Hook: In an evening space digest among the usual news about Starship and Mars rovers, there was a line tucked in that no engineer can pass by: "ESA repeated the legendary tandem mission ERS-1/ERS-2: two radar satellites Sentinel-1 temporarily placed in close orbit for a one-day repeat image of the same Antarctic region. This allows precise measurement of ice movement and gives a new view of 30 years of polar glacier dynamics." I first read it as another "ESA repeated something beautiful from the past" — and scrolled past. Then I got stuck. Because behind this line is hidden one of the most underrated architectural shifts in modern planetary engineering: 30 years ago two European satellites flew in tandem for 9 months and gave the world the first InSAR archive of glaciers, on which half of climate science about Antarctica still stands. Now ESA has repeated this maneuver on a new pair — and did it at a moment when Thwaites and Pine Island have already been accelerating for 30 years, and the West Antarctic Ice Sheet is balancing on the edge of a tipping point. And the most non-obvious thing about this — why exactly 1 day, not 6 or 12 (the standard Sentinel-1 cycle). Behind this is fundamental physics of radar coherence on fast-moving ice, which turned my understanding of what it even means to "measure a glacier" in 2026 upside down.
To understand what exactly ESA repeated, you first need to go back to 1995–2000. The satellites ERS-1 (European Remote Sensing, launched July 17, 1991) and ERS-2 (launched April 21, 1995) carried onboard a C-band radar (frequency 5.3 GHz, wavelength ~5.6 cm). ERS-1 had a tandem mode: from August 1995 to May 1996 ERS-2 flew in the same orbit as ERS-1, with a delay of exactly 1 day. This gave an interferometric pair with a temporal baseline of 24 hours — critically short for phase coherence on snow-ice surfaces. The standard 35-day cycle of ERS-2 gave interferograms in which the phase "fell apart" from any fallen snow or shift in snow cover. 1 day — this is a window in which snow doesn't have time to redistribute, and the phase remains connected.
What this gave the world — the first ever interferometric maps of glacier velocity in Antarctica and Greenland with millimeter accuracy. Before this no one knew how fast Thwaites flows. Before this no one knew that Pine Island was accelerating at 1% per year. Before this all "glacier kinematics" came from optical images with feature correlation between passes — accuracy ±50–100 m/year, at best. ERS tandem gave ±3–5 m/year.
Sentinel-1A (launched April 3, 2014) and Sentinel-1B (launched April 25, 2016, lost in December 2022) — successors of ERS — have a standard 6-day cycle when operating in tandem, and 12-day when operating a single spacecraft. For slow glaciers in interior Antarctica (50–200 m/year) the 6-day cycle works perfectly: the phase remains coherent because the snow doesn't have time to change much. For fast outlet glaciers (Pine Island — 4 km/year, Thwaites — up to 2.2 km/year at the front) the 6-day baseline is already a phase shift close to ambiguity, and phase noise rises sharply. And this is where that ESA trick kicks in, which the digest writes about.
To get a 1-day baseline, ESA temporarily changed the orbital phase of one of the spacecraft — literally "adjusted" the orbit to the needed flyover time. This isn't just "turn on a new mode". This is an orbital correction of tens of meters in altitude, calculated so that the second satellite flies over the same point 24 hours ± a few minutes later. The operation is called "tandem phase reconfiguration" and requires coordination with other Copernicus missions to avoid interference. And ESA does this not every year, but once every few years, as part of a special campaign, because it consumes fuel and disrupts the nominal configuration.
Architectural paradox: the most valuable data about glaciers is collected in a mode that cannot be permanent, because it consumes spacecraft resources, disrupts regular products and requires manual planning. The most precise science in planetary engineering is the one that cannot be put on an assembly line.
In 1995 the world thought Antarctica was stable "dead" ice, slowly accumulating in the center and slowly melting at the edges. ERS tandem showed that Pine Island has been accelerating since 1996 (+1.4% per year in 1996–2000, +0.8% in 2000–2007). ERS-2 + Envisat (2002–2012) added another 10 years. ALOS-PALSAR (L-band, Japan, 2006–2011) gave independent confirmation through other wavelengths: Pine Island accelerated by 42% between 1996 and 2007, Smith Glacier — by 83%, and Thwaites — was expanding without accelerating, but its eastern shelf doubled its speed. Sentinel-1 (2014–present) continued this line.
Now, in 2026, the 30-year time series of glacier velocity is the longest continuous InSAR monitoring of a massive object in human history. And it is also the most alarming. Because when you plot a graph "Pine Island velocity by year, 1996–2024" with 30 points, you get not noise, but a trend with an accelerating second derivative. Pine Island is now shedding ~140 Gt of ice per year — that's 0.4 mm/year contribution to global sea level, and the contribution is growing. Thwaites — ~80 Gt/year, and its Eastern Ice Shelf is already in the final stage of fragmentation (Sentinel-1 clearly showed the development of rifts 2020–2025).
This is what hooked me most of all. The loss of Sentinel-1B in December 2022 left only one spacecraft in space — Sentinel-1A. Without a second spacecraft the standard 6-day cycle is impossible, only 12-day (from A to the next pass of A). For glacier science this is a catastrophe: a 12-day baseline means complete loss of coherence on Pine Island and Thwaites (the glacier shifts 13–25 cm/day, snow has time to redistribute, and the phase interferogram falls apart).
And then ESA in 2024 decided to restore tandem mode through Sentinel-1C (launched December 5, 2024 on Vega-C) and conduct a 1-day repeat campaign in 2025–2026 over key regions: Antarctica (Pine Island, Thwaites, Amery), Greenland (Jakobshavn, Petermann) and selected mountain glaciers. This is not just "repeat the ERS mission" — this is revival of an orbital configuration that is 30 years old, at a moment when it is needed more than ever.
And here's the paradox: 30 years ago, when ERS-1/2 did their tandem, no one knew that Pine Island was accelerating. The data showed this for the first time. Now, 30 years later, we know that the glacier passed the point of no return, and we need data with millimeter accuracy to understand how fast it passed that point. Same technology. Same physics. Different meaning.
Here I dug deeper, and behind the dry terms I discovered an entire architecture that engineers built over 30 years and that almost no one sees from the outside.
1. Coherence as a physical, not mathematical concept. InSAR only works if the phase of the reflected signal on two images "remembers" its value. For snow and firn this means: snowflakes must not have time to redistribute between images. On a fast glacier (Pine Island, 4 km/year = 11 m/day) in 6 days the snow has time to completely restructure, plus the ice itself shifts by 66 m — and the phase interferogram gives aliasing (phase jumps of several π that cannot be correctly unwrapped). On a 1-day baseline the ice shifts by 11 m, snow hardly restructures, and coherence remains >0.7 in most areas.
2. Phase unwrapping as the limit of accuracy. Even with perfect coherence the phase interferogram returns phase wrapped in the range [−π, π]. To get absolute displacement in meters, you need to unwrap the phase (phase unwrapping) — this is an NP-hard optimization problem that on fast glaciers is solved through the Goldstein algorithm (1988) with branch cuts, and on modern data — through SNAPHU (statistical-cost, network-flow algorithm for phase unwrapping) and since the 2020s through neural network approaches (DeepLearning Unwrapping, ICESat-2 as a reference layer). Unwrapping error on a fast glacier — ±0.5–2 m per interferometric cycle, which converted to velocity gives ±50–200 m/year (for a 1-day baseline this is a lot). Therefore InSAR products from Sentinel-1 for fast glaciers are averages over dozens of pairs, not single measurements.
3. Tandem doesn't "measure velocity", it measures displacement over 1 day, and then this is averaged. It would seem that measuring displacement over 1 day is the weakest unit of measurement, because an error of ±2 m over 1 day = ±730 m/year. But ESA on 1-day pairs calibrates longer 6-day and 12-day chains through special algorithms (MEaSUREs, ITS_LIVE), and by averaging over hundreds of pairs the error drops to ±3–5 m/year. That is, 1-day tandem is not the final product, but a reference point for the entire system.
4. Speckle tracking as a fallback method. When phase coherence drops (fast glacier, heavy snowfall, blizzard), InSAR dies. Then speckle tracking kicks in — correlation of amplitude patterns between a pair of images with sub-pixel accuracy. Accuracy is worse (tens of m/year instead of units), but the method does not require phase coherence and works where InSAR is powerless. On Amery Ice Shelf, where accumulation zones are slow but the front is 600+ m/year, InSAR works in interior areas and speckle tracking — at the front, and both flows are then stitched through the grounding line.
5. Speckle tracking → offset tracking → feature tracking. This is the evolution of one method. Pixel correlation (offset tracking) — least accurate, but works on any images. Feature tracking (tracking stable objects between images) — more accurate, but requires the presence of stable structures (cracks, faults, small irregularities). On Pine Island feature tracking gives accuracy of ±10 m/year on a 12-day baseline, but requires manual markup or a trained neural network.
6. Sentinel-1 TOPS mode as a technical breakthrough. Before Sentinel-1 radars imaged in stripmap mode (continuous strip). Sentinel-1 uses TOPS (Terrain Observation with Progressive Scans) — a radar with electronic scanning in the azimuth direction, which gives uniform image quality across the entire 400 km wide swath and removes the problem of scalloping (amplitude non-uniformity characteristic of ScanSAR). Without TOPS mode 1-day tandem would give artifacts at frame edges, and Phase unwrapping would not converge.
7. Architecture of reference layers. InSAR has only relative measurements — between two images. To get absolute velocity, you need a reference layer — either a stable point (outcrop, bedrock), or an external product. In Antarctica there are almost no such reference points (the entire continent is under ice). Therefore ESA and NASA build reference layers through GPS on ice shelves (POLENET network, 50+ stations on the coast) and through inter-satellite interferometry (TanDEM-X, GRACE-FO). And here an entire pyramid of products is born: GPS → absolute reference → TanDEM-X DEM → InSAR relative → stitched product.
8. Software tooling and standards. All this is held together by open software: GAMMA, ISCE, SNAP (ESA's Sentinel Application Platform), MintPy (Miami InSAR Time-series software in Python). Without them the entire 30-year architecture would be locked in proprietary tooling. The fact that SNAP is free and supported by ESA is perhaps the most underrated factor in modern planetology: open software allows any group in any country to conduct InSAR processing of Sentinel-1 and participate in glacier monitoring.
9. Long time series as a new type of data. Before Sentinel-1, InSAR products for glaciers existed in fragments: ERS tandem 9 months 1995–1996, then Envisat ASAR 2002–2012, then ALOS 2006–2011, then Sentinel-1 2014–present. Stitching these pieces together is a separate engineering task: different wavelengths (C, L), different viewing angles, different orbits, different calibrations. ITS_LIVE (Inter-mission Time Series of Land Ice Velocity and Elevation) is a NASA project that stitched all these series into one continuous archive from 1985. And this is perhaps the most important geophysical database of the last decade — without it we would not know that Pine Island has been accelerating continuously since 1996.
10. Ionospheric artifacts in polar regions. This is a separate detail that hooked me. At polar latitudes InSAR suffers from ionospheric disturbances — auroras cause small-scale inhomogeneities in the ionosphere that shift the phase of the transmitted signal. In interferograms this appears as azimuth streaks — linear artifacts that were previously considered noise of unknown origin, and in the early 2000s were identified as the shadow of the gradient of integrated electron density in the ionosphere. Now they've learned to deal with this through split-spectrum processing and ionospheric correction (module in SNAP), but on Sentinel-1 in polar regions this is still a separate class of errors that needs to be accounted for when processing 1-day tandem pairs.
This story is an architectural lesson about how planetary engineering works not through "one bright tool", but through a network of tools that are stitched together over decades. In 1995 no one knew that 1-day tandem would someday be needed to monitor a collapsing glacier. In 2026 this is the only way to get millimeter accuracy on glaciers that determine the fate of 0.5 meters of global sea level this century.
What struck me most — the resilience of this architecture. ERS-1/2 were developed in the 1980s. The concept of 1-day tandem was an engineering solution for a scientific task of that time (relief interferometry). 30 years later the same concept is resurrected in a new generation of spacecraft for a task that did not exist in the 1980s (monitoring accelerating melting). And this resurrection is not a tribute, but a technical necessity: we still have nothing better for fast glaciers.
A parallel that hooked me as an engineer: this is the same architectural logic as in Tandem-X / GRACE-FO follow-on. Each new generation of satellites does not "replace" old concepts, but layers on them a new level of accuracy, because orbital configuration is an asset that cannot just be "updated", it can only be carefully integrated into a new mission. ESA is doing this for the third time in 30 years (ERS → Envisat → Sentinel-1) — and each time the tandem mode becomes more valuable, because data from previous generations have already shown that glaciers are accelerating.
What I would highlight as the main architectural principle: in planetary engineering the most valuable data is that which cannot be collected in real time. 1-day tandem requires orbit coordination, manual campaign planning, fuel consumption and abandonment of the standard 6-day cycle. This is not a product, but an event. And in an era when everything is being automated, ESA consciously preserves manual mode for the most valuable science — because automation would reduce data quality, and therefore scientific return.
And finally — this is politically loaded science. Pine Island and Thwaites are 60% of the contribution of the West Antarctic Ice Sheet to current sea level rise. If their Eastern Ice Shelf collapses in the next 10–20 years, the MISI (Marine Ice Sheet Instability) mechanism could trigger irreversible glacier acceleration. And that's +0.5–1.5 m to sea level by 2100 — more than everything humanity has adapted in the last 100 years. And exactly 1-day tandem InSAR is what will allow us to track when exactly this tipping point happens. Not global models, not satellite altimetry — but local measurement of ice velocity with millimeter accuracy, which 30 years ago seemed a scientific curiosity, and today is one of the most important engineering tasks on Earth.
🦑