14th International Symposium on Particle Image Velocimetry – ISPIV 2021
August 1–5, 2021

Large Scale Infrared-Based Remote Sensing of Turbulence
Metrics in Surface Waters: Going Beyond Mean Flow

Seth A. Schweitzer1∗, Edwin A. Cowen1

1 DeFrees Hydraulics Laboratory, School of Civil & Environmental Engineering, Cornell University,
Ithaca, NY ∗ seth.schweitzer@cornell.edu

Abstract
In recent years field-scale applications of image-based velocimetry methods, often referred to as large
scale particle image velocimetry (LSPIV), have been increasingly deployed. These velocimetry mea-
surements have several advantages—they allow high resolution, non-contact measurement of surface
velocity over a large two dimensional area, from which the bulk flow can be inferred. However, visible-
light LSPIV methods can have significant limitations. The water surface often lacks natural features
that can be tracked in the visible and generally requires seeding with tracer particles, which creates
concerns regarding the fidelity with which tracer particles track the flow, and introduces challenges in
achieving sufficient and uniform seeding density, in particular in regions with appreciable velocity ac-
celerations such as turbulence. In LSPIV, image collection is generally limited to daylight hours, and
can suffer from non-uniformity of illumination across the camera’s field of view. Due to these issues
LSPIV often requires spatio-temporal averaging, and as a result is generally able to extracting the
mean, but not the instantaneous, velocity field, and hence is often not a suitable tool for calculating
turbulence metrics of the flow.

Figure 1: Instantaneous velocity field measured by
IR-QIV over several hundred m2 at the surface of a
river. Background color indicates the local instan-
taneous velocity. Arrows indicate the perturbation
from the mean velocity, i.e., u⃗′, the difference be-
tween the local instantaneous and mean velocity.

We present an alternative to visible-
light LSPIV that avoids these seeding and
illumination issues: infrared quantitative
image velocimetry (IR-QIV), which uses
infrared (IR) images to accurately cap-
ture temperature patterns at the water sur-
face with high thermal and spatial resolu-
tion (Schweitzer and Cowen, under review).
Motion of the patterns is tracked over time,
from which the surface velocity field is cal-
culated (figure 1). In natural flows small
temperature differences are created in the
surface skin due to spatial heterogeneity in
turbulent stirring and heat exchange be-
tween air and water. These spatial differ-
ences in temperature form a rich texture of
patterns on the water surface that are ob-
servable in IR images (figure 2). IR images
are a record of radiation emitted by the wa-
ter surface, with no external illumination,
and no concerns of possible differences be-
tween motion of tracer particles and the
flow. In contrast to LSPIV, IR-QIV is able
to extract the instantaneous velocity field
reliably and robustly. These measurements
can be used for a range of applications, in-
cluding interaction of fish with structures
and flow features, and non-contact esti-
mation of bed stress, bathymetry and dis-

charge. In this presentation we will provide an overview of our developed IR-QIV technique and



detail the environmental conditions required for high quality IR image collection, and dealing with
images collected under sub-optimal environmental conditions. These issues are key for optimizing
IR-QIV for accurate turbulence measurements.

We use the minimum quadratic difference (MQD) method for pattern matching (Gui and Merzkirch,
1996, 2000), as it is more suitable than cross-correlation for images containing gradients of intensity
and not discrete particles (Cowen et al., 2010). This method generates, at each interrogation loca-
tion, a surface D(m,n) = 1

ṀN ∑M
i=1 ∑N

j=1(g(i, j)−g′(i+m, j+n))2, where g(i, j) represents a subwindow
of dimensions M,N with a corner at pixel-coordinates (i, j), and g′(i+m, j+ n) represents a similar
subwindow from a subsequent image, displaced by (m,n) pixels. Lower values of the surface D indicate
a greater similarity (smaller quadratic difference) between the pattern of pixel intensities in the two
images at that displacement. We use a method that considers image entropy to optimize parameters
such as the subwindow dimensions M,N, and the temporal separation ∆t between images.

Figure 2: IR image of a river surface under environmental
conditions leading to high (left, top), and low (left, bot-
tom) dynamic range image, and representative surface D
for a single location in each of the images (right). In both
examples there a local minimum at the coordinate corre-
sponding to zero displacement (marked with a red circle).

An important metric of image
quality and suitability for IR-QIV
is pixel dynamic range (the range
of pixel intensities present in the
image). In images where the dy-
namic range is low the signal (of
temperature differences at the wa-
ter surface) can be overwhelmed by
noise in the camera’s optical and
digitization path, known as fixed
pattern noise (FPN). Since FPN
does not change significantly be-
tween consecutive images, it leads
to a match between subwindows at
a displacement value of zero pix-
els. A local minimum in match val-
ues at a zero displacement is always
present (figure 2). When the dy-
namic range is low this peak can be
lower than the match due to com-
parison of patterns created by tem-
perature patterns at the water sur-
face, leading to incorrect velocime-
try results. Since this minimum
will always be at a coordinate cor-
responding to zero displacement it
can be identified and filtered from
the velocimetry record.

The dynamic range of IR images is controlled by the range of temperature differences at the
water surface, as well as the camera’s sensitivity and position. In the measurement described in
this presentation, IR images exhibited high dynamic range when the difference between bulk water
temperature and wet-bulb air temperature was greater than ∼3 °C. Smaller air-water temperature
differences (corrected for latent heat) lead to low dynamic range in the image, and as a result to a large
number of zero velocity results. The environmental conditions required for sufficient image quality to
avoid zero-displacement issues will vary between measurements, depending on factors including the
camera’s sensitivity and internal noise level, optical setup (e.g., viewing angle, distance, and optics),
and environmental conditions such as wind velocity and turbulence intensity of the flow.

Acknowledgements
Funding for this work was provided by the California Department of Water Resources (DWR).



References
Cowen EA, Dudley RD, Liao Q, Variano EA, and Liu PLF (2010) An insitu borescopic quantitative

imaging profiler for the measurement of high concentration sediment velocity. Experiments in Fluids
49:77–88

Gui L and Merzkirch W (1996) A method of tracking ensembles of particle images. Experiments In
Fluids 21:465–468

Gui L and Merzkirch W (2000) A comparative study of the mqd method and several correlation-based
piv evaluation algorithms. Experiments in Fluids 28:36–44

Schweitzer SA and Cowen EA (under review) Instantaneous river-wide water surface velocity field
measurements at centimeter scales using infrared quantitative image velocimetry. Water Resources
Research