Accurate predictions of snowfall require good knowledge of the microphysical properties of the snow ice crystals and particles. Shape is an important parameter as it strongly influences the scattering properties of the ice particles, and thus their response to remote sensing techniques such as radar measurements. The fall speed of ice particles is another important parameter for both numerical forecast models as well as representation of ice clouds and snow in climate models, as it is responsible for the rate of removal of ice from these models.
We describe a new ground-based in situ instrument, the Dual Ice Crystal Imager (D-ICI),
to determine snow ice crystal properties and fall speed simultaneously.
The instrument takes two high-resolution pictures of the same falling ice particle from two different viewing directions.
Both cameras use a microscope-like setup resulting in an image pixel resolution of approximately 4
The instrument has been tested in Kiruna, northern Sweden (67.8
Accurate knowledge of atmospheric ice crystals and snowflakes, or snow particles, is needed for meteorological forecast and climate models (see, e.g.
To retrieve precipitation amount reaching the ground, the microphysical properties of the snow particles on their way down have to be known. Fall velocity plays a significant role for modelling of the microphysical processes. It determines the number of particles present in the measuring volume and the snowfall rate, or the rate of particle removal from clouds.
Other important microphysical properties of snow particles are their shape and morphology, not only for investigating growth processes. Snow particle shape and morphology affect radar measurements
Snowfall has long been monitored by ground-based instruments. However, instruments that can measure size, shape, and fall speed simultaneously are still scarce.
Instruments can be classified into different categories depending on what is measured primarily.
Disdrometers, for example, measure shadow or attenuation as droplets or snow particles pass one or several light beams.
Fall speed can be estimated either from the duration between the two beam passages, in case of instruments that have two parallel beams with known vertical spacing, or from the duration of attenuation.
Three common disdrometers are
Parsivel
Another category of instruments uses camera systems for optical imaging of snow particles.
One example is SVI (Snowflake Video Imager,
There are instruments designed for aircraft that have also been used on the ground for snow measurements.
CIP
Holographic imaging has the advantage of a larger depth of field when compared to so-called “in-focus imaging”.
Shadow-like images of out-of-focus particles can be reconstructed and their position determined.
Holographic Detector for Clouds (HOLODEC) is an aircraft instrument
PHIPS (Particle Habit Imaging and Polar Scattering) uses a combination of optical imaging and scattering (with polar nephelometer).
A first version of the instrument had a high pixel resolution, better than the 2
This work presents a novel instrument that uses two cameras for simultaneous particle imaging and fall speed measurement. It is called the Dual Ice Crystal Imager (D-ICI) and is a development of ICI
The cross-sectional area as seen from the top is better related to the particle drag and terminal fall velocity
than the area determined from the side view.
Additionally, particle size and area from the top view are also more useful when comparing to optical remote sensing,
which often uses vertical viewing geometries too.
Sections
D-ICI uses passive sampling with a vertically pointing inlet.
Its setup can be seen schematically in Fig.
Schematic cut views of the setup of D-ICI.
Ice particles, i.e. small ice crystals, snow crystals, and snowflakes, falling into the inlet will
fall down the sampling tube and traverse the optical cell.
In the centre of the optical cell is the sensing volume.
If a particle is falling through the sensing volume, it is detected by the detecting optics.
Upon detection, the ice particle is optically imaged from two different directions.
Figure
Two examples of ice crystals imaged in two viewing geometries: top view and side view.
The ice crystal shown in panel
One of the two viewing directions looks horizontally from the side, called the side view,
and the other vertically from the top, called the top view.
The former will allow to measure the fall speed, if using a double exposure (see Sect.
Photograph of D-ICI (door of enclosure is removed).
Similarly to the ICI probe
In the sensing volume (see Sect.
The tube lens of the side-view optics is
a positive achromatic doublet (AC254-045-A, Thorlabs)
with the same focal length as its microscope objective (45 mm).
As tube lens of the top-view optics, the same achromatic doublet as for its objective is used.
Thus, the resulting magnifications are the same for both systems (
Both imaging systems use bright-field illumination from the back. This
is achieved by a light-emitting diode (LED)
with a simple focusing lens optics allowing for an even illumination of the FOV.
Each of these two lens–LED configurations is arranged along the optical axis of the respective imaging optics
on the opposite side of the sensing volume (see Fig.
The top-view optical system uses a mirror between the sensing volume and the objective lens. This allows to fold its optical axis so that it is parallel to the optical axis of the side-view system for a simpler mechanical setup.
The sensing volume, i.e. the volume in which particles are detected and imaged, is defined as the intersection of the laser beam for detection with the overlapping FOVs of the imaging systems. The laser beam, which has a wavelength of 650 nm and power of 1 mW, is aligned perpendicular to the optical axes of both imaging optics. It is shaped by an aperture to about 1 mm horizontal width, which corresponds approximately to the depth of focus of the side-view camera. The laser beam is further shaped by a cylindrical lens (LJ1960L1, Thorlabs) with focal length of 20 mm such that its vertical height, originally about 3 mm, is focused to approximately 0.1 mm in the centre of the FOV of the side-view camera. Thus, the laser beam forms a light sheet with width of approximately 1 mm and height of 0.1 mm. Both the side- and top-view cameras are focused so that their focal planes are aligned with this resulting laser sheet. As a consequence, all detected particles are in focus for both images.
To determine the snowfall rate or the snow crystal number concentration,
the sensing area, i.e. the area through which detected particles fall,
needs to be known rather than the sensing volume.
The sensing area is the horizontal cross section of the sensing volume
(i.e. the cross section perpendicular to the vertical falling motion).
The area is therefore given by the product of the width of the FOV of the cameras
and the sum of laser beam width (1 mm) and particle size.
This sum has to be used instead of laser beam width only,
because particles that are only partially in the laser beam will be detected too.
Thus, the sensing area is size dependent (larger particles have a larger sensing area).
When assuming a constant sensing area corresponding to a particle size of 500
Scattered light from the part of the laser beam within the sensing volume is collected and focused on a photodiode
(FDS010, Thorlabs) by two plano-convex detector lenses (LA1951-A, Thorlabs).
The photodiode is located along the axis of the laser beam, which is stopped by a light trap mounted in the centre of the first lens.
The diameters of the light trap and the lens tube holding the detector lenses are such that
the photodiode detects light scattered by ice particles in the sensing volume
in near-forward direction in the range of scattering angles
between approximately 10 and 32
The current of the photodiode is converted to a voltage and amplified
(effective current-to-voltage amplification of 2.2 M
Both imaging systems are triggered by the same signal (see Sect.
Both computers are connected to a network via ethernet cables. This allows to synchronize them with each other. Consequently, corresponding side- and top-view images can be recognized by their time stamp, which is part of the file name. Both computers can be accessed through an additional laboratory or office computer, which is connected to the same network via cable or internet, if the network provides internet access. Data can then be retrieved using this laboratory computer.
Alternatively, the SD cards can be collected to copy the image data.
Then, the image data will be processed by the laboratory computer as described in Sect.
While the focus of D-ICI is high-resolution images for shape and fall speed measurements, snowfall rate and number concentration can also be determined from the acquired data.
For that, snowfall rate
The size dependencies of the sensing area and the probability of the particle being
partially outside the FOV cancel out to a good approximation (see Sect.
The assumption of constant fall speed
An additional uncertainty in estimating the effective sensing area
resulting from the uncertainty in determining the laser beam width,
which may be on the order of
These uncertainties
affect both
The images have pixels with grey levels between 0 (black) and 255 (white).
An automated image processing algorithm is applied to all top-view images to retrieve ice particle size, area, area ratio,
and aspect ratio.
It first removes non-particle features from the background.
Then, the particles on the images are detected and their edges are found.
This algorithm has been used by
Automated image processing steps shown for an example image.
Panel
A background image without any ice particle is used to correct for uneven background illumination, i.e. remove non-particle features from the background. For this, the difference between background and image to be analysed is determined. The difference is positive where the presence of a particle makes the image darker than the background. For regions where the image is brighter than the background, the resulting negative values are set to zero. These are usually only regions within an ice particle where transmitted light can appear as a brighter spot, surrounded by darker features or the edge of the particle. Now, images are rejected from further analysis if no particle was captured on them, i.e. images that are very similar to the background. For this, a lower threshold is applied to the difference. The image is rejected if the difference does not exceed the threshold for any pixel. A suitable threshold is 30; images with ice particles exceed this by a large margin.
Then, for the remaining images,
the difference to the background is first scaled to increase the dynamic range of grey values.
This is done for each pixel individually, so that the possible maximum difference (when the image pixel is black),
at any background pixel becomes 255.
Effectively, the scaling factor at any pixel is
The following steps in the image processing apply to the cleaned image resulting from the background removal described above.
For detecting in-focus particles, two thresholds are applied: a grey-level threshold and a gradient threshold.
The grey-level threshold is used to find particles and their edges, and the gradient threshold is used to reject out-of-focus particles.
First, images that do not have any pixel darker than the grey-level threshold are discarded.
This rejects particles that are much out of focus.
Then, a binary mask, i.e. a black-and-white image,
of the same dimension as the original image is created where logically True entries represent
image pixels that are darker than the grey-level threshold.
The binary mask is then smoothed to remove variations at the 1-pixel level,
which are considered to not reflect the actual variations in the edge of the ice particle.
The smoothing is achieved by first dilating each True pixel in the binary mask so that the four neighbouring pixels
(above, below, right, and left) will also be True.
Then, the dilated binary mask is eroded, to restore its original size,
by setting the four neighbours of each False pixel to be also False.
Between the dilation and erosion steps, the binary mask is also filled, i.e. all pixels that are False but completely enclosed by True pixels are converted to True. This will include the brighter spots,
which many ice crystals show on the images, to the particle they belong to.
Then, on the resulting black-and-white image (see example in Fig.
Firstly, out-of-focus particles are rejected.
For this purpose, a gradient matrix is computed from the image.
The values of this matrix
are used as a parameter indicating in-focus or out-of-focus particles.
For computing the gradient values, the image is filtered (using the Matlab function
Secondly,
particles with apparent problems are marked with quality flags.
A particle that is in part out of focus can sometimes have parts of the edge not being detected,
yielding an apparently fragmented edge with narrow gaps.
Similarly, if thin ice particle features result in brighter pixels than the grey-level threshold,
a fragmented edge is the consequence.
To account for this, two or more detected particles that appear very close to each other are joined and the resulting particle is marked
as being “fragmented”. The area of such a particle as determined from the detected pixels will be too small.
The resulting error is not large, because the gaps are only small, and
by joining the fragmented pieces, the particle may still be considered.
However, being marked, it can also easily be excluded from further analysis.
An example of an ice particle detected with fragmented edge is given in Fig.
Detected edges of processed ice particle images.
The edges are shown in red and have been enlarged to a thickness of 3 pixels for better visibility in this figure.
One example, panel
Examples of ice particles flagged as“on-border” (on the right side of the image) and “in-darkregion” (on the left side of the image).
The original image
Lastly, area and size information is determined for each detected ice particle.
As a parameter describing a characteristic size of the detected particle
we are using maximum dimension, i.e.
the diameter of the smallest circle that completely encloses that particle on the image
(see Fig.
As this method is the same as used for the imager described by
The side-view camera can be operated in a fall speed mode, in which
the falling ice particle is captured twice on the same image by using a double exposure.
This concept has been tested with ICI in a configuration without inlet, so that ice particles could freely
fall through the instrument
The vertical fall distance
While falling, the difference of the horizontal coordinates is usually close to zero.
Such a difference could be caused by sideway or rotating (tumbling) motion.
Horizontal winds, which affect other instruments, with an open sampling volume,
such as PIP and MASC,
do not cause a sideway motion in the enclosed sensing volume of D-ICI.
Thus, only a tumbling particle can be responsible for a difference of the horizontal coordinates,
and tumbling of ice particles is not often seen (see Sect.
While side-view images are not processed automatically,
the top-view images are (see Sect.
According to the design, the pixel resolution should be equal to the pixel size
of the CCD cameras, 3.75
Figure
Ice particles as imaged in two viewing geometries: top view and side view. Each ice particle is shown as a pair of these two views, with
the top view always above the corresponding side view.
Eight particles are shown in two rows
These detailed images of ice particles allow to recognize their shapes.
On 23 October 2014, the ice particles had predominantly bullet-rosette and similar shapes, but also
plate-like and capped-column shapes (see Fig.
Example images of the two shapes, graupel
Figure
Example side-view images of doubly exposed falling ice particles.
The fall speed is determined from the vertical separation of the two instances of the
particle on the same image.
Panel
One of the particles shown in Fig.
Using the top-view images, the ice particles' projected area in the fall direction
(i.e. area projected on a surface perpendicular to the vertical fall direction) can be determined.
Figure
For the data shown in Fig.
Area
When the ice particles are classified according to their shapes,
power laws
can be fitted to the resulting subsets of data
to find relationships describing area for specific shapes.
On 19 October 2014, two dominant shapes were observed: graupel and rimed needles
(see Fig.
Fall speed versus maximum dimension
The images from 19 October 2014 also show a few drizzle droplets,
which can be seen in Fig.
When looking at the area–dimensional relationship for a certain shape, the fit to the power law can be very good.
An exception here are rimed needles.
However, they still have a fairly good fit, better than the fit to all data with one common power law, which would predict poorly the area for any of the shapes here, droplets, graupel, and rimed needles (see Fig.
Figure
The figure also shows the fall speed measured for the drizzle droplets.
As expected, the droplets have the strongest dependence on size.
With increasing complexity of particle shape, from droplets over graupel to rimed needles,
the size dependence becomes weaker, the spread in data larger, the speed (at same size) slower, and
We have described the Dual Ice Crystal Imager (D-ICI), a ground-based in situ instrument to determine snow ice crystal properties and fall speed simultaneously. Dual images are taken of detected snow particles using two CCD cameras that image along a horizontal and close-to-vertical viewing direction, respectively. The horizontal, or side, view is used to determine fall speed from images taken with double exposures. The close-to-vertical, or top, view is used to determine size and area.
Both cameras use the same pixel resolution of approximately 4
Snow particles fall some distance vertically through the sampling tube before images are taken, from which speed is derived.
Therefore, the fall speed measurements of D-ICI are not affected by
the vertical component of the wind speed or by turbulence close to the ground.
The accuracy of fall speed measurements has been discussed and is mainly limited by tumbling of snow particles.
However, tumbling is not observed frequently.
Rejecting particles that tumble with a rotation of more than 10
Snow particle size and area are determined from top-view images, i.e. as projected along the vertical fall direction. These properties are more appropriate than the same properties determined from a horizontal view, as done by most instruments, when studying relationships to the fall speed or comparing to vertically pointing remote sensing measurements.
The presented data are available at the Swedish National Data Service (SND-ID: SND 1129;
TK designed and built D-ICI. TK prepared the analysis methods with contributions from SV-M. TK carried out the measurements in 2014–2016; SV-M carried out the measurements from 2017 on with contributions from TK. TK analysed data from 2014. SV-M analysed data used to evaluate frequency and extent of tumbling. TK prepared the manuscript with contributions in text and figures from SV-M.
The authors declare that they have no conflict of interest.
We thank the Kempe Foundations (Kempestiftelserna, SMK-1024) for financial support of hardware, the Swedish National Space Agency (SNSA) for funding during the prototype development (grant Dnr 85/10), and the Graduate School of Space Technology at Luleå University of Technology for additional financial support.
This research has been supported by the Kempe Foundations (grant no. SMK-1024) and the Swedish National Space Agency (grant no. Dnr 85/10).
This paper was edited by Alexander Kokhanovsky and reviewed by Kevin Hammonds and Timothy Garrett.