CC BY
8ЭКОЛОГИЧЕСКИЕ АСПЕКТЫ ИСПОЛЬЗОВАНИЯ АЛЬТЕРНАТИВНОЙ ЭНЕРГЕТИКИ, ЭКОЛОГИЯ МЕГАПОЛИСОВ, МАЛЫХ ГОРОДОВ, ДЕРЕВЕНЬ
ECOLOGICAL ASPECTS OF ALTERNATIVE ENERGY AND ECOLOGY OF MEGAPOLISES, CITIES AND VILLAGES
ALGORITHMIC SYSTEM FOR IDENTIFYING BIRD RADIO-ECHO | AND PLOTTING RADAR ORNITHOLOGICAL CHARTS i
a О
"it
L. Dinevich, Y. Leshem |
a H
George S. Wise Faculty of Natural Sciences, Dept. of Zoology |
Tel-Aviv University, 69978, Israel J
E-mail: dinevich@barak-online.net; leonid@post.tau.ac.il g
The proposed algorithmic system for identifying bird radio-echo against the background of reflectors of other types was developed within a novel approach based on the analysis of echo movement characteristics. A long-term implementation of the previously designed algorithm (Dinevich et al. 2004) has demonstrated its ability for identifying bird echo with high confidence. At the same time, this work enabled to determine the directions of further research, aimed at: a) significant reduction of computation time; b) increasing echo identification accuracy in cases of weak echo and of large dense bird masses; c) plotting radar ornithological charts on-line.
In the course of the present study, a comparative analysis was carried out of radio-echo typical of different categories of reflectors. As a result, a set of characteristics was obtained that distinctly specify bird echo and distinguish it from echoes of other types of reflectors. The algorithmic system based on this set of characteristics enables to determine whether a radio-echo movement belongs to one of the four patterns: a) straightforward at non-uniform velocity; b) straightforward at uniform velocity; c) significant deviation from a straight line, non-uniform velocity and d) chaotic undirected shifts. The data on echo movement pattern were used for plotting bird (bird group) flight vectors. In order to filter off false vectors, a special algorithmic procedure was devised based on a number of additional echo characteristics, including the threshold value, the extent of chaotic status in the direction of closely located vectors, the maximum and minimum velocities etc. Another proposed algorithmic procedure enables to make a prompt and accurate (at least 80% confidence) decision on the "bird-not bird" origin a particular echo on the basis of its fluctuation pattern. The system enables online plotting of operational ornithological charts every 12-15 min, including charts that combine meteorological and bird monitoring data, and thus is as an efficient means of maintaining air traffic safety in complicated meteorological and ornithological conditions. In view of the fact that MRL-5 radars are located in many countries and cover an extremely vast territory, it appears expedient to connect them into a network. Using the algorithmic system for bird echo identification by means of MRL-5 radar, such a network could perform intercontinental bird monitoring in the real-time mode, contributing to providing collective air traffic safety.
1. The Objectives of the Study
The present study has been carried out within the novel approach to bird echo identification based on an algorithmic analysis of specific parameters characterizing echo movement. It was shown in the previous work (Dinevich et al., 2004) that during bird migration, the coordinates of bird echo centers form almost straight lines over flight legs. This property was used as the basis of the identification algorithm. A long-term implementation of the algorithm has demonstrated not only the advantages of the method, but also the directions of its further improvement. In some cases, the previously developed techniques proved inapplicable, e.g. when echo from small birds were of power below
the noise level and, therefore, were not registered
in some of the scans within the scan cycle pre- i
scribed by the algorithm. In these cases, straight t
line segments that should have served as the basis |
for echo identification were disrupted. Another ^
problem arose when the mass of birds was espe- §
a
cially dense and rapidly moving, which sometimes |
resulted in the system's mistakenly taking sum- I
mated bird echoes for reflections from ground clut- p
ter, thus excluding those echoes from the further i
analysis. Yet another problem to be solved was g
reducing the computation time. 0
The objective of the present study was to reveal additional parameters that are distinctively characteristic of bird echo and use them to develop a more accurate and prompt algorithmic system
Статья поступила в редакцию 13.09.2005. The article has entered in publishing office 13.09.2005.
for echo identification. As in the previous study, the data was obtained by means of MRL-5 radar.
MRL-5, which is a high-grade meteorological radar designed mainly for cloud monitoring, has a capability of full azimuth scanning (0°-360°), the elevation range of minus 2° to +90° in the upper <t hemisphere and a symmetric narrow beam operat-^ ing simultaneously at two wave-lengths (3.2 cm and
5 10 cm). The application of MRL-5 radar for bird u monitoring and the system of primary data pro-| cessing are described in a number of studies (AbI shayev et al., 1980; Abshayev et al., 1984; Dinev-^ ich et al., 2000; Dinevich et al., 2004).
c
cu
£ 2. The basic principles of bird echo identification underlying the proposed algorithm
Comparative analysis of echoes reflected from different categories of objects shows that there are sets of distinctive properties typical only of a particular category.
Ground clutter echoes are characterized by considerable extension in space, high signal power, wide spectra of fluctuation, amplitude and frequency, as well as by relative immobility (Atlas, 1967; Hajovsky et al., 1966; Chernikov, 1979; Dinevich et el., 2001).
Considerable extension in space is also typical of echoes from clouds of different types, including convective and stratus clouds, that tend to shift in the direction of dominating wind flow. In contrast to bird echo, the echo of atomized clouds and precipitation is larger at 3 mm wavelength than that at 10 cm, while its polarization signal characteristics are typical of spherical targets. The value of differential reflexibility which is the ratio of horizontally reflected echo (at horizontally polarized illumination) to vertically reflected echo (at vertically polarized illumination) is close to unit1 for small drops. Echoes from large-drop clouds at 3cm and 10 cm wavelengths are considerably more powerful than bird echoes (Atlas, 1967; Shupiatzky, 1959; Chernikov, 1979; Doviak and Zrnic, 1993; Dinevich et al., 1994; Zrnic and Ryzhkov, 1998).
It is typical of echoes from visually unobserv-able atmospheric inhomogeneities to have low power, chaotic direction pattern while moving in space <t and polarization parameters that are close to those ¡S of spherical hydrometeors (Shupiatzky, 1959; Bat-^ tan, 1963, 1973; Lofgren and Battan, 1969; Doviak I and Zrnic, 1993; Zrnic and Ryzhkov, 1998; Ven-1 ema et al., 2000).
| Aircraft echoes are characterized by high pow-
| er and velocities (Daniel et al., 1999; Skolnik, ° 1970). Insects reflect low-power echoes whose di-
6 rection and velocity usually coincides with those g of the wind (Hajovsky et al., 1966; Glover and 8 Hardy, 1966; Skolnik, 1970).
Finally, bird echoes are of relatively low power (Z < 30 dBZ), their movement being forward and relatively straightforward. The maximum amplitude fluctuations occur in the low frequency band (below 10 dB within 2-50Hz band). Bird echoes at 10 cm wavelengths are more powerful than those at 3 cm wavelength, their polarization character-
istics being typical of horizontally oriented targets and differential reflexibility considerably exceeding unit (Lack D., 1959; Houghton, 1964; Eastwood, 1967; Chernikov and Schupjatzky, 1967; Chernikov, 1979; Bruderer, 1969, 1997A, 1997B; Bruderer B., 1992; Ganja I., Zubkov M., Kotjazi M., 1991; Russell and Gauthreaux, 1998; Gauthreaux et al., 1998; Miller et al., 1998; Buurma, 1999; Larkin et al., 2002; Gudmundsson et al., 2002; Komenda-Zehnder et al., 2002; Gauthreaux, Bels-er, 2003; Larkin et al., 2002).
In order to obtain data on specific properties of bird echo, we analyzed echo fields obtained by photographing the radar screen with an open objective, within horizontal scans performed at constant vertical angle, both during one scan (Fig. 1a) and during 18 scans (3 min; Fig. 1b).
Fig. 1c shows the same echo field after the digital data processing and data summation over 18 scans.
Comparative analysis of the three figures shows that the main common characteristic of bird echo is its specific movement, resulting in transformation of dotty radio-echo (Fig. 1, a) into streaks (Fig. 1, b). The streaks are relatively straightforward (Fig. 1, b, c). The increments in the length of the streaks take place as the result of the echoes' forward movement from scan to scan. At the same time, the number of echo dots forming most streaks is smaller than the number of scans, hence the straightforwardness of streaks is disrupted by a change of direction. To sum up, bird echoes do move, and this movement has pronounced distinguishing characteristics.
The goal of the present study was:
a) to reveal those characteristics, related to: uniformity of direction and velocity; changes in movement linearity; specificity of echo fluctuation etc.;
b) to analyze the revealed characteristics as a system of parameters and to use these parameters as the basis for an algorithmic system for prompt and accurate bird echo identification.
The first step of the procedure was aimed at identifying moving echoes and setting up the conditions for collecting the relevant data.
3. Primary radio-echo identification on the basis of echo movement
The primary source of data relevant for the task of bird echo identification is radio-echo fields obtained by summation of data collected over a prescribed number of consecutive azimuth scans at several prescribes angles (Fig. 1c), the duration of a single scan being 20 sec.
During the echo field plotting, the primary echo processing is carried out, part of it being the exclusion of noises from the subsequent analysis by the low threshold value (Dinevich et al., 2000).
On the basis of the multiple measurements carried out in the study, a cycle of 8 consecutive scans was found to be the most efficient one. Performing fewer scans leads to loss of valuable data, while more scans mean a significant increase in computation time without seriously contributing
Fig. 1a. Radar screen photograph over a single scan (10-sec exposure)
Fig. 1b. Radar screen photograph over a18 scans (3-min exposure)
to data enrichment (Table 1). The data obtained over 8 scans are taken as 100 % and processed by means of the algorithm developed in the study. The computation time was measured in minutes.
The number of prescribed angles was chosen in such a way that to cover all the bird echo at all
Table 1
Comparison of computation time and data volume obtained over different number of scans (on the basis of 10 measurement cycles)
Number of scans The quantity of identified birds, % Computation time, %
8 100 100
7 70-90 80-90
9 110 120
10 110 150
flight heights that are reachable at the given beam | width. Normally, for 10 cm-wavelength (beam width ^ 1.5°), 4-5 scans are enough during daylight hours, | while at night it was often necessary to perform ^ 6-7 scans. §
Totally, 120 scans were performed during day- ® light hours on different dates in order to trace bird echo movements on the radar screen. The digital processing of the data showed how many times, within a 8-scan cycle (160 sec), the radar registered the same echoes on the basis of their movement at a given coordinate position: once — 82 % cases; twice — 15 % cases and three times — 3 % cases. Repeated registering of the echo from the same bird (bird group) at the same coordinate position, notwithstanding the bird's uninterrupted moving in space, can be due to several reasons: the size of the scanned area, high bird concentrations, the relationship between flight velocities and the discontinuity parameters of the digital system analyzing the echo fields (in our study — 60 m-distance and 0.176° in radial direction). It might happen, for example, that a bird leaves the scanned area during the first scan, while another bird enters the area by the time of the second scan. In yet another example, a slow-flying bird whose echo has been once registered, might not have enough time to leave the scanned area before the next scan, thus its echo is registered for the second time.
Completely different data were obtained on the recurrence of echoes reflected from ground clutter and clouds at the same coordinate position. These echo are usually repeatedly registered at the same point, scan after scan. The only exception is areas yielding weak or heavily fluctuating echo, such as ground clutter/cloud boundaries or areas of poor radar reflectivity. Such areas can be a < source of false echoes (i. e. echoes mistakenly tak- ^ en for bird reflections), and their identification | poses a special problem that has been solved by the > | proposed algorithm (see Stage 3 below). s
On the basis of these findings, the directions | and conditions were determined for collecting the £ relevant data on bird echo for subsequent algorith- | mic processing, as well as the objectives of the first ^ stage of bird echo identification at each coordinate § position of each of the scans, were determined. ©
4. The Algorithmic Procedure
a. Stage 1
For each elevation angle value, radio-echo fields that were obtained over eight scans were summed up, and two data files were built.
Fig. 1c. Radio echo field (identical to that in Fig. 1b) after digital processing and summation over 18 scans at the constant elevation angle. Lines formed by dotted echo represent bird flight tracks, other lines are echoes from hills
The first data file contains charts plotted on the basis of summation of all the radio-echoes obtained during all the scans and at all the prescribed elevation angles (Output 2 in Fig. 2). These charts describe the total presence of all the reflectors under observations, as well as their movements in space. ¡< Fig. 3 shows a sample of such a chart registering ^ the situation at 12:56, August 8, 2002; the radar is S located in the center of the chart. The upward u direction is oriented northward; the scan radius is S 60 km; the long line in the west marks the sea frontier. Bird echoes are represented by:
a) the streak of about 100 km long, southwardly northward oriented;
b) the short southward oriented streaks in the § east sector of the chart and
® c) radio-echoes in the form of separate dots.
Streaks are frequently formed when birds are flying through a cloudless bottom mesofront layer caused by a breeze (Alpert et al., 2000). Echoes in the form of areas, as well as of wide or narrow radial-oriented streaks and in some cases those in the form of separate dots, are echoes reflected from ground clutter.
Being a valuable source of information, charts of this type, however, cannot be sufficient for
Fig. 3. Chart showing summated echo at 12.56, October 8, 2002. Areal and radial-orientated radio echoes are reflections from ground clutter. The long streaks in the center of the chart and the short streaks in the east orientated southwards, as well as dotty signals, are bird echoes. The long streak is about 100 km long
Isolation of bird signals by the fluctuation parameter
Output 2
The threshold for the linearity and
forwardness criteria is raised.
An aaditional criterion for the chaotic character of direction is introduced
Fig. 2. The flowchart of the algorithm for bird identification and estimation of flight velocities
identifying bird echoes, as using them implies a specific knowledge of the terrain.
The second file contains data related to echo movements. Echoes registered at the same coordinate position more times than it is prescribed by a selected number are excluded from the further analysis. This procedure enables to exclude echoes from most of the ground clutter and of dense clouds.
The primary digital processing results in representing echoes in the form of «spots» were formed by a number of signals (points), each point having its coordinates and power value (Dinevich et al., 2004, Section 2, Table 2). Taking into account the power values of the signals forming a particular «spot», the coordinate position of the center for each «spot» is determined by the following dependencies
n In n In
Xc = Z PjxJ Z fj; Yc = Z PfyJ Z Pf, (1) / / / / / / where n is the number of signals (dots) forming the spot, Pi is the power value of the signals, and i is the number of a dot within the «spot». The upper value of i is restricted by the technical capabilities of a particular radar and the digital processing method, being 27 in our study. The low value of i is determined by the noise threshold, being i > 4 in our study according to the measurements. At values i < 4, the number of echo «spots» multiplies by several factors, which significantly increases the computation time and eventually leads to a drop-in of false signals at the end of the filtering procedure.
At Stage 1, as a result of the filtering procedure based on presence/absence of echo movement, a data file is obtained that contains the coordi-
nates of the centers of «spots» representing echoes reflected from moving and/or heavily fluctuating objects. Most of such objects are flying birds, while some may be echoes from ground clutter and clouds, reflecting weak signals.
Further filtering aimed at excluding non-bird echoes is based on specific properties of bird echo movement.
b. Stage2
In course of migration, some birds fly straightforwardly at a uniform speed, while other birds frequently change both the direction and the velocity of flight (Bruderer, 1992, Ganja, et el., 1991). The latter is especially typical of birds taking advantage of thermics. According to our observations, a weak bird echo can periodically escape from the observation range, getting above or below the noise level. ESA (effective scattering area) of an echo from the same bird may change by a factor of 10, depending on its orientation toward the radar beam (Houghton, 1964; Eastwood, 1967; Bruderer, 1969, 1997A, 1997B;). According to the measurements of bird echo ESA, performed inside an anechoic chamber at various angles in relation to the radar beam, the maximum echo value lies within the range of 65° to 115° (Zavirucha et el., 1977). These values correspond to the beam's orientation toward the bird's side (0° corresponding to orientation toward the bird's beak). Bird's wing-flapping may also cause ESA variations, from increasing it by a factor of 10 (relative to the average value) to lowering it almost down to 0. The frequency of the fluctuations in these cases is within the 2-50 Hz range (Chernikov, 1979).
In order to study the ratio between bird echoes reflected head-ward and side-ward, a special experiment was carried out in the present study (photo 1). The 3-cm signal generator that is a part of the regular radar measuring instrument was utilized as a transmitter, in combination with the MRL-5 radar receiver. A recently euthanized pigeon used as the reflector was placed on a platform, in a position providing minimal background echoes. By rotating the platform to have head-ward and side-ward exposures, echo ratios were multiply measured. The maximum echo was obtained at side-ward exposure, and the minimum echo at head-ward/tail-ward exposures (the latter two echo values were found to be quite close). The average ratio was found to be 4.6. Hence, a radio echo of a bird is a function of not only its size and dielectric properties, but also of its wing span and its spatial position relative to the direction of the probing beam.
In view of these considerations, at each elevation angle the proposed algorithm performs three echo-filtering operations taking into account the abovementioned echo movement properties.
b.1. Operation #1: plotting straightforward echo movement segments by a prescribed number of points located within consecutive scans
The coordinate of the center of each echo «spot» obtained in the first scan is assumed to be the point
of origin for a coordinate system. Out of this central point, a bundle of n straight lines can be plotted. The end of each line should not be located beyond the distance range that can be theoretically covered by a bird during the given time of observation (160 sec, 8 scans in our case). The range is determined by the minimum flight velocity (Vmin) and the maximum flight velocity (Vm), which we assumed to be 8 kph and 70 kph, respectively, according to generally known ornithological parameters and taking into account the impact of wind.
Then ¿min = Vmin □ t; Lmax = V^ □ where L is the minimum/maximum distance that can be covered by a bird during the time between the 1st and the 8th scans. Any «spot» obtained in the scans that lies within this distance range is assumed to be the end of a straight line originating from each of the «spots» in the first scan. It is important to take into account that the initial echo from a bird may not necessarily appear within the 1st scan, but within any subsequent scan. In view of this consideration, the procedure is repeated on the data obtained in the subsequent scans, assuming the centers of newly obtained «spots» to be point of origin for bundles of n lines. The procedure is repeated for scans 1-4.
Each of the n straight lines can be described by the dependency
y = ax + b, (2)
where a = (y - y2 )/( - X2 ); b = y2 - axx and yv У2, x^, X2 are the coordinates of echo «spots» in the first and the last scans, respectively. In this coordinate system, the point of origin represents the location of the radar.
As a result of this procedure, a set of straight-line bundles is obtained, having their points of origins in the centers of echo «spots» that were registered for the first time.
At the next step, out of each bundle of n lines, the algorithm chooses m lines that meet certain requirements. Those are the lines on which not less than k signals (k is a prescribed number) are supposed to fall during the subsequent scans. On the basis of our multiple measurements, the following values of k (kmax = 8) were prescribed in our study: within the distance range 0-15 km k = 6; within the range 15-30 km k = 5; within the range of 30-6-km k = 4. These are the values that are most consistent with the radar screen readings, taking into account the controlled deletion provisioned in the procedure of radar signal correction by the distance squared.
b.2. Operation #2: implementing the movement uniformity criterion
Having obtained, as a result of Operation #1, m straight lines plotted from each initially registered echo, the algorithms now selects only those of these lines where the distance between the «spots» centers grows with each subsequent scan, i. e. l1 < l2 < l3 < ... < ln, li being the distance between the «spots» centers obtained in two subsequent scans.
By the lp l2, ... ln dependencies between subsequent scans, we can evaluate the uniformity of
birds' flight velocities and set up certain values for the velocities that can serve as specific criteria. To do this, the system calculates the average bird velocity over the entire observation period and uses it as a benchmark to compare velocities within certain flight legs.
In our study, the 20 % value was chosen on the basis of multiple measurements. In case the deviation of flight velocity from the benchmark value is below 20 % over at least one flight leg, the flight movement was considered to be a uniform one. In contrast, when the said deviation exceeded 20 %, the movement was regarded as a non-uniformed one.
b.3.Operation #3: implementing the movement linearity criterion
Among the set of lines that have remained within the analysis after Operation #2 filtering, the algorithm must now select the unique line where the diametral deviations of echo «spot» centers in consecutive scans are within a certain range of values prescribed by the linearity criterion. This value for a particular echo is assumed to be a certain prescribed length fraction of the direct line that connects the echo from the first scan to its «counterpart» in the last scan. The linearity criterion is fulfilled when the position of echo centers within the intermediate scans do not exceed this prescribed value. In our study, we set a 10 % deviation to be the benchmark for the movement linearity criterion, namely, a deviation up to 10 % of the line length means the movement is straightforward, while values from 10 % to 40 % characterize a non-straightforward movement (it should be noted here that we did not observe deviations of direction exceeding 40% of the said length).
In case of a straightforward movement, the centers of radio-echo fall in close proximity to a direct line. The algorithm searches for the maximum proximity between a line and a set of radioecho centers obtained over a scan cycle. The required line is described by the dependency y = ax + b, where the coordinates of two points are known (y1,
Xp y2, x2), hence a = (y - y2 )/(x - x2 ); b = y2 - ax1, thus the problem is reduced to finding at least two more points meeting the above -stated requirements. To find those points, we drop a perpendicular from the point with coordinates (Xn, Yn) on the said line; the coordinates of the base of this perpendicular
being Xi = (Xn + aYn - ab))l + a2, Yi = aXi + b, where Xi, Yi are the coordinates of the point of intersection of the said line and the perpendicular dropped on it from the point X, Y .
c n n
Hence,
di =J(Xn - x, )2 + {¥„ - y, )2
(3)
where di is the distance between a point and the related line.
This distance should not exceed the value prescribed by the linearity criterion, thus di < < (10^40%) • L, L being the distance between the echo centers in the first and the last scans. There might be several perpendiculars meeting this re-
quirement, out of which the algorithm chooses the one which is the shortest for a given scan.
This procedure completes the filtering that results in plotting a straight line representing a bird's flight. This line has a pronounced direction and, therefore, can be considered to be a vector.
c. Stage 3
At this stage, filtering is aimed at identifying and rejecting false line segments. According to our multiple observations, false line segments are frequently formed in areas of weak echo from clouds, visually non-observable breeze and mountain-plane mesofronts, sub-inversion thermics etc. Echoes from such objects shift in a chaotic, non-directed pattern, in contrast to lines formed of straight segments that are typical of bird echo (Dinevich et al., 2001). Sometimes, however, there might be a certain direction in the shifts of non-bird objects coinciding with wind directions.
We introduced a parameter for assessing the extent of state of chaos in the direction of echo movement (referred to hereafter as «state of chaos» parameter) and used this parameter for identifying possible sources of false echo, thus completing echo filtering. The «state of chaos» parameter shows to what extent the movement of an object monitored by a radar follows a distinct direction and is assessed by the following procedure.
The system looks for a spot which is the point of origin of a vector. Having detected such a «spot», the algorithm assumes it to be the center of a control circle, and plots the circle of a prescribed diameter (300 m in our study, on the basis of multiple measurements). The circle is subdivided into 8 sectors, which are numbered 1-8; number 1 is given to the sector with 0e direction that expands clockwise, number two is given to the adjacent sector clockwise, and so forth.
The orientation of all the vectors located within the control circle is analyzed. At the first step, the system determines the number of vectors which, on the basis of their direction, were found within sectors 1-4. If we assume the number of such vectors to be bi, i being the number of a sector, the total number of vectors found in sectors 1-4 is
(4)
In a similar procedure, the number bj of vec-
4
tors found in sectors 5-8 is calculated, B2 = ^bj.
The ratio between the two sums is 4 /4
k1 =1 b IV
(5)
At the next step, assuming the same echo being the center of the control circle, the circle is turned clockwise, sector #1 now becoming sector #2 and so forth, thus shifting the 4-sector arrangement by one sector; k2 = (n2 + n3 + n4 + n5)/ (n6 + n7 + n8 + n1). In a similar way, k3 and k4 are calculated. Using the values kj, k2, k3, k4, parameter K representing the degree of the «state of chaos» in echo direction is determined
K £ + £2 + + £4
(6)
If K is equal to 1, it indicates that the vectors throughout the control circle are oriented toward all the directions with equal probability, which is defined as a «state of chaos» incompatible with bird migratory flight pattern.
The value of K in case of bird migration is to differ from unit as much as possible. Hence, the system identifies vector with K = 1 as false bird echoes, but does not exclude them from the vector filed before the analysis of all the echo spots is completed. Only after calculating K values for all the vectors, the system establishes the areas of chaotic movement and excludes all the vectors plotted within these areas from the final data file.
The n echo «spots» remaining after this operation are sent for a repeated procedure that calculates K value over n +1 «spots».
5. Determination of bird flight velocities and plotting flight vectors
As a result of Stage 2 and Stage 3 of filtering, straight-line segments are obtained, which are plotted over a prescribed number of echo «spot» centers that are moving within a given period of time. The technique used to plot this line suggests that:
- in case moving of an echo is represented by segments of lines meeting the above-described requirements, this echo belongs to a bird (a bird group);
- positions of echo centers in proximity to the general line which they form are determined by the flight pattern of a particular bird (bird group);
- the orientation of lines relative to the coordinate system whose point of origin is the radar itself is considered to be the bird flight direction.
Having the coordinates of each vector and the time measurements corresponding to both its origin and terminus, we can calculate the average velocity of a bird (bird group) flight over the observation time. Having the coordinates of each echo center and the exact time when the echo was first registered on the radar screen, we can establish the direction and the velocity of flight and determine whether the direction is or is not straightforward.
Using the values of X, Yi, t, root-mean-square linear regression dependencies X(t), Y(t) are built in the centers of the two echoes. The first echo center is related to time t1, the second point is related to time t2, t2 > t1.
The slope ratios of the obtained dependencies X(t), Y(t) provide evaluations of the bird's velocity
components Vx and V. The actual bird (bird group)
x y
velocity is
Vxy =
(7)
The averaged velocity of all the birds within the scanned area is assumed to be the mean velocity
of bird migration flow Vsum and is calculated as
К =У v
sum / >
(8)
It is important to note that thus calculated radar-related bird flight velocity is, in fact, the
sum of two components: a) the velocity of a bird's flight powered by its wing labor and b) the velocity and direction of the wind flow relative to the direction of the bird's flight.
The mean direction of bird migration flow is defined as the geometrical mean of all the flight vectors and is calculated by composition of vectors.
6. Graphic representation of bird flight data against the background of ground clutter and atmospheric inhomogeneities
Upon completing all the stages of echo filtering, a file containing only bird-related data is obtained. After summation of the totality of the data over all the scan angles, we obtain a sum horizontal plane projection of all the vectors that represent the flights of all the birds within the scanned area. If we then chart scale cursors, residence sites, roads, the coast line and other terrain elements on the said projection, we obtain a chart that can be called a radar ornithological chart (by analogy with a weather chart).
A sample of such a chart is presented in Fig. 4 (data obtained at 12:56, October 8, 2002). The chart presents the same ornithological situation as that in Fig.3 after the algorithm has performed the filtering procedure.
As can be seen, the volume of information in Fig. 4 is significantly larger than that in Fig.3. Bird echoes are presented in a vector form. The scan radius is 60 km. The length of the band representing migrating birds is about 100 km and consists of segments differing in vector density. The total quantity of birds (bird groups) is 1711, among them 982 are flying with steady direction, as is demonstrated by the corresponding vectors.
The flight velocity spectra show that the maximum bird velocity (taking into account the wind velocity) is about 70 kph, while the minimum velocity is just over 10 kph. As can be seen from the direction rose, the sum vector is oriented pronouncedly toward the south (182°). The four above-mentioned patterns of echo movement are actually plotted in different colors, but in the monochromatic image in Fig. 4 the colors are substituted for designating symbols shown in the framed scaled-up A section.
The same sum data base is used for plotting the graph of birds' distribution over height within a prescribed scanned area.
A sample of such a chart is shown in Fig. 4 (Section C), where the X-axis indicates the quantity of birds (bird groups) and the Y-axis indicates the height. The graph caption mentions the sub-area within which the distribution over height is shown.
There are several paired arrows in the graph, terminating with numbers. The length of each arrow is proportionate to the number of birds within the corresponding 500m-high layer. The number terminating the bottom arrow in a pair is the number of birds that fly with frequently changing directions, i. e. non-migrating birds; the number terminating the upper arrow in a pair is the number of birds whose flight vectors have been plotted by the
Fig. 4. Ornithological chart. Horizontal distribution of birds over the total flight height. Section A: An enlarged fragment of the vector field. Symbols put at the origin of a vector designate: v — chaotic movement; • — straightforward movement at uniform velocity; x — straightforward movement at non-uniform velocity; ■ — movement with deviations from a straight line and non-uniform velocity. Section C: A fragment of bird distribution over height in 350-15° sector. On the right below: two distributions containing 1) the velocity spectrum and 2) the direction spectrum. On the left below: the total quantity of birds, including those with steadily directed movement, and data reflecting the maximum and the minimum flight velocity values
algorithm, i. e. migrating birds. The graph enables to determine both the height of the maximum bird concentration and the maximum flight height.
Fig. 5 shows a sample of volume visualization of bird distribution over certain terrain areas. The arrow taken out of the drawing indicates the northward orientation; the large dot indicates the position of the radar. In the west, the terrain changes into the coastline, in the east there are hills. The horizontal planes are positioned at different heights.
The proposed algorithm provides on-line determination of the exact coordinate position of
Fig. 5. A sample of volume visualization of bird distribution. The radar is located at the intersection x = 0, y = 0. The horizontal planes are drawn at the heights of 1 km, 2 km, 3 km and 4 km. The arrow is northward oriented. The sea is in the west, is the east there are hills
each bird echo. The databases constituting the two abovementioned files (see Stage 1) make it possible to superpose ornithological and meteorological data and to obtain combined charts that provide the bird monitoring data (in a vector form) simultaneously with the data on ground clutter and atmospheric in-homogeneities.
Fig. 6 (a) shows a sample of a combined chart obtained at 8.20, October 10, 2002. For the sake of comparison, Fig. 6 (b) shows the corresponding chart showing radio-echoes after Stage 1 of the filtering procedure, i. e. echoes selected by the movement characteristic alone, without the subsequent filtering accurately identifying bird echoes.
Fig. 6 (a, b) presents a finalized chart based on the data obtained throughout all the stages of the above described filtering procedure. It contains a vast corpus of various data demonstrating the advantages of the proposed algorithm, as can be seen from the list below.
1. Total quantity of birds within 30 km-radius from the radar is 2770 (charts can be plotted up to 60 km-radius).
2. Number of vectors (i. e. birds flying in distinct direction) is 1670.
3. Maximum radar-related flight velocity is approximately 68 kph; minimum radar-related flight velocity is approximately 13 kph.
4. The majority of the birds fly at 45-50 kph velocity.
5. Average flight velocity is 44 kph.
6. Most of the birds fly at varying velocity, deviating from a straight line; however, there is a significant number of birds that deviate from a straight line but fly at uniform velocity.
7. Within the scanned area, one can distinguish groups differing in bird concentration.
8. Birds tend to by-pass clouds, flying around the perimeter or «diving» into gaps between separate cloud cells (as can be seen in the area occupied by two cloud cells in the southern part of the chart).
9. Within 180-270° sector, the maximum bird concentration is observed in the ground — 500 m-high layer (209 bird groups); a significant number of birds fly within the 500-1000 m layer (150 bird groups); a small number of bird groups were observed within 1000-3000 m height, only 13 bird groups were registered at heights of 2000-3000 m.
10. Preferred flight direction is 177°, while there is a small number of birds flying in the reverse direction.
The latter phenomenon has been observed in a number of studies (Komenda-Zehnder et al., 2002).
All the data can be obtained from the scanning of the whole hemisphere (360°) every 10-15 minutes and delivered to users on-line. It is note-
worthy that the described charts provide only the general information on atmospheric inhomogeneities, namely, their location and shift dynamics. More complete data on clouds, precipitation etc. is provided by radar meteorological charts of a different type. The proposed algorithm plots these charts in parallel with the ornithological charts, while radar data for the former charts are collected by means of a specifically designed software. Within this task, one of the conditions for data collection is antenna elevation up to 85°. Fig. 7 shows the horizontal distribution of echo from clouds and precipitation at the level of maximum echo power. The right part of the figure shows the distribution of echo powers over the vertical structure of a cloud cell.
7. A technique for identifying bird
echo by the fluctuation parameter
In certain tasks in the field of air traffic control, there arises a necessity of making a prompt (1-2 min) and highly accurate decision concerning the ori-
Fig. 6a. Radar ornithological chart. Date: 21.10.02. Time: 8:20. The designating symbols area the same as in Fig. 4.
nal of maximum amplitude, to store it in the memory and to accumulate such signals into a database. Using this database, the corresponding software program plots amplitude and frequency spectra over 10-20-second samples.
Since MRL-5 operation is characterized by high recurrence of direct impulses (500 impulses per second), it is possible to obtain signal power spectrum of 5000-10000 signals over each sample. This data enabled to calculate specific spectra typical of different categories of objects.
Fig. 8 (a, b, c,) shows typical oscillograph patterns of echoes reflected from: a single bird (a), a cloud (b), ground clutter (c). As can be seen in the figure, each of the three oscillographs has its individual characteristic fluctuation pattern. The frequency estimation of the oscillographs showed that in the low frequency area (below 50 Hz) the
Fig. 6b. Radio-echoes from all the categories of reflecting objects: arrows indicate zones of echo from clouds; echoes in the form of «spots», of dotty and radial-orientated lines are reflections from ground clutter; dotted and firm lines represent bird echoes
gin of a specific echo registered on the radar screen. As has been shown in the present study, it can be effectively done by means of the proposed algorithm in the mode of halted antenna.
On the basis of data on the fluctuation radio echoes from different categories of reflectors obtained in the previous study (Dinevich et al., 2004), a special electronic device was designed. It enables, within a preset 200 m-long strobe and within each reflected direct impulse, to isolate the sig-
Fig. 7. Radio-echo chart within the horizontal section of clouds and precipitation at the level of maximum reflective power. The right part of the chart shows the vertical section of a cloud cell. The graph presents vertical distribution of echo power within this cell
spectrum of amplitude fluctuations of bird echo is considerably wider than other spectra. A specially adjusted low-frequency filter allowed, with confidence above 80 %, to classify echo fluctuation spectra as «bird' or «not bird». Moreover, in cases when the signal is reflected from a single bird, the confidence reaches 95 %. Fig. 9 (a, b) presents a sample of data obtained by the low-frequency filter over two typical echo spectra.
In order to perform the echo fluctuation analysis, it is necessary to collect the required number of echoes from a desired target with radar antenna being halted for 10-20 sec. This mode is not compatible with the conditions of echo collecting for identifying birds over vast areas. For this reason, it appears reasonable to perform the echo fluctuation analysis while dealing with particular tasks demanding a prompt and highly accurate «bird/not bird» decision concerning one specifically targeted echo. The proposed algorithmic system includes a software device for such solutions.
8. Results and discussion
a. Control procedure providing data verification
In order to assess the accuracy of the filtering process identifying bird echo against the background of other reflectors, we designed s special
09:47:03 18/10/2002 Time changes in radioecho amplitude (oscillogram)
70 60
x104
Bird
12:51:46 18/10/2002 Time changes in radioecho amplitude (oscillogram)
2.5 2.6 Time, c
11:37:21 31/03/2002 Time changes in radioecho amplitude (oscillogram)
2.4 2.5 2.6 Time, c
c
Fig. 8. Characteristic oscillographs of echoes reflected from: a — birds; b — clouds; c — ground clutter. The X-axis shows time of signal accumulation, the Y-axis show the signal amplitude in relative units
0 200 400 600 800 1000 1200 1400 1600 1800 D:\fluctdat\2002.10.15\Birds\2002.10.15 09.35.01 Ch.2 D.13.5 km.dat
Not a bird
2000
1500
500
0 200 400 600 800 1000 1200 1400 1600 1800 D:\fluctdat\2002.10.14\Clouds\2002.10.14 12.44.32 Ch.2 D.37.5 km Az = 23 deg Um =12 deg.dat b
Fig. 9. Typical frequency spectra of birds and non-bird echoes
control algorithmic procedure enabling to trace back the echo movement and to repeat plotting of individual vectors, as well as of the entire vector field. This procedure can be used to verify data obtained by the proposed technique against any control data, when needed.
Fig. 10 (a) shows a comprehensive ornithological chart in the form of a vector field plotted 29.08.2003, when a large stork flock flew over Israel. The length of the bird band exceeds 100 km. The small square in the southern sector of the chart presents a sample framed for more detailed analysis, chosen for the sake of simplicity as it contains only two vectors. The large framed square presents those two vectors together with echo 'spots' whose centers were used to plot them. The calculated echo centers are marked with dots. The point at the origin of each vector coincides with the center of the initially registered «spot», i. e. with the moment the bird was first registered in the given observation series. The length of a vector is proportionate to the flight velocity. Fig. 10 (b) (drawings 1-6) shows the dynamics of these echo movement from scan to scan. The first (northern) vector was plotted over five echoes «spots» that were moving uniformly but not straightforwardly. It disappeared in the 3rd scan, being apparently below the noise level. The movement of this bird (bird group) is straightforward but not uniform velocity-wise. The second (southern) vector is plotted over six echo «spots» moving uniformly from scan
a
2500
0
3
4
lime, c
a
b
Fig. 10. Stages of plotting bird flight vectors. Ornithological chart obtained at 12.11, August 29, 2003: a — the scanned area; b — the patterns of echo movement used as the basis of plotting two vectors; drawings 1-6 show the patterns of echo movement from scan to scan
to scan. The movement of this bird (bird group) is uniform but not straightforward.
Analysis of this kind enables to trace back the filtering procedure within any sector in order to check the values of each parameter used for decision-making during the filtering.
b. The Scheme of the Algorithm
Fig. 2 shows the flowchart of the algorithm developed in the present study for identifying birds and estimating their flight velocities. To sum up, the main stages of the algorithmic processing of radio-echo fields, described in detail in the preceding sections of the paper, are as follows:
- summating the totality of radio-echo (that are above the noise level) over a prescribed number of scans;
- isolating each bird (bird group) echo against echoes from other reflectors on the basis of echo movement and specific parameters of this movement;
- calculating flight velocity for each bird (bird group);
- excluding false vectors by implementing a special analysis of vector fields on the basis of additional parameters;
- identifying, semi-automatically and promptly, individual target selected radio-echoes («bird/non-bird») on the basis of the fluctuation parameter in halted antenna mode.
Having completed the identification procedure, the algorithm plots ornithological and meteorological charts of various types.
c. Testing the algorithmic system The proposed method was tested for accuracy, using photography of the radar screen as a source of control data for comparison.
Fig. 1 (a, b) shows a sample of radar screen photography. In the first case (Fig. 1, a), the bird echoes are represented by separate dots. In the second case (Fig. 1, b), the bird echo dots form tracks, while echoes from static reflectors remain in the dot form. Comparing the two photographs enables to identify bird echoes with high confidence, the error being caused solely by the radar capabilities and the precision of the photography. Therefore, juxtaposing the data obtained via photographing with those obtained by the digital echo filtering system, one can estimate the comparative error. In addition, one can compare the calculated velocities of bird flights.
To do this, a prescribed number of tracks are randomly chosen in the photograph, after which the lengths of the tracks are calculated. The ratio of track lengths to the total scan time (3 minutes) yields the mean flight velocity. However, we should not expect here a complete agreement of data, due to a number of reasons. First, there is a 35 min time gap between the computer data collection and obtaining the screen photograph. Second, counting the number of birds in a photo is performed manually. Finally, the radar screen representation in a photograph depends significantly on the quality of photography.
An example of juxtaposing the data obtained by the two techniques is presented in Table 2. The data were obtained and processed by the algorithm at 8.30 p.m. and 9.15 p.m. 21.10.2002 at two elevation angles (1.0° and 3.0°). At the same angles, with a time interval of 3-5 min, the corresponding echoes were photographed.
Table 2 shows a good agreement between the data obtained by the two techniques, which suggests that the proposed algorithm identifies bird echo with high confidence. The total time required for the algorithm to identify bird echoes according to all the criteria described above, to calculate flight vectors and to plot different kinds of ornithological charts does not exceed 12 min in the daytime and 15 minutes at night when bird migration is intensive. It should be noted here that a similar procedure performed by the previously developed algorithm (Dinevich et al. 2004) took considerably more time, sometimes by factor of 10.
Table 2
Comparing the data on the quantity of birds and their flight velocity: digital techniques vs. photography
Data items First test, 8.30 a.m. Second test, 9.15 a.m.
Identifying technique Digital Photographic Digital Photographic
Quantity of birds 1.2S5 1.154 194 153
Maximum flight velocity 65.3 kph 59.4 kph 46.4 kph 43.0 kph
Minimum flight velocity 12.2 kph 10.3 kph 11.6 kph 11.2 kph
Mean velocity 40.5 kph 3S.5 kph 35.3 kph 35.l kph
Í Í CU C
a u
"ñE
E:
cu 9. Conclusions
1. The study enabled to establish the follow-& ing characteristics of bird radio echo that distin-g guish it from radio echo of other origins, namely. @ - bird echoes shift with continuity and forwardness;
- in case of bird migration, the coordinates of echo centers form an approximately straight line over separate flight legs;
- amplitude fluctuations of bird echo reach their maximum values at low frequencies (amplitudes below 10 dB are within the 2-50 Hz frequency range).
2. These characteristics, confronted with typical characteristics of radio echoes from other categories of objects described by a number of researches, were used as the basis for the proposed algorithm. The central procedure of the algorithm is on-line plotting of vector fields that represent the movement of a singe bird or a group of birds.
3. The technique of vector field plotting enables not only to identify bird echoes, but also to assume, with high confidence, that a given bird (group of bird) belongs to a certain species, e. g. frequent shift in flight direction characterize local non-migrating birds; straightforward flight at steady velocity is observed in wild ducks, geese, cranes, pelicans; straightforward flight at nonvariable velocity is typical of passerines, oscines; flight with repeating deviation from a straight line and variable velocity — of storks and eagles.
4. Comprehensive charts plotted on the basis of the algorithm data enable to obtain diverse and relevant information on the ornithological situation within the area of 60-km radius from the radar position, including:
a) the total current quantity of birds in the ® air, specifically migrating birds; | b) maximum and minimum flight velocity values;
,jj c) distribution of birds' mass throughout the
§ height;
I d) the spectra of flight directions and veloci-
ty ties, including the sum direction vector; p e) vector fields of bird movement juxtaposed with
I" current meteorological status and local terrain; g f) bird distribution by the flight pattern (the
™ degree of straightforwardness and velocity steadiness);
g) data on clouds, precipitation and visually unobservable atmospheric inhomogeneities, including their evolution with time, i. e. from scan to scan.
It is especially important to note that the whole process of collecting and analyzing the radar data, plotting the comprehensive charts and delivering
the results on-line to air traffic control services aviation does not exceed 12 min in the day time and 15 min at night when bird migration is especially intensive.
5. Bird identification by the fluctuation parameter can be used in tasks where there is a need for an especially prompt (1-2 min) and accurate «bird/not bird» decision.
6. In view of the fact that MRL-5 radars are located in many countries and cover an extremely vast territory, it appears expedient to connect them into a network. Using the algorithmic system for bird echo identification by means of MRL-5 radar, such a network could perform intercontinental bird monitoring in the real-time mode, contributing to providing collective air traffic safety.
7. The algorithm for bird identification proposed in the study can be implemented in other types of high-grade potential incoherent radars whose antennas produce narrow-directivity beams.
Acknowledgements
The authors express their gratitude to the Ministry of Defence of Israel for financing the project. Our thanks go to A. Kapitannikov, Dr. M. Pinsky and Dr. A. Sterkin for their assistance in fulfilling a number of project tasks; to O. Sikora and V. Garanin for providing radar maintenance; to Dr. Roger Reinking and Dr. Elena Negnevitsky for reviewing the paper and making valuable remarks.
References
Abshayev M., Burtsev I., Vaksenburg S., She-vela G. (1980). Guide for use of the MRL-4, MRL-5 and MRL-6 radars in urban protection systems. Leningrad: Hydrometeoizdat (in Russian).
Abshayev M., Kaplan L., Kapitannikov A. (1984). Form reflection of meteorologic targets at the primary processing of the meteorologic radar signal // Transactions of VGI. Vol. 55. P. 26-37 (in Russian).
Alpert P. et al. (2000). Migrating soaring birds align along sea-breeze fronts; First evidence from Israel // Bull. Amer. Meteor. Soc. Vol. 81(7). P. 1599-1601.
Atlas D. (1967). Advances in Radar Meteorology. Leningrad: «Hydrometeoizdat» (in Russian).
Battan L. (1963). The vertical velocities of angel echoes // Proc. 10th Weather Radar Conf., Am. Meteorol. Soc., Boston. P. 309-315.
Battan L. (1973). Radar Observation of the Atmosphere. University of Chicago Press: Chicago.
Bruderer B. (1969). Zur Registrierung und Interpretation von Echosignaturen an einem 3-cm-Zielv-erfolgeradar // Ornithol. Beob. Vol. 66. P. 70-88.
Bruderer B. (1997A). The study of bird migration by radar. Part I. The technical basis / / Naturwissenschaften. Vol. 84 P. 1-8.
Bruderer B. (1997B). The study of bird migration by radar. Part II. Major achievements / / Naturwissenschaften. Vol. 84. P. 45-54.
Bruderer B. (1992). Radar studies on bird migration in the south of Israel // BSCE. Vol. 21. P.269-280.
Buurma L. (1999). The Royal Netherlands Air Force: Two Decades Of Bird Strike Prevention «En Route» // Int. Seminar on Birds and Flight Safety in the Middle East, Israel, 25-26 April. P. 71-83.
Chernikov A. (1979). Radar clear sky echoes. Leningrad: Hydrometeoizdat. P. 3-40 (in Russian).
Chernikov A., Schupjatzky A. (1967). Polarization characteristics of radar clear sky echoes // Proc. USSR Academy of Sciences: Physics of the Atmosphere and the Ocean. Vol. 3(2). P. 136-143 (in Russian).
Daniel C. J. et al. (1999). Accuracy of 404 MHz radar profilers for detection of low-level jets over the central United States // J. Appl. Meteor. Vol. 38(9). P. 1391-1396.
Dinevich L., Kapitalchuk I., Schupiatzky A. (1994). Use of the polarization selection of radar signals for remote sounding of clouds and precipitation // 34th Israel Annual Conf. on Aerospace Science. P. 273-277.
Dinevich L., Leshem Y., Gal A., Garanin V., Kapitannikov A. (2000). Study of birds migration by means of the MRL-5 radar // J. Scientific Israel - Technological Advantages. Vol. 4. P. 94-104.
Dinevich L., Leshem Y., Sikora O. (2001). Radar Observation Analysis of Season Bird Migration in Israel at Night // J. Scientific Israel -Technological Advantages. Vol. 3. P. 77-101.
Dinevich L., Leshem Y., Pinsky M., Sterkin A., (2004). Detection of Flying Birds and Estimation of their Velocity Vectors by MRL-5 Meteorological Radar // J. Scientific Israel - Technological Advantages. Vol. 6. P. 31-46.
Dinevich L., Leshem Y., Pinsky M., Sterkin A., (2004). Detecting Birds and Estimating their Velocity Vectors by Means of MRL-5 Meteorological Radar. The Ring // Int. Ornitological J. Vol. 26(2). P. 35-53.
Doviak R. J., Zrnic D.S. (1993). Doppler Radar and Weather Observation. San Diego: Academic Press Inc.
Eastwood E. (1967). Radar Ornithology. London: Methuen and Co.
Gauthreaux S. A., Belser C. G. (2003). Radar ornithology and biological conservation / / Auk. Vol. 120(2). P. 266-277.
Ganja I., Zubkov M., Kotjazi M. (1991). Radar Ornithology. Stiinza. P. 123-145 (in Russian).
Gudmundsson G. A., Alerstam T. et al. (2002).
Radar observations of Arctic bird migration at the Northwest Passage, Canada // Arctic. Vol. 55(1). P.21-43.
Gauthreaux S. A., Sidney A. Jr., David S. Mizrahi, Carroll G. Belser (1998). Bird Migration and Bias of WSR-88D Wind Estimates // Weath- | er and Forecasting. Vol.13. P. 465-481.
Glover K., Hardy K. (1966). Dot angels: in- f sects and birds // Proc. of the 12th Weather Ra- 8 dar Conf., Amer. Met. Soc., Boston. P. 264-268. |
Houghton E. (1964). Detection, recognition | and identification of birds on radar / / World Conf. ^ Radio Met., Amer. Met. Soc., Boston. P. 14-21. |
Hajovsky R., Deam A., La Grone A. (1966). * Radar reflections from insects in the lower atmo- S sphere // IEEE Trans. On Antennas and Propaga- ® tion. Vol. 14. P. 224-227.
Komenda-Zehnder S., Liechti F. et al. (2002). Is reverse migration a common feature of nocturnal bird migration? An analysis of radar data from Israel // Ardea. Vol. 90(2). P. 325-334.
Lack D. (1959). Migration across the North Sea Studied by Radar. Part 1. Survey through the year // Ibid. Vol. 101(2). P. 209-234.
Larkin R. P., Evans W. R. et al. (2002). Nocturnal flight calls of Dickcissels and Doppler radar echoes over south Texas in spring. Vol. 73(1). P. 2-8.
Lofgren G., Battan L. (1969). Polarization and vertical velocities of dot angel echoes // J. Appl. Met. Vol. 8. P. 948-951.
Miller M.A., Verlinde J. et al. (1998). Detection of nonprecipitating clouds with the WSR-88D: a theoretical and experimental survey of capabilities and limitations // Weather and Forecasting. Vol. 13(4). P. 1046-1062.
Russell K. R., Gauthreaux S. A. (1998). Use of weather radar to characterize movements of roosting purple martins // Wildlife Society Bulletin. Vol. 26(1). P. 5-16.
Skolnik M. (1970). Radar Handbook. McGraw-Hill, 1970.
Shupiatsky A. (1959). Radar dispersion by non-spherical particles // Proc. CAO. Vol. 30. P. 39-52 (in Russian).
Venema V., Russchenberg H. et al. (2000). | Clear-air scattering observations: downdraft and t angles // Phys. and Chem. of the Earth. B: Hy- £ drology. Vol. 25(10-12). P. 1123-1128.
Zavirucha V., Saricev V., Stepanenko V., She- § pkin U. (1977). Study of the dispersion characteris- I tics of the meteorological and ornitological objects | in echo-free cameras / / Proc. Main Geophysic Ob- | servatory. Vol. 395. P. 40-45. £
Zrnic D. S., Ryzhkov A. V. (1998). Observa- g tions of insects and birds with a polarimetric ra- & dar // IEEE Transactions on Geoscience and Remote Sensing. Vol. 36. P. 661-668.
