Photography and Snowflakes

It is generally believed that no snowflake is the same because of one man and his obsession with looking at snow crystals: Wilson Bentley. Over the course of his life, Bentley photographed over 5000 snowflakes which were collected from around his home in Vermont over his lifetime (1865-1931). This impressive collection sparked curiosity among the scientific community into how snowflakes form and cemented Bentley’s place as the most dedicated observer of snow in history. I came across Bentley’s work whilst reading a book about snow, and his photographs and story have captivated me.

A snowflake will have one of a specific number of structures – such as dinner plates, branch networks, columns and ‘flower’ patterns – but with different detailing that makes it unique to any other. Up close, these internal symmetries and dendrites appear intricate and beautiful and is what captured the attention of Wilson Bentley as a fifteen year old boy when he first put a snowflake under a microscope.

Wilson Bentley's pictures of a series of plate snowflakes, circa 1902. Source: The Guardian

‘Under the microscope, I found that snowflakes were miracles of beauty and it seemed a shame that this beauty should not be seen and appreciated by others’. A series of plate snowflakes captured by Wilson Bentley. Source: The Guardian

Wilson Bentley lived all of his life in Mill Brook Valley, a small place in Vermont (New Hampshire, USA) which lies adjacent to the Green Mountains of Vermont and commonly receives high snowfall in the year. On receiving a microscope as a birthday present from his parents, Bentley became obsessed with looking at snow crystals gathered from around the town. A couple of years later he attached a bellows camera to his microscope and began photographing snow crystals. Microscope photography is now referred to as photomicrography, and Bentley was the first person in the world to perfect this technique.

Bentley’s set-up consisted of the bellows camera on top of the microscope, with a series of  attached pulleys and strings that controlled the focal length and focus of the camera. He would take a three-inch plate and create the luminous white shape of a snowflake on a field of black by scratching off the black emulsion from the photograph negative. Over a period of 50 years, he photographed more than 5000 snow crystals and eventually published a selection of these in ‘Snow Crystals’ in 1931 which propelled him to fame and attention from the scientific community.

Wilson Bentley in action with his set-up for photographing snowflakes (source: The Guardian)

Wilson Bentley in action with his set-up for photographing snowflakes (source: The Guardian)

Even when he was beginning to be acknowledged for his work in the 1920s, many people in the town thought he was mad for isolating himself in his garden shed obsessively studying snowflakes, including his father and his brother. At this point, he had been published in news outlets such as the New York Tribune and the Boston Herald, and was even featured in a short film called ‘Mysteries of the Snow’. He didn’t massively profit from this success, instead being content in doing the thing he loved. Shortly after his book ‘Snow Crystals’ was released, he died of pneumonia after insisting on walking back to his home through a blizzard.

Future studies showed that Bentley had only scratched the surface on snow crystal structures, partly because he exclusively studied snowflakes from Vermont. After his death, the scientific community became interested in exploring snow crystal structures, growing snow crystals in laboratory condition. This was largely led by Japanese nuclear physicist Ukichiro Nakaya, who could fine-tune temperature, pressure and moisture content in a controlled chamber to grow different snow crystal structures. Bentley’s six-sided stars and plates are just a few of the many varieties of structures that exist – prisms, columns, needles, triangular crystals, twelve-branched stars and irregular shapes are just some of the structures that can be grown under specific environmental conditions and are also found all over the world. This probably wouldn’t be known if it wasn’t for the work of Wilson Bentley.

'Every crystal was a masterpiece of design and no design was ever repeated'. A photograph of one of Wilson Bentley's snowflakes. Source: The Guardian.

‘Every crystal was a masterpiece of design and no design was ever repeated’. A photograph of one of Wilson Bentley’s snowflakes. Source: The Guardian

Today in Mill Brook Valley there is a small museum dedicated to Bentley and his work, with walls lined with his photography and his original contraption for taking these photographs. I would love to see this one day. What really captivated me about Bentley and his work was his unrelenting curiosity and his drive to share the beauty of snow crystal  structures that would eventually be scientifically translated. This is why, for similar reasons, I like using time-lapse photography to capture the dynamics of glaciers – images are not only scientifically valuable, but also resonate with everyone regardless of their knowledge of snow and ice.


Further reading

This article and this article from the Guardian on Bentley’s photography

The Snow Tourist by Charlie English which contains a detailed chapter on Bentley’s life work and also more generally on studies of snowflakes. I plan on writing a review when I have finished the book – it’s very good so far!

Advertisements

PhD Update: September 2016

I have been in Svalbard (again) for most of September, collecting images from our 14 time-lapse cameras that we have based in Kongsfjorden and Tempelfjorden. We haven’t seen these cameras since May 2016 (Kongsfjorden cameras) and August 2015 (Tempelfjorden cameras) so it was quite nerve-racking to go back and see if everything had worked. We had a couple of disappointments but generally the retrieval was a success, with approximately 130,000 photos collected in total.

Ten time-lapse cameras were deployed in Kongsfjorden last May (click here for more info on the deployment). Eight of these were installed on Collethøgda, overlooking Kronebreen, a fast-flowing marine-terminating glacier at the end of the fjord.  It is hoped that the close array of images from these cameras can be used to generate three-dimensional time-lapse sequences using a technique called Structure-from-Motion (SfM) which uses images from multiple angles to generate 3D point clouds of a target.

The other two cameras were installed by Kongsbreen and Kongsvegen, the two glaciers adjacent to Kronebreen. The data from these cameras form part of a longer-term project to monitor glaciers in the Kongsfjorden area. It was of particular importance to the influence of submarine melting on glacial retreat in this area.

Our camera at Kongsbreen... survived and still working! (September 2016)

Our camera at Kongsbreen… survived and still working! When were first started installing cameras in Svalbard, we would bolt them into the bedrock and use guide wires to stabilise them. Over time we have learnt that building cairns around the tripod legs is just as effective and takes much less time. We had a particularly long time around this camera site in May to build a large cairn… and do some sunbathing.

Due to bad weather, it proved difficult to access the camera sites and were limited to only two days of helicopter time to retrieve data. We had hoped to have enough time to survey and maintain each of the cameras so that they could run over the winter season, but alas! that is the beauty of fieldwork – you have to work with the weather you have. We managed to retrieve all of the memory cards from the cameras in the end, but couldn’t complete the camera surveying.

Previously we had deployed 7 time-lapse cameras in 2014 and 8 cameras in 2015, so we knew we were being ambitious with 10. In total, 6 worked through the entire season collecting images either every 10 minutes or every 30 minutes. We have had better success in the past (5/7 in 2014 and 6/8 in 2015) so we were a little disappointed that we weren’t able to beat our personal best! All of the problems were related to the power supply – temperamental solar controllers did not recharge the batteries from the solar panels, and there were poor connections in the camera boxes that had developed over the duration in the field.

Overall, 48 000 images were collected from the cameras at Kongsfjorden – we have a good sequence from Kongsbreen showing multiple submarine plumes creating inlets in the ice front, good coverage over the front of Kronebreen to look at calving activity and surface velocities over the summer season, and a good sample dataset to begin looking at constructing 3D SfM time-lapse sequences.

As the weather was so limiting on our helicopter time, we also accessed the shoreline next to Kronebreen by boat, where we set up our 4K video camera (left over from the CalvingSEIS project last month) to record calving activity. We recorded an 11-hour 4K video sequence which provides some awesome close-ups on isolated calving events such as the one in the video below.

This work is so important to ensure the safety of tourists and scientists alike. Currently the minimum safe distance from a calving front is 200 metres, but accidents do still happen. The distance that ice can be thrown from a calving event is thought to be controlled by the height of the calving origin and the impact with the water. With this in mind, the minimum safe distance should be different for each calving glacier front in Svalbard. We hope that we can track projectiles from such calving events in this sequence to re-assess the distance that boats should be from calving  glacier fronts in Svalbard. It is likely that glacier calving fronts require different categories of risk based on calving activity (frequency and volume), ice cliff height and ocean temperature.

Preparing for our helicopter ride over to Tunabreen (September 2016)

Preparing for our helicopter ride over to Tunabreen. Mats (pilot, left) is ‘composing himself for the flight’ whilst Harold (technician, centre) is doing routine checks on the helicopter. Chris Borstad (UNIS, right) joined us on this trip to check the time-lapse cameras and survey the glacier surface using a laser scanner to look at crevasse propagation rates in the upper section of the glacier tongue.

After finishing in Kongsfjorden, we flew back from Ny Ålesund to Longyearbyen and got a lucky opportunity to fly to Tunabreen and collect data from 4 time-lapse cameras that have been there for over a year now (see here for information on the installations and other work in this area). They were meant to be collected in September 2015, but the plan had to be abandoned due to poor weather. We also planned to retrieve them at the beginning of this year, but the warm winter had left Tempelfjorden first without sea ice for snow scooter transport and then with too much sea ice for the boat season.

After some manic negotiations with the Sysselmannen (Governor of Svalbard), we got permission for two helicopter landings on Ultunafjell, where the time-lapse cameras were installed. When flying over, it was impressive to see how much the calving front has changed, even over the past couple of months. Normally there is one consistent submarine plume at the west side of the calving front (near to the camera in the image below) that is active throughout the melt season, creating an inlet in the calving front. This year though, it appears that a second strong plume at the east side of the calving front  has created a marked inlet (see far inlet in the image below). The upper section of the glacier tongue has also changed, with the crevasse field extending much further up-glacier than in previous years. Both the growth of the crevasse field and the change in submarine plume activity could indicate a change in the subglacial conditions at Tunabreen.

The calving front of Tunabreen (September 2016)

The calving front of Tunabreen. The muddy water in front of the glacier is where the submarine plume has been strong enough to entrain sediment from the sea bed to the surface through turbulent mixing of freshwater and seawater. This promotes melting of the ice below the waterline, which has created two inlets in the calving front this year – the first is closest to the camera with a very visible plume adjacent, the second is the marked bay on the far side.

Three of the cameras on Ultunafjell were entrained on the calving front and lower section of Tunabreen, all of which had captured images till now. Unfortunately, due to unknown circumstances, two of the cameras were taking images that were out of focus when they came back on in the spring (after hibernating over the winter). It is likely that either someone has been up there, taken a look at the cameras and accidentally knocked the focusing ring on the lenses; or that high winds caused vibrations in the camera box that gradually shifted the focusing ring.

It’s a new set of circumstances for us anyway! From now on, we will fix the focusing ring in position by taping each lens. Luckily, one camera did not experience focus drift so we have three sets of images from August till November 2015, and one set of images from May to September 2016. This is plenty to work with and will give us a nice dataset to extract velocities and calving rate from.

The fourth camera is positioned further up the ridge, looking at the upper section of the glacier tongue where crevasses are begin to form and propagate. It was installed to monitor an array of strain meters that were set out on the glacier surface, measuring the rate at which crevasses were opening and the rate of longitudinal stretching. The relative distance between each strain meter in the images can be used to ensure that the strain meters are accurately measuring changes at the glacier surface. This camera has captured images every 10 minutes from August 2015 till September 2016, which is a great success. The camera has been surveyed and will now continue to take photos through the rest of 2016 into 2017, providing a complimentary dataset for Chris Borstad and the University Centre in Svalbard (UNIS) to use with other on-glacier instruments.

sampleimg

Sample image from the time-lapse sequence at the upper section of Tunabreen. These images monitor an area that is 497.6 m x 331.7 m. The strain meters are difficult to find in this image, with each strain meter box only represented as a 2 x 2 pixel square!

sampleimg_closeup

A close-up of the strain meters. The people in the image are myself and Doug Benn, part of the team that installed the seven strain meters on Tunabreen in August 2015.

So, in total, we have collected approximately 130,000 images in this month, providing us with a third consecutive year of time-lapse data at Kronebreen, and new insights into processes at Tunabreen and Kongsbreen. From these images, we should be able to extract a record of surface velocities, calving rate, submarine plume activity and crevasse propagation from each glacier. I am now back from Svalbard for this rest of this year to begin processing this data and enjoy Edinburgh in the autumn/winter season.

A tame ptarmigan at Tunabreen (September 2016)

A ptarmigan at Tunabreen. This little guy was walking along with me to visit our third time-lapse camera. Ptarmigans often nest around our cameras at both Tunabreen and Kronebreen – this makes for lovely images, but is a hinderance for photogrammetry processing!


With thanks to the following for making this fieldwork possible: Richard Delf (University of Edinburgh), Jack Kohler (NP), Chris Borstad (UNIS), our helicopter pilots Mats Larsen and Harold Edorsen, our skipper Wojtek Moskal,  and all in the NP Sverdrup station in Ny Ålesund.

A detailed look at the front of Kronebreen, Svalbard

Thanks to some nice boat trips this September (and a good zoom lens on my camera*), we were able to take a detailed look at the calving front of Kronebreen, a fast-flowing glacier which terminates into Kongsfjorden, an inlet on the west coast of Svalbard. Since 2011, Kronebreen has been retreating significantly faster (approx. 100m per year) and the front (where the ice meets the fjord water) has changed drastically. What is interesting is that the calving front does not appear uniform – the distinctiveness of each section indicates that different processes are active. If this intrigues you then please read on! If not, then sit back, relax, and enjoy the pictures at least!

All photos were taken on the Sony NEX-5R with the Sony E 55-210 mm f/4.3-6.3 lens and the Sony E 20 mm f/2.8 lens.

Kronebreen (centre) viewed from the west. Kronebreen shares its southern (right) margin with Kongsvegen, a slow-moving surge-type glacier that has been fairly inactive for the past couple of years. The glacier adjacent to Kronebreen, separated by the mountain Collethøgda (left), is called Kongsbreen. Kongsbreen has been retreating from the fjord onto land since approximately 2014 (September 2016)

Kronebreen glacier (centre) viewed from the west this September (2016). Kronebreen shares its southern (right) margin with Kongsvegen, a slow-moving surge-type glacier that has been fairly inactive for the past couple of years. The glacier adjacent to Kronebreen, separated by the mountain Collethøgda (left), is called Kongsbreen. Kongsbreen has been retreating from the fjord onto land since approximately 2014.

Kronebreen (‘crown glacier’) is a heavily crevassed glacier tongue fed from a large ice field called Hotledahlfonna, flowing between 1-5 metres per day into Kongsfjorden, an inlet on the west coast of Svalbard (see here for more information on Kongsfjorden and Svalbard). The front of Kronebreen is approximately 3 km, forming an impressive ice face that deposits ice into the fjord to form icebergs. This process is called calving and is one of the main processes by which ice is transferred from land to sea. For more information on Kronebreen, please refer to another post I wrote on why small glaciers are important to study here.

In many settings (such as at the large glaciers found in Greenland), the front of a glacier is a tall cliff of ice from which large sections of ice calve off. From studying the front of Kronebreen though, it is apparent that the ice face is not uniform. Certain sections of the ice front look entirely different from one another, suggesting that the controls on calving are also different. There are five sections that are particularly interesting:

Numbered sections of the calving front of Kronebreen (Landsat image from 09/07/2016, downloaded from USGS LandsatLook Viewer)

Numbered sections of the calving front of Kronebreen (Landsat image from 09/07/2016, downloaded from USGS LandsatLook Viewer)

  1. The glacier margin next to Collethøgda
  2. The retreated region around a large submarine plume that forms during the summer melt season each year (since at least 2011)
  3. The pinnacle – the middle section of glacier that is the furthermost point of the ice front
  4. The ‘traditional’ tall ice cliff at the front of the fastest-flowing section of the glacier
  5. The shared margin with Kongsvegen

1. The glacier margin next to Collethøgda

The north margin of Kronebreen. This section tends to move much slower as the ice flows against the land which generates shear stress. A lot of the land (at the foot of Collethøgda) is ice-cored - as the glacier has retreated, ice has been left isolated at the margin which becomes buried over time due to landslides and rockfalls (September 2016)

The north margin of Kronebreen. This section tends to flow much slower as the ice shears against the land from lateral drag. A lot of the land (at the foot of Collethøgda) is ice-cored – as the glacier has retreated, ice has been left isolated at the margin which is buried over time due to landslides and rockfalls from above.

A close-up of the north margin (September 2016)

A close-up of the north margin. The ice face appears broken up because it has been subject to compressive flow deformation further upstream – marginal crevasses form as the ice shears. These crevasses form weaknesses in the ice front when the ice is transferred downstream. Generally the calving activity in this margin zone is caused by exploitation of these weaknesses, producing ice falls and small collapses.


2. Region around a large submarine plume

Undercutting at the waterline exposed during low tide. Turbulence at the interface between the ice face and the fjord water causes undercutting. During low tide, this undercut section is exposed which creates a weakness in the front and promotes calving activity driven by gravitational potential energy (September 2016)

Undercutting at the waterline exposed during low tide. Turbulence at the interface between the ice face and the fjord water causes undercutting. During low tide, this undercut section is exposed which creates a weakness in the front (from the undercutting) and the ice flow marginally increases (from the reduction of hydraulic head). It is thought that there is a correlation between calving activity and tidal patterns, but it has been challenging to thoroughly test this hypothesis. This undercutting occurs over the entire front of Kronebreen, with a tidal range of roughly 1 metre (excluding spring/neap tides). Marked undercutting is noted to occur in areas where a submarine plume is active.

A significant undercut chasm located where a submarine plume emerges from consistently year on year. The submarine plume generates a marked amount of turbulence in the water column which cuts/melts the ice column beneath the waterline and causes small collapses of ice from above the waterline (September 2016)

A significant undercut chasm located where a submarine plume emerges from Kronebreen consistently year on year. In the summer melt season, meltwater is channeled through large tunnel networks in the glacier, exiting into the fjord as large bodies of freshwater. The freshwater rises as it comes into contact with the fjord as freshwater is lighter than salt water, creating a turbulent column of water that promotes melting/cutting in the adjacent ice column below the waterline. The plume is visible at the surface when it is strong enough to generate turbulence through the entire water column (click here to see a visible plume at Tunabreen in August 2015). Turbulence from the plume also can cut back into the ice column as we see at Kronebreen, forming chasms in the ice front where melt-back beneath the waterline causes small collapses above the waterline.


3. The pinnacle – the middle section of glacier that is the furthermost point of the ice front

The 'pinnacle' of the calving front at Kronebreen. The face is not the usual ice cliff you tend to see in marine-terminating settings. Instead, the calving front is composed of ice columns which look separated from above but are connected beneath the waterline (September 2016)

The ‘pinnacle’ of the calving front at Kronebreen (viewed from the north side). The face is not the usual smooth, clean-cut ice cliff you see in marine-terminating settings. Instead, the calving front is composed of ice columns which rest at different orientations to one another. Ice flow from behind the calving front is pushing these columns into the fjord, often causing these columns to calve into the fjord with rotations… and big splashes!

A close up in to the pinnacle section of the calving front. Internal collapses are common at Kronebreen, where ice behind the front collapses (September 2016)

A close up in to the pinnacle section of the calving front. Internal collapses are common at Kronebreen, where ice behind the front collapses. This has been known to form columns of ice which appear isolated from the rest of the calving front, similar to geological sea stacks. Such ice columns are attached to the rest of the front beneath the waterline.

The pinnacle, viewed from the south side of Kronebreen. The front is approx. 50 metres high in this section. For some sense of scale, see the bird perched on the top of the ice column in the centre of the image (September 2016)

The pinnacle, viewed from the south side of Kronebreen. The front is approximately 50 metres high in this section. For some sense of scale, see the bird perched on the top of the ice column in the centre of the image. It is uncertain exactly why this preserved pinnacle exists at Kronebreen. It is likely that the section is pinned to a region of the sea bed, and its shape is exaggerated by two submarine plumes which are active either side of the pinnacle.


4. The ‘traditional’ tall ice cliff at the front of the fastest-flowing section of the glacier

Tall ice cliffs at the fastest-flowing section of the glacier front at Kronebreen (September 2016)

These tall ice cliffs are a common feature of marine-terminating glaciers. The steep profile of the face is maintained by the fast delivery of ice to the front.  At Kronebreen, this part of the ice face is at the front of the fastest-flowing section, reaching velocities of 5 metres per day  in the peak summer melt season. Calving activity is infrequent here, but events are often large with failures through the entire column of ice (at least above the waterline). Click here for an example of a calving event in this area captured in September 2015.


5. The shared margin with Kongsvegen

Dirty ice at the shared margin between Kronebreen and Kongsvegen (September 2016)

At the divide between Kronebreen and Kongsvegen is a section of dirty ice. This is a lateral moraine. As Kongsvegen has flowed into the side of Kronebreen, sediment has been squeezed and thrust to the surface between these two glaciers. The layers of sediment between the ice can easily be exploited and creates weaknesses at the ice front. There are often small ice falls and collapses in this area, with few large failures.

Where Kronebreen meets Kongsvegen (September 2016)

The shared margin from above. This large section of dirty ice is the lateral moraine where Kronebreen (left) meets Kongsvegen (right). A large inlet has formed (since approx. 2014) just to the south of the lateral moraine at the front of Kongsvegen. This inlet is roughly 200 metres deeper than the rest of the calving front, thought to be the result of weaknesses at the lateral moraine. This could be the beginning of Kongsvegen’s separation from Kronebreen. Another explanation is that a large meltwater channel exits the glacier here, exploiting weaknesses in the lateral moraine to create a large channel that has promoted calving activity in the immediate vicinity. From looking at this photograph, it is apparent that part of the inlet has now retreated to land (see the two small black sediment beaches that are in the inlet). The gap between these two beaches could be the location of this channel opening.


* All photos were taken from approx. 500m from the glacier front. It is generally advised to be AT LEAST 200 METRES from the front of any calving glacier in Svalbard. Being close to calving glaciers is dangerous and could result in serious injury or even death if a calving event were to occur. See here for more information issued by the Governor of Svalbard and Norsk Polarinstitutt, or google ‘glacier calving accident’ if you want to scare yourself silly.


Further reading

AntarcticGlaciers.org article on stress and strain in glacial environments

Mauri Pelto’s AGU blog on the mutual margin of Kronebreen and Kongsvegen

Calving activity at Kronebreen by Anne Chapuis et al. (2010)

Submarine plumes in Greenland by Kristin Schild et al. (2016)

PhD Update: August 2016

The month started with troubleshooting ongoing uncertainties with projection of measurements from the two-dimensional image coordinates to real-world positions. This is a crucial step in photogrammetry for gaining measurements from images.

The CalvingSEIS group camp set-up besides Kronebreen glacier. The two tents contain valuable equipment - mainly the radar systems and the lidar system which were being used to scan the calving front of the glacier (August 2016)

The CalvingSEIS group camp set-up besides Kronebreen glacier. The two tents contain valuable equipment – mainly the radar systems and the lidar system which were being used to scan the calving front of the glacier (August 2016)

Our software (called PyTrx) is currently projecting points slightly off from their actual position. We have been looking at determining the position of the glacier front and comparing this to LandSat satellite imagery to better understand why this offset is occurring. Knowing the terminus position through time at high-frequency intervals is also useful in determining where calving rates are highest in relation to the upglacier surface velocities.

The second half of August saw us begin the transition from time-lapse photography to videography, with our first attempt to capture calving activity at Kronebreen glacier (Svalbard) using video cameras as part of the CalvingSEIS project. The main aim of CalvingSEIS is to better understand calving activity at ocean-terminating glacier fronts using a variety of high-precision, high-frequency techniques – seismic detection, radar scanning, lidar scanning, submarine acoustic monitoring and echo-sounding, and time-lapse photogrammetry/videography. The fieldwork involved setting a base camp on the shore next to Kronebreen and having all instruments running simultaneously each day to capture calving activity. It was very intense but I am so pleased to have been a part of it and working with such a great group of people. I hope to do a separate post fully outlining the aims, the fieldwork and the outcomes of the CalvingSEIS project.

PiM (Pierre-Marie Lefeuvre, UiO) and I setting up one of the time-lapse cameras high up on the moraine to capture calving events at Kronebreen glacier. These cameras were capturing images every three seconds (August 2016)

PiM (Pierre-Marie Lefeuvre, UiO) and I setting up one of the time-lapse cameras high up on the moraine to capture calving events at Kronebreen glacier. These cameras were capturing images every three seconds (August 2016)

Next month I have to go back to Svalbard yet again to download data from our cameras which have been monitoring glaciers in the Kongsfjorden area since May 2016. I feel like the fieldwork is never going to end… and I really hope it doesn’t!

The Expedition: Solving the Mystery of a Polar Tragedy by Bea Uusma | Book Review

Normally the last thing I want to do after a hard day’s work focused on a polar subject is… read yet more polar-related stuff. There are multiple books gathering dust on my bookshelves that come under this category – Antarctica, polar explorers, Arctic history and folklore, popular science – lovingly given to me by friends and family who presumed that my interest was unrelenting. A small challenge I have now set myself is to read these, and this was encouraged by one book, which I picked up on a whim from the museum in Svalbard because I had nothing else to read at the time.

The Expedition: Solving the Mystery of a Polar Tragedy by Bea Uusma tells the true story of three Swedish men who vanished in 1897 whilst attempting to be the first to cross the North Pole in a hot air balloon. Thirty three years later, their bodies were stumbled upon at the shores of the harrowing White Island, which lies to the north-east of Svalbard. There was no concluding evidence to suggest how they had ended up there and why they died – they were found with ample provisions, warm clothing, functioning weapons and plenty of ammunition. Many speculated on the cause of death, from eating polar bear liver (known to be toxic) to carbon monoxide poisoning to suicide. But there has never been consensus on which theory is more likely, with little supporting evidence for each one.

The crashed balloon in 1897 (source: Wikipedia)

The crashed balloon in 1897, with expedition members Salomon August Andrée and Knut Frænkel. Photographed by the third expedition member, Nils Strindberg. Photography from the expedition was recovered from White Island in 1930 by a Norwegian group who stumbled across remains whilst on an expedition studying the glaciers of the Svalbard archipelago (source: Wikipedia).

Bea Uusma’s first encounter with the story occurred in the nineties when she began reading a book at a boring party about the subject. The author’s obsession grew from this as she spent fifteen years trawling through museums to find missing clues, compiling past theories, visiting White Island… and eventually providing her own theory with convincing evidence.

The narrative jumps from details of the expedition to Uusma’s own experiences trawling through museum after museum for evidence, and documenting her several attempts to visit White Island. The level of her obsession is obvious, with documented diary entries and thorough record of their diets in the lead up to their deaths. This is nicely broken up with images, tables, maps and diagrams. It kept me captivated and I was continually wondering whether she could conclude what really did happen rather than just add another theory to the mix. The ending conclusion is worth sticking around for. In my opinion, her evidence is very conclusive and a much stronger argument than any of the other theories out there. I’m bursting to write about it but I don’t want to spoil it for anyone else!

Frænkel (left) and Strindberg (right) with their first polar bear kill (source: Wikipedia)

Frænkel (left) and Strindberg (right) with their first polar bear kill. The three expedition members encountered many polar bears on the pack ice and on the shores of White Island. They lived off a diet of polar bear and seal for the majority of their time after the crash, which first began speculation that they died of dietary-related issues such as Trichinosis (a parasite found in undercooked meat), Vitamin A poisoning (from eating polar bear liver), lead poisoning (from their canned food), scurvy and diarrhoea (source: Wikipedia)

Overall, this book has a great balance of detailed documentation and the author’s personal exploration. Uusma gracefully navigates the trap of producing a very dry record of events. As I said, this book has really sparked an interest in reading more about early polar exploration, which I am quite embarrassed to say that I know little about (I hang my glaciology head in shame). Hopefully whatever I read next will be as good as this!

PhD Update: July 2016

The majority of July has been focused on widening the applications of the Area class in PyTrx, which is being developed to determine real world areal data from oblique photography. Specifically I have been working on detecting visible plume extent. Unlike surface lake extent, plume detection has proved difficult to automate due to poor contrast. However, I have made an alternative set of functions to manually distinguish plume extent which look promising for yielding high-frequency, high-resolution areal data.

A submarine plume is an upwelling of freshwater from an ocean-terminating glacier terminus. A sediment-rich, ‘muddy’ area of water forms in cases where this upwelling reaches the ocean surface. This upwelling is largely caused by the density difference between fresh water and salt water, creating a column of turbulent water that promotes melting of the glacier front in its immediate vicinity. This submarine melting is considered to be one of the main controls on ice calving in tidewater settings. Plumes also provide significant feeding grounds for birds, seals and other local wildlife in the area – small organisms (which survive in saltwater) are stunned when they come into contact with freshwater from the glacier and are transported to the surface via the upwelling plume column. Birds feed off these stunned organisms at the surface whilst seals and other sea life feed from organisms entrained in the upwelling plume column.

tuna_plume

A submarine plume emerging at the ocean surface in front of Tunabreen glacier in Svalbard (August 2015)

Monitoring the areal extent of plumes is thus vital to understanding current and future glacier dynamics and wildlife feeding tendencies. Building on the automated detection of surface lakes (for more information, click here), I first attempted to automate the detection of plume extent from oblique time-lapse images based on pixel intensity (i.e. the RBG signature that denotes the ‘colour’ of a pixel). Unlike the detection of lakes, the contrast between the plume and surrounding fjord water is often quite poor. Sunlight glare off the water surface can also create false extents and obscure the plume (see example in the GIF sequence below).

A submarine plume at Kronebreen, with one image taken every hour.

A sample of our time-lapse images from Kronebreen glacier, Svalbard. This sequence covers twelve hours (real-time) from 9am-9pm during the peak summer melt season, with one image taken every hour. A submarine plume can be seen in the foreground growing and shrinking through the course of the day.

In most instances the plume extent needed to be manually defined, hence I created a collection of functions within PyTrx* for manual extent definition. The extent is defined by manually plotting points onto the image that denote the plume’s boundary. The first GIF below shows those points plotted on to each image, forming polygons that show the plume’s extent over a twelve hour period (9am – 9pm, July 2015).  The plume extent grows and subsides over the course of a day during the summer melt season.

These points are projected from the image plane to the real-world scene using a process called georectification, for which the following parameters need to be known:

  1. Camera location
  2. Camera pose (represented as the camera’s yaw, pitch and roll)
  3. Discrepancies between the image plane and the real-world scene created by assymetry in the camera lens and misalignment between the camera lens and camera sensor (called the radial and tangential distortions)
  4. Known geographical points in the landscape, such as peaks and geographical features
  5. Digitial Elevation Model (DEM) of the landscape.

These parameters are used to mathematically represent the translation between the image environment and the real-world environment. The real world coordinates of the points can then be used to calculate the actual area of the plume extent, simply by forming polygons from the set of coordinates.

plumeextentgif

plumepolygif

Plume detection in the image scene and projection to the real-world environment (polygon overlaid onto a scene from Landsat). Note: the projection is slightly wrong as you can see the plume origin is slightly off-centre from the ice embayment. We are still trying to diagnose why this is happening. It is likely to be an unknown anomaly in the projection, or a mismatch between the polygon and Landsat datums.

This short twelve-hour sequence shows promising results and the method will next be implemented on larger image sequences to observe longer-term variations in plume extent. It is expected that the plume extent is controlled by melt volume, the rate at which melt is transferred to the glacier front, fjord temperature/salinity and prevailing wind direction. Over the twelve hour period studied here, it is likely that the plume extent is controlled by changes in melt volume, which is understood to increase and decrease on a daily basis. There are few longer-term observations of plume dynamics at such a high temporal resolution, so it will be interesting to see exactly how a plume evolves over a summer melt season.

Graph showing plume extent over the course of a twelve-hour period in the summer melt season at Kronebreen glacier, Svalbard.

Graph showing plume extent over the course of a twelve-hour period in the summer melt season at Kronebreen glacier, Svalbard. These areas have been calculated using PyTrx and the set of functions for manual extent detection and projection.

 

* PyTrx is a set of photogrammetry tools specifically designed to obtain measurements (length, area and velocity) from oblique photography in glacial environments. PyTrx is programmed in Python and largely uses functions from the OpenCV toolbox. The first version of PyTrx will be made freely available at some point in the near future. 

 

PhD update: June 2016

I have spent this month focused on developing PyTrx, our analysis software for oblique time-lapse image sequences. In particular, I have been working on a set of functions that will automatically detect areal features in glacial environments, such as surface lakes and submarine plumes, and transform pixel regions to real world areas. I have been testing this on an image sequence from Kronebreen glacier in Svalbard, where we captured several surface lakes filling and draining over a summer melt season.

June kicked off to bad start when I agitated an old knee injury whilst at the gym. I’ve had bad knees for a large part of my life and thought most of the problem had been sorted after surgery in 2008. Unfortunately I am now looking at intensive physiotherapy and perhaps another round of surgery after rupturing a ligament… so I have been off my feet for a lot of this month.

But, turning a negative into a positive, it has meant that I have been able to concentrate on developing an additional module to PyTrx. PyTrx is the software that I have been developing as part of my PhD to obtain measurements from oblique time-lapse imagery – ‘Py’ signifying that it is coded in Python, an open access computing language; and ‘Trx’ to indicate that the software can be used to track measurements through sequential images in time. This was initially implemented to obtain velocity measurements from a time-lapse images of a glacier in Svalbard. These time-lapse images also showed several lakes pooling at the surface of the glacier, which simultaneously drained. What is unknown is how the drainage of these lakes affect the velocity of the surrounding region. A separate set of functions was needed to measure the surface area of these lakes.

Supraglacial lakes filling and draining in the upper section of Kronebreen glacier, Svalbard. The sequence covers June to July 2015, one image per day.

Supraglacial lakes filling and draining in the upper section of Kronebreen glacier, Svalbard. The sequence covers June to July 2015, one image per day.

Lakes are distinguished in an image based on the intensity of each pixel, which is determined by the three colour channels – red, blue and green – also referred to as the RGB values. Given an RBG range of the ‘lightest’ and ‘darkest’ regions of the lakes, they can be automatically distinguished and masked. Knowing where the camera is located and its pose, the pixel location of the lakes in the image can be translated to their position in the real world. For their position in the real world, we can determine their surface area.

The lake extents, as seen in real world coordinates from above, filling and draining as they move downglacier (to the left of the image)

The lake extents, as seen in real world coordinates from above, filling and draining as they move down-glacier (to the left of the image).

The software effectively runs, showing the main lakes from the front of the time-lapse image sequence growing and connecting to cover a 100 m area (approximately). There are still two problems that need rectifying though:

1. The lake areas are quite noisy, with ‘false’ lakes detected in some of the images. The next step will be to apply a filter to exclude areas that are too small to be lakes

2. Part of this sensitivity is due to changes in illumination from image to image. The most apparent illumination differences can be seen, with spikes in the total surface lake area associated to detection of these ‘false’ lakes. Unfortunately it is very difficult to change this in the software as it relies on the RGB value of the pixels, which are affected by the change in illumination. The only options are to either select images with more consistent illumination or apply an averaging algorithm to smooth out irregular area data.

Plot showing the surface area of all the surface lakes from image to image (02/06 - 13/07/2016)

Plot showing the surface area of all the surface lakes from image to image (02/06 – 13/07/2016). The three spikes in the data show where ‘false’ lakes have been detected due to changes in image illumination.

There have also been a couple of issues with the translation from pixel areas to real world areas (which is done here using a process called georectification), which has been ongoing for a while now. This process is not only used to determine real world areas, but also other real world measurements such as distance and velocity. It is hoped that when these problems are solved then PyTrx will be made freely available for others to use. Despite these limitations, I’m quite happy with this part of PyTrx and hopefully next month it can be used to measure the area of other glacial features, such as submarine plume extent.

So. Broken knees, working code… and Brexit (I wrote a blog post on my thoughts regarding the EU referendum here). That sums up my month. It’s not been the greatest month but I’ve tried to make the most of it. Hopefully July will go a little better.