After expeditions to the subantarctic Bounty and Antipodes Islands during the past two penguin breeding seasons, we are now sitting on thousands of images taken by drones flying above penguin colonies, capturing a snapshot in time. These images allow us to count penguins on the ground and thereby obtain a reliable and, most importantly, reproducible population estimate of Erect-crested penguins.

But how do we analyze these data? For the first trip in 2022, we conducted full drone surveys of both the Bounty and Antipodes Islands. It took our team more than a year to manually annotate the images of the penguin colonies photographed from above. It was a long and protracted process that kept us busy even while we were back on the Antipodes last season, recording even more survey data!

More recently, our partners at DroneDeploy added a deep learning algorithm to their web service using our data to train their AI. The model is currently in a beta state and is improving, but it needs more work.

There is also a third option—the power of the crowd. Specifically, the countless volunteers who devote some of their time to annotate images and videos on the citizen science platform Zooniverse. The platform is already familiar with penguin data as it is used by our friends from Oxford University that have been running ‘Penguin Watch‘ on the platform for a few years now. They use trail camera images, so why not do something similar with our drone footage?

We few weeks ago, we started to put together our own Zooniverse project – Penguins From Above. Last Tuesday, we were ready to have a selected group of volunteers test our project. We only uploaded data from a survey of a single colony—a total of 317 images that combined to form a massive aerial photo of the Orde Lees colony on Antipodes Island.

Looking at the entire image, it is impossible to make out any individual penguins. However, when you zoom in, things become clearer.

So, we chopped up the full aerial photo into 1000 x 1000 pixel chunks and uploaded them to our Zooniverse project. On Tuesday night, the beta of our project was launched. We planned to advertise the beta the next morning on our social media channels to get the results more quickly. To our astonishment, when we woke up on Wednesday morning, it was all over. In a matter of hours, 85 volunteers had placed 85,000 annotations on the images we provided.

By converting the annotations to GPS positions, we ended up with a massive point cloud indicating the location of each penguin in the colony. Since each image chunk was analyzed by multiple users, we ended up with point clusters for each penguin.

Using the centroid of each point cluster, we obtained the information we were after—single points that mark each penguin in the image.

From downloading the data to having the first results took us maybe 30 minutes. The Zooniverse users counted 11,944 penguins—only 14 penguins off from our highly curated counts that took our team weeks to complete.

Clearly, the Zooniverse data needs to be groomed and validated. But we’re still excited about the possibility of using a different form of AI—‘Actual Intelligence’—for the survey data we have collected and will collect in the future. Now, we just need to get all our drone imagery ready for Zooniverse and let the amazing netizens help us count penguins.

By the way – everyone can help! Just head over to https://www.zooniverse.org/projects/tawakitom/penguins-from-above!

Note: We are terribly behind with updating our website! However, fund raising, data analysis and juggling several parts of the project (and students) at the same time, and maintaining an ever growing number of social media channels makes it a struggle to keep the website up to date. Sorry for that! We’ll try our best to update tawaki-project.org at some stage soon!

With that out of the way… why not listen to what we do instead?

During the past two expeditions to the subantarctic islands, our team was joined by journalists from both Radio New Zealand and NZ Geographic. During the 2022 expedition, NZ Geographic publisher James Frankham and award winning photographer Richie Robinson documented our work on the Bounty Islands, while in January 2024, Claire Concannon came along when the team was picked up at the Antipodes and briefly visited the Bounties. After NZ Geographic published a fantastic feature article last year, today the “Voice of Tangaroa” podcast covering our work was published.

https://www.rnz.co.nz/programmes/voice-of-tangaroa/story/2018922478/a-tale-of-two-islands-erect-crested-penguins

Voice of Tangaroa is a collaboration between RNZ’s award-winning science podcast Our Changing World & New Zealand Geographic, exploring the state of our oceans & the extraordinary variety of life that calls it home. New Zealand Geographic reporting for the Voice of Tangaroa is Public Interest Journalism funded through NZ On Air.

At 8.30am this morning, the Tawaki Project launched its crowd funding campaign towards TawakiCam, a live video stream solution to observe tawaki in their natural habitat.

Rather than writing oodles here, why not head straight over to the campaign page and see if this is something you want to support:

HTTPS://TAWAKI.CAM

Klemens, Hannah and Thomas start the day watching Germany’s second first round match at the Football world cup. After losing the first game against Japan, the German team must now beat Spain to stay in the game. The tight match ends up being a 1:1 draw. Luckily, Japan lost it’s second match so that Germany still had a chance to make it to the second round.

After the match, Thomas, Robin and Dave gear up to drone the final colonies of the Northwest coast while Jeff and Hannah head back to Crater Bay to do the same with the penguin colonies missed out on the other day. It is another gorgeous day with breath taking vistas. After droning the inaccessible penguin colonies, Thomas attempts to record a vertical panorama to survey a sooty albatross colony that appears to be glued to a vertical cliff covered in wet moss.

Both teams are back with accomplished missions before 5pm. After dinner, two camera loggers are prepared with AxyDepth dive loggers and deployed on two unsuspecting females from the rock platform colony closest to the hut. Fingers crossed we will get the first video data tomorrow!

Clear day with westerly conditions. Jeff, Hannah & Thomas make their way to Alert Bay. Encounter finger post above the cliff.

Fly drone manually way below cliff.

Smallish Erect-crested penguin colonies with a few individual Rockhopper nests on the fringes. Spotted Dougall Stream waterfall.

Then up the hill and across to the point separating Alert and Crater Bay. Flew more manual missions from the point way down below the launch spot.

One hour slog up to the top of the ridge.

Relatively easy going from there. Past giant petrel nests along the ridge almost all the way to the hut.

Easy drop to trail and home by 6pm.

The team today splits up in two group. One heading towards Mirounga Bay to assess the penguin situation – and access to them – consisting of Jeff. Robin, Bianca and Klemens.

The second group – Dave, Thomas and Hannah – has Orde Lees as their destination. The day is supposed to be clear so a good opportunity to get some droning done. But first, it is foggy. The drone team takes the route up towards Mt Galloway in an attempt to circumnavigate the worst of the tussock-fern nightmare between Orde Lees and the North Plains. By the time they reach to tops of Mt Galloway, they already start to regret their choice as the trek up the mountain already drained most of their energy.

Thomas carrying the drone kit, as well as a rotary hammer to install a time lapse camera at Order Lees certainly has had his fill before the day has even begun to clear. What’s worse, however, is that the traversal from the mountain top to Orde Lees turns out to be just as nightmarish as the direct route would have been: sections with shoulder deep tussock and fern patches.

The final climb up the ridge that separates Orde Lees from the plains pulls the last bits of energy out of the team. To their horror, they realize that they needed to cross one more saddle to the south to have direct access to the penguin colony. As it turns out, they now have to sidle along a steep slope to get over to the penguins. Given that they have already 4 1/2 hours of struggling through the difficult terrain, Dave decides that this is as far as he goes, while Thomas and Hannah push on. The slope is overgrown with chest deep tussock grass that is extremely effective at issueing paper cuts to whoever decides to hold on to it. With the grass being the only hand hold available, Thomas and Hannah’s hands and arms soon look like they have been through a torture session.

However, once they reach the Orde Lees colony, they are rewarded with the most amazing sights. Literally thousands of penguins  occupy expansive rock platforms at the foot of a bizarre pointy limestone ridge that an abstract painter could have dreamed up. Orde Lees islet a stone pyramid rising up to more than 100m as a backdrop completes the picture.

The noise in the penguin colony is deafening. Every now and then, a Skua flies over the penguins which creates an acoutsic wave of outraged penguin cries that washes from one end of the colony to the other. Almost every penguin on in the colony seems to be associates with an active nest. Some birds still sitting on eggs, others tending chick that are in their second week – little grey fluffy tennis balls with a penguin head stitched on at the top.

Thomas and Hannah fly the first planned mission on the Antipodes. It is difficult to set the mission boundaries on the drone controller screen as the underlying satellite imagery does not really show elevated portions of the map very well. The mission is a true nail biter as the plastic bee sometimes gets dangerously close to the hillside and once even shaved off the tips of the tussock that grows on the steep slope at the southern end of the penguin colony. But the drone manages to get the mission done without intervention from the pilot required.

At around 3pm, Thomas and Hannah make their way back to Dave. He has been lying spreadeagled in the tussock for the better part of two hours now, snoozing in the sun. Still, the prospect of having to walk back to the hut for another eternity does not seem appealing at all. As it turns out, the walk back to the hut isn’t quite as bad. Somehow the team managed to stay out of the thick tussock and fern by following a stream and then just zigzagging from albatross nest to albatross nest. The volume of water the three consume is outrageous and just underlines how strenuous this trip has become. They refill their bottles and water bladders three times.

Exhausted the three stumbled back into the hut around 6pm where the Mirounga Bay team is already busy preparing dinner. Dave sums up his day as follows:

“By midday I thought I was gonna die. But then it got worse.”

The mist is back with northeasterly conditions.

We are having problems getting the solar power to work. For the time being we rely on the generator to power our many devices which means our 60 l of petrol may not last the entire four weeks we are here. There is a stash of another 60 l, so we should be fine, but still annoying that the solar does not seem to do anything. Could be that the battery banks are too old. Or maybe there is a gremlin in the system we haven’t worked out yet. Or maybe it’s the constant fog?

With the weather not being very conducive of expeditions away from the hut, we busy ourselves with data analysis, primarily bird counts on orthomosaics from the Bounties. Processing the data is hampered by upload issues. The raw photo data is uploaded to the Drone Deploy servers and processed online. However, the upload often cuts out due to Starlink drops, and for some reason upload resumes don’t seem to work. Another problem that needs solving over the next few days.

Robin heads over to the Reef Point colony to do a ground count of active nests – she finds 374 nests. A subsequent drone survey of the colony flown by Hannah shortly after Robin’s return finds 886 individuals on the ground. Dave and Bianca head over to the Anchorage Bay colony around the corner (i.e. not the bit below the ladder) to count nests and ended up with an average 316 nests.

In the late afternoon, the first three GPS dive loggers are deployed on female penguin attending their chicks on the rock platform next to the ladder in Anchorage Bay. The wind picks up towards the evening and it start raining in the night.

Finally the mists have lifted. A strong breeze is blowing in the morning, but not too bad for flying the drone. But first a celebratory breakfast for Dave’s birthday. Thomas’s marble cake turns out slightly under cooked but the it is the good will that counts. Probably hoping to scavenge some of the birthday food, a couple of Antipodes parakeets blocked the way to both the toilets and the penguin colony.

In the penguin colony, hatching is underway. About half of the nests in the Anchorage Bay study colony have now small chicks. Which means that we can start our GPS tracking work relatively soon.

Therefore, Jeff and Robin are heading over to the South to see what the working conditions will be for us down there. Hannah and Thomas this time stay behind to take care of the drone surveys around the hut.

Therefore, Jeff and Robin are heading over to the South to see what the working conditions will be for us down there. Hannah and Thomas this time stay behind to take care of the drone surveys around the hut. The reduced southern party leaves at 9.30am and radios the others around 11am informing them that they have reached the summit of Mt Galloway. From here on, only InReach communication will be working. Jeff and Robin reach the south coast around 12.30pm and find penguins reasonably numerous and stretched widely along the coast. Some of the nests have chicks that are about a week old. The two also spot a half-leucistic penguin with a snow-white lower back, as if the bird was dipped in a tub of white paint.

Rain starts to set in soaking the two by the time they decide to walk back in the afternoon. They reach the top of Mt Galloway around 4.30pm where a gale is blowing making it difficult for them to stay on their feet. They are back at the hut by 6pm.

Meanwhile, Thomas and Hannah fly the drone from the hut to survey the penguin colony at Reef Point, just across Hut Cove. The mission is completed without any incident within 15 minutes. They then head over to Anchorage Bay to survey the length of the bay. Using pre-planned missions turns out to be very difficult to fly due to the steep cliffs bordering the penguin colonies. As a result, Thomas decides to fly missions manually which is a less ideal solution as the overlap and coverage of the missions often leave a lot to be desired.

Using a video feed, they discover a small Rockhopper penguin colony wedged into cliff crevasses behind Erect-crested penguin aggregations. Pretty much impossible to survey from above, so some re-thinking of the drone surveys are required here as well. Just before lunchtime, the team makes it across to Stella Bay to fly a brief penguin survey there. After a few manually shot survey photos, Thomas tries to record footage of Light-mantled Sooty albatross breeding in the cliffs above Stella Bay, directed by Klemens and Bianca who both spotted the birds the day before with binoculars.

Locating the nests high above the beach using the drone screen turns out to be rather difficult. Yet another drone problem we need to solve in the next four weeks. Just as the team prepared to head towards Crater Bay after lunch, rain started to set in, bringing the drone aspirations or today to an end.

In the evening, the team assembles around the hut’s table to process Bounty Island data.

Misty. Again.

We accounted for rain and wind to be problematic for our drone missions. Fog wasn’t on our radar. After yesterday’s slog for the southern party, we decides to take it slow today. Robin, Bianca and Klemens get started with counting Bounty Island orthomosaics using DotDotGoose.

Thomas, Hannah and Dave walk over to Anchorage Bay to fly the drone mission for our main study site. But halfway through the relatively short mission, fog wafts over the cliff and into the bay putting an end to this endeavour.

Meanwhile, Robin finished counting birds on Molly Cap and found that nearly 6,000 Salvin’s albatross were present during our drone mission (15.11.). Surprisingly, she found more than 200 penguins up there too; how the birds get there is anyone’s guess.

In the afternoon, Thomas scrambles through the tussock over to the penguin colony at reef point. That colony is so close, that drone missions can be flown from the hut. When he gets back to give it a first run, the fog is back.

First day on the mission. A team of four – Jeff, Robin, Thomas & Hannah – head toward South Bay to establish camp there before returning in the evening. Klemens and Bianca leave shortly thereafter to survey the cliffs between Stella and Anchorage Bay for Light-mantled Sooty Albatross nests. Dave will map the penguin colony in Anchorage Bay and determine progress of the breeding season, before laying out a network of tracking tunnels to monitor potential mouse incursion after the successful eradication in 2014.

But first… complete our daily sched with the mainland that we are all okay. While this used to be done via VHF radios or more recently satellite phones, Starlink offers a new convenient method – Whatsapp.

The South Bay team leave around 10am and after 30 minutes of reasonable progress run into the infamous Antipodes Islands tussock country that makes walking a nightmare. The tussock is growing so tall that it gives the walker two options – to balance on top of the tussock to try to take one wobbly step after the other to get from tussock to tussock or walk between the tussock plants that are often chest high. Both options make for very slow progress and drain the energy out of the party quick. Thick fog and unfamiliarity with the island make it difficult for the team to find the most suitable route towards the South.

By midday, they hear a penguin cacophony somewhere in the fog which means that the Orde Lees colony in the west of the island is not too far off. Not at all the direction they had planned to go, but nevertheless enticing enough to push towards the penguin noise. Just after midday, the team find themselves on the top of a cliff overlooking the penguin colony in the distance. The colony looks impressive and takes up a rock platform at the base of a steep cliff.

The majority of nests are Erect-crested penguins with a group of Rockhopper penguins lined up at the base of a rock overhang some way up the cliff. Overall a rough estimate of 2000 pairs is made. Thomas and Hannah decide to fly a manual drone mission, but the extent, shape and topography of the colony make it difficult for the pilot to get bearings. It is, however, clear that the drone missions must be rethought as the terrain poses significant problems for the standard way drone missions are flown. Just before 1pm the team continues their trek to the South.

Progress is slowed not only by tussock areas but also increasingly fern patches that are just as difficult to traverse. By 2pm, the team realizes that the involuntary detour to Orde Lees cost them the couple of hours it would need to get to the South today. They decide to reach the central plateau and stash the camping gear about two thirds of the way to the South before making their way back to the hut. As the afternoon progresses, the difficult walking conditions slowly but surely drain the energy out of the team. They develop a general rule of thumb by which albatross nests dotted around in the vegetation are a good indicator for reasonably walk-able terrain, while the absence of open seabird burrows is a sure indicator for terrible conditions.

Each an every step they take is followed by Reischek’s and Antipodean parakeets, the latter potentially eyeing up one or two of the tired team member for later consumption, should they succumb to the hardship of the trail.

Close to the top of Mt Galloway, the team comes across what looks like a Giant Petrel crèche where several petrel chicks have bunched up in the tussock with three or four adults standing guard.

During a much needed water break, the team is approached by an adult Antipodean albatross which after several attempts takes flight into the clearing mist. The last two hours back to the hut remain an incredible slog. The southern party returns to the hut exhausted and smelly by 7pm.

Klemens and Bianca also have an eventful day. Scrambling along the top of the cliffs between Stella Bay, Reef Point and Anchorage Bay turns out just as arduous as what the southern team experiences. Over the course of three hours, the two find a total of 3 light-mantled sooty albatross nests.

Meanwhile, Dave mapped the part of the Anchorage Bay Erect-crested penguin colony closest to the hut. About two thirds of the nests are still on eggs, the other nests have generally very small chicks that have just hatched. So it will be another three or four days before we can start deploying GPS dive loggers on chick rearing female penguins. The Rockhopper penguins are about two weeks behind. In the afternoon, Dave ran several tracking tunnel lines radiating out from the hut. Hopefully these tunnels will not pick up any mustelid activity.

In the evening the whole team discusses how to proceed from here. While the work on the Bounty Islands was straight forward and could be completed as planned, we will have to improvise quite a bit more on the Antipodes. For one, access to the south of the island is far more difficult that anticipated. That will have an effect on what we can do down there. We definitely want tracking data form down there. But at the same time, we want to reduce the number of trips we need to do across the island to n absolute minimum. Blood sampling will also have to be done with no frills – carrying centrifuges across the island is a less then appealing prospect. Drone missions will have to be toned down to the essential sites first before we can even start to think about covering additional sites around the island. How accessible these sites will be is another question.