite imagery have been available since well before the start date of this study, and these ancillary datasets are frequently used to verify touchdown locations, which can increase positional accuracy.

Second, the Getis-Ord statistical test for the detection of clusters identifies counties in the United States with high tornado activity surrounded by other counties with high activity. Sampling biases can occur when the probability of observing phenomena depends on the shape and size of the geographic area (counties). These biases were reduced by ensuring that every county was analyzed with at least one neighbour, and larger counties primarily occur in western states where tornado activity is reduced. Edge effects can also occur for areas that do not have physical neighbours (e.g., counties along the coast), but since most of the tornado activity in the United States occurs inland [48], edge effects are limited.

Lastly, the Fujita and Enhanced Fujita scales used to assign magnitude values to tornadoes are damage-rating systems, not intensity-rating systems [49]. While damage and intensity are highly correlated, there can be differences in magnitude depending on the environment through which the tornado passed. For example, tornadoes in urban areas are more likely to obtain higher F-scale (or EF-scale) ratings than rural areas because the potential for damage is greater [50]. Aside from the lowest magnitude tornadoes, we did not find significant differences between different magnitudes for the environmental and land cover variables, so this discrepancy is not likely impacting our analysis disproportionately.


In this study, we investigated the relationship between tornado formation and land surface heterogeneity in Tornado Alley and Dixie Alley using binary logistic regression and mapped the results of that analysis to probability surfaces of tornado formation. Our key findings are highlighted below:

Relationships between land surface heterogeneity and tornado formation vary from region to region.

Increases in elevation increase the likelihood of EF0-EF1 storms but decrease the likelihood of EF2-EF5 storms in Tornado Alley. Increases in elevation increase the likelihood of EF0-EF1 storms in Dixie Alley but have no significant relationship with EF2-EF5 storms in that region.

Increases in distance to rivers (i.e., being further from a river) increase the likelihood of EF0-EF1 storms decreases in Tornado Alley and Dixie Alley but decrease the likelihood of EF2-EF5 storms in Dixie Alley. There was no significant relationship between distance to major rivers and EF2-EF5 storms in Tornado Alley.

Increases in slope (i.e., steeper slopes) slightly increased the probability of EF0-EF1 storms in Tornado Alley and EF2-EF5 storms in both regions but did not have a significant impact on EF0-EF1 storms in Dixie Alley.

Urban land was associated with a significant decrease in the likelihood of tornado formation for all tornadic storms (EF0-EF1 and EF2-EF5) in both regions.

Forest land cover was associated with a significant decrease in the likelihood of tornado formation for EF0-EF1 storms in both regions but was not a significant predictor of significantly EF2-EF5 storms in either region.

Wetlands were associated with a decrease in the likelihood of EF0- EF1 storms in both regions but an increase in EF2-EF5 storms in Tornado Alley.

Aspect does not appear to have a strong relationship with tornado formation.

The results of this study are not intended to pinpoint locations where tornadoes will form but are meant to aid scientists in developing more accurate warning systems with increased information to reduce false alarms and provide adequate lead times for sheltering. In the future, more accurate information on the relationships between tornado formation and land surface heterogeneity may help researchers transition from the current ‘warn on detection’ model to an improved ‘warn to forecast’ system, ultimately leading to reduced fatalities.


This research was supported by a National Science Foundation (NSF) EPSCoR Track 2 Research Improvement Infrastructure grant 1539070 for CLOUD-MAP: Collaborative Leading Operational UAS Development for Meteorology and Atmospheric Physics. Some of the computing for this project was performed at the OSU High Performance Computing Center at Oklahoma State University supported in part through NSF grant OCI–1126330.


Select your language of interest to view the total content in your interested language
Post your comment

Share This Article

Relevant Topics

Recommended Conferences

Article Usage

  • Total views: 163
  • [From(publication date):
    August-2017 - Sep 23, 2017]
  • Breakdown by view type
  • HTML page views : 134
  • PDF downloads :29

Post your comment

captcha   Reload  Can't read the image? click here to refresh

Peer Reviewed Journals
Make the best use of Scientific Research and information from our 700 + peer reviewed, Open Access Journals
International Conferences 2017-18
Meet Inspiring Speakers and Experts at our 3000+ Global Annual Meetings

Contact Us

Agri, Food, Aqua and Veterinary Science Journals

Dr. Krish

[email protected]

1-702-714-7001 Extn: 9040

Clinical and Biochemistry Journals

Datta A

[email protected]

1-702-714-7001Extn: 9037

Business & Management Journals


[email protected]

1-702-714-7001Extn: 9042

Chemical Engineering and Chemistry Journals

Gabriel Shaw

[email protected]

1-702-714-7001 Extn: 9040

Earth & Environmental Sciences

Katie Wilson

[email protected]

1-702-714-7001Extn: 9042

Engineering Journals

James Franklin

[email protected]

1-702-714-7001Extn: 9042

General Science and Health care Journals

Andrea Jason

[email protected]

1-702-714-7001Extn: 9043

Genetics and Molecular Biology Journals

Anna Melissa

[email protected]

1-702-714-7001 Extn: 9006

Immunology & Microbiology Journals

David Gorantl

[email protected]

1-702-714-7001Extn: 9014

Informatics Journals

Stephanie Skinner

[email protected]

1-702-714-7001Extn: 9039

Material Sciences Journals

Rachle Green

[email protected]

1-702-714-7001Extn: 9039

Mathematics and Physics Journals

Jim Willison

[email protected]

1-702-714-7001 Extn: 9042

Medical Journals

Nimmi Anna

[email protected]

1-702-714-7001 Extn: 9038

Neuroscience & Psychology Journals

Nathan T

[email protected]

1-702-714-7001Extn: 9041

Pharmaceutical Sciences Journals

John Behannon

[email protected]

1-702-714-7001Extn: 9007

Social & Political Science Journals

Steve Harry

[email protected]

1-702-714-7001 Extn: 9042

© 2008-2017 OMICS International - Open Access Publisher. Best viewed in Mozilla Firefox | Google Chrome | Above IE 7.0 version