I’ve recently seen discussion about applying the tried and tested methods of Arctic radar altimetry to Antarctica. Of course, people have already done this with mixed results (e.g. Schwegmann et al., 2016; Kacimi and Kwok, 2020), but I think people might now fancy making a product (in addition to the ESA CCI one which isn’t updated). Radar-based sea ice thickness products for the Arctic have a long history, but generally rely on the assumption of total radar penetration of the overlying snow. This is a really important assumption, that I don’t think is portable to Antarctica.
The basic idea underpinning most radar-satellite retrievals of sea ice thickness is that Ku-band radar waves emitted from altimeters on EnviSat and CryoSat-2 penetrate a dry, cold snowpack and return to the detector after bouncing off the ice surface. By working out how far the ice is sticking out of the water, we can estimate sea ice thickness.
What are the challenges in Antarctica?
There are a couple of things about Antarctica that mean this may not work consistently there. They mostly have to do with there being a lot more snow on Antarctic sea ice. Some sources put Antarctic snowfall at three times more than the Arctic (Holland et al., 2021). This is for a few reasons: the ice drifts more freely allowing leads to open up, and it’s surrounded by the Southern Ocean which means moisture transport is more abundant. This produces a number of asymmetries by comparison to the Arctic:
The ice freeboard is sometimes (often?) zero or sub-zero
The snow is sometimes so deep and heavy that it pushes the sea ice to the waterline and even below it. When that happens, we can’t convert a freeboard measurement into thickness. For observations and descriptions of zero or negative freeboard see Wever et al. 2021, Perovich et al. 2004, Lange et al. 1990, Weissling et al., 2011, Nicolaus et al., 2009, Massom et al.. 1997, Eicken et al. 1995, and Ozsoy-Cicek et al. (2013). It’s worth saying that there’s also an observational bias, where ice camps and helicopter landings are much more feasible and safe on thicker ice with larger freeboards. So observational experience would lead you to underestimate the prevalence of zero freeboards.
Brine wicking blocks radar waves
Figure and caption from WIllatt et al., 2010. Shows brine wicking and negative freeboard.
Because of the larger weight of snow, when the sea ice surface is submerged the overlying snowpack can flood. This can cause brine to wick up into it. Basic electromagnetic theory tells us that radar waves from satellite altimeters can’t penetrate brine wetted snow – as such, we can’t measure the height of the ice surface. The radar waves just don’t make it. This isn’t so much because of the salt scattering the radar waves itself, but because the presence of salt means liquid water can exist in the snowpack at temperatures well below zero. Snow wetness above 1% is generally considered to be a show-stopper. Brine-wetted snow is again often observed in the Antarctic, although its prevalence is hard to quantify (Wever et al. 2021; Willatt et al., 2010; Rapley and Lytle, 1998; Massom et al., 1998; Massom et al., 2001).
Figure and caption from Massom et al., 2001
However, a lack of understanding of this might make you think that you’re taking a real sea ice freeboard measurement. Because brine can travel up fairly high in the snow and freeze there to form snow-ice, and your ice freeboard is typically low, you may retrieve radar freeboards that look normal by Arctic standards. Even though the radar waves that return to your detector have not returned from the ice surface.
However, there are some areas such as the Weddell Sea where the sea ice is thicker, so less prone to negative freeboards and brine wicking. Here we have a separate issue:
Snow Metamorphism
A core of some snow on Antarctic sea ice (my photo). There’s a continuum from large-grained, icy, highly scattering snow to basically clear superimposed ice at the base.
Snow in Antarctica often doesn’t all melt away in summer (Andreas and Ackley, 1982). I recently encountered this in the Weddell Sea, where we found a thick layer of icy snow, and superimposed fresh ice at the base of the snow left over from the previous year.
When snow is “left lying around” over the summer, it becomes impenetrable to Ku-band radar waves – the snow I saw was impenetrable to even a knife. That's because its grains undergo metamorphosis, and become extremely large (see top of core in picture). And any surface melting percolates down to the bottom of the snow and refreezes in a layer of superimposed ice (see bottom of core in picture).
Again, if your radar waves encounter a large-grained snow layer, or a superimposed ice layer like this, they’re very likely to scatter back to your detector before reaching the sea ice. This might make you think that you’re measuring the sea ice surface, but actually you’re just measuring the height of the surface of last summer’s snow. This might help you retrieve a “sensible” value, but it won’t actually represent the sea ice thickness.
What we need
In order to actually believe an Antarctic SIT product, I would like to see three things:
Sea ice thickness data should meaningfully incorporate radar data.
I’d like to see that the radar freeboards actually impact the sea ice thickness figure. Some sea ice thickness retrievals in Antarctica have used the “zero freeboard” assumption (e.g. Kurtz and Markus, 2012), and estimate sea ice thickness assuming the ice surface is always at the waterline. This works much better for laser-altimetry than radar. With radar, your SIT is entirely determined by how much weight of snow is assumed to be sitting on the ice and the ice density, rather than your very low CS2 freeboard measurement.
Because there’s a seasonal evolution in the amount of snow that’s assumed to be on the ice, this imparts a seasonal evolution in the sea ice thickness – all without a radar measurement. This means that we’re not really using radar to retrieve the thickness at all, unless we’re using radars to calculate the weight of snow. Which takes me onto my next point:
A decent and interannually varying snow cover product
Ku-band radar waves often don’t reach the Antarctic sea ice surface in my opinion, which is sometimes below the waterline anyway. This makes it much harder to use “dual-frequency” methods of snow depth estimation than in the Arctic. It’s tempting to just take the ranging differences between CryoSat-2 and AltiKa as the snow depth and argue that it matches Operation Ice Bridge snow measurements. There are two potential problems with this: Firstly, to do that you should evaluate against an OIB product that’s not systematically biased low by sidelobes (like the NSIDC quicklook product). Garnier et al. (2021) did this, and it makes sense that their low snow depths from underpenetration of CryoSat match the low snow depths of the quicklook-OIB product which originate from sidelobes (Kwok and Haas, 2015, Kwok et al., 2017, Webster et al., 2014). Secondly, OIB snow depths are a radar product, so as well as sidelobes they’ll suffer from very similar underpenetration biases as the satellite-altimeters when it comes to brine wicking. This would lead to false agreement and artifically good evaluation against a dual-frequency product.
To fully capture the interannual variability and trends, we’ll also need to develop an interannually varying snow product (I wrote a paper about the drawbacks of snow climatologies in the Arctic here). See Steer et al. (2016) for how much snow depths in East Antarctica varied from year to year.
Evaluation against thickness data (not freeboard data) over several years (not just replicating the seasonal cycle)
I’ve already mentioned the issues with evaluating a satellite radar product with an airborne radar product (OIB). What really needs to be done is the final thickness product needs to be evaluated against an independent thickness (or draft) product such as that from upward looking sonar buoys.
Furthermore, any new thickness product should be required to outperform a seasonally evolving climatology. It’s easy to get the seasonal cycle of sea ice thickness from climatologically accumulating snow and assuming that it’s hiding more and more ice over the winter. What will really indicate whether a sea ice thickness product is “working” is whether it captures year-to-year variability in measured thickness or draft – not the seasonal cycle.