Volume 8, Issue 1 p. 18-31
Original Research
Open Access

Bioscatter transport by tropical cyclones: insights from 10 years in the Atlantic basin

Matthew S. Van Den Broeke

Corresponding Author

Matthew S. Van Den Broeke

Department of Earth and Atmospheric Sciences, University of Nebraska-Lincoln, 126 Bessey Hall, Lincoln, Nebraska, 68588 USA


Matthew S. Van Den Broeke, Department of Earth and Atmospheric Sciences, 126 Bessey Hall, University of Nebraska-Lincoln, Lincoln, NE 68588. Tel: 402-472-2663; Fax: 402-472-4917; E-mail: [email protected]

Search for more papers by this author
First published: 09 July 2021
Citations: 1


Tropical cyclones (TCs) can transport birds and insects near their center of circulation. In this study, we examined the maximum altitude, area and density of the radar-derived bioscatter signature across a set of 42 TC centers of circulation sampled from 2011 to 2020. All TC events contained at least one time when a bioscatter signature was present. More intense hurricanes with closed eyes typically had taller and denser bioscatter signatures, and sometimes larger areas dominated by bioscatter. This indicated a larger number of organisms within the circulation of more intense hurricanes, supporting the speculation that those storms were most likely to trap birds that do not want to risk flying through their eyewall thunderstorms. Larger and denser bioscatter signatures, indicating a larger number of birds, tend to occur when fall migration brings a large bird population to the Gulf and East Coasts where most storms were sampled. TC formation location was not related to bioscatter characteristics, but storms sampled in the Gulf of Mexico and Florida tended to have larger and denser bioscatter signatures.


Since the late 1990s, data from the Weather Surveillance Radar-1988 Doppler (WSR-88D) network have been used to monitor the presence, movement and density of biological targets in the airspace above the contiguous United States (e.g. Farnsworth et al., 2004; Gauthreaux & Belser, 1998; Russell et al., 1998). Since the WSR-88D network was upgraded to polarimetric capability from 2011 to 2013, a new set of products allows more certain distinction of biological targets (e.g. Park et al., 2009; Stepanian et al., 2016; Van Den Broeke, 2013). The primary challenge of using radar to monitor biological targets is the frequent lack of ground truth and the so-far relatively weak ability to distinguish specific groupings of organisms (e.g. Gauthreaux & Diehl, 2020). Regardless, radar remains one of the best options to study movement and density of airborne organisms on large spatial and temporal scales.

Recent methodological advances have allowed the quantification of biological signatures in radar data. Dokter et al. (2011) and Chilson et al. (2012) present a method to relate radar reflectivity factor (ZHH) to the number of scatterers present, assuming a reasonable estimate of scatterer radar cross section. This method has been applied, for example, by Stepanian and Wainwright (2018) to estimate changes through the WSR-88D record in the population and phenology of Brazilian free-tailed bats (Tadarida brasiliensis) at a known bat cave. Similar radar methods can be used to quantify bird migration on continental scales (e.g. Dokter et al., 2018; Lin et al., 2019) and may be combined with citizen science data for insight on which migratory communities may be present (Horton et al., 2019). Often the radar cross section of targets is not known, for example, if scatterer identity is unknown or if scatterers are mixed species. When a value for the radar cross section cannot be reasonably assigned, analysis can instead focus on the total bioscatter density (cm2 of biological cross section per km3 of air). This approach has been applied, for example, by Hansen et al. (2020) and Van Den Broeke and Gunkel (2021). Additional progress in this area may be provided by more advanced modeling (e.g. Stepanian et al., 2018), linking biological and atmospheric models to forecast bird movement (e.g. Van Doren & Horton, 2018) and machine learning methods (e.g. Gauthreaux & Diehl, 2020; Lin et al., 2019).

While some seabirds are known to avoid tropical cyclone centers of circulation (TCCC) (e.g. Weimerskirch & Prudor, 2019), transport of organisms by TCs particularly at their centers of circulation, has been known in the literature for nearly 140 years (e.g. Garriott, 1902; General Weather Service of the United States, 1882; Young, 1921; also see Van Den Broeke, 2013 and references therein). This includes transport of birds (e.g. Coronas, 1925; Hurd, 1933; Parry, 1930) and insects (e.g. Freeman, 2003; Herring, 1958; Tannehill, 1936). The advent of polarimetric capability allows the differentiation of rain and biological targets in TCs, and bioscatter has been observed near their centers of circulation (e.g. Van Den Broeke, 2013). Including these initial observations, radar data have now been collected over 10 Atlantic Basin TC seasons. These datasets, including 42 separate data collections from 33 unique Atlantic TCs, allow us to address the following key questions regarding biological transport by TCCC:
  1. How often are biological signatures present in association with TCCC, and what are their typical maximum altitude, area and density?
  2. How are biological signatures in TCs related to time of year and to geography, including formation location, landfall location and TC track region?
  3. How are biological signatures in TCs related to select characteristics of the TC, including intensity, size of the wind field and inner core structure?

These questions can now be addressed more quantitatively since new methodologies are available for estimating quantitative aspects of the biological signature, expanding our understanding of how organisms respond to TCs and leading to a preliminary understanding of how avian population dynamics may be affected by TCs via transport by TCCC.

Materials and Methods

Radar is a valuable tool for observing the atmosphere and organisms therein over large spatial and temporal scales. Many countries have operational radar networks which allow nearly continuous monitoring, such as the WSR-88D network in the United States. Variables obtained from these networks typically include ZHH, a measure of scatter received from targets in the airspace; radial velocity, a measure of target motion toward or away from the radar; and several newer polarimetric variables that can be used to increase confidence that targets are biological. In this study, we examined TCs that came within 140 km (base scan beam height ~2.75 km assuming standard beam propagation; e.g. Brown et al., 2005) of a polarimetric WSR-88D. Given when the radar network was upgraded to polarimetric capability, this included storms which approached the US coast or Puerto Rico from 2011 to 2020. Pacific TCs were not included. Storms were set aside for analysis if they contained a well-defined base scan center of circulation. Some weak systems did not contain a clear center of circulation and were discarded. A clear center of circulation was required since bioscatter has been closely associated with the TCCC in prior observations (Van Den Broeke, 2013) and it is hypothesized that organisms are preferentially carried along with a TC at its relatively calm center. This sorting retained 42 datasets from 33 separate TCs (Table 1); several TCs were sampled by multiple radars at the required range. Two datasets abruptly ended when the TC destroyed the radar [Maria (2017) and Laura (2020)].

Table 1. Tropical cyclone datasets analyzed in this study.
Year Storm Date Radar Analysis period (UTC)
2011 Irene 27 Aug KMHX 0800-1748
2012 Beryl 28 May KJAX 0004-1116
Sandy 29-30 Oct KDIX 2201-0130
2013 Andrea 7 Jun KCLX 0701-0748
2014 Arthur 4 Jul KMHX 0002-0526
2015 Ana 10 May KLTX 0200-0839
Bill 16 Jun KCRP 1801-1859
Erika 28 Aug TJUA 0121-0328
2016 Bonnie 31 May KLTX 2102-2222
Julia 13 Sept KMLB 0710-1357
Julia 14 Sept KJAX 0025-0324
Matthew 7 Oct KMLB 0632-1320
Matthew 8 Oct KCLX 0801-1411
Matthew 8 Oct KLTX 1752-2230
2017 Cindy 22 Jun KLCH 0554-0826
Emily 31 Jul KTBW 1116-1456
Harvey 25-26 Aug KCRP 1403-1150
Irma 6-7 Sept TJUA 2041-0005
Irma 10 Sept KBYX 0657-1213
Maria 20 Sept TJUA 0625-0950
Nate 8 Oct KMOB 0404-0938
2018 Alberto 28 May KEVX 1734-2127
Alberto 29 May KMXX 0801-0945
Florence 14 Sept KMHX 0002-1214
Gordon 5 Sept KMOB 0140-0259
Michael 10 Oct KEVX 1506-2003
2019 Barry 13 Jul KLCH 1446-1654
Dorian 5-6 Sept KLTX 2213-0520
2020 Cristobal 7-8 Jun KLIX 2300-0140
Fay 10 Jul KDIX 2031-2234
Hanna 25-26 Jul KCRP 1352-0121
Hanna 25-26 Jul KBRO 1859-0339
Isaias 2 Aug KAMX 0730-0847
Isaias 2-3 Aug KMLB 1850-0043
Isaias 4 Aug KLTX 0116-0516
Laura 27 Aug KLCH 0158-0554
Sally 16 Sept KMOB 0703-1047
Beta 21-22 Sept KCRP 2114-0627
Delta 10 Oct KPOE 0151-0611
Zeta 28-29 Oct KLIX 2039-2338
Eta 9 Nov KAMX 0203-0608
Eta 9 Nov KBYX 0415-0917
Bioscatter was identified in MATLAB (version following prior studies (e.g. Park et al., 2009; Van Den Broeke, 2013; Van Den Broeke & Gunkel, 2021). Bioscatter voxels were required to have non-negligible ZHH and correlation coefficient (ρhv) <0.80. Once MATLAB identified an area of low ρhv, I manually checked the same area to ensure that differential reflectivity (ZDR) generally >2 dB was collocated. This ZDR check could not be readily automated because of the large noise present in ZDR estimates at low ρhv. Having a ZDR check was valuable, for example, to remove ground clutter when some storms in this study approached or moved over land. Given the relatively rare sampling of TC intensity and relatively slow speed of motion, temporally evenly spaced radar scans were selected every 10–15 min throughout the analysis periods indicated in Table 1. For these radar scans, the domain was modified to include the TC center of circulation, readily identified by the passage of the zero isodop, surrounding curved precipitation bands or TC eyewall and relative lack of precipitation. Figure 1 includes examples of this domain for several TCs in varying stages of organization, including reflectivity (left column) and radial velocity (right column, indicating the zero isodop). Bioscatter may also be present outside of this area (e.g. between the outer rainbands) and may even be transported by the TC while there, but we did not consider such transport here. Within the TC center of circulation, several variables were calculated including:
  1. Maximum altitude (km) to which the bioscatter signature could be identified, which was the maximum altitude at which the signature was apparent in the highest elevation scan displaying the signature.
  2. Bioscatter area (km2) from the 500-m CAPPI after the ρhv and ZDR filtering described above.
  3. Bioscatter signature density from the 500-m CAPPI (cm2 km−3; Van Den Broeke & Gunkel, 2021).
Details are in the caption following the image
Examples of TC centers of circulation with varying levels of organization. (A and B) A closed circulation [Hurricane Eta, 9 November 2020 at 0611 UTC, viewed from the Key West, Florida, WSR-88D (KBYX)]; (C and D) a mostly closed circulation [Hurricane Irene, 27 August 2011 at 1204 UTC, viewed from the Morehead City, North Carolina, WSR-88D (KMHX)]; (E and F) a partly open circulation [Tropical Storm Bonnie, 31 May 2016 at 2202 UTC, viewed from the Wilmington, North Carolina, WSR-88D (KLTX)]; (G and H) a mostly open circulation (Hurricane Matthew, 8 October 2016 at 1932 UTC, viewed from KLTX); (I and J) an open circulation [Hurricane Delta, 10 October 2020 at 0500 UTC, viewed from the Fort Polk, Louisiana, WSR-88D (KPOE)]. The left column is radar reflectivity (dBZ) and the right column is radial velocity (kt). In each panel, the white rectangle denotes the center of circulation.

These assume that all the reflectivity from ‘bioscatter voxels’ is actually bioscatter. Given the requirement of ρhv <0.80 collocated with ZDR >2 dB, this is a reasonable assumption and rules out ground clutter. There may be a small overestimate of bioscatter area where very light rain is present; for example, near the edges of a TC eyewall where ρhv may be low due to noise. For each radar scan, the mean distance from the radar to the bioscatter signature was also recorded. Calculation of the variables noted above followed the procedure and equations described by Van Den Broeke and Gunkel (2021), which are built on the foundation of Chilson et al. (2012) and Dokter et al. (2011). For TCCC near the radar, a 500-m constant altitude plan position indicator (CAPPI) was used to eliminate the effects of low-level clutter, for example, due to radar beam interactions with the coastline. Thus, the lowest altitude data considered were at 500 m, though base scan could be much higher if the storm was far from the radar.

Distance between the radar and bioscatter signature was expected to potentially influence observed bioscatter signature characteristics, since observation at closer range is more likely to properly represent local maxima and to detect bioscatter if present at low concentration. For each dataset with bioscatter detected over at least a 10 km change in range (n = 33; n = 32 for maximum altitude), Pearson’s correlation was calculated between distance and bioscatter signature maximum altitude, area and density (across all radar scans for each dataset; each dataset yields one correlation coefficient value for each of these three variables, n = 4–63, n < 10 in 3 cases).

Characteristics of each TC analyzed were determined using archived data, including:
  1. TC intensity (minimum central pressure and maximum sustained wind speed, at both the time of analysis and extreme values over the TC’s life) obtained from the National Hurricane Center’s (NHC’s) season summary documents and advisory archive, which contain official values for these quantities derived from dropsonde data, ship reports, oil platform data and a wide variety of land-based stations.
  2. Initial formation location, also obtained from NHC’s season summaries and advisory archive. This was identified as the location where NHC first began tracking the system, and was assigned one of the following classifications (Fig. 2):
    1. Open Atlantic: all areas east of 55°W.
    2. Windward/Leeward Islands: within 55°–65°W, from the northern coast of South America to 20°N.
    3. Caribbean: from 65°W to Central America, and from northern South America to Hispaniola and Cuba.
    4. Gulf of Mexico: bounded by the land from Florida to the east coast of Mexico, and south to the Yucatán Peninsula and northern Cuba.
    5. Atlantic Coast: a region north of Hispaniola and Cuba and along the east coast of the United States (Fig. 2).
    6. Western Atlantic: region between the Open Atlantic and Atlantic Coast regions (Fig. 2).
  3. Proportion of TC track within 100 km of land was estimated to the nearest 10% from formation to landfall using NHC track information. This assumes that the quantity of bioscatter which may become incorporated into the TC circulation differs depending on the distribution of birds, which may be higher near land than over open water.
  4. Diameter and hurricane diameter of each TC were recorded through the analysis times. This information was obtained from archived NHC advisories, which contain the wind radii at each advisory time. Diameter of a TC was defined as twice the average radius of tropical storm-force wind (34+ kt.) over the four quadrants of the TC circulation, and hurricane diameter was defined as twice the average radius of hurricane-force wind (64+ kt.) over the four quadrants. Non-hurricane intensity storms were indicated as having a hurricane diameter of 0 km.
  5. TC inner circulation convective structure was recorded for each analysis time (radar scan) according to the following classifications (Fig. 1):
    1. Closed eye: TCCC was 100% enclosed by precipitation. Throughout this classification system, ‘precipitation’ refers to non-biological scatter with ZHH ≥20 dBZ.
    2. Mostly closed eye: TCCC was 75–99% enclosed by precipitation.
    3. Partly open eye: TCCC was 50–74% enclosed by precipitation.
    4. Mostly open circulation: TCCC was 25–49% enclosed by precipitation.
    5. Open circulation: TCCC was <25% enclosed by precipitation.
Details are in the caption following the image
Map of TC formation regions defined for this study.

For quantification, these categories were assigned values ranging from 1 (closed eye) to 5 (open circulation). This scheme was developed to test the hypothesis that TCs with closed or mostly closed eyes are more likely to contain large bioscatter density, since the intense thunderstorms surrounding a TCCC would not be favorable for birds to fly through.


First, a quantitative description of the bioscatter signature is presented across the TC datasets. Bioscatter was ubiquitous within these storms, but large/deep bioscatter signatures were relatively uncommon. Associations between the bioscatter signature and time of year were examined, and July–October storms were found to generally contain larger, deeper bioscatter signatures. Associations between geographic location and TC structure were examined next, since they could affect time of year when the signature was most pronounced. Where storms form and how much time they spend near land generally does not predict bioscatter signature characteristics, though TC intensity and structure may strongly control the depth, density and area of the bioscatter signature.

Characteristics of bioscatter in TCs

First, a descriptive look is presented of bioscatter characteristics across the 42 datasets. Potential influence of radar–TC center distance is also addressed, since larger and taller bioscatter signatures could be expected closer to the radar.

Histograms of the bioscatter signature maximum altitude, area and density are shown, containing one point from each of the 42 TC datasets (Fig. 3). Each point represents the average over the three radar scans when the centroid of the bioscatter signature made its closest approach to the radar. This does not indicate the maximum values of those variables; for example, the bioscatter signature might decrease in areal coverage after landfall, which was frequently observed when the TCCC was closest to the radar. It could be assumed that closest approach values are the highest quality available for each dataset, but this is also not necessarily true since factors such as sea clutter and beam blockage can become more problematic as storms get closer to the radar.

Details are in the caption following the image
Bioscatter signature statistics averaged over three radar scans centered on the time of closest approach of its centroid, for each dataset used in this study. (A) Distance (km) of bioscatter signature centroid from the radar at the time of closest approach, (B) the highest altitude (km) at which a bioscatter signature was detected, (C) area (km2) of the bioscatter signature, and (D) bioscatter signature density (cm2 km−3). Notable storms are labelled in each panel.

Distance between the bioscatter signature centroid and the radar ranged from 6.9 to 57.5 km (Fig. 3A), with most signatures 18–40 km from the radar at closest approach and nearly symmetrically distributed (skewness = 0.386). The maximum altitude at which the bioscatter signature was detected (Fig. 3B) was generally ≤2.5 km, with many storms in the 1–1.5 km range. A few notable storms had bioscatter detections at an altitude >3 km, contributing to a moderate positive skewness (skewness = 0.688). These storms were often intense and/or had relatively closed centers of circulation, though associations between bioscatter signature altitude and storm characteristics are explored below. Area (Fig. 3C) and density (Fig. 3D) of the bioscatter signature are distinct measures. Area can be large, but the signature can have low ZHH (indicating relatively few scatterers), leading to a low density. Bioscatter area could also be small but there could be many scatterers present, leading to high density. Both area and density were generally characterized by relatively low values (area <500 km2 and density <2500 cm2 km−3), though each variable had substantial outliers with much higher values (Fig. 3C and D) leading to large positive skewness values (skewness = 1.756 for area and 2.167 for density).

The influence of radar–bioscatter distance was investigated via distributions of Pearson’s correlation coefficients between distance and bioscatter signature characteristics, as described in Materials and Methods (Fig. 4). Observations of large altitude are slightly skewed toward times when TCCC are far from the radar, though both the mean and median of this distribution of correlation coefficients were <0.3 (Fig. 4A) indicating that distance is not critically important to observations of bioscatter signature height. Area of the bioscatter signature was strongly dependent on distance (Fig. 4B), with larger areas typically observed as TCCC became closer to the radar site. This indicates that most bioscatter is concentrated relatively close to the 500-m CAPPI level selected for this study. Bioscatter density (Fig. 4C) showed a relatively uniform distribution, indicating that it is a more distance-resistant measure of the bioscatter characteristics of a TCCC. These results indicate that radar–bioscatter distance can be one important factor in determining how the bioscatter signature appears, especially for its area.

Details are in the caption following the image
For all TC centers of circulation whose distance to the radar varied by >10 km, distributions of Pearson’s correlation coefficient values between distance and (A) altitude, (B) area and (C) density of the bioscatter signature, calculated over all valid analysis times (e.g. one count = one dataset). Gold vertical line is the mean value, and purple vertical line is the median.

Bioscatter transport and month

Variability of the bioscatter signature was investigated across the months when TCCC were sampled (Fig. 5; May–November) using the maximum altitude, largest area and highest density observed for each dataset. Though a Kruskal–Wallis H test did not indicate a significant difference in these variables between months (altitude: P = 0.331; area: P = 0.066; density: P = 0.175), a few general trends were apparent. Maximum altitude of the bioscatter signature was higher during the more active portion of the TC season (July–October). Area of the bioscatter signature was much greater and generally similar from July to October (Fig. 5B), while bioscatter density was largest August–October (with large density being especially common in September and October, given the influence of one very high-density storm in August; Fig. 5C). Overall, large and tall bioscatter signatures are more frequent from July to October, while shallower and smaller signatures are associated with the beginning and end of the TC season. Whether this may be related to differing storm intensity and/or convective structure is investigated below.

Details are in the caption following the image
Distributions of bioscatter (A) maximum altitude (km), (B) area (km2) and (C) density (cm2 km−3) for each month. Dots represent individual datasets, and stars represent the monthly average value.

Bioscatter transport and geographic considerations

Given the variable distribution of birds across the Atlantic and Caribbean and between land and open ocean, several potential geographic factors were examined as possibly contributing to characteristics of the bioscatter signature (Table 2). As in the preceding section, the maximum altitude, area and the density from each dataset were used. First, initial TC formation region was identified (Fig. 2). When datasets were sorted by TC formation region, few trends were noted and Kruskal–Wallis P-values resulting from comparisons across formation regions were not significant (Table 2). Maximum altitude of the bioscatter signature was comparable across most formation regions, with lower altitudes among the generally weaker storms that formed near the Atlantic coast and Bahamas. Signature area varied widely with no regions clearly associated with much larger or smaller bioscatter signatures. Signature density was generally large for storms which formed open the open Atlantic and generally small for storms forming in the Gulf of Mexico.

Table 2. Average bioscatter signature characteristics for several potential geographic factors.
n Altitude (km) Area (km2) Density (cm2 km−3)
Formation region
Open Atlantic 11 2.81 (1.05) 984.93 (727) 5360 (7161)
Windward/Lee. Islands 6 2.87 (1.21) 609.23 (651) 2820 (4250)
Caribbean 7 2.45 (0.74) 527.96 (431) 3027 (4415)
Gulf of Mexico 10 2.80 (1.15) 846.32 (705) 569 (533)
Atlantic Coast 8 1.77 (0.69) 760.11 (1055) 1585 (1533)
Kruskal–Wallis P-value 0.221 0.558 0.397
Percent of storm track over land
0–20 14 2.73 (1.07) 981.27 (732) 4013 (6250)
21–40 15 2.31 (0.89) 501.04 (500) 1687 (3143)
41–60 6 3.44 (0.94) 805.38 (632) 3384 (3942)
61–100 7 1.82 (0.81) 979.01 (1207) 1069 (1880)
Kruskal–Wallis P-value 0.052 0.395 0.349
Location of observation
TX, LA, AL 15 3.03 (1.02) 844.49 (632) 2562 (4702)
FL 12 2.47 (1.19) 899.27 (985) 4420 (5864)
SC, NC 10 2.33 (0.79) 731.13 (599) 1158 (2520)
NJ 2 1.64 (0.56) 430.45 (423) 77 (73)
PR 3 2.05 (0.53) 299.40 (43) 3309 (4682)
Kruskal–Wallis P-value 0.231 0.812 0.2985
  • Number in parentheses after each value is the standard deviation. The Kruskal–Wallis P-value reported below each column is for a comparison across all groups (e.g. all formation regions).

Percent of storm track within 100 km of land, estimated to the nearest 10% as described above, also showed little consistent association with bioscatter signature characteristics (Table 2). A Kruskal–Wallis H test P-value of 0.052 suggests potential for a difference in maximum altitude of the bioscatter signature for different proportions of time spent over land, but sample size was likely too small for a robust conclusion. Density of the bioscatter signature generally decreased as storms spent more time near land (Table 2), but a Kolmogorov–Smirnov test rejected the hypothesis that density in storms in the 0–20% category was different from density in storms in the 61–100% category (P = 0.188).

Sampling location was also examined since the current location of a TC may influence the amount of bioscatter present. Sampling locations were divided into the categories shown in Table 2, generally consisting of Puerto Rico, the northern Gulf Coast (Texas, Louisiana and Alabama), Florida, the Carolinas and the Atlantic coast north of North Carolina (in this dataset, two storms which affected New Jersey). Although the number of samples was small for northern Atlantic coast and Puerto Rico storms, it appears that storms in the northern Gulf Coast region often have relatively high-altitude bioscatter signatures. The largest signatures were associated with the Gulf of Mexico and Florida, and the smallest signatures were found in storms near Puerto Rico and along the northern Atlantic coast. Density of the bioscatter signature was typically high near the Gulf of Mexico/Florida and Puerto Rico, and low near the northern Atlantic coast (Table 2), but the small number of cases in some regions precludes robust analysis. Kruskal–Wallis P-values indicate that the differences between groups are not significant (Table 2).

The bioscatter signature sometimes markedly increased as the TCCC passed over coastal islands or reached the coast [e.g. Fig. 6 shows an example from Tropical Storm Beta (2020)]. For storms with sufficient data, I examined whether bioscatter characteristics in the scans prior to landfall were substantially different from those in the scans after landfall. It could be hypothesized, for example, that bioscatter would be reduced post-landfall as trapped birds landed. The number of pre- and post-landfall scans was always the same, and the number of scans selected for each was the maximum possible value between 5 and 8. Twelve datasets contained sufficient scans. Two of these were observations of Hurricane Hanna (2020), sampled by both the WSR-88D at Corpus Christi (KCRP) and that at Brownsville (KBRO). For these datasets, I calculated the P-value for the Wilcoxon–Mann–Whitney rank sum test for a comparison of bioscatter signature altitude, area and density pre- versus post-landfall. For example, I compared the set of area values pre-landfall to the set of area values post-landfall. A P < 0.05 indicates that the samples have a 95% chance of being taken from populations with different medians (e.g. there was a significant change in bioscatter characteristics across the time of landfall). Maximum altitude of the bioscatter signature was significant in three cases (two increase and one decrease in altitude), area was significant in three cases (two decrease and one increase) and density was significant in four cases (three increase and one decrease). This indicates that the bioscatter signature does not repeatably change in a significant way across the time of landfall.

Details are in the caption following the image
Example of a large increase in bioscatter around the time of landfall of Tropical Storm Beta (2020). (A and B) are ZHH (dBZ); (C and D) are CC. Data are from the Corpus Christi WSR-88D (KCRP) at 0201 UTC (left column, pre-landfall) and 0253 UTC (right column, post-landfall) on 22 September 2020. White ovals indicate area of bioscatter.

Bioscatter transport and TC characteristics

TC intensity and structure were investigated as potentially influencing bioscatter signature characteristics. Such associations may indicate the types of TCs which are most important for bioscatter transport. All correlation values reported in this section are Spearman’s correlation given the potential non-normal distributions of variables examined.

Estimates of minimum central pressure, maximum sustained winds and size of the surface wind field (>34 kts, tropical storm strength; >64 kts, hurricane strength) are generally available every 3 h. To assess the influence of storm intensity, average values of minimum pressure and maximum sustained wind speed were calculated over the analysis period. Since minimum pressure and maximum wind were strongly correlated (rs = −0.90; P = 5.33 × 10−16; n = 42) only the results for maximum wind are shown here, related to maximum altitude, area and density of the bioscatter signature (Fig. 7). Maximum altitude (Fig. 7A) and density (Fig. 7C) of the bioscatter signature increased markedly with maximum sustained wind (altitude: rs = 0.55, P = 3.55 × 10−4, n = 38; density: rs = 0.51, P = 9.54 × 10−4, n = 39). Area of the bioscatter signature (Fig. 7B) did not increase as markedly but was still slightly larger with higher wind speeds (rs = 0.31, P = 5.64 × 10−2, n = 39).

Details are in the caption following the image
Bioscatter characteristics versus average sustained wind speed (kt.) through the analysis period: (A) maximum altitude (km), (B) area (km2) and (C) density (cm2 km−3). Each point indicates one hurricane event. R-values indicated are from a Pearson’s correlation.

TC structural factors were also examined (Table 3). Average values of the wind field radius (>34 and >64 kt.) and of the open/closed nature of the circulation (described above) were calculated over the analysis period. Size of the tropical storm-force wind field was not related to bioscatter signature characteristics (Table 3). As the radius of hurricane-force winds increased, however, altitude, area and density of the bioscatter signature also increased. Finally, storms with a center of circulation enclosed by convection (e.g. an eye structure) were more likely to have dense and high-altitude bioscatter signatures (Table 3).

Table 3. TC structural variables related to bioscatter signature characteristics.
Variable 1 Variable 2 rs P d.f.
Size Altitude 0.29 0.07 37
Area 0.23 0.17 38
Density 0.08 0.61 38
HurrSize Altitude 0.55 <0.001 37
Area 0.50 <0.01 38
Density 0.43 <0.01 38
ConvStruc Altitude -0.34 0.04 37
Area -0.05 0.77 38
Density -0.50 <0.01 38
  • Here, rs is Spearman’s correlation and P is the associated P-value. ‘Size’ is the average radius of the tropical storm-force wind field (34+ kt.); ‘HurrSize’ is the average radius of the hurricane-force wind field (64+ kt.), and ‘ConvStruc’ is the average value of the convective structure variable described in Materials and Methods (1 = closed circulation and 5 = open circulation). ‘d.f.’ indicates the number of degrees of freedom for each comparison.

Four storms were well sampled by multiple radars at different times, ranging from several hours to several days apart. These datasets did not show similar bioscatter altitude, area or density characteristics, indicating that bioscatter can change markedly within a single storm over hours to days. Two storms [Hanna (2020) and Eta (2020)] were sampled at the same time by multiple radars, presenting a unique opportunity to compare two sets of observations taken at different distance from the bioscatter signature. For these storms, radar scans were as closely temporally matched as possible and the bioscatter altitude, area and density compared. Altitude is expected to correspond well between overlapping datasets since the top of the bioscatter signature should be similarly observable at different distances, while area/density is not expected to correspond as strongly because they are more dependent on radar–bioscatter distance. Hanna was a hurricane with a closed eye during the >3-h sample overlap, while Eta was a strong tropical storm with a closed or mostly closed eye (Fig. 1A and B) during a ~1.5-h sample overlap. Among the bioscatter characteristics, altitude showed the best correspondence between datasets. Hanna’s bioscatter had an average altitude of 3.42 km from KBRO and 3.40 km from KCRP, despite an appreciable difference in radar–bioscatter distance between portions of the overlapping dataset. Eta’s bioscatter had an average altitude of 1.20 km sampled from KAMX and 0.95 km sampled from KBYX, also in good agreement. Area and density of the bioscatter signature did not correspond as well, as expected (percent differences 33–97% for these variables; not shown).

Hurricanes Harvey (2017) and Michael (2018) were sampled while strengthening rapidly prior to landfall. They present the best opportunity in this set of storms to examine bioscatter signature characteristics in storms whose intensity rapidly changes while over water, but the comparison is limited since both were intense hurricanes with completely closed eyewalls throughout the analysis period (average sustained wind over the analysis period of 106 kt. in Harvey and 129 kt. in Michael). These storms were remarkable for their high-altitude bioscatter signatures (Harvey and Michael were in the upper tail of the distribution, Fig. 3B) and their large-density bioscatter signatures (Harvey was the second highest in the dataset and Michael was the third highest, Fig. 3D). Their bioscatter signatures did not have large area, which is expected for closed-eye TCs. Temporal changes in bioscatter signature characteristics through these datasets could not be robustly assessed since both storms were rapidly approaching the radar.


Over a 10-year archive, polarimetric radar observations offer a unique opportunity to examine the interactions of organisms in the airspace with their environment on large spatiotemporal scales, including near and within TCs. In this study, bioscatter signatures were highly variable across a sample of 42 Atlantic-basin TCs that occurred from 2011 to 2020. Although several datasets had times without a bioscatter signature, all contained at least one time when a bioscatter signature was present, indicating it is ubiquitous in Atlantic TCs. The relatively high ZHH values comprising the bioscatter signature, its frequent high altitude and the little interaction with land experienced by most storms in the dataset suggest that a vast majority of the observed signature is due to birds rather than insects. TC bioscatter can be present near the center of circulation and elsewhere, for example, between the outer rainbands. While bioscatter not associated with the TCCC was not considered in this study, it may also sometimes be important. The process leading to the presence of bioscatter in a TC may also differ, with a tendency for entrainment or trapping followed by the displacement of organisms within the center of circulation and a tendency for displacement to dominate outside of the TCCC.

Distributions of maximum altitude, area and density of the bioscatter signature were commonly right-skewed (Fig. 3) with higher end outliers. These outliers, especially altitude (Fig. 3B) and density (Fig. 3D), typically represented intense hurricanes. This indicates that birds in these systems are more likely to be contained by the TCCC. Thus, though bioscatter density may be quite high in some intense hurricanes, it may be present over a relatively small area. Distance from the radar has little effect on observed maximum altitude of the bioscatter signature (Fig. 4A) but has a large effect on estimated signature area (Fig. 4B). This is an expected result since the altitude is a vertical measure of bioscatter usually extending above the base radar scan, while the area is measured on a horizontal plane and typically decreases with altitude (e.g. fewer birds reach high altitude). Maximum altitude of the bioscatter signature, particularly in TCs with eye features, may be a function of the thermal structure of the TC eye including inversion height (e.g. Willoughby, 1998). Although investigating potential associations between the bioscatter signature and TC thermal structure was beyond the scope of this paper, it may be beneficial for meteorologists if a link can be established between inversion height and the TC life cycle, including periods of intensification and weakening.

Mature hurricanes frequently undergo eyewall replacement cycles (ERCs; e.g. Houze et al., 2007). During these 24–36 h cycles, the inner core structure of a hurricane changes markedly (e.g. Sitkowski et al., 2011) as the inner eyewall is gradually replaced by an outer eyewall. During this transition, there are often multiple convection-free areas near the TCCC, and TC intensity often decreases before increasing once the ERC is complete (Kossin & Sitkowski, 2012). Given these changes, one could expect corresponding changes in bioscatter signature characteristics across an ERC (e.g. the signature’s area may increase as an outer convection-free area opens, and then decrease as the new eyewall consolidates). Since inversion height changes across ERCs are unknown, it is unknown how altitude of the bioscatter signature may respond. There appeared to be a large increase in altitude of bioscatter across an ERC in Hurricane Matthew (2016), while the opposite was noted across an ERC in Hurricane Maria (2017) (not shown). Future work with the bioscatter signature may reveal the details of hurricane inner core structure during ERCs.

Larger and denser bioscatter signatures were often present in late summer and fall TCs (Fig. 5B and C). Some of this trend could be related to a corresponding increase in insects, but given the reasoning in the first paragraph of the Discussion, the majority of this bioscatter is likely to be birds. The larger and denser nature of the signature likely indicates a larger number of birds. We speculate that this late-season increase may correspond to autumn migration (e.g. Walsh et al., 2017). Some TCs may entrain seabirds which are normally present over the water, but some birds which become incorporated into TC circulations could possibly also be migrating land birds. For example, large flights of land birds are common over the Gulf of Mexico during fall migration (e.g. Rappole & Ramos, 2010; Ward et al., 2018), and if this occurs prior to the arrival of a TC, these birds may be entrained into the TC circulation. The degree of any land bird contribution to the signature remains unknown, however, given the lack of direct observations, and may also be a function of prevailing weather conditions prior to hurricane arrival (e.g. had north winds over the Gulf of Mexico encouraged migration in the day preceding hurricane arrival?). Future work may investigate the potential interaction between autumn land bird migration and the TC bioscatter signature. Potential for typical bird distributions to be altered in the vicinity of a TC could also be investigated in future work.

Geographic factors including TC formation location and portion of TC track near land were not strongly related to bioscatter characteristics (Table 2). Stronger associations were found between sampling location and bioscatter characteristics (Table 2). Storms sampled on the Gulf Coast and in Florida generally had larger and denser bioscatter signatures. TCs near the coast are passing through the ranges of many more species (e.g. there is larger bird species diversity near the Gulf and East Coasts of the United States; Somveille et al., 2013) and a much larger total number concentration of birds (e.g. Northeast Ocean Data, 2021). Observed larger diversity and number of birds present are consistent with larger and denser bioscatter signatures in these regions. This result may be biased by the prevalence of more intense TCs in this region, and may also be influenced by prevailing weather conditions at the time of hurricane arrival. Sometimes a TCCC passing over a coastal island or reaching the coast was associated with a large increase in the bioscatter signature, but this was not repeatable.

TC intensity (minimum central pressure and maximum sustained wind speed) was among the variables most strongly associated with the bioscatter signature. Maximum altitude and density of the bioscatter signature, and to a lesser extent its area, increased with TC intensity (Fig. 7). This result indicates that birds are more likely to become trapped near the TCCC in intense TCs, whether because they are there when the circulation forms or because they are not as likely to attempt to leave through intense wind surrounding the calm center of circulation. The same appeared to be true in rapidly strengthening hurricanes, likely reflecting an aversion for birds to fly through an intense eyewall and the likelihood that once birds are present, they will remain within the calm eye. These radar-derived results are consistent with prior observer reports of birds and insects in the eyes of intense TCs (e.g. Coronas, 1925; Halverson, 2004; Young, 1921). We speculate that a mature hurricane could transport a bird for as long as the hurricane is over water. Indeed, observer reports indicate that birds in the eye may be exhausted, seeking to rest on ships (e.g. Mayhew, 1949). Hurricane inner core structure also influenced bioscatter characteristics. TCs with larger hurricane-force wind fields had taller and larger bioscatter signatures, while TCCC closed by convection had taller and denser bioscatter signatures (Table 3). These findings support the speculations that birds are less likely to leave a TCCC if it is ringed by thunderstorms and if the TC is intense. Birds may seek to avoid thunderstorms in this case, especially since their fall movements are less likely to be obligatory [e.g. Van Den Broeke and Gunkel (2021) found that birds are more likely to avoid mid-latitude thunderstorms in fall than in spring].

Bioscatter characteristics can vary substantially over time. It is currently unknown whether these changes can indicate anything beneficial about inner core structural changes, but this is a potential avenue of future research. Such research should carefully consider the influence of radar–bioscatter distance changes through time. The most range-resistant bioscatter characteristic is maximum altitude, since this variable measures vertical aspect of the signature which will be sampled by any radar able to scan that altitude, provided there are enough birds present to be detected at that distance.

This sort of work also has great potential for elucidating bird behavior in the presence of varied weather conditions. It is especially helpful for understanding how birds respond to extreme weather events, which may become more frequent as the climate system shifts. It can also add insight to how extreme weather events influence population dynamics, for example, by destroying existing populations and potentially founding new colonies via long-range transport to new locations. Future interdisciplinary work incorporating biological and radar data will likely lead to broader awareness of how organisms in the airspace interact with their environment, which will be a crucial perspective as that environment changes.


The authors thank Phil Stepanian for contributing MATLAB code used in the analysis, and two anonymous peer reviewers for their comments which substantially improved the manuscript.

    Funding Information

    This project and manuscript are partially funded by the National Science Foundation (grant 1545261) and partially by an academic appointment at the University of Nebraska-Lincoln. No funders had input into the content of the manuscript or required their approval prior to submission or publication.

    Data Availability Statement

    Radar data are available for free at the National Centers for Environmental Information (https://www.ncdc.noaa.gov/has/HAS.FileAppRouter?datasetname=6500&subqueryby=STATION&applname=&outdest=FILE) or the Amazon Web Services Level 2 NEXRAD archive (https://s3.amazonaws.com/noaa-nexrad-level2/index.html). Historic tropical cyclone data are available from the National Hurricane Center (https://www.nhc.noaa.gov/data/).