The Case for a New Radar
Research radars are expensive. The “KuKa” radar that I worked with in Churchill in 2021, the Weddell Sea in 2022 and Rothera in 2023 costs hundreds of thousands of USD. This type of radar is also big! KuKa with its battery on its sled weighs around 150 kg, and ships in a truly gigantic box that needs a pallet jack or a forklift truck to move. In defence of KuKa, it is a science machine in a literal sense, but also metaphorically. It was custom-designed by a company full of hardware and software experts just for research into snow on sea ice, and has been a success story in terms of what we’ve found so far. And we’re not done; the data from the three campaigns listed above will keep us all busy for a long time.
But when you work with big, expensive science machines you do wonder: could we trade their costs against their benefits to our advantage? If we cut the cost by 95% and still learned more than 5% of what we would in the counterfactual, that would be a “good deal”, right? It’s more complicated than that because our time is valuable, and the fieldwork costs money too. But there’s clearly the potential for exploring lighter, cheaper machines for research.
A Raspberry-Pi is a small, low-cost computer. You can buy them from as little as $10 USD. As well as being small and lightweight, they’re also super versatile. For instance, they have a large array of input/output pins that can exchange analogue and digital information with sensors, displays, and mechanical machines. My first project with a Raspberry-Pi involved linking it to Twitter and a water-pump, and having it water a houseplant when you tweeted at its own dedicated account. It would then reply to your tweet with a video of it watering itself, along with a measurement of the soil moisture content before and after. I was messing around at the time, trying to get more familiar with Linux (the operating system that runs Pis, but also research supercomputers) and Python (the programming language that I use for science, but also scripting Pi commands).
It also turns out that radars don’t have to be expensive. In fact, they’re pretty ubiquitous in industry. They monitor the liquid levels in tanks, and tell you when you’re about to back your car into a fence. They estimate your car’s speed when you go past a speed camera (which triggers the flash-mechanisms that actually provides the legal measurement), and open automatic doors for you as you approach. Radar devices that do these tasks often operate in the K-band of the electromagnetic spectrum, with a rough frequency of 24 GHz. There’s a huge industrial appetite for these radars, which means they’re quite cheap. Historically, the radars have been operated with proprietary machines and the interfaces have only “spat out” summary statistics rather than the underlying data. But with the rise of the Raspberry Pi (and a similar but competing microcomputer called an Arduino) K-band radars aimed at industry are increasingly sold as standalone hardware that the user combines with the microcomputer. So we’re now in a position where we have cheap radars that can be controlled by a microcomputer, allowing all the “raw data” to later be analysed by a scientist.
If you’re a radar altimetry enthusiast/nerd, K-band will ring some bells. The altimeter aboard the CryoSat-2 mission operates in the Ku-band (~13 GHz) and the one aboard AltiKa is in the Ka-band (~36 GHz). The European Space Agency is planning a mission for launch in 2028 that will operate in the Ku and Ka-bands, and KuKa (our sled-mounted science machine) also uses those bands, hence the name. So these K-band radars sit right in the middle of those two frequencies. It would be nice if they actually used one of them, but you can’t have it all. This is not just unfortunate, it’s by design. You can’t just build a radar of any frequency and sell it to anybody, because the buyer could use it in a way that interferes with important, even strategic, communication devices etc. So there are certain “bands” of frequency that are carved out for free-for-all purposes, such as the K-band.
So a K-band radar mounted on a Raspberry Pi is not a “satellite simulator” for Ku/Ka-band missions, because its frequency lies between the two bands and not at or within them. However, I should say that KuKa is also not a satellite simulator, because its beam geometry is very different. That’s quite jargonistic, but it suffices to say that we can’t learn about how a satellite-mounted radar works in practice by dragging the same radar about on the ground. The intuitive explanation for this is that the curvature of a surface-based instrument’s wavefront is tighter, so it experiences a given flat surface as much rougher that it is. That is to say, points on a flat surface near the edge of the footprint appear further away in range. In contrast, the wavefront of a pulse from a satellite-mounted altimeter is extremely flat at the point where it reaches the surface, so a satellite-mounted radar will see a geometrically flat surface as actually flat, and therefore more specular. The surface will therefore probably be “seen” by the instrument as brighter. If you (quite reasonably) don’t understand this paragraph, then it’s not important; the key takeaway is that surface-based radars are not satellite simulators even when they match the frequency of the satellite radar.
So these K-band radars have the “wrong” frequency, and the “wrong” beam geometry. I should throw in that the range resolution is not very good either (it’s comparable to CryoSat-2, and much worse than KuKa). So what do they have to offer us as scientists? Firstly, they’re a great tool to explore how a radar works. The ability to hold a radar in your hand and point it at stuff (and yourself!) is super interesting. And then there’s the processing. It’s possible to look, line by line, at the code for many of these radars and understand exactly what every Fourier transform does, and how (for instance) spectral windows and zero-padding impact the result (again, these technincal terms not necessary to the point). I’ve learned a lot from mine, and I think you could use them as part of a taught course for students.
The second use case is understanding the physics of how radar waves interact with snow and ice, particularly as a function of incidence angle. This is still a relatively open question in satellite remote sensing of sea ice. In this respect, the K-band radar actually has an advantage over KuKa. Large radar instruments are generally set up on pivoting pedestals that reorientate the radar antennas at different angles, while keeping the radar in the same place. This means that they measure different snow at different incidence angles. This is compounded by the issue that they look at different snow, at different angles, *from different ranges*. To rigorously establish the control of incidence angle on radar backscattering, we would prefer to look at the same snow, at the same range, at different angles. You need a smaller, lighter, more nimble instrument to do this because it requires arcing the radar antenna radially around the patch of test snow.
Fieldwork in Svalbard
I applied for a grant from the World Climate Research Programme’s Climate and Cryosphere project to test one of these radars in Svalbard during a visit to the University Centre in Svalbard (UNIS). I was interested in setting up the a consumer K-band radar in the cheapest, simplest possible way and seeing if it worked. I only used “off the shelf” components for the setup, such a a standard 5V USB power bank, which powered a travel monitor (both bought from amazon). I housed the Pi and radar in a small pelican case, and drilled cable ports (also from amazon) into it with a dremel. The whole thing sat on a consumer camera tripod that I bought from Clas Ohlson in Tromsø which allowed it to be orientated at arbitrary incidence angles. I set the incidence angle with an inclinometer app on my phone. I operated the radar in the field with a standard, wired computer mouse which I had lying around my office. Data was written to a micro-SD card mounted to the Pi. It would have been possible to “spec out” the radar much more robustly, but I wanted to make the whole exercise as cheap, easy and intuitive to non-scientists as possible.
I was kindly hosted by Eero Rinne, who coordinated my visit with another field campaign and helped me join it. That was a campaign by scientists at NORCE testing out a drone-mounted ultrawideband radar, and connecting their radar measurements with in-situ observations and measurements from the ICESat-2 lidar altimeter. As well as being very aligned with my research interests, they also hailed from Tromsø! To complete the team, we were also helped and guided in the field by two of Eero’s masters students.
My visit began by visiting and radaring snow on land in Adventdalen, Gangskaren and Fardalsbakken. These were day trips by skidoo. In the second week of my visit we went on a multi-day trip to the sea ice near the old town of Svea, where UNIS has nearly finished building a new biology lab for teaching. We stayed in a nearby cabin, also owned by UNIS. We were at maximum occupancy, taking up all eight of the cabin’s beds. With no running water and an unheated toilet that needs a rifle to visit, it was my first real experience of “Cabin life”, which is a big thing in Norway. I’m glad to say I very much enjoyed it! We had a great rapport, and the science we all ended up doing individually was very complementary and synergistic.
We were also lucky to be visiting Svea at the same time as a UNIS graduate course in bio-optics and a graduate research project using an underwater ROV with a hyperspectral camera. I got to hang out with the ROV team on the first day of the Svea trip, which was super interesting! On the second day I joined the NORCE team to take K-band measurements on landfast ice further down the fjord. I was lucky to find some snow covered sea ice near the tide crack that was partially flooded, right next to some that wasn’t and scan both at multiple incidence angles, following up with snow-pits. I think this is going to provide a really interesting comparison. On the third day I joined the NORCE team again, and after a brief visit to the sea ice we visited a frozen-over lake where I repeated my radar measurements. As well as allowing me to contrast sea and lake ice, I’m fairly confident that the K-band radar could “see” the bottom of the freshwater ice over a range of incidence angles, which is cool (and scientifically interesting!).
Overall I think the campaign was a big success. The radar setup performed really well, and it was great to work so closely with new colleagues from Tromsø. I was very impressed by their ultrawideband snow radar and learned a lot about the more technical aspects of my field. I was also very impressed by UNIS itself and its capacity to facilitate high quality polar fieldwork. The same goes for Sacha and Oliver, the UNIS grad students who guided and assisted our work - I was stunned by how capable they were in the field, and they helped us all collect a lot more, and a lot higher quality data.
Over the next month I’ll be analysing the data from the Pi radar, and comparing it to the snow pit data that I and others gathered. As well as being scientifically interesting on its own merits, I think the results might open up interesting avenues for further research (and the funding that would support that research). Stay tuned for more K-band action in future.