Mountain glacier mass balance from repeat optical stereo DEM differencing
For the world's ~200,000 mountain glaciers, the geodetic method, differencing repeat stereo DEMs and converting volume change to mass, offers the only practical regional-scale mass-balance estimate. Cloud, density assumptions, and coregistration error set the honest limits.
Sensors
- ASTER / Terra: 15 m along-track stereo pair (nadir + 27.6° aft), archive from 2000 to present. The ASTER-SRTM DEM pair (SRTM acquired February 2000) is the standard geodetic baseline for multi-decadal change studies across High Mountain Asia and the Andes. Vertical accuracy typically ±10–20 m on steep terrain before coregistration.
- Pléiades 1A/1B: 0.5 m panchromatic, tri-stereo mode available. DEM post-spacing of 1–2 m achievable; vertical accuracy after bundle adjustment and coregistration can reach ±0.3–1 m on low-slope glacier surfaces. Narrow swath (~20 km) means large glacierised basins require multiple strips.
- SPOT-6/7: 1.5 m panchromatic, stereo and tri-stereo modes. Swath 60 km, making it practical for entire glacier catchments in a single pass. DEM vertical accuracy typically ±1–3 m after coregistration. Useful intermediate resolution between ASTER and Pléiades.
- SkySat constellation (Planet Labs): 0.5 m panchromatic, video and collect modes support in-track stereo. Constellation of ~21 satellites allows flexible revisit. DEM quality on glaciers is competitive with Pléiades but radiometric calibration and stereo geometry vary by collect mode.
- SRTM (NASA/NGA, February 2000): C-band radar interferometry, ~30 m posting globally. Penetrates dry snow by 0–10 m depending on density, introducing a systematic bias when used as a baseline against optical DEMs acquired on bare ice or wet snow. The penetration correction is a known, partially resolvable uncertainty.
Why geodetic mass balance is the only scalable answer
GRACE and GRACE-FO gravimetry resolve glacier mass loss at the scale of large mountain ranges, not individual glaciers. ICESat-2 photon-counting lidar gives superb along-track elevation accuracy but track spacing at mid-latitudes leaves most glacier area unsampled between repeat cycles. In-situ stake networks cover fewer than 500 glaciers worldwide, a fraction of the roughly 200,000 catalogued in the Randolph Glacier Inventory.
The geodetic method sidesteps all of those constraints. Acquire two DEMs of the same glacier separated by years or decades, subtract one from the other, integrate the volume difference over the glacier area, and apply a density conversion. The physics is elementary; the difficulty is in the error budget, which is dominated not by the sensors but by what you assume about the ice.
What a floating roof gives away: the volume-to-mass conversion
Converting a measured volume change to a mass change requires an assumption about the density of the material that disappeared or accumulated. Pure glacier ice is approximately 900 kg/m³. Fresh snow is closer to 50–200 kg/m³. The firn layer that sits between them has a density that varies with compaction history, temperature, and accumulation rate in ways that are not directly observable from orbit.
Huss (2013) established the widely adopted value of 850 ± 60 kg/m³ for the conversion of geodetic volume change to mass change, applicable when the observation period is long enough for the firn column to be in approximate equilibrium. That ±60 kg/m³ uncertainty translates directly into a mass-balance uncertainty of roughly ±7 per cent for a typical thinning signal. Over short windows of two to three years, the assumption breaks down: a single anomalous accumulation season can shift the firn column enough to make the density assumption the dominant error source, larger than DEM vertical uncertainty or coregistration residuals.
This is not a solvable problem with better sensors alone. It is an irreducible physical ambiguity that should appear in any honest mass-balance product.
Coregistration: the error you control, and the one you do not
Two DEMs of the same glacier will disagree for three reasons: real surface change, sensor geometry differences, and misalignment. Coregistration, aligning the DEMs on stable off-glacier terrain before differencing, removes the third source. The Nuth and Kääb (2011) method, which minimises elevation differences on stable slopes as a function of aspect, is the published standard. Applied carefully, it reduces horizontal misalignment to sub-pixel levels and cuts apparent elevation-change noise by factors of two to five.
What coregistration cannot fix is the SRTM C-band radar penetration bias. When SRTM is the earlier DEM, the radar signal penetrated dry snow and firn by an estimated 0–10 m depending on glacier zone, meaning SRTM elevations are systematically lower than a contemporaneous optical DEM would have been. Studies across High Mountain Asia apply corrections of 1–3 m on average, but the correction is spatially variable and not directly measurable from the archive. Users relying on SRTM as a baseline should report this as a systematic uncertainty, not a corrected value.
Cloud cover is the operational constraint that does not appear in the error budget but dominates tasking cost. Glaciers in the monsoon-influenced Himalaya or the maritime ranges of Patagonia may offer only a handful of cloud-free acquisition opportunities per year. Pléiades and SkySat tasking requests must be timed against seasonal weather patterns, and even then a campaign may require two or three years of attempts to obtain a usable stereo pair.
Resolution floors and what they hide
ASTER at 15 m is adequate for glaciers larger than roughly 1 km², which covers the vast majority of glacier area by volume but misses the long tail of small glaciers that are numerically dominant in the RGI. Pléiades at 0.5–1 m post-spacing resolves crevasse fields, supraglacial ponds, and debris-cover boundaries that ASTER smears into a single pixel. That matters because debris-covered ice thins at a different rate than clean ice, and a coarse DEM averages over the two.
Even at Pléiades resolution, the geodetic method cannot distinguish dynamic thinning from surface melt. A glacier losing mass through accelerated flow and calving will show the same DEM signal as one losing mass through increased ablation. Separating the two requires velocity data, typically from optical feature tracking or SAR offset tracking, as a companion product.
Building a mass-balance time series: archive depth and cadence
ASTER has operated continuously since 2000, and the full archive is freely accessible through NASA Earthdata. For any glacier with cloud-free ASTER coverage, a 20-plus-year geodetic record is in principle retrievable today. The practical limit is the availability of scenes with sufficient snow-free ice exposure and low sensor-pointing error. Many High Mountain Asia glaciers have five to ten usable ASTER stereo pairs in the archive; others have fewer than three.
Pléiades archive depth is shorter (operational since 2012) but the geometric quality is high enough that a single well-timed pair can anchor a change estimate against the SRTM or ASTER baseline. The combination of a 2000 SRTM baseline, a circa-2010 ASTER mid-point, and a current Pléiades DEM gives a two-segment time series that can detect acceleration or deceleration in mass loss, which is the scientifically and policy-relevant question.
Satellize can structure tasking campaigns and archive retrieval pipelines for this kind of multi-sensor time series. The analytical workflow we apply follows published geodetic methods of the kind used in the High Mountain Asia mass-balance assessments, and we report density-conversion uncertainty explicitly rather than absorbing it into a single headline figure. Our Tonga crop-estimation programme is a different domain, but the underlying principle is the same: quantify what the data can and cannot tell you before presenting a number to a client.
What a buyer should ask for, and what to be sceptical of
A credible geodetic mass-balance product should state the observation period, the DEM sources and their vertical accuracies on stable terrain, the coregistration method and residual, the glacier area used (and whether it is from a contemporaneous outline or the RGI), and the density assumption with its uncertainty. Any product that reports mass balance to three significant figures without a density-conversion uncertainty range is concealing the dominant error.
For policy applications, such as downstream water resource planning or climate-treaty reporting, the observation window matters as much as the precision. A five-year window with the ±60 kg/m³ density uncertainty may yield a mass-balance uncertainty comparable to the signal itself for a slowly changing glacier. Ten years or more is the practical minimum for a reliable geodetic estimate in most mountain ranges. Shorter windows are useful for detecting rapid change events, such as a surge or a sudden acceleration, but should be labelled as indicative rather than climatological.
Typical figures
| Spatial resolution (DEM post-spacing) | ASTER: ~15 m; SPOT-6/7: ~5–10 m; Pléiades / SkySat: 1–2 m achievable |
| Vertical accuracy on stable terrain (post-coregistration) | ASTER: ±5–15 m; SPOT-6/7: ±1–3 m; Pléiades: ±0.3–1 m (slope-dependent) |
| Minimum glacier area reliably resolved | ~1 km² for ASTER; ~0.1 km² for Pléiades / SkySat |
| Archive depth | SRTM baseline: February 2000; ASTER: 2000–present; Pléiades: 2012–present; SPOT-6/7: 2012–present |
| Density conversion uncertainty | ±60 kg/m³ (Huss 2013 standard); dominant error source for observation windows under ~5 years |
| SRTM C-band penetration bias | 0–10 m into dry snow/firn; typical correction applied in High Mountain Asia studies: 1–3 m |
| Revisit / tasking cadence | ASTER: ~16-day repeat (opportunistic); Pléiades / SkySat: on-demand tasking, cloud permitting |
| Spectral bands used | Panchromatic or pan-sharpened visible for stereo matching; ASTER VNIR bands 1–3 for stereo |
| Recommended minimum observation window | 5 years for indicative estimates; 10+ years for climatological mass-balance rates |
| Delivery formats | GeoTIFF DEM difference grids, glacier-averaged mass-balance table (CSV), uncertainty report (PDF) |
Analytics Satellize can run
| Geodetic mass-balance estimate | DEM differencing on stable-terrain-coregistered stereo DEMs (Nuth & Kääb 2011), volume-to-mass conversion at 850 ± 60 kg/m³ | Per-glacier and basin-aggregated mass-balance table with full uncertainty breakdown, delivered as CSV and PDF report |
| Multi-epoch elevation change map | Pixel-wise DEM subtraction after coregistration, filtered for outliers on steep terrain and shadow zones | GeoTIFF difference raster with per-pixel uncertainty layer, clipped to RGI glacier outlines |
| Debris-cover extent and thinning partition | Spectral classification of debris vs. clean ice on optical imagery, applied as mask to elevation-change grid | GIS polygon layer distinguishing debris-covered and clean-ice thinning rates within each glacier |
| Supraglacial pond and ice-cliff detection | NDWI thresholding and morphological filtering on Pléiades or SkySat imagery co-registered to DEM epoch | Annual pond area time series (CSV) and mapped pond polygons (GeoPackage) |
| Mass-balance acceleration assessment | Two-segment geodetic time series (e.g. 2000–2010 vs. 2010–present) compared against a common baseline DEM | Report quantifying change in mass-loss rate between periods, with confidence intervals on acceleration signal |
| Tasking campaign design and archive audit | Cloud-frequency climatology analysis (ERA5 or MODIS cloud fraction) to identify optimal acquisition windows; archive search across ASTER, Pléiades, SPOT-6/7 catalogues | Tasking brief specifying recommended acquisition dates, sensor priority, and fallback options by season |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.