Journal list menu
- Advances in information technology are increasing the use of radar as a tool to investigate and monitor bird migration movements. We set up a field campaign to compare and validate outputs from different radar systems.
- Here we compare the pattern of nocturnal bird migration movements recorded by four different radar systems at a site in southern Sweden. Within the range of the weather radar (WR) Ängelholm, we operated a “BirdScan” (BS) dedicated bird radar, a standard marine radar (MR), and a tracking radar (TR).
- The measures of nightly migration intensities, provided by three of the radars (WR, BS, MR), corresponded well with respect to the relative seasonal course of migration, while absolute migration intensity agreed reasonably only between WR and BS. Flight directions derived from WR, BS and TR corresponded very well, despite very different sample sizes. Estimated mean ground speeds differed among all four systems. The correspondence among systems was highest under clear sky conditions and at high altitudes.
- Synthesis and applications. While different radar systems can provide useful information on nocturnal bird migration, they have distinct strengths and weaknesses, and all require supporting data to allow for species level inference. Weather radars continuously detect avian biomass flows across a wide altitude band, making them a useful tool for monitoring and predictive applications at regional to continental scales that do not rely on resolving individuals. BirdScan and marine radar’s strengths are in local and low altitude applications, such as collision risks with man-made structures and airport safety, although marine radars should not be trusted for absolute intensities of movement. In quantifying flight behaviour of individuals, tracking radars are the most informative.
Radar is a powerful tool to observe and track animals. It requires no tags or handling and can be used to remotely observe the movements of free-flying animals (birds, bats and insects). Radar is particularly suitable for monitoring migratory movements, as these typically take place at high altitudes and during the night, which are ideal conditions for radar monitoring but make other types of observations difficult.
Technological advances have made both radars and radar data more accessible, leading to an increased use of marine radars (MR), dedicated bird radars and weather radar (WR) networks to monitor animal movements and migration passage, especially in environmental assessment studies. Common applications include monitoring of bird movements in relation to collision risks with man-made structures, in particular wind farms (Fijn, Krijgsveld, Poot, & Dirksen, 2015; Plonczkier & Simms, 2012); bird strike prevention, with radars being used at airports to avoid collisions during take-off and landing (Gerringer, Lima, & Devault, 2016); and identifying hotspots of animal movement to inform airport management, as well as measuring high-altitude migration intensities for subsequent issuing of flight restrictions for military training flights (Van Belle, Shamoun-Baranes, Van Loon, & Bouten, 2007).
The most recent development in radar ornithology has been an increased focus on using WR for bird movement studies (Bauer et al., 2017; Shamoun-Baranes et al., 2014). Consequently, WR data are increasingly utilized by biologists, supplying a completely new spatial and temporal coverage of bird migration movements, and offering new possibilities for monitoring applications (Bauer et al., 2017; Dokter et al., 2011; Horton, Shriver, & Buler, 2014). These new applications include monitoring of flyways for dispersal of pests and disease (Bauer et al., 2017), large scale attraction of migrants to artificial light (Van Doren et al., 2017), identifying stopover sites for informing conservation (Buler & Dawson, 2014) and using WR as a monitoring tool for assessing long-term population changes (Bauer et al., 2017).
Despite the recent popularization of radar monitoring, extensive cross-validation of animal movement data obtained by different radar systems have been sparse (Dokter, Baptist, Ens, Krijgsveld, & van Loon, 2013; Dokter et al., 2011). Several studies have compared small-scale radars with visual observations and infrared detection (e.g., Gauthreaux, Belser, & Welch, 2006; Schmidt, Aschwanden, Liechti, Wichmann, & Nemeth, 2017). We present the first large-scale, co-located calibration campaign with several radars dedicated to tracking biological targets at a single site. We evaluate the strengths and weaknesses of four different radar systems and provide recommendations for using radar systems to monitor bird movements, with particular focus on nocturnal migration.
2 MATERIALS AND METHODS
During September–October 2015, we deployed three small radar systems dedicated to extracting bird signals at a site approximately 22 km from the WR Ängelholm (Figure 1, Table 1). The site (56°16′51N, 12°31′38E) is part of the Kullaberg nature reserve located on the southern slope of the Kullaberg ridge, in southern Sweden.
|Weather radar (WR)||BirdScan (BS)||Marine bird radar (MR)||Tracking radar (TR)|
|Recording method||Horizontal scanning (360°)||Vertical pointing||Vertical scanning (180°)||Tracking single targets|
|Operation range for birds||5–25 km||0.05–2 km||0.1–3 km||0.3–10 km (size dependent)|
|Bird data output||Vertical profiles of biomass density and ground speed||Multiple continuous individual tracks||Multiple individual tracks built from repeated scans||Continuous single individual tracks|
|Rain filter of bird data||Automatic||Automatic, manual check||Manual||Not applicable|
|Bird echo classification||Radial velocity pattern and echo strength||Wing-beat pattern specific size classes||Distance, speed||Wing-beat pattern|
|Assumptions for bird migration quantification||Standard bird size (RCS), bird movements mainly well directed||Distance-dependent detection probability for each size class||Constant detection probability||Representative sample of speed and direction|
|Availability of equipment/data||High/low depending on country and meteorological institute||High||High||Low|
Due to ground clutter interference, we limited most of the analyses to data from two altitude intervals where there was good coverage from all systems: 200–800 m above sea-level (a.s.l.) and 800–1,400 m a.s.l. (“low” and “high,” respectively). This will exclude some low and high-flying migrants. We also limited the analysis to night time, where nights were defined as starting at 17:00 hr and ending at 08:00 hr, local time (CEST). Sunset/sunrise occurred at 19:48/06:30 hr at the start and 17:31/08:16 hr at the end of the sampling period (9 September to 31 October 2015). Throughout this study we use migration traffic rate (MTR) to compare migration intensity among the different systems. MTR represents the number of birds passing over a virtual transect, perpendicular to the migration direction, of 1 km within an hour (Bruderer, 1971; Lowery, 1951). We chose MTR as the main way of describing migration intensity as it was reliably available from all radar systems (except the tracking radar [TR], which was not used for intensity comparisons). MTR is a flux measure combing both bird density (birds per volume) and bird speed, reflecting the number of birds passing through a given area.
2.1 Weather radar
The dual-polarization WR Ängelholm (56°22′3″N, 12°51′6″E, Figure 1, Table 1) is part of the Swedish WR network. It operates at C-band (5.35 cm wavelength) and the antenna is 209 m a.s.l. The radar operates in 5 min cycles, in which the atmosphere is scanned at 10 different elevation angles ranging from 0.5° to 40°. Radial velocities of objects detected by the radar are collected as well as radar reflectivities.
Vertical profiles of birds were calculated following Dokter et al., 2011; and only briefly described here. Reflectivity factors (dBZ) were converted to reflectivity η (cm2/km3), and averaged into 200 m altitude bins from resolution volumes identified as containing biological scattering only, including ranges up to 25 km. Lowest altitude bin includes samples from 244 to 400 m a.s.l., with 244 m being the lowest surveyed altitude at 5 km range. Bird speed and direction were calculated using a volume velocity processing technique (Holleman, 2005; Waldteufel & Corbin, 1978). Bird density was obtained by dividing the averaged η value by a radar cross section of 11 cm2, which was the average cross section of nocturnal migrants determined during a validation campaign spanning a full autumn and spring in western Europe (Dokter et al., 2011). As opposed to the other radar systems used in this study, scattering due to rain (as well as insects) is removed automatically, using criteria based on reflectivity and radial velocity texture for target identification. One additional postprocessing step was applied to minimize the risk of rain contaminations. When 80% of the profile in the 0–2 km range measured a reflectivity factor of >7 dBZ (a conventional lower threshold used by meteorologists for precipitation) we assumed it was raining and no bird data were calculated.
For each vertical bird profile, MTR (individuals km−1 hr−1) in each altitude bin was calculated by multiplying bird density (individuals/km3), flight speed (km/hr) and the height of the bin (0.2 km). The MTR of altitude bins in each altitude interval (low and high) were summed to obtain the total MTRs in these two larger altitude bands of interest. Finally, we averaged these band-specific MTRs into nightly averages.
For calculating the mean ground speed and mean flight direction per night and altitude interval (low and high) only nights with 5 or more altitude bins containing a bird density higher than 5 birds/km3, were included. This excluded 16 out of 55 nights in the low altitude interval and 31 out of 55 nights in the high altitude interval.
A BirdScan-MR1 ornithological radar (BS) from the Swiss Ornithological Institute was operated during the entire campaign, from the 1 September–5 November, 24 hr/day. BS-MR1 is a newly developed vertical-looking radar system designed to monitor bird movements in real-time (Swiss-BirdRadar.com). BS-MR1 is a 25 kW pulsed X-band radar (9.4 GHz) based on a commercial MR (Table 1). The radar was operated in short pulse (65 ns, range resolution 7.5 m, PRF 1,800 Hz) and long pulse (750 ns, range resolution 110 m, PRF 785 Hz) modes. With a nutation of 2°, the rotating antenna tracks objects within the radar-beam and retrieves information on flight direction and ground speed. BS-MR1 uses characteristics of the echo signature to classify tracks as bird or nonbird, and further classifies birds based on the wing-beat pattern as “passerine-type” and “wader-type” (Zaugg, Saporta, van Loon, Schmaljohann, & Liechti, 2008). We computed MTRs, accounting for distance (height) dependent detection probabilities for the different sized classes (Schmaljohann, Liechti, Bächler, Steuri, & Bruderer, 2008).
In this study, the BS-MR1 radar detected echoes using four operation modes of 15 min each: static short-pulse, rotating short-pulse, static long-pulse, and rotating long-pulse. We restricted the computation of MTR to echoes detected using short-pulse at 200–800 m a.s.l. because the maximal detection range of small birds under short-pulse does not exceed 800 m. At 800–1,400 m a.s.l., we used echoes detected using long-pulse only. We computed no MTR if the effective monitoring time fell below 5 min per 30 min protocol period (short or long pulse) because of rain or technical shut-down. Data on flight behaviour are only retrieved under rotating mode. Means per night were used in this study if at least 10 bird tracks were available. See Supporting Information for additional technical details of the BS-MR1 system.
2.3 Vertical scanning marine bird radar (MR)
The MR system operated from 5 to 17 October 2015, 24 hr/day. The radar (manufacturer: GEM, Italy) is a 25 kW X-band radar (9.1 GHz), with a 2.17 m T-bar antenna (nominal beam width of 22° in elevation and 1° in azimuth) rotating with 34 revolutions per minute (Table 1). The antenna was oriented vertically (horizontal rotation with an additional antenna was not possible due to ground clutter), with the rotation plan along North-South (the expected main flight direction) in order to detect longer trajectories of the birds. With only vertical rotation, it is not possible to determine the direction of a bird flying across the radar beam, which sets limitations on the use of some of the produced information (in particular the track length and speed). During data collection, the radar operated in long pulse mode (200 ns and PRF 1,000 Hz).
It is not possible to access the raw data from the MR, as the acquisition software ExtraSea (from the radar manufacturing company GEM) automatically preprocesses the raw data, directly returning the visual result of this processing (green moving echoes on the screen) (see Supporting Information for details). The visual output of the acquisition software was recorded continuously by using a screen capture software (NCH). We processed the recorded video using the r-package RadR (R Core Team, 2017; Taylor, Brzustowski, Matkovich, Peckford, & Wilson, 2010) to reconstruct bird tracks from the subsequently recorded echoes potentially originating from the same individual bird trajectories. To exclude insects, we ignored tracks shorter than 200 m and with less than four consecutive echoes. In addition, we also excluded tracks within 300 m from the radar and tracks characterized by a ground-speed lower than 30 km/hr and higher than 100 km/hr (Bruderer & Boldt, 2001; Schmaljohann et al., 2008).
The number of tracks processed by RadR (not the number of echoes) theoretically corresponds to the number of detected objects. However, with increasing track duration, RadR tends to split tracks of single objects into more than one track. This trend is intensified with an increasing number of simultaneous echoes. Thus, an overestimation of the number of tracks can occur, leading to a greater increase in the numbers of tracks as the number of actual targets increases.
To calculate MTR, we assumed that all birds crossed the beam parallel to the rotation axis of the radar (N-S). We used the estimated beam width at 100 m altitude layers to weight the number of echoes and compute MTR. To calculate the ground speed, we also assumed that the birds flew parallel to the rotation axis of the radar. We thus underestimate the track length, and thereby ground speed, for birds with flight direction that deviate from the N-S axis.
2.4 Tracking radar
A manually operated TR was operated during 8 nights (7, 9, 28, 30 September and 3, 7, 11, 14 October). The TR tracks individual birds, following one target at a time; it is a mobile 200 kW X-band radar (0.25 μs pulse duration, PRF 504 Hz, 1.5° beam width). Targets are located manually by an operator scanning the sky and then automatically tracked from 1 to 10 min. During tracking the exact position of the target is recorded every second, giving precise measurements of flight altitude, ground speed and track direction. Targets are classified as nonbird, bird, passerine or flock by the operator based on the characteristics of the echo signature (e.g., temporal variation in the echo intensity) representing, in case of a single bird, the wing-beat pattern. Methods closely resembled those in Karlsson, Nilsson, Bäckman, and Alerstam (2012) and Bäckman and Alerstam (2003).
All birds, passerines and flocks are included in this comparison. Only nights with more than 10 birds tracked were included in the nightly means. The TR was used in the comparisons of track directions and ground speed. The manual selection of object to track and duration of tracking may introduce biases in the numbers of targets tracked. Therefore, the TR data were not used to estimate migration intensity or for comparisons of altitude distributions.
2.5 Falsterbo ringing
Falsterbo (FBO) (55°22′27″N, 12°48′29″E, Figure 1) bird observatory has a long-standing ringing regime, with standardized mist net captures since 1980 (Karlsson, 2009). Mainly actively migratory birds are caught as the immediate area is not suitable for stopover and has few resident birds (Zehnder & Karlsson, 2001). As an approximate estimate of migration intensity, the total number of birds ringed in the lighthouse garden during the morning immediately following the night in question was used (e.g., for the night between 5 and 6 September, ringing on the morning of 6 September was used). Ringing starts half an hour before sunrise and continues for at least 6 hr (Karlsson, 2009). All species caught are included in the total sum of birds, but note that this sample excludes species that do not shelter, and hence are not caught, in the shrub and woodland habitats at the ringing station.
2.6 Weather stations and rain filtering
We retrieved hourly rain measurements from two SMHI weather stations in the nearby area (http://opendata-catalog.smhi.se/explore, Figure 1): Hallands Väderö (56°26′58″N, 12°32′49″E, 18 km North from the field site), and Helsingborg A (56°1′49″N, 12°45′55″E, 30 km South-East from the field site). We used data from the station with the largest amount of rain recorded per night in the comparisons, and a night was counted as a “rain night” if any precipitation was measured at either station during the night. This was to make sure that also nights with very light rain would be included, as light rain could pose a challenge to the bird detection algorithms.
There are principal differences in how the WR bird algorithm, BS and MR filters out precipitation (Table 1). For the WR, the algorithm extracting bird echoes filters out events with precipitation automatically. Cases with light precipitation are most challenging to filter out, especially when reflectivity values are similar to those observed in bird migration. Precipitation may therefore be classified as birds on some rare occasions. BS works with a threshold of occupied cells, above which track detection is stopped. However, before the threshold is reached, false tracks are recorded and sometimes wrongly classified as birds. Therefore, the raw track time series is checked manually to exclude events with notable false echoes. For the MR data, events with precipitation were manually excluded from the analysis by visual inspection. The TR did not operate during rain events.
To investigate under which circumstances the relative patterns of MTR among systems was most robust, we used model II major axis regressions in r package lmodel2 (Legendre, 2018). We compared the match among systems on nights with and without rain and, for all nights, at low and high altitudes. Differences in absolute MTRs among systems were tested with Wilcoxon signed rank tests using r version 3.4.1 (R Core Team, 2017).
To investigate flight speeds, we tested the measured mean ground speed per night among the different radar systems in pairwise t tests (R Core Team, 2017). Correlations of flight directions over the season were tested with circular correlations using the r package “circular” (Agostinelli & Lund, 2013). The nightly mean directions were tested against each other with Moors paired test for circular data in Oriana 4.0 (Kovach Computing services, Anglesey, UK).
3.1 Migration intensity
We compared the measured intensity of migration among three of the radar systems (WR, BS, MR) based on the mean MTR per night (Figure 2 and Table 2). The TR was not included in the comparison of intensities. The relative intensity of migration and detection of peak nights corresponded well among the three radar systems from which intensity measures were available (WR, BS, MR), Figure 2 and Table 2. Absolute MTRs differed significantly among all three systems at low altitude (Wilcoxon signed rank tests; WR-BS: v = 268, p < 0.001, WR-MR; v = 78, p < 0.001, MR-BS; v = 78, p < 0.001). Absolute MTRs corresponded well between BS and WR at high altitudes, but the MR differed significantly from both (Figure 2; Wilcoxon signed rank tests; WR-BS: v = 548, p = N.S., WR-MR; v = 78, p < 0.001, MR-BS; v = 78, p < 0.001). The MR provided generally much higher MTRs than the other systems (note the secondary y-axis in Figure 2). Correlations were stronger for the high altitude interval (Table 2, above the diagonal). The mean MTRs also matched reasonably well with the total number of ringed birds at Falsterbo bird observatory (Figure 2, Table 2). It is important to note that there are many reasons to expect significant differences between birds sampled in the air during active migration and birds caught on the ground at a site further south, and we do not expect them to match perfectly. However, a high correlation has been found between number of birds ringed and birds aloft at this site (Zehnder & Karlsson, 2001). On nights with no rain, the measured migration intensities from all the systems, including ringing at Falsterbo, clearly matched better than on the nights with rain present (Tables S1 and S2).
|n = 55||n = 12||n = 60|
|n = 55||n = 12||n = 59|
|n = 12||n = 12||n = 12|
|n = 60||n = 59||n = 12|
3.2 Altitude distribution
We compared the distribution of migration intensity (MTR) across altitude to see whether the vertical profiles differed among systems (Figure 3). We compared the WR and BS over the entire season, and the WR, BS and MR during the period of the MR deployment (5–16 October). The relative mean MTRs at different heights were highly correlated among all systems (model II major axis regressions, entire season: WR & BS r2 = 0.91, time of MR deployment: WS & BS r2 = 0.92, WR & MR r2 = 0.97, BS & MR r2 = 0.85). The BS showed a higher proportion of MTRs at the lowest altitude bin (200–400 m a.s.l.) (Figure 3). The difference in mean MTR between the BS and the marine and WR was also much greater in the lowest altitude bin. The difference between WR and BS is more prominent during clear nights than during nights with rain (Figure S1).
3.3 Ground speed
Mean ground speed per night varied considerably among all the systems (Figure 4). Speeds derived from the BS were significantly higher than those from the WR at low (all tests pairwise t tests; t(df=23) = −10.35, p < 0.000) and high altitudes (t(df=13) = −5.93, p < 0.05). TR mean speed differed from WR data at low (t(df=4) = −4.50, p < 0.05) and high altitudes (t(df=3) = −4.15, p < 0.05). Results from the TR did not differ from the BS estimates at low altitude (t(df=3) = 1.21, p = 0.31), and there were not enough nights to test at high altitude. The MR speeds did not differ significantly from the WR at low (t(df=9) = 1.55, p = 0.11) or high altitudes (t(df=4) = 2.02, p = 0.16). There were not enough nights to test the MR with the other systems.
The overall mean groundspeed, and standard deviation, was, at low altitude, WR: 8.6 ± 2.2 m/s, BS: 12.8 ± 3.4 m/s, MR: 8.7 ± 1.2 m/s, TR: 14.4 ± 3.0 m/s and at high altitude WR: 11.9 ± 2.8 m/s, BS: 15.5 ± 2.6 m/s, MR: 8.8 ± 0.2 m/s and TR: 12.8 ± 3.6 m/s. The WR gives only the average speed of the entire scan volume at a specific height interval, while the other radars measure speeds of individuals (directly or indirectly). WR thus measures the average ground speed of many individuals, which is lower than the ground speed of individuals when individuals fly in varying directions within the scan volume. To estimate the size of this effect, we used the system with likely the most reliable speed measurements (TR) and calculated mean speeds per night by averaging the Cartesian speed components of the individuals per night, as well as the individual speeds. The average difference between these two methods was 0.96 m/s per night (8 nights, SD = 0.66), which only partially accounts for the low speeds on the WR (mean absolute differences at low altitude: WR and BS: 4.38 m/s [24 nights], WR and TR 4.68 m/s [5 nights], WR and MR: 0.56 m/s [10 nights]; high altitude: WR and BS: 4.00 m/s [14 nights], WR and TR 3.01 m/s [4 nights], WR and BS: 2.58 m/s [5 nights]).
3.4 Flight direction
Mean track directions per night were well correlated among the three systems (WR, BS, TR) at both altitudes (Figure 5, Table S3) with R2 values ranging from 0.67 to 0.84. Overall mean directions and circular standard deviations at low altitude were WR: 204° (n = 25, SD = 37°), BS: 195° (n = 24, SD = 26°) and TR: 199° (n = 5, SD = 26°) and high altitude: WR: 194° (n = 15, SD = 23°), BS: 196° (n = 14, SD = 20°) and TR: 190° (n = 4, SD = 11°). Paired tests showed that the WR and BS were significantly different (Table S3), however they were still highly correlated with very similar mean directions during most nights (Figure 5, Table S3). The overall mean directions fit well with the expected migration direction in the area (Sjöberg & Nilsson, 2015). There is more variation in directions at low altitude compared to high altitude, both between nights and within nights (Figure 5). The WR shows less variation within nights (smaller SD) than the BS and the TR. This is expected as the WR bird profile only gives an average direction for each scan volume, while the TR and BS are based on individual directions.
4.1 Migration intensity
The monitoring of the intensity of bird movements requires an unbiased method that can account for distance-dependent detection probabilities (Schmaljohann et al., 2008). In this study, we show that WR, BS and MR provide reliable measures of relative MTR over the season, and WR and BS provide reliable measures of absolute MTR.
Overall the WR and BS matched well, but there were some discrepancies at low altitude. We should keep in mind that the air volume scanned by the WR is very much larger than the volume scanned by the other systems. For instance, the range of the WR (radius of 25 km) extended out over sea, whereas the BS (radius 500 m) detected only birds that flew over land. The overall good agreement of absolute migration intensity retrieved from the WR and the BS confirm previous results comparing two similar radar systems (Dokter et al., 2011).
In the WR measurements, stationary components, such as residual clutter contributions to the signal of resolution volumes (e.g. due to imperfect Doppler filtering), as well as nonmigratory bioscatter (e.g. bats foraging around a roost), have an average radial velocity near zero, which will bias speeds downward, but also bias densities upward by the same proportion. The product of speed and density, the MTR, is therefore expected to be largely free from stationary components, and thus we recommend using MTR to report migration intensity.
The most serious outlier in terms of absolute MTR values was the MR, and two issues contribute to the exaggerated MTR values. Firstly, the MTR calculation is sensitive the alignment of the vertical rotation axis and the main flight direction of birds. The MR was oriented N-S, while the mean flight direction during this period was SSW-SW (Figure 2). The radar thus surveyed a narrower air column relative to the birds tracks than if the beam would have been aligned perpendicular to the mean flight direction (Figure S2). Quantitative measurements with vertically rotating MR are more reliable when the axis of rotation is adjusted perpendicular to the expected main flight direction, because the theoretical length of the transect varies in relation to the sine of the angular difference between flight direction and rotational axis. The same deviation therefore causes much less variation in the transect length at 90° than around 0°. Whenever possible, nightly mean flight directions should be considered for calculating migration intensities with MR data. Secondly, the processing software used, RadR, tends to split long tracks into several tracks, causing an overestimation of the number of tracks, also inflating the MTR. This is more likely to happen with high migration intensity, because the automatic algorithm for a proper allocation of blips of consecutive scans to individual tracks seems to be overstrained. This accords well with the serious overestimates during peak migration nights (Figure 2).
Precipitation had a negative impact on the correlations among the systems, which we believe is mainly due to differences in the exclusion of these events among the systems. The decreased match on nights with rain could be due to either rain contamination of the actual measurements, or that the lower migration intensity on rain nights in itself decreases the match among the systems. At low migration intensities there might also be more spatial structure in the migration, leading to variation between small-scale (BS and MR) and large-scale systems (WR). Heavy rain situations are usually well filtered out, and as migration seldom occurs during heavy rain situations (e.g., Erni, Liechti, Underhill, & Bruderer, 2002), there is little risk of excluding significant migration. Light rain can pose more of a problem, as it can produce weak and varying targets that may sometimes be mistaken for co-occurring migration (however, manual checking of the data easily identifies cases like this). Mainly for nonpolarimetric WRs, variable rain patterns within the volumes are in some cases hard to automatically distinguish from light migration. This distinction is greatly simplified in the new generation polarimetric radars (in this study, no polarimetric products were used). However, with respect to the impact of weather on extracted migration intensities, we encourage manual plausibility checking of processed data by trained researchers for all three types of radar systems. The correlation with Falsterbo ringing data also decreases on nights when rain is present. This could be due to rain contamination in the data, but it could also be that ringing in Falsterbo and the passage of migrants over the Kullaberg area are less well correlated on nights with unfavourable conditions for migration and lower overall migration activity.
Relative altitudinal distributions matched quite well among the three systems compared (Figure 3). Only at the lowest height bin investigated (200–400 m a.s.l.) did BS show higher intensities than the WR, especially on nights without rain (Figure S1).
The WR could potentially have reduced coverage in the lowest scans because masks used to remove ground clutter could also mask low flying migratory movements. The topography can locally influence the height distribution of migratory birds and the surveyed area of the WR includes important variation in topography and a prominent part over sea. Hence, the height distribution observed by the BS may not be representative of the entire area covered by the WR.
Even though not obvious from this study, the upper detection limit of MRs could make them miss some high altitude migration, see Dokter et al. (2013). Although the MR sampled the same area as the BS, the MR showed proportionally lower movement intensity at the lower altitude bin than the BS. Ground clutter and low sensitivity settings generally used to mask ground clutter could also reduce the detection probability of small nocturnal passerines migration in the MR. The relative migration intensity of the WR and MR matched well, also in the lowest altitude bin.
Accurate information of migration intensity at low altitude (below 200 m above ground) is crucial for impact assessment studies aiming to estimate collision rates with human-made structures. In that perspective, the vertical-looking antenna of the BS provides a clear advantage to monitor low-flying migration movements, as it minimizes the effect of ground clutter.
4.3 Ground speed
Ground speed showed variation among all systems (WR, BS, MR and TR), and should be interpreted with caution at the moment. Since the TR measures speeds of individuals directly, we are confident that the speeds registered by the TR reflect “true” ground speeds. However, the TR samples only a small proportion of the total migration and may not be fully representative of all migration movements. For instance, it is possible that larger targets are slightly overrepresented, leading to an upwards bias in the speeds measured by the TR.
In general, the WR showed lower ground speeds than the TR. Some underestimation of ground speed from the WR is expected, as the calculation of ground speed is based on the radial velocities of all birds included in a measurement volume. Only if all birds flew in exactly the same direction, would true mean ground speed be measured. This issue is similar for both the WR and the MR in vertical mode, and they also show quite similar speeds. We estimated this effect by calculating mean ground speed in a similar way with the TR data, but found that there was still a difference even with this effect taken into account. The lower speed on WR would also increase with larger scatter in flight directions, as typically observed at lower altitude where more nonmigratory movements take place and when bird movements are influenced by topography. At low altitudes, it is also possible that a limited amount of clutter mixed in with the relatively weak bird signals (collected at close ranges from the radar) can explain some of the lower speeds detected by WR. We conclude that mean ground speeds derived from WR are reliable when directional scatter is small.
Ground speeds provided by the BS matched TR data at low altitude, but were overestimated at high altitudes. The overestimated ground speeds somewhat exceed previously observed values from former studies in this area (e.g., Nilsson, Bäckman, & Alerstam, 2014). The estimated speed depends on the measured transit-time of the bird within the beam (duration of echo), as well as the estimated beam width at the flight altitude. At low altitudes, the beam width is well defined; in contrast, towards the edge of the detection range small differences in the echo size can provide important differences in the estimated beam width. Without going into further details, the overestimated speeds indicate that the true beam width at high altitude should be somewhat smaller than applied in our calculations. However, the beam width not only varies with altitude, it also depends on the birds’ detection probability (which varies with size, shape and behaviour), leading to uncertainty in the calculated ground speeds. Until further improvements have been made to estimate the true echo size and the beam width, BS estimates of ground speeds for high-flying birds should be interpreted cautiously.
4.4 Flight directions
All systems where directions were available (WR, BS, TR) showed consistent, well-correlated mean directions. The TR and BS both showed larger scatter of flight direction at low altitudes than at high-altitude, corroborating earlier reports in the study area (Sjöberg & Nilsson, 2015).
This means that WR, BS and TR would all be appropriate for investigating flight directions, and MRs operating in a horizontal mode can also measure direction (see Table 3).
|Type of data||WR||BS||MR||TR|
|Relative migration intensity over time||–|
|Absolute migration numbers||–||–|
|Large spatial coverage||–||–||–|
|Detailed site information||–|
|Long time series||–||–|
|Data in (near) real time||–||–||–|
|Overall direction of migration||a|
|Relative flight speeds over time (GS)|
|Absolute flight speeds (GS)||Conditional||Conditional||Conditional|
|Flight speed of individuals||–||Conditional||–|
|Tracks of individuals||–||–||–|
|Relative height distribution||–|
|Low altitude migration||–||a||–|
|Wing beat pattern||–||–|
- a Depending on operation mode.
- b If combined with visual observations.
4.5 Target identification
In general, species identification of targets is not possible with any of the systems used in this study, except when combined with visual observations or under special circumstances (Dokter et al., 2013; Panuccio, Stanzione, Catoni, Santini, & Dell’Omo, 2016). Combining with visual observations is possible at a very local scale with the TR, BS and MR, but is difficult with the WR as it covers large areas. Broad species group classification based on wingbeat patterns is available in the BS and TR.
The possibility of insect contamination will be an issue at certain sites, especially during insect migration periods. The BS separates insects from birds based on echo characteristics and the TR does not track objects as small as insects. We do not expect that insects had a significant effect on our comparison, as mass southward migrations of insects in north-west Europe typically occur in August and early September (Chapman et al., 2010; Hu et al., 2016) before the large peaks of bird migration observed in this study, and previous studies (Alerstam et al., 2011; Chapman et al., 2015, 2016). Depending on site, time of season as well as the time of day, insect contamination needs to be carefully taken into account, especially for the MR and the WR.
The different systems differ in accessibility for applied use (see Table 1). Access to WR data differs depending on the meteorological institute involved and their data policy, though open data policies are becoming more common (c.f. the United States and the Netherlands). WR data are of course also limited to the geographical area surrounding the WR stations, limiting coverage for example offshore. The use of WRs to monitor animal movements have so far mainly been explored in continental US and Europe, but it has the potential to be used in other countries with extensive WR networks, like Russia, China and India. In the US the entire data archive of all 143 continental NEXRAD WR stations are publicly available (Ansari et al., 2018) and in Europe the European Network for the Radar Surveillance of Animal Movement together with the Operational Program for the Exchange of Weather Radar Information (OPERA) is in the process of making bird profiles from European WRs available (Shamoun-Baranes et al., 2014).
BirdScan (Swiss-BirdRadar.com) and MRs, as well as other similar types of scanning radars (such as MERLIN Avian Radar Systems [DeTect, Inc, USA] and ROBIN [ROBIN Radar Systems, the Netherlands]), are commercially available products. They have the advantage of being able to be placed at almost any site, also offshore.
Tracking radars, like the one used here, have extremely limited availability and are not commercially available. However, some dedicated bird radars, and MRs operated in horizontal mode, also have tracking functions.
In this study, we show a high degree of agreement among the different radar systems in describing the relative bird migration intensity and flight directions, and to a reasonable extent the absolute migration intensity and flight speed. The differences observed in absolute migration intensity and flight behaviours highlight the strengths and weaknesses of the different radar systems for different applications (see Tables 1 and 3). The choice of the most appropriate radar will depend on the spatial, temporal, and taxonomic scale of the study (Table 3).
Of the three radars providing reliable migration intensity measures, the WR is best suited to investigate large-scale flows of migration, such as mapping flyways to identify important stopover sites or predicting spread of pests and disease (see Table 3). The extensive coverage, and the possibility of obtaining long time series makes the WR data well suited for planning and evaluating effects of large constructions and developments, such as major infrastructure projects. The possibility of obtaining historical data (for example the US NEXRAD originating in 1991) also makes WR data particularly valuable for planning, conservation and monitoring of long-term changes. As the WR data does not contain species information, it is most appropriate for investigating effects at the assemblage level, for example the effect of artificial light structures on all passing nocturnal migrants (McLaren et al., 2018; Van Doren et al., 2017) or identifying which stopover areas are used in large numbers (Buler & Dawson, 2014). If species composition is deemed important, the WR data can be complemented by other methods such as connecting to bird counts (Sullivan et al., 2014) or acoustic monitoring of flight calls (Farnsworth, 2005).
The highly mobile small scale radar system such as BS, MR and TR can temporally monitor site-specific animal movements aloft. A BS type radar is more appropriate for investigating intensity of movements on a local scale, such as the risk of airstrikes in the immediate area surrounding an airport or the local impact of a wind farm. MRs also operate on the local scale, but are, depending on software used, appropriate for investigating relative patterns, rather than absolute migration intensity. Ground clutter and the placement of the radar generally determines as to how low altitude a radar can give reliable data. A vertical pointing radar, such as the BS, and to some extent MRs, will be less affected by ground clutter and are therefore appropriate for applications that require low-altitude information, such as most collision risks with human-made structures. A BS type radar also has the advantage of recording wingbeat patterns, which makes it possible to assign targets to certain species groups (Bruderer, Peter, Boldt, & Liechti, 2010).
For detailed investigations of flight behaviour the WR is best suited to investigate over larger areas, while BS type radars, MR and TRs all can give reliable information on flight directions (as well as amount of variation and changes in flight direction) at a single site. Only a radar with tracking capabilities can however provide a detailed view of individual bird’s reactions and flight paths.
In conclusion, all radar systems we investigated have the potential for being useful to investigate and monitor bird movements and migration, however careful attention should be given to which questions can be answered by which system.
We acknowledge the support provided by COST – European Cooperation in Science and Technology through the Action ES1305 ‘European Network for the Radar Surveillance of Animal Movement’ (ENRAM) in facilitating this collaboration. The study received financial support from Gyllenstierna Krapperup ́s Foundation and the Centre for Animal Movement Research (CAnMove) financed by a Linnaeus grant (349-2007-8690) from the Swedish Research Council and Lund University. We thank M. Cianchetti and C. Catoni for fieldwork and Ornis italica for travel and logistic support. D. Früh, at Zürich University of Applied Sciences, provided the BS radar. We want to thank Länstyrelsen Skåne and Kullabergs Naturrum as well as landowners Berit and Bernt Sjögren. This is report nr 317 from Falsterbo ringing station.
F.L., J.W.C., C.N. and J.B. planned and organized the study; C.N. and J.B. collected and processed the tracking radar data; M.S. and G.D.O. collected and processed the marine radar data; A.M.D., L.V., G.H. and H.L. extracted, processed and analysed the quality of the weather radar data; B.S. and F.L. collected and processed the BirdScan data; C.N. analysed the data and wrote the paper with substantial input from all authors and all authors approved the final version.
|jpe13174-sup-0001-SupInfo.pdfPDF document, 195.5 KB|
Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.
- 2013). R package ‘circular’: Circular Statistics (version 0.4-7).
- 2011). Convergent patterns of long-distance nocturnal migration in noctuid moths and passerine birds. Proceedings of the Royal Society B, 278, 3074–3080. https://doi.org/10.1098/rspb.2011.0058
- 2018). Unlocking the potential of NEXRAD data through NOAA’s Big Data Partnership. Bulletin of the American Meteorological Society, 99, 189–204. https://doi.org/10.1175/BAMS-D-16-0021.1
- 2003). Orientation scatter of free-flying nocturnal passerine migrants: Components and causes. Animal Behaviour, 65, 987–996. https://doi.org/10.1006/anbe.2003.2119
- 2017). From agricultural benefits to aviation safety: Realizing the potential of continent-wide radar networks. BioScience, 67, 912–918. https://doi.org/10.1093/biosci/bix074
- 1971). Radarbeobachtungen über den Frühlingszug im Schweizerischen Mittelland. Der Ornithologische Beobachter, 68, 89–158.
- 2001). Flight characteristics of birds: I. Radar measurements of speeds. Ibis, 143, 178–204.
- 2010). Wing-beat characteristics of birds recorded with tracking radar and cine camera. Ibis, 152, 272–291. https://doi.org/10.1111/(ISSN)1474-919X
- 2014). Radar analysis of fall bird migration stopover sites in the northeastern U.S. The Condor, 116, 357–370. https://doi.org/10.1650/CONDOR-13-162.1
- 2010). Flight orientation behaviors promote optimal migration trajectories in high-flying insects. Science, 327, 682–685. https://doi.org/10.1126/science.1182990
- 2016). Adaptive strategies in nocturnally migrating insects and songbirds: Contrasting responses to wind. Journal of Animal Ecology, 85, 115–124. https://doi.org/10.1111/1365-2656.12420
- 2015). Detection of flow direction in high-flying insect and songbird migrants. Current Biology, 25, R751–R752. https://doi.org/10.1016/j.cub.2015.07.074
- 2013). Bird radar validation in the field by time-referencing line-transect surveys. PLoS ONE, 8, e74129. https://doi.org/10.1371/journal.pone.0074129
- 2011). Bird migration flight altitudes studied by a network of operational weather radars. Journal of the Royal Society, Interface, 8, 30–43. https://doi.org/10.1098/rsif.2010.0116
- 2002). Wind and rain govern the intensity of nocturnal bird migration in central Europe – A log-linear regression analysis. Ardea, 90, 155–166.
- 2005). Flight calls and their value for future ornithological studies and conservation research. The Auk, 122, 733–746. https://doi.org/10.1642/0004-8038(2005)122[0733:FCATVF]2.0.CO;2
- 2015). Bird movements at rotor heights measured continuously with vertical radar at a Dutch offshore wind farm. Ibis, 157, 558–566. https://doi.org/10.1111/ibi.12259
- 2006). Atmospheric trajectories and spring bird migration across the Gulf of Mexico. Journal of Ornithology, 147, 317–325. https://doi.org/10.1007/s10336-006-0063-7
- 2016). Evaluation of an avian radar system in a midwestern landscape. Wildlife Society Bulletin, 40, 150–159. https://doi.org/10.1002/wsb.614
- 2005). Quality control and verification of weather radar wind profiles. Journal of Atmospheric and Oceanic Technology, 22, 1541–1550. https://doi.org/10.1175/JTECH1781.1
- 2014). A comparison of traffic estimates of nocturnal flying animals using radar, thermal imaging, and acoustic recording. Ecological Applications, 25, 390–401.
- 2016). Mass seasonal bioflows of high-flying insect migrants. Science, 354, 1584–1587. https://doi.org/10.1126/science.aah4379
- 2009). Vingar Över Falsterbo ( 2nd ed.). Falsterbo: Falsterbo fågelstation.
- 2012). Nocturnal passerine migrants fly faster in spring than in autumn: A test of the time minimization hypothesis. Animal Behaviour, 83, 87–93. https://doi.org/10.1016/j.anbehav.2011.10.009
- 2018). lmodel2: Model II Regression. R package version 1.7-3.
- 1951). A quantitative study of the nocturnal migration of birds. Kansas City, KS: University of Kansas.
- 2018). Artificial light at night confounds broad-scale habitat use by migrating birds. Ecology Letters, 21, 356–364. https://doi.org/10.1111/ele.12902
- 2014). Seasonal modulation of flight speed among nocturnal passerine migrants: Differences between short- and long-distance migrants. Behavioral Ecology and Sociobiology, 68, 1799–1807. https://doi.org/10.1007/s00265-014-1789-5
- 2018). Data from: Field validation of radar systems for monitoring bird migration. Dryad Digital Repository, https://doi.org/10.5061/dryad.nr1h6t0
- 2016). Radar tracking reveals influence of crosswinds and topography on migratory behavior of European honey buzzards. Journal of Ethology, 34, 73–77.
- 2012). Radar monitoring of migrating pink-footed geese: Behavioural responses to offshore wind farm development. Journal of Applied Ecology, 49, 1187–1194. https://doi.org/10.1111/j.1365-2664.2012.02181.x
- R Core Team. (2017) R: A language and environment for statistical computing. Vienna, Austria: R Core Team.
- 2008). Quantification of bird migration by radar – A detection probability problem. Ibis, 150, 342–355. https://doi.org/10.1111/j.1474-919X.2007.00797.x
- 2017). Comparison of visual bird migration counts with radar estimates. Ibis, 159, 491–497. https://doi.org/10.1111/ibi.12473
- 2014). Continental-scale radar monitoring of the aerial movements of animals. Movement Ecology, 2, 9. https://doi.org/10.1186/2051-3933-2-9
- 2015). Nocturnal migratory songbirds adjust their travelling direction aloft: Evidence from a radiotelemetry and radar study. Biology Letters, 11, 20150337. https://doi.org/10.1098/rsbl.2015.0337
- 2014). The eBird enterprise: An integrated approach to development and application of citizen science. Biological Conservation, 169, 31–40. https://doi.org/10.1016/j.biocon.2013.11.003
- 2010). radR: An open-source platform for acquiring and analysing data on biological targets observed by surveillance radar. BMC Ecology, 10, 22.
- 2007). An operational model predicting autumn bird migration intensities for flight safety. Journal of Applied Ecology, 44, 864–874. https://doi.org/10.1111/j.1365-2664.2007.01322.x
- 2017). High-intensity urban light installation dramatically alters nocturnal bird migration. Proceedings of the National Academy of Sciences of the USA, 114, 11175–11180. https://doi.org/10.1073/pnas.1708574114
- 1978). On the analysis of single-Doppler radar data. Journal of Applied Meteorology, 18, 532–542.
- 2008). Automatic identification of bird targets with radar via patterns produced by wing flapping. Journal of the Royal Society Interface, 5, 1041–1053. https://doi.org/10.1098/rsif.2007.1349
- 2001). Do ringing numbers reflect true migratory activity of noctural migrants? Journal fur Ornithologie, 142, 173–183. https://doi.org/10.1007/BF01651785