Early aerial expedition photos reveal 85 years of glacier growth and stability in East Antarctica

The East Antarctic Ice Sheet (EAIS) holds over 52 m of potential sea-level equivalent and has shown signs of increased vulnerability in recent decades, with mass losses concentrated in some marine-based catchments of Wilkes Land. Terrestrial basins, largely grounded above sea level, have shown mass gains from increased accumulation, partially offsetting losses elsewhere. However, the scarcity of long pre-satellite observations in East Antarctica limits the ability to distinguish recent changes from natural variability. Historical aerial imagery provides a critical bridge between geological or ice-core records and satellite-era observations by offering regional spatial coverage with useful temporal resolution. Prior Antarctic reconstructions from historical imagery focused on West Antarctica and the Antarctic Peninsula from 1947 onward, leaving a gap in East Antarctica. This study rediscovered and utilized the earliest large-scale aerial photographs from 1936–37, complemented by mid-20th-century Australian campaigns and modern satellite data, to reconstruct glacier evolution for 21 marine-terminating outlet glaciers across three regions of East Antarctica from Lützow-Holm Bay (38°E) to Ingrid Christensen Coast (79°E). The aim is to quantify long-term changes in glacier elevation, velocity, and terminus position to determine whether modern trends exceed natural variability and to contextualize recent dynamics within an 85-year perspective.

Historical aerial imagery has been pivotal in Greenland and Svalbard for reconstructing multi-decadal glacier changes and calibrating models, but equivalent East Antarctic applications have been rare. In Antarctica, prior use of historical imagery has mainly been for West Antarctica and the Antarctic Peninsula (from 1947), showing near-frontal lowering and inland stability since 1960 on the Peninsula, and stable conditions at Byrd Glacier over decades. Geological and geomorphological records provide longer timescales but with large temporal uncertainties, while ice-core derived SMB is highly local. Climate reanalyses before the satellite era (pre-1970s) are uncertain due to sparse data, complicating trend detection. Recent syntheses suggest increased Antarctic snowfall over the last two centuries, with regionally variable signals and potential links to atmospheric warming, ozone depletion, and SAM shifts. Basin-scale studies indicate mass gains in some EAIS terrestrial sectors in recent decades, balancing losses from marine sectors.

- Data sources: Rediscovered Norwegian Thorshavn IV aerial photographs (1936–37; oblique Zeiss; 128 images selected from ~2200), Australian ANARE and National Mapping campaigns (1954–1973; K17/Eagle V trimetrogon and Wild RC9; 155 + 9 images), and modern satellite datasets (REMA, ASTER/Landsat, ITS_LIVE velocities, ICESat/ICESat-2 altimetry). Total ~300 aerial images used. - Study area: 21 marine-terminating outlet glaciers spanning ~2000 km from Lützow-Holm Bay (38°E) to Ingrid Christensen Coast (79°E), within basins totaling ~2.6 M km³ of ice (SLE 7.23 m); focused sub-regions ~0.42 M km³ (SLE 1.15 m). - Photogrammetry: Structure-from-Motion (SfM) and multi-view stereo (Agisoft Metashape v1.7.4). Fiducials detected/validated; images aligned (up to 50,000 keypoints/photo; Medium accuracy; manual tie points as needed). Bundle adjustment solved camera parameters. - Georeferencing: Ground control points (GCPs) placed on stable bedrock, coordinates extracted from REMA reference DEM strips. Same GCPs reused when possible. Sparse/uneven bedrock limited optimal GCP distribution. - DEM and orthomosaic generation: Dense clouds at High quality (1:2 resolution), filtered (Metashape; additional filtering in CloudCompare). Gridded DEMs at 6–25 m pixel size (1936–37 least resolved); orthophoto mosaics at 2–10 m. - DEM co-registration and uncertainty: Co-registered to REMA using Nuth & Kääb method. Accuracy assessed using NMAD over stable bedrock; elevation uncertainty 1.6–9.9 m (largest for 1936–37). Spatial variograms fitted (xDEM package) to characterize autocorrelation and propagate uncertainties to mean elevation change estimates. - Elevation change (dH) estimation: Calculated relative to REMA to minimize data gaps, both across whole glacier areas and within multiple 150-m radius circles located within GCP-bounded overlap zones. Conservative 1 m uncertainty added for REMA. Inter-period dH derived via differencing against the REMA strip (composite equation). Post-2010 elevation changes from differenced co-registered REMA strips. Sensitivity tests performed by subsampling/shifting circles. ICESat/ICESat-2 dH (2003–2021) extracted near grounding lines. - Frontal positions: Manual digitization from orthomosaics, georeferenced aerial images, and historical maps (1937–1973); ASTER/Landsat mapping (1974–2022). Centerline distance changes referenced to glacier-specific maximum extent; crevasse-defined front for highly fractured termini. - Velocities: Manual feature tracking (distinct crevasses) between sets of orthomosaics for Hoseason, Taylor, Jelbart, and Utstikkar in the 1950s–60s; 4–13 features per glacier; intervals 4 months to 4 years. Uncertainty combines MSPE (from GCP residuals) and 1-pixel manual placement error. Modern velocities from ITS_LIVE annual mosaics (2006–2018) extracted over comparable areas. - Climate/SMB: ERA5 monthly means for austral summer (DJF) air temperature and mean annual snowfall (0.25°; 1940–2022). Station summer temperatures from Syowa, Mawson, and Davis (SCAR READER). RACMO2.3p2 SMB (1979–2022) aggregated over the same regional masks.

- Lützow-Holm Bay (six glaciers): Net frontal retreat from 1937 to the 1980s with an almost simultaneous minimum associated with complete fast-ice breakup; additional retreat phases mid-2000s (except Shirase) and 2016–2018. Shirase showed the largest frontal variability (~90 km range, 1963 max to 1988 min). Langhovde and Hovdebreen fronts varied <1 km over 1937–2023. Honnörbrygga experienced >10 km retreat by 1974; partial regrowth since. Despite frontal changes, grounded ice surface elevations remained essentially constant from 1937–2020: e.g., Honnörbrygga mean dH rates +0.01 ± 0.10 m/yr (1937–2016) and +0.24 ± 1.08 m/yr (2016–2020); Langhovde +0.01 ± 0.10 m/yr (1937–2016) shifting to −0.43 ± 0.74 m/yr (2016–2020); Hovdebreen +0.03 ± 0.10 m/yr (1937–2016) to −0.25 ± 0.74 m/yr (2016–2020). Satellite altimetry (2003–2021) shows limited change (+0.02 to +0.06 ± 0.05 m/yr), except Shirase thickening at +0.53 ± 0.27 m/yr. - Kemp & Mac Robertson Land and Ingrid Christensen Coast (15 glaciers): No coherent regional long-term trend in frontal positions (1937–2022); mixed advances/retreats with distances <0.1 to 13.5 km and intervals of 3–50 years. Notable: Mulebreen advanced ~13.5 km since the 1980s; Jelbart remains ~3.5 km behind its 1937 extent. - Long-term thickening: Widespread glacier surface elevation increase in Kemp & Mac Robertson Land (1937–2021) and Ingrid Christensen Coast (1960–2021). Examples: Hoseason +0.23 ± 0.07 m/yr; Taylor +0.11 ± 0.05 m/yr (since 1937). Lesser thickening at Utstikkar +0.06 ± 0.04 m/yr and Jelbart +0.04 ± 0.04 m/yr; excluding dynamic frontal zones increases rates (e.g., +0.11–0.14 m/yr since 1937 for Jelbart/Utstikkar; +0.13–0.42 m/yr since 1973 for Utstikkar). Ingrid Christensen Coast: Shennong +0.17 ± 0.06 m/yr (1960–2011); Flatnes/Hovde/Brown +0.06 to +0.11 ± 0.02–0.05 m/yr. Between 2011–2021: Hovde and Flatnes thin (−0.23 ± 0.15 and −0.07 ± 0.15 m/yr), Shennong thickens (+0.26 ± 0.15 m/yr), Brown ~0; all show net elevation gains over 1960–2021/22. - Velocities: Historical flow speeds (1950s–60s) at Taylor, Jelbart, Hoseason, and Utstikkar are consistent with modern ITS_LIVE velocities (2006–2018) within uncertainties (historical velocity uncertainty ~7.2–10 m/yr), indicating long-term dynamic stability. - Climate linkage: ERA5 indicates mean annual snowfall increased ~50% in Kemp & Mac Robertson Land since 1940 (~17.3 mm w.e. per decade) and ~15% along Ingrid Christensen Coast (~5.3 mm w.e. per decade), while Lützow-Holm Bay snowfall remained nearly constant. Summer air temperatures show no significant long-term warming and rarely exceed 0 °C, implying limited melt influence. Regional snowfall trends correspond to observed long-term glacier thickening or stability. - Overall: Stability and growth in ice elevations seen in recent decades extend back at least to the 1930s, with low-magnitude decadal variability superimposed on century-scale trends.

Frontal-position changes in Kemp & Mac Robertson Land and Ingrid Christensen Coast lack a pronounced regional trend over the last ~85 years, aligning with prior East Antarctic observations of cyclic or regionally variable terminus behavior. In Lützow-Holm Bay, episodes of simultaneous retreat correlate with land-fast sea-ice breakup (1980s and 2016–2018), but these frontal variations did not translate into long-term thinning of grounded ice, indicating limited decadal-scale buttressing by floating tongues at some glaciers. Detection of warm deep water beneath Langhovde during a breakup event suggests localized oceanic influences. The long-term thickening in Kemp & Mac Robertson Land and along Ingrid Christensen Coast, combined with stable velocities and generally stable frontal positions, points to increased snowfall as the primary driver rather than dynamic acceleration or enhanced buttressing. ERA5 snowfall trends since 1940 mirror the elevation increases, while summer temperatures remained mostly below freezing, minimizing melt contributions. Basin-wide altimetry confirms elevation increases since the mid-1980s in the studied sub-regions, though the Prince Olav Coast signal diverges from the Lützow-Holm Bay’s long-term constancy, suggesting spatial heterogeneity. The results underscore the importance of century-scale records to avoid misinterpreting short-term variability as sustained trends and indicate that mass gains in these terrestrial basins have partially mitigated recent losses from marine-based sectors of East and West Antarctica.

By reconstructing glacier elevations, velocities, and terminus positions from the earliest large-scale Antarctic aerial photographs (1936–37), mid-20th-century Australian campaigns, and modern satellite data, this study extends observational records for 21 East Antarctic outlet glaciers back 85 years. It reveals century-scale stability of grounded ice elevations in Lützow-Holm Bay and long-term thickening in Kemp & Mac Robertson Land and along the Ingrid Christensen Coast, consistent with increasing regional snowfall. Historical velocities are broadly unchanged since the 1950s, supporting a snowfall-driven surface mass increase rather than dynamic changes. These long-term perspectives clarify that recent decades of growth in terrestrial EAIS basins are part of a multi-decadal to centennial trend, highlighting the need to contextualize short-term signals. Future work should expand historical reconstructions to additional regions, further integrate altimetry and SMB modeling, and extend climate records to better separate long-term trends from natural variability.

- Historical imagery constraints: Many 1936–37 images lacked sufficient metadata, had oblique geometry, low contrast, overexposure, or were acquired far from the coast, limiting coverage and DEM quality; extensive manual selection and geolocation were required. - Ground control limitations: Sparse and uneven distribution of visible bedrock restricted optimal GCP placement, affecting absolute georeferencing and increasing uncertainties. - DEM accuracy and gaps: Historical DEMs have relatively large uncertainties (1.6–9.9 m), with the oldest DEMs least accurate; data gaps due to low texture/overlap required sampling strategies (150-m circles) that may not capture full spatial variability. - Sample size and representativeness: Elevation changes quantified for 12 glaciers and historical velocities for 4, potentially limiting regional generalizability; some frontal regions are dynamically complex with short-term undulations (±20 m) that complicate dH interpretation. - Climate data uncertainties: Pre-satellite-era (pre-1970s) reanalysis data, especially snowfall, are uncertain; disentangling trends from background variability requires longer series or larger signals. - Attribution limits: Given the modest magnitude and localized nature of observed changes and uncertainties in historical climate forcing, precise attribution of drivers beyond a snowfall linkage remains limited.