Springer 2006

Aquatic Ecology (2006) 40:463479

DOI 10.1007/s10452-005-6567-z

-1

Basics and principles of particle image velocimetry (PIV) for mapping biogenic
and biologically relevant ows
Eize J. Stamhuis
Department of Marine Biology, University of Groningen, P.O. Box 31, 9750 AA Haren, The Netherlands;
(e-mail: [email protected]; phone: +31-50-3632078)
Received 9 September 2004; accepted in revised form 26 April 2005

Key words: Benthic boundary layer, BIOFLOW, Biogenic ows, Flow analysis, PIV

Abstract
Particle image velocimetry (PIV) has proven to be a very useful technique in mapping animal-generated
ows or ow patterns relevant to biota. Here, theoretical background is provided and experimental details
of 2-dimensional digital PIV are explained for mapping ow produced by or relevant to aquatic biota. The
main principles are claried in sections on ow types, seeding, illumination, imaging, repetitive correlation
analysis, post-processing and result interpretation, with reference to experimental situations. Examples
from the benthic environment, namely, on lter feeding in barnacles and in bivalves, illustrate what the
experiments comprise and what the results look like. Finally, alternative particle imaging ow analysis
techniques are discussed briey in the context of mapping biogenic and biologically relevant ows.

Introduction
Particle image velocimetry (PIV) has been developed from the early 1980s onwards to map complete ow elds instantaneously (Adrian 1991).
The novelty and great advantage of PIV are that it
delivers high resolution ow velocity vector
information of a whole plane in the ow in one
time. All other methods known until the 1980s
delivered point values, or low resolution vector
diagrams of whole planes after repeating the
experiment several times. Consequently, PIV has
given new impulse to uid dynamics research, in
particular at unsteady ows that were dicult to
map instantly beforehand (Rael et al. 1998).
PIV has been developed in the elds of fundamental and applied uid dynamics research and
mechanical engineering. It has therefore mainly
been used to study experimentally induced ows,
such as water ow in conned areas or around
streamlined objects, air ow around wing proles

and plane models, etc. (Adrian 1991; Stanislas

et al. 2000). Since 1993, PIV has been applied to
biogenic or biologically important ows, using
systems that were either developed for this purpose
(e.g. Stamhuis and Videler 1995; Muller et al.
1997) or adapted or tuned to the particular circumstances (e.g. Drucker and Lauder 1999; Zirbel
et al. 2000). Due to its non-intrusive character,
PIV can be characterised as an ideal tool for
studying animal-generated ows, restricting disturbances and thereby non-natural behaviour to a
minimum.
For the benthic zone, information on methods
for ow studies, on ow facilities and on results
are being exchanged within the European BIOFLOW network. This paper contributes to this
exchange by explaining the basics and principles
of PIV, being one of the more recently developed
ow mapping methods that can well be applied to
study ows in the benthic zone. It will focus on 2dimensional (planar) PIV and is geared to scien-

464
tists that have no real experience with PIV and do
not have a degree in physics, uid dynamics or
mathematics. For detailed background information on PIV, the following references may prove
valuable: Particle Image Velocimetry, a practical
guide (Rael et al. 1998), Particle Image Velocimetry (Hinsch 1993), and the Special Issue on
PIV of Measurement Science and Technology,
Volume 8 Nr 12 (Jones 1997), in particular the
rst paper in this volume, on the Fundamentals of
Digital PIV (Westerweel 1997). Digital PIV, theory and application (Westerweel 1993) is recommended for the theoretical backgrounds of
Digital PIV (DPIV). Details on DPIV applied to
animal generated ows and on the specic problems that one can run into when working with
animals in ow studies, can be found in Stamhuis
and Videler (1995), and Stamhuis et al. (2002).
The nature of ow phenomena around biota in
general and the basic relationships between
organisms and their uid environment are very
well explained by Steven Vogel in his book Life in
Moving Fluids, the physical biology of ow (Vogel
1994) which is recommended for basic background reading.
The following subjects will be covered successively: (1) 2D-PIV in a nut-shell, describing what
2D-PIV basically comes down to; (2) The basics
and principles of 2D-PIV, with explanatory sections on every step when applying 2D-PIV; (3)
Biological examples, shortly covering two examples from our studies on biogenic ows, (4) Other
PIV-techniques, with an overview and discussion
of recent developments such as Micro-PIV, Stereo-PIV, Holographic PIV, Super Resolution PIV
and 3D Particle Tracking Velocimetry (3D-PTV);

and (5) Conclusion. The treatment of sections 1,

2 and 3 is based mainly on of the authors own
experience with reference to the literature and
section 4 is a brief overview of the relevant literature.

PIV in a nut-shell
2D-PIV can be dened as Mapping of average
displacements of groups of tracer particles over a
short time interval in a uid ow by correlating
sub-images of two successive images of one illuminated plane of that ow. In essence, 2D-PIV
requires a ow seeded with particles, bright illumination of one plane of that ow and imaging of
the particles in the plane. Successive or doubleexposed images are analysed for displacements of
groups of particles to derive vector ow elds. The
ow elds can be post-processed to derive additional parameters describing e.g. the local character of the ow. A generalised PIV set-up is shown
in Figure 1.
To nd the displacements of groups of particles
over a short time interval, two successively recorded images are compared. A small sub-area of
the rst image, usually called interrogation areas
(IA), is compared with a sub-area at the same
location in the second image using a cross-correlation technique. This results in the most probable
displacement vector for that particular particle
pattern (Figure 2). The process is repeated for all
interrogation areas of the image pair resulting in a
complete vector diagram of the ow studied.
(Adrian 1991; Willert and Gharib 1991; Stamhuis
and Videler 1995).

Figure 1. Diagram of a generalised PIV setup showing all major components: the experimental tank with the particle seeded uid, a
laser producing a light sheet that illuminates only one plane in the uid, a camera imaging the particles in the sheet in the area of
interest, and a PIV system that analyses the images to derive a vector representation of ow eld.

465

Figure 2. Diagram of the steps in PIV analysis of successively recorded particle patterns in a ow: two sub-images from the same
location of two frames are compared in a cross-correlation procedure, resulting in a 2D probability density distribution which shows a
peak at the most probable displacement. The peak is located with high precision and a velocity vector, representing the average
displacement of the particles in sub-image 1 compared to sub-image 2 is calculated.

Basics and principles of PIV

Types of ow
Particle image velocimetry can be applied to virtually any kind of ow, as long as the uid is transparent to enable imaging of suspended particles.
The uid may be moving, as in a ow tank or wind
tunnel, or standing still, as in e.g. an aquarium in
which an organism itself is moving. The uid may be
liquid or gaseous. Liquids such as water, but also
mineral oil or glycerine, are easily seeded with neutrally buoyant particles and sinking of particles may
only become relevant at very low ow speeds. Liquids allow a broad Reynolds range (Stamhuis et al.
2002), ranging from 10 2 for small structures in still
water, or even lower in oil or glycerine, to 107 for
intermediate speed ow tanks (Vogel and LaBarbera 1978; Vogel 1994). Gaseous media such as air
are of course necessary when the ow around life
organisms normally living in air is to be studied, as
in e.g. ying birds (Spedding et al. 2003).

Seeding
To visualise ow for PIV purposes, the uid has to
be seeded with particles. To follow the ow accurately, these particles need to be neutrally buoyant

and small with respect to the ow phenomena

studied, (Rael et al. 1998). For liquids, a whole
series of polymer powders or particulate half-products for plastics with a range of densities are available, and can be matched to the uid density. In
water, e.g. polystyrene (unexpanded), polythene,
nylon, PVC or pliolite particles work well and can
be purchased from chemical companies (e.g. BASF,
Dow, Dupont, Goodyear) in a range of diameters.
The typical particle size for uid ow applications
ranges from 5 to 200 lm, depending on ow speed
and camera magnication. Also hollow glass balloons in micrometer sizes (e.g. Nortek A.S, Potters
Industries), sometimes coated with a reective layer,
can be used in aqueous ows.
Seeding is more dicult in low speed gas ows
because particles tend to sink due to the relatively
high density dierence. The seeding particles thus
need to be very light and small, e.g. very small
polystyrene (expanded) beads (diameter ca. 10
5 lm). Alternatively, Helium lled soap bubbles of
a few mm in diameter have been used to map ow
around ying birds (e.g. Spedding 1986). In high
velocity air ow, e.g. in wind tunnels, vegetable oil
or water droplets produced with an aerosol generator in diameters between 1 and 10 lm are
commonly used (Rael et al. 1998; Spedding et al.
2003). For biological applications, the particles
should not be toxic, or clog up the respiratory

466
structures of the organisms involved (see also
Stamhuis et al. 2002). In all cases the particles
should preferably be highly reective, to yield
good particle images on the PIV recordings.

Illumination
In PIV ow studies, illumination is usually provided by a laser light sheet. A narrow sheet of light
is used to dene the plane of investigation. It
prevents blurry imaged particles just above or below the focussed plane, which would pollute the
PIV images and thereby adversely aect the analysis result. A laser is used because it can deliver a
bright sheet with almost constant thickness without aberration or diusion, due to the coherent
and monochromatic character of the emitted light.
The slightly diverging light beam produced by a
laser is usually transformed into a sheet by converging it with a weak positive lense and subsequently make the beam fan out in one plane to a
sheet by an additional cylindrical lense (Figure 3).
This results in a sheet with a slightly converging
thickness, enabling the experimenter to select a
certain sheet thickness tuned to the experimental
conditions. Sometimes additional optics are added
to further condition the light sheet to the application (Rael et al. 1998). As a consequence of the
sheet illumination, the main ow direction and the
light sheet have to be aligned. This prevents
the particles that are illuminated and pictured in
the rst image from disappearing from the illuminated plane and being replaced by other, differently arranged particles in the second image.
The types of lasers used for PIV illumination can
be divided in two categories: continuously emitting

lasers, usually called CW (= continuous weight)

lasers, and pulsating lasers. The CW lasers, e.g. HeliumNeon and Argon lasers, produce relatively low
power light (Pmax = 0.01 50 W) of good beam
quality and are easy to set up and maintain. The light
sheet is continuously visible to the experimenter and
to the experimental animals. Pulsed lasers, e.g.
Nd:YAG lasers, however, can produce much higher
illumination levels per pulse with very short intervals
between two pulses. They are, therefore, highly recommended in high speed applications where short
illumination times and a high pulse frequency are
mandatory. Pulsed lasers such as Nd:YAG lasers are
usually more expensive and more dicult to set up
due to the added timing and synchronisation
equipment compared with CW lasers.

Imaging
The illuminated plane in the particle seeded ow is
imaged by a camera with its optical axis perpendicular to the plane, in order to record the whole plane
in focus. High quality lenses are generally necessary
because the light levels are usually relatively low and
large apertures have to be used. In normal 2D-PIV,
a black-and-white (B/W) camera suces.
Until about 10 years ago, photographic cameras
were used to record the PIV images, so that evaluation and analysis could only be done after the
lm was developed. The analysis included scanning
of all interrogation areas of the (double-exposed)
image with a separate light source and an electronic
camera, taking a considerable amount of time.
Today, electronic cameras, with a CCD or CMOS
pick-up device, are the main imaging media. They
provide immediate feedback on image quality and

Figure 3. Basic optical arrangement to produce a light sheet from a laser beam: The mildly divergent laser beam coming from the left
(b) is collimated by a weak positive lense (L) and subsequently fanned out in one plane only by a cylindrical lense (C). This results in a
sheet (s) with a slightly converging thickness.

467
Table 1. Recommended camera system depending on the speed
and the repeatability of the event studied, and the illumination
intensity level.
Event
speed

Illumination
int.

Repeatability

Camera type

Fast

High

Slow

Low

Good
Poor
Good

Double-exposure
HS-video
Progr. scan video
(normal/MS)
Progr. scan video
(normal/MS)

Poor

enable instant analysis (Willert and Gharib 1991).

Electronic cameras are available in a whole range
of resolutions, sensitivities, shutter speeds, frame
rates and image transfer protocols (analog or digital). Experimental characteristics such as event
speed, repeatability and illumination intensity need
to be evaluated for choosing a camera for a specic
application, often in combination with the illumination system. Table 1 gives a summary of the
most occurring experimental situations and the
recommendable camera system. The choices are in
general: progressive scan normal video rate camera, medium speed progressive scan camera, high
speed video system, or double exposure camera.
A progressive scan normal video rate camera
provides a stream of video images at a rate of
25 30 fps. The images are usually recorded in
(dedicated) memory on a host computer which
also controls the camera. Each pair of images out
of this stream can be used for ow analysis. Due
to the relatively long inter-frame period (0.03
0.04 s), this camera can only be used for low
speed PIV applications. A medium speed progressive scan camera can record a stream of video
images at a rate up to (ca.) 100 fps. Because such
a camera is often an enhanced model of the
normal progressive scan camera, recording and
control are usually performed in the same way.
Frame intervals of 0.01 s are possible, enabling
higher event speeds than the normal camera. A
high speed video system can usually record up to
1000 fps at full sensor resolution and much
higher rates at reduced image sizes. This camera
is suitable for fast PIV events, especially those
with poor repeatability, which is often the case
for animals. Any image pair of a series can be
used for ow analysis, enabling mapping of unique ow velocity distributions in both space (2D)
and time. High speed video in combination with a

PIV analysis system is also known as timeresolved PIV.

A double-exposure camera is capable of
recording two frames with a very short inter-frame
period. After recording, both frames are transferred to the host system and the camera is prepared for the next image pair. Inter frame times of
below 500 ns are possible, but inter pair times can
be quite substantial, in the order of 0.1 0.5 s (!),
depending on the image transfer system. Consequently, double-exposure cameras are well suited
for high speed repeatable ow events, or when
only snap-shots of a ow are desired. Due to the
architecture of the camera shuttering system, a
double-exposure camera has to be used in combination with a pulsed laser.
Most of the camera types mentioned are
available with sensor resolutions of 512 512
pixels or 1K 1K, or higher. An image with a
high resolution usually results in vector ow eld
with a high resolution or high accuracy. However, alternative analysis schemes, such as discrete
window osetting (Westerweel et al. 1997) or
recursive local correlation (Hart 1999), can
greatly enhance the results from a lower resolution recording system.
A low-cost alternative for a double-exposure
camera + pulsed laser, combining the advantages
of double-exposure, continuous video stream and
a CW laser, may be to use an external fast shutter.
The shutter should allow illumination of every rst
image just before the imaging device (e.g. CCD) is
blocked and the second image immediately after
the imaging device is opened again after frame
transfer. Such a system needs external triggering
tuned to and synchronised with the normal or
medium speed progressive scan camera, making it
somewhat more complicated. The shuttering may
result in very low image illumination levels and
therefore a powerful CW laser is essential.

PIV-analysis
The principle of PIV-analysis of a pair of images
recorded successively from one illuminated plane
of a particle seeded ow is depicted in Figure 2.
The steps in the analysis process are: (1) selection of interrogation areas or sub-images for
analysis, (2) correlation analysis, (3) nding the
displacement peak, (4) calculating the velocity

468
vector, and (5) repeat for next set of sub-images
(back to 1). These steps will be explained in
detail below.
Selection of sub-images for analysis
To map local ow velocities, sub-images (interrogation areas) are selected containing a relatively
small number of particles. In principle, an area
containing only one single particle would be ideal,
but it is hard to nd that single same particle back
in the second image, especially in a densely seeded
ow as for PIV-recordings. The probability for a
good analysis result is highest when the interrogation area contains about 8 10 particle images
(Keane and Adrian 1992; Hinsch 1993). Such an
area is selected from frame 1 (IA-1), and compared
with an area at exact the same location in frame 2
(IA-2). Because the particles have moved in the
inter-frame period and some of them may have left
area 1, area 2 may be selected larger, e.g. twice the
size, depending on the mathematical procedure
used (see later).
Correlation analysis
To get the most probable displacement of the
particle pattern in the interrogation area, a mathematical correlation procedure is applied. One can
imagine this procedure as moving area 1 over area
2 until the best matching is found. The expression
best matching is used because in practice there is
never a 100% matching due to particles that have
left or entered the imaged area in the second image
compared with the rst.
Mathematically, there are two methods used for
image correlation analysis: Fourier transformations and convolution ltering. The results of both
methods are comparable, but their applicability
and input formats vary, causing a somewhat less
accurate result from the Fourier path. Details on
the mathematics of both methods can be found
elsewhere (e.g. Rael et al. 1998). Here only the
essence of each method is covered briey. Convolution ltering comes closest to the description
of moving the images over one another for the
best match. A new correlation image of the same
size as the interrogation area is built up by calculating a new value for each pixel from the sum of
the products of all pixel values from interrogation
area 1 and interrogation area 2 for that particular
location (Figure 4). After normalisation, the new
image is a visible representation of the 2D proba-

bility-density function of the level of matching

between the 2 sub-images (Figure 5a).
The procedure for Fourier analysis of the image
pairs is in short: each sub-image is transformed
from the real to the complex domain using fast
Fourier transforms (FFT). In the complex
domain a conjugate multiplication between the
transform results from both images takes place,
and the product is transformed back to the real
domain using an inverse FFT. This yields after
normalisation about the same image representation of the 2D probabilitydensity function of the
level of matching between the 2 sub-images
(Figure 5b).
The two methods dier in that in general the
Fourier-pathway is faster, especially when the
FFT routines are optimized for this kind of
analysis, compared with the convolution-pathway. Fourier analysis needs interrogation area
formats of a power of 2, e.g. 16 16, 32 32 and
64 64 pixels and both input images must have
the same format when applying FFTs. While
using convolution ltering, any interrogation area
format is possible, enabling ne tuning of interrogation window size to e.g. number of particle
images or average particle displacement. Also, the
second sub-image can be chosen larger (see
(Figures 2, 4 and 5a), to make sure that as many
particles in the interrogation area (or IA-1) as
possible are also found back in the sub-image (or
IA-2), resulting in a more reliable probability
density distribution. This prevents the implementation of repair routines such as windowosetting in which a rst FFT-cycle is used to get
a rst estimate of the particle displacement and a
second cycle at the estimated displaced position is
used to get an accuracy comparable to convolution ltering.
Finding the displacement peak
The high intensity peak in the PIV-analysis result
image has to be located to know the particle displacement. Because the peak is usually larger than
one pixel, all pixel information of the peak can be
used to calculate the peak position, and obtain
sub-pixel accuracy. Methods used so far are centroiding, centre of gravity B/W, centre of gravity
weighed to grey value, Gaussian curve tting and
parabolic curve tting. Centroiding, in which all
the pixels of the peak outline are used to calculate
its centre, works well for larger peaks but is not

469

Figure 4. In convolution ltering, the resultant image (bottom right) has the same number of pixels as the interrogation area from the
rst image. Each new pixel is calculated from the sum of the products of all individual overlapping pixels from the interrogation area
from image 1 (Area 1, light grey) and the sub-image from image 2 (Area 2, white). The sum of products of the top-left position results
in a new value for pixel 1 in the result image (dark grey pixel). E.g. in formula for the rst result pixel that is:
R(1,1)=A1(1,1) mA2(1,1) + A1(2,1) A2(2,1) + A(3,1) B(3,1) + + A(7,7) B(7,7). The new (virtual) image is normalised
to the same grey value range as the original images and can then be displayed and/or used for further processing.

very accurate for small ones, and the same is true

for calculation of the centre of gravity B/W
(=after thresholding). Calculation of the centre of
gravity weighed to grey value, in which lighter
pixels contribute more than darker ones, gives
good and accurate results for average size peaks (3
pixels diameter) and even better for larger peaks.
Gaussian and parabolic curve tting work well
and accurate on average size peaks (3 pixels) but
are less reliable on smaller or larger peaks.
Depending on the peak size and the method used,
accuracies of less than one tenth of a pixel are
possible. With displacements of 3 pixels or more, a
condence level of much higher than 95% can thus
be achieved.
Calculating velocity vectors
The displacement information in pixels is translated to real world units (metres or millimetres)
after calibration. Subsequently, velocity vectors
are calculated by dividing the calibrated displacements by the inter-frame or inter-pulse time. The
resulting vectors can each be represented by ui and
vi, being the components in X- and Y-direction of

the ith vector, respectively, or by a magnitude, and

an angle with the horizontal with respect to a
dened origin.
Repeat for next set of sub-images
The process of selection of an interrogation area,
cross-correlation, peak-nding and calculation of
the velocity vector is repeated over the whole image, resulting in a complete vector representation
of the ow eld (Figure 6). To obtain more
velocity information out of the images, adjacent
interrogation areas are often overlapping. Any
overlap is in principle possible, but one should
keep in mind that a high degree of overlap also
means that the same information is sampled multiple times, resulting in over-sampling. An overlap
of 50% is recommendable because of a good balance between information density and over-sampling (Hinsch 1993).
Post-processing
Post-processing of velocity vector arrays resulting
from PIV analysis comprises two main components:

470

Figure 5 (a) Cross-correlation by convolution ltering in practice, being the lter product of 2 sub- images from 2 successively
recorded images displayed as mathematical product, resulting in a probability density distribution with the displacement peak. (b)
Cross-correlation by Fourier analysis in practice: the sub-images from the 2 successively recorded images are transformed to the
complex domain using FFTs, the products are subjected to a complex conjugate multiplication of which the product is translated back
to the real domain with an inverse FFT, eventually resulting in a probability density distribution with the displacement peak.

Figure 6. A set of two successive images is analysed by repetitive sampling and subsequent PIV analysis, often with overlap between
neighbouring interrogation areas (in this example the overlap is 50% of the interrogation area size in X and Y direction) to get an
accurate result without over- or under-sampling.

(1) data-validation and replacement of erroneous

vectors, and (2) analysis and visualisation of velocity gradients. They are both essential to get reliable
and well-understandable results that help to quantify, comprehend and interpret the ow phenomena
studied, the ultimate goal of a PIV ow study.

Data validation and error correction

The resultant vector set of automated PIV analysis
often includes a certain number of incorrect vectors that are usually obvious in the vector diagram.
This type of erroneous vectors are usually caused
by imperfections in the input images, caused

471
mainly by local variations in seeding density, local
over-illumination due to an object or wall in the
light sheet, strong out-of-plane-ow, local low
illumination close to the image borders, or crippled interrogation areas next to the image border.
These problems cause lack of correlation in the
normal way, and a background (lower) peak is
then recognized as if it were the displacement
peak, resulting in an erroneous vector. Incorrect
vectors usually stand out clearly with respect to
surrounding vectors, due to their dierent orientation and often dierent read much higher
magnitude (Figure 7a). This enables relatively easy
recognition to the observer as well as to any
automated validation routine. Automated validation routines usually compare every vector with
the surrounding ones in e.g. a 3 3 matrix, and
use statistical parameters (e.g. the standard deviation or a histogram distribution function) or uid
dynamical parameters (e.g. vorticity or divergence)
to discriminate between correct and erroneous
vectors. The details of some of these validation
procedures are discussed elsewhere (e.g. Rael
et al. 1998). Sometimes, especially close to the
image border or when erroneous vectors show up
in clusters, manual removal is necessary because
automated validation does not work due to lack of
reliable neighbours. This has to be done with great
care, of course, and often needs an experienced
evaluator.
Once the incorrect vectors are recognized and
removed, empty places should preferably be lled
by new vectors for continuous calculation of local
derivatives and gradient parameters (see later).
The new vector should represent the local ow
velocity as close as possible. The most common
way to do this is by 2D interpolation (Figure 7b).
Although simple linear or polynomial interpolation yields reasonable results, the best and also
principally the most reliable results have been
obtained with 2D cubic natural spline interpolation (Spedding and Rignot 1993; Stamhuis and
Videler 1995). Cubic natural spline ts leave the
original neighbouring data unaected and nd the
missing points according to the most uent line or
plane, i.e. with minimisation of the local curvature
(second derivative). In non-compressible or subsonic compressible uids, gradients are hardly ever
sudden, which illustrates why spline ts give the
best results. But even in shock waves in transonic
ow, spline interpolation may perform very well.

Analysis and visualisation of velocity gradients

The vector ow eld resulting from the PIV analysis and validation + interpolation procedure
gives an enormous amount of information on local
velocity and direction of the ow. Sometimes it is
dicult to see the actual ow phenomena in the
vector diagrams. Rotationality, for example, may
be masked by the overall ow speed, and so may be
other gradients. A rst tool to reveal hidden ow
structures is to subtract the average or background
ow from all vectors. This is often done when
studying the ow phenomena caused by a still or
moving object, e.g. a wing section, in a ow tank or
wind tunnel (Figure 7c vs. b). Additionally, the
velocity magnitude (absolute or relative) shown as
a colour code superimposed on the vector diagram
is also quite revealing (Figure 7d).
To visualise other gradients in the ow, a
number of parameters can be derived, of which
vorticity, rate of shear, rate of strain, divergence
and a vortex locator are most common and most
clarifying. These parameters are calculated from
the local velocity derivatives (see Table 2) (Vollmers et al. 1983; Lighthill 1986; Stamhuis and
Videler 1995; Vollmers 2001). Two of them are
illustrated in Figure 7e and f, showing the same jet
pulse with accompanying circular ow areas
(without background ow) as in Figure 7b, but
now with visualisation of the vorticity (Figure 7e)
and the vortex cores (Figure 7f) through colour
coding. The rate of shear often helps to assess the
velocity proles and pressure/force distributions
close to objects or walls. The rate of strain indicates acceleration or decelerations of the ow.
High strain rates in a 2D analysis are often indicative for a signicant 3D ow component,
which can be checked with the divergence. Because
they are computed from the velocity derivatives,
these parameters are insensitive to masking eects
of a high average ow speed or background ow.
This makes them very useful in revealing the real
but sometimes visually hidden ow morphology,
e.g. as in turbulence in high speed ows or impinging jets.
The distribution of vorticity and other derived
parameters give insight in the relative importance
and the distributions of velocity gradients in the
ow eld. That is above all necessary to assess and
understand the morphology of the ow phenomena studied. Moreover, velocity vectors and derived parameters can be used for further

Figure 7. Post-processing PIV-results from a pulse-jet. top row: Vector diagram before (left, a) and after (middle, b) validation and interpolation, and when masked by a
background ow (right, c). lower row: Colour coded vector diagrams showing the velocity magnitude (left, d), the vorticity (middle, e) and the vortex locator (right, f), and a 3D
interpretation of the results from this analysis (far-right, g).

472

473
Table 2. Gradient parameters that can be derived from the ow velocity information resulting from 2D-PIV, and aid in quantication
and interpretation of the ow phenomena studied.
Parameter

Stands for:

Equation

Vorticity
Shear rate

Rotationality of the uid

Strength of velocity gradient perpendicular to the local velocity
(sliding of adjacent water layers)
Strength of velocity gradient in the direction of the local velocity
(acceleration within one water layer)
Strength of out-of-plane ow rate (water leaving or entering the illuminated plane)
Finds centre of vortex (e.g. through discriminant for complex eigenvalues)

x = v/x u/y
h = v/x + u/y

Strain rate
Divergence
Vortex locator

processing or comparison, e.g. between successive

analyses, between comparable setups, or to validate theoretical predictions.
Interpretation
To fully understand PIV results, proper interpretation is essential. A rst challenge in interpreting
2D PIV results (and of other types of PIV) is that
we are basically looking at one 2D plane or one
cross-section of a 3D ow phenomenon. On top of
that, the ow pattern may be changing in time,
which adds an extra dimension to our problem.
For example, the mass of water travelling from the
upper-left corner towards lower-right with its
neighbouring elevated vorticity areas as displayed
in Figure 7af is in fact a cross-section of a jet
pulse with accompanying vortex ring (Figure 7g).
The exact shape and morphology of the jet + ring
combination can only be found when using a real
3D visualisation technique or when repeating the
experiment and making a series of parallel and
perpendicular cross sections. That may not always
be possible, especially when working with biota.
For a good interpretation a good background
knowledge of the uid dynamics involved, and
experience are important. The rst of these two
can be found in uid dynamics text books and in
the literature. Experience can only be built up by
directly working with the technique. When studying 3D free ow structures, ows close to objects
or boundaries, sudden gradients, etc., one should
be aware of the three-dimensionality of the ow
and know the limitations of 2D representations.

Examples from Biology

The application of 2D-DPIV to animal generated
ows is illustrated with two examples, one on acorn

e = u/x
H = u/x

v/y
v/y

barnacle lter feeding and one on lter feeding in

four dierent bivalve species. All these animals are
benthic and were chosen because of the emphasis of
the BIOFLOW network on benthic ow phenomena. These experiments serve as examples here, and
the results will be discussed only briey because
that is not in the scope of this paper.
The normal habitat of benthic animals is often
dominated by the external ow regime (waves and
tides), but these animals may also generate their
own ow phenomena. Here we concentrate on the
ow patterns generated by the animals themselves,
with some detail on the PIV setup, recording and
analysis (for details see Stamhuis et al. 2002;
Stamhuis and Vos in prep.). All PIV analyses were
performed using SWIFT 4.0 (Dutch Vision Systems), a commercially available package derived
from the system originally developed in our laboratory, using convolution ltering for the crosscorrelation analysis. Illumination was provided by
a red CW Krypton laser at a wavelength of
647 nm (Pmax = 1 W), delivered as a sheet at the
recordable area through an armed glass bre and a
light sheet probe. The probe allows adjustment of
the sheet thickness in the range 0.15 mm.

Acorn barnacle lter feeding

Acorn barnacles from the North Sea area are
relatively small crustaceans living in protective
calcareous tests, that colonize hard substrates
such as rocks, dikes, breakwaters and bivalve
shells. They have relatively long leg-like structures
called cirri that can be extended and swept
through the water to lter it for food, such as
detritus particles and small planktonic organisms.
Due to the small size of the cirri (lengths around
6 mm, diameter <0.2 mm) and their low beat
frequencies (around 1 beat/s), they are confronted

474
with viscous as well as inertial ow aspects. The
barnacles used in the experiments (Balanus
crenatus) are only actively beating at very low
ambient ow velocities, and were studied while
feeding in still water.
To map the ow around and caused by the
beating cirri, single animals that had settled on
mussel shells were mounted in a small aquarium
with sea water, free from any wall or the surface
within 5 cm. To reduce reections and diusion by
the white calcareous tests, the animals were covered with a small black cloth cloak, leaving only
the test orice open for protrusion of the cirri. The
ow was recorded in a series of horizontal and
vertical sheets to get a good 3D impression of the
ow morphology. The water was seeded with 5 lm
hollow glass spheres (Nortek A.S.) which were
neutrally buoyant in sea water. The particle seeded
ow was illuminated with a light sheet of 0.4 mm
thickness produced by the krypton laser with sheet
probe. Images were recorded using an Adimec
MX12 camera equipped with a Nikon macroobjective at 25 frames/s and 512 512 pixels resolution. The images were digitized and stored
uncompressed in real time. PIV analysis and
evaluation was performed after the recording was
nished. Figure 8 shows a diagrammatic representation of the setup.

The results from kinematic and PIV analyses

showed that the cirri accelerate fast during the rst
part of the power stroke and that water moves
through the cirri at that time at a cirral Reynolds
number of about 3 (based on cirrus diameter).
During the 2nd half of the power stroke, the water
catches up with the cirri, the Reynolds number
drops to 0.3, and the cirral lter starts to function
as a paddle. During the recovery stroke, the cirri
are folded back towards the test opening and rubbed along the mouthparts to collect and ingest retained particles, and then the cirri ick back to the
initial starting position without much interference
with the water. Figure 9a shows the ow pattern of
an active barnacle in side view halfway the power
stroke (to the left in the gure). The volume of
water accelerated and ltered during the previous
beat cycle is still visible as a big yellowred blob in
the left part of the image. The results indicate that
the barnacle actively lters the water for food
particles during the rst half of the power stroke.
Because the cirral function changes from a lter to
a paddle approximately halfway the power stroke,
the amount of water that has just been ltered is
swept away from the barnacle, probably to make
sure that it does not lter the same water again.
This seems to guarantee a good and continuous
food supply to the animal.

Figure 8. PIV setup for mapping the ow patterns produced by the beating 6 mm long cirri of the acorn barnacle Balanus crenatus.
The picture in the upper-right corner shows the horizontal and vertical planes that have been illuminated and analysed.

475
Bivalve lter feeding strategies
On the coastal intertidal ats of the Dutch Wadden Sea, several species of bivalves can be found
buried in the mud. Some species, e.g. the cockle
(Cerastoderma edule) and the soft-shelled clam
(Mya arenaria) are found close to the gullies with
relatively fast tidal currents, other species, such as
the blue mussel (Mytilus edulis) and the Baltic
tellin (Macoma balthica) are found higher up the
ats in areas with slower currents and higher
detritus sedimentation rates. All four species, like
all bivalves, use their internal gills to lter the
water for food particles, letting the water in and
out through inhalant and exhalant siphons. From
these four species, the Baltic tellin is regarded as a
facultative suspension/deposit feeder, able to inhale particles suspended in the overlaying water as
well as deposit from the sea oor, with its long and
exible siphons (Kamermans 1994; De Goey and
Luttikhuizen 1998) although it recently has been
considered to be an exclusive deposit feeder with
consequent low pumping rates (Riisgaard and
Kamermans 2001).
The ow velocity and ow morphology just
outside of the inhalant and exhalant siphons were
studied with PIV in a setup similar to Figure 8.
Animals of comparable size (1.5 2.0 cm) were

placed upright on a rubber ring on the bottom of

the aquarium and illuminated with a vertical laser
sheet of 0.5 mm thickness. The water was seeded
with 25 lm Pliolyte (BASF) particles and images
were recorded and stored digitally in real time at
25 fps with a 512 512 resolution using an Adimec MX12 camera with a Nikon macro-lense.
Suspension feeders sometimes reduce their
pumping rate in response to high particle loading.
All animals were oered the same conditions to
make the results of this experiment comparable
with one another. The pumping rates may, however, in general be lower than in natural situations. The seeding particles did not seem to
hamper the experimental animals except for
occasional backwash of the lter (Stamhuis et al.
2002).
The results for the comparable size specimens of
all four species are summarized in Table 3. The
Baltic tellin showed very low feeding current
velocities and pumping rates. Both the cockle and
the clam displayed diuse inow patterns and jetlike outow patterns and processed relatively large
volumes of water. Their ltration eciency
appeared to be so high that the middle part of the
outow jet showed up as a dark band, because
most particles had been ltered out. The outow
patterns were therefore partially reconstructed

Figure 9. Samples of the PIV-results from the biological examples. Left panel (a): The vector ow eld with colour coded velocity
magnitude from a lateral view (with a vertical light sheet) of the acorn barnacle Balanus crenatus, showing the moving 6 mm long cirri,
the water accelerated by the cirri, and the water volume on the left that was accelerated and shed during the previous beat cycle; imaged
area=24 30 mm; cirri beat towards left. Right panel (b): The diuse outow pattern of the exhalant siphon of an actively lter
feeding mussel Mytilus edulis; imaged area = ca. 30 0 mm. colour coding (a + b): red = rel. high velocity, blue = rel. low velocity.

476
using spline interpolation (for more details see
Stamhuis et al. 2002). The mussel was intermediate in both pumping rate and outow pattern
(Figure 9b). Thus, the two species living close to
the gullies had a diuse inow pattern, a jet-like
outow and high ltered volumes. The two species
living higher up the ats had lower ow rates.
Cockle and clam apparently optimized for high
ltering rates to compensate for low particle food
loading of the dilute water of the gullies, whereas
the mussel lives and feeds higher up the ats with a
higher detrital loading, and the tellin grazes from
an almost slurry-like food source very high up the
ats and therefore does not need high pumping
rates. This may support the conclusion that the
Baltic tellin is mainly a deposit feeder. In conclusion, ltering rate seems inversely related to particle concentration in the water in their natural
habitat.

Alternative PIV techniques

In a number of cases, e.g. with animal induced
ow phenomena that cannot be reproduced, or
with a highly variable 3D ow component, other
ow analysis techniques may be necessary. StereoPIV, Holographic PIV (H-PIV) or modern 3DPTV variants are potential candidates when it
comes to mapping 3D ow phenomena. Plain PTV
may be valuable when very low seeding densities
are required, and Micro-PIV may assist in resolving ow phenomena at the very small scale.
Stereo-PIV is often also called 3C-PIV, the 3C
standing for three components or three velocity
components. This emphasizes the main dierences
between 2D and Stereo PIV: Stereo-PIV enables
one to resolve the third velocity component in the
illuminated plane. In Stereo-PIV a second camera
is added and both cameras and their imaging optics are arranged in such a way that the whole area
of interest within the illuminated plane is projected

sharply onto the imagers. Additionally, the laser

sheet is made somewhat thicker, to prevent the
particles that travel in the out-of-plane 3rd
dimension from leaving the illuminated area before the next picture is recorded. Finally, after
calibration to correct for image distortion, two
pairs of images recorded in the stereo setup yield
the distribution of ow in X, Y and Z direction (u,
v and w, respectively) in one 2D plane. The big
step forward in comparison with 2D-PIV is thus
resolution of the out of plane ow. This information is, however, available for one plane at a
time only.
In Holographic PIV, a certain volume in the 3
dimensional space is illuminated as uniformly as
possible and imaged together with a reference
beam at two successive moments onto a hologram
(Hinsch 1993). This hologram holds in principle all
information of particle displacements in all three
dimensions. The way to retrieve this information
from the hologram is to read virtual planes from
the hologram with a CCD camera, and perform
regular 2D PIV on the resulting image. By combining parallel and perpendicular planes afterwards, the whole volume can be mapped
(Hermann et al. 2000). The advantage and great
power of HPIV is that all 3D uid movements are
recorded instantaneously with a high resolution
and are available for later evaluation. A disadvantage, at least from the experimenters point of
view, is that setting up an HPIV experiment
and performing HPIV analyses needs a lot of
experience as well as expensive equipment such as
special optics and usually a powerful laser. High
illumination levels or high energy laser pulses
sometimes invoke avoidance reactions or may even
be harmful to experimental animals. Up to now
only a few laboratories have specialised in HPIV,
mainly for fundamental uid dynamics research.
Recently, the 3D ow eld of a swimming copepod
has been mapped using HPIV (Malkiel et al.
2003).

Table 3. Summary of the results from a comparative PIV study at 4 species of bivalve to quantify their ltration activity with respect to
their normal habitat on the Dutch coastal ats.
Species

Flow speed (mm/s)

Flow rate(ml/h)

Pattern

Habitat

Balthic tellin (M.balthica)

Cockle (C.edule)
Clam (M. arenaria)
Mussel (M.edulis)

2.2
11
13
ca. 5

6.2
76
330
ca. 50

diuse
jet
jet
intermediate

ats
gullies
gullies
intermediate

477
Three-dimensional particle tracking velocimetry
(3D-PTV) is a technique that has already been
used for quite some time and has also been applied
to complex biological ows (e.g. Spedding 1986).
In early 3D-PTV studies, the experimenter had to
manually trace particles in e.g. 2 stereo images
from the same scene, allowing only low seeding
densities and involving labourious analysis. The
resulting 3D vector diagrams showed a resolution
high enough to get a good impression of the ow
phenomena and local ow velocities, but too low
to derive additional descriptive and analytical
parameters. In the recently developed 3DPTV
techniques, often more than 2 views from the same
scene are taken, the seeding density of the uid has
been increased to get higher spatial resolutions and
the analysis has been automated. Three views from
the same scene yield 3 sets of stereo images instead
of 1 set, and, when placing the cameras in a xed
frame of reference, triple triangulation can be
performed to locate each particle with a high
accuracy . Newly developed image analysis algorithms, which are mainly based on statistical procedures and forecasting (Cowen and Monismith
1997; Van der Plas and Bastiaans 1998), may
manage to track 1500 particles or more in one
volume. The resulting vector diagrams show a
reasonable to good spatial resolution, high enough
for derivation of e.g. vorticity. 3D-PTV may see a
revival with these new recording and analysis
methods, and the technique seems to be very
suitable for biogenic ows, providing that the
recording system has the proper speed and resolution.
When the seeding density is not high enough for
automated (3D-) PTV or PIV analysis due to
limitations on the animals side, one can revert to
plain PTV as it has already been used for decades
(e.g. Strathman 1973; Joergensen 1975; Sponaugle
1991; Larsen and Riisgard 2002). Here, the particles are usually located visually and manually, and
the result is a ow eld with irregular spaced
vectors at a relatively low density. The velocity
vectors may be very informative, but the low
vector density makes it dicult to derive gradient
parameters (shear, vorticity) without a more
complex mathematical procedure. The accuracy
depends highly on the analysis procedure used,
which may range from drawing on a foil on the
monitor screen to centroiding or peak curve tting at particle images in HR digital pictures.

At the very small scale (ca. 1 mm), the limited

focal depth of the camera- microscope-combination may be used to dene the plane of interest in
PTV or PIV studies (e.g. Klank et al. 2001). Such a
system is usually classied as a Micro-PIV system.
However, a very thin lightsheet (e.g. 0.1 mm
thickness) may still help to prevent blurred particle
images just above or below the plane of interest
(e.g. Van Duren et al. 1998), that increase the
noise and thereby lower the accuracy. Micro-PIV
(or PTV) may undergo some improvements in
optical as well as analytical aspects which increase
the accuracy and make this technique more
applicable for biological micro-applications.
One of the latest developments in pursuit of
higher resolutions and higher accuracy in PIV
analyses is called Super Resolution PIV (SR-PIV).
In SR-PIV, two routes have been followed to
achieve higher resolution: a PIV route and a PTV
route. The PIV route can be summarized as rst
performing normal PIV analysis with a normal
interrogation window size, and subsequently taking
smaller and smaller segments of the interrogation
window using the displacement information from
former steps, to increase accuracy and resolution,
almost down to the single particle image size (Hart
1999). The other route combines PTV algorithms,
sometimes similar to those used in the latest
3D-PTV methods, with classic PIV. A PIV step
provides a rst estimation of the ow velocity,
followed by PTV which provides information on
gradients within the interrogation area using the
velocity information from the PIV step to improve
particle pairing probabilities (Keane et al. 1995;
Cowen and Monismith 1997; Bastiaans et al. 2002).
Both routes yield comparable results, the vector
density may increase more than 10-fold, but the
PIV route with recursive correlation seems capable
of even more spectacular increases (Hart 1999).

Conclusion
Two-dimensional digital particle image velocimetry appears to be a very useful and versatile technique that can be applied to biogenic or
biologically relevant ows relatively easily. Even
irreproducible and very fast ow events can be
tackled, when using the right illumination and
recording techniques. Application depends on
tolerance to the relatively high level of (laser)

478
illumination, and to the presence of densely seeded
particles in the ow, that may hamper the organism or prevent it from behaving naturally. PIV
ow analysis can deliver 2D vector ow elds in an
instant, and a number of derivable parameters
assist in understanding and describing the ow
phenomena studied in a qualitative as well as a
quantitative way. Interpretation of the 3D ow
patterns from the 2D PIV cross sections is, however, not trivial and has to be done with care.
When doing so, 2D DPIV can greatly assist in
understanding how an organism interacts with a
uid and what it actually does with that uid, or
e.g. how it manages to stand up to external
currents.

Acknowledgements
Jan Drent assisted in recording of the bivalve ow
elds, Michiel Vos recorded the barnacle ow
elds, Sandra Nauwelaerts recorded the vortex
ring ow elds (cooperation with the University of
Antwerp), Luca van Duren and Per Jonson are
acknowledged for the invitation to the BIOFLOW-2002 symposium. The comments of Dr
H.U. Riisgaard and 2 anonymous referees
enhanced the paper greatly.

References
Adrian R.J. 1991. Particle imaging techniques for experimental
uid mechanics. Ann. Rev. Fluid Mech. 23: 261 304.
Anderson J.D. 2000. Introduction to Flight (4th int. ed.).
McGraw-Hill, Boston.
Bastiaans R.J.M., der Plas G.A.J. and Kieft R.N. 2002. The
performance of a new PTV algorithm in super-resolution
PIV. Exp. In Fluids 32: 346 356.
Cowen E.A. and Monismith S.G. 1997. A hybrid digital particle
tracking velocimetry technique. Exp. In Fluids 22: 199 211.
Drucker E.G. and Lauder G.V. 1999. Locomotor forces on a
swimming sh: three-dimensional wake dynamics quantied
using digital particle image velocimetry. J. Exp. Biol. 202:
2393 2412.
Goey P. de and Luttikhuizen P. 1998. Deep-burying reduces
growth in intertidal bivalves: eld and mesocosm experiments
with Macoma balthica. J. Exp. Mar. Biol. Ecol. 228: 327
337.
Hart D.P. 1999. Super-resolution PIV by recursive localcorrelation. J. Visualisation 10: 1 10.
Hermann S., Hinrichs H., Hinsch K. and Surmann C. 2000.
Coherence concepts in holographic particle image velocimetry. Exp. In Fluids 29(Suppl.): S108 S116.

Hinsch K.D. 1993. Particle image velocimetry. In: Sirohi R.S.

(ed.), Speckle Metrology. Marcel Dekker Inc, New York, pp.
235 342.
Joergensen C.B. 1975. On gill function in the mussel Mytilus
edulis L. Ophelia 13: 187 232.
Jones J.D.C. (ed.), 1997. Particle image velocimetry. Meas. Sci.
Technol. 8(12): 13791583.
Kamermans P. 1994. Similarity in food source and timing of
feeding in deposit- and suspension-feeding bivalves. Mar.
Ecol. Prog. Ser. 104: 63 75.
Keane R.D. and Adrian R.J. 1992. Theory of cross-correlation
analysis of PIV images. Appl. Sci. Res. 49: 191 215.
Keane R.D., Adrian R.J. and Zhang Y. 1995. Super-resolution particle image velocimetry. Meas. Sci. Technol. 6:
754 768.
Klank H.G., Goranovic G., Kutter J.P., Gjelstrup H., Michelsen J. and Westergaard C.H. 2001. Micro PIV measurements in micro cell sorters and mixing structures with three
dimensional ow behavior. CD-ROM proceedings 4th Int.
Symp. on PIV, Gottingen, Paper, Paper 1161, pp. 1 12
Larsen P.S. and Riisgard H.U. 2002. On ciliary sieving in
pumping bryozoans. J. Sea Res. 48: 181 195.
Lighthill J. 1986. An Informal Introduction to Theoretical
Fluid Mechanics. Oxford University Press, Oxford.
Malkiel E., Sheng J., Katz J. and Strickler J.R. 2003. The threedimensional ow eld generated by a feeding calanoid
copepod measured using digital holography. J. Exp. Biol.
206: 3657 3666.
Muller U.K., den Heuvel B.L.E., Stamhuis E.J. and Videler J.J.
1997. Fish foot 23 prints: morphology and energetics of the
wake behind a continuously swimming mullet (Chelon labrosus- Risso). J. Exp. Biol. 200: 2893 2906.
Rael M., Willert C. and Kompenhans J. 1998. Particle Image
Velocimetry, A Practical Guide. Springer-Verlag, Berlin.
Riisgaard H.U. and Kamermans P. 2001. Switching between
deposit and suspension feeding in coastal zoobenthos. In:
Reise K. (ed.), Ecological Comparisons of Sedimentary
Shores. Ecological Studies Vol. 151. Springer Verlag, Berlin,
pp. 73 101.
Spedding G.R. 1986. wake of a Jackdaw (Corvus monedula) in
slow ight. J. Exp. Biol. 125: 287 307.
Spedding G.R. and Rignot E.J.M. 1993. Performance analysis
and application of grid interpolation techniques for uid
ows. Exp. In Fluids 15: 417 430.
Spedding G.R., Hedenstrom A. and Rosen M. 2003. Quantitative studies of the wakes of freely ying birds in a lowturbulence wind tunnel. Exp. In Fluids (in press).
Sponaugle S. 1991. Flow patterns and velocities around a suspension-feeding gorgonian polyp: evidence from physical
models. J. Exp. Mar. Biol. Ecol. 148: 135 145.
Stamhuis E.J. and Videler J.J. 1995. Quantitative ow analysis
around aquatic animals using laser sheet particle image
velocimetry. J. Exp. Biol. 198: 283 294.
Stamhuis E.J., Videler J.J., Van Duren L.A. and Muller U.K.
2002. Applying digital particle image velocimetry to animalgenerated ows: Traps, hurdles and cures in mapping steady
and unsteady ows in Re regimes between 10 2 and 105. Exp.
In Fluids 33: 801 813.
Stanislas M., Kompenhans J. and Westerweel J. (eds) 2000.
Particle Image Velocimetry, Progress Towards Industrial
Application. Kluwer Academic Publishers, London.

479
Strathmann R. 1973. Function of lateral cilia in suspension
feeding of lophophorates (Brachiopoda, Phoronida, Ectoprocta). Mar. Biol. 23: 129 136.
der Plas G.J. and Bastiaans R.J.M. 1998. Accuracy and resolution of a fast PTV- algorithm suitable for HiRes-PV. In:
Grant I. and Carlomagno G.M. (eds), Proceedings of the 8th
International Symposium on Flow Visualisation. Sorrento,
Italy, pp. 1/1 1/12.
Van Duren L.A., Stamhuis E.J. and Videler J.J. 1998. Reading
the copepods personal ads: increasing encounter probability
with hydromechanical signals. Phil. Trans. R. Soc. Lond. B
353: 691 700.
Vogel S. and LaBarbera M. 1978. Simple ow tanks for
research and teaching. Bioscience 28: 638 643.
Vogel S. 1994. Life in Moving Fluids, the Physical Biology of
Flow (2nd ed.). Princeton Press, New Jersey.
Vollmers H., Kreplin H.-P. and Meier H.U. 1983. Separation
and vortical-type ow around a prolate spheroid evaluation

of relevant parameters. NATO-AGARD Conference Proceedings No. 342, pp. 14/1 14/14.
Vollmers H. 2001. Detection of vortices and quantitative evaluation of their main parameters from experimental velocity
data. Meas. Sci. Technol. 12: 1199 1207.
Westerweel J. 1993. Digital Particle Image Velocimetry Theory and Application. Delft University Press, Delft.
Westerweel J. 1997. Fundamentals of digital PIV. Meas. Sci.
Technol. 8(12): 1379 1392.
Westerweel J., Dabiri D. and Gharib M. 1997. The eect of a
discrete window oset on the accuracy of cross-correlation
analysis of PIV recordings. Exp. In Fluids 23: 20 28.
Willert C.E. and Gharib M. 1991. Digital particle image velocimetry. Exp. In Fluids 10: 181 193.
Zirbel M.J., Veron F. and Latz M.I. 2000. The reversible eect
of ow on the morphology of Ceratocorys horrida (Peridiniales, Dinophyta). J. Phycol. 36: 46 58.