2. Data and methods

A total of 11.4 km of ground-penetrating radar (GPR) profiles were recorded for the three largest glaciers in the Pyrenees; 2.6 km on Ossoue in late summer 2006 (Marti and others, Reference Marti2015), 2 km on Monte Perdido in early fall 2016 (López-Moreno and others, Reference López-Moreno2019) and 6.8 km on Aneto in early summer 2020 (Vidaller and others, Reference Vidaller2023). To obtain continuous ice thickness mapsfrom the sometimes sparsely collected GPR profiles, we applied an interpolation method using the radial basis function (RBF) kernel to generate the baseline ice thickness maps for Ossoue (2006), Monte Perdido (2016) and Aneto (2020; Vidaller and others, Reference Vidaller2023). The Ossoue survey (2006) covered ∼50% of the glacier’s area in 2006 (Marti and others, Reference Marti2015); to account for sparse coverage and after subtracting elevation losses (2006–11; −8.2 m in total), we applied RBF interpolation constrained by bedrock topography from LIDAR data (Marti and others, Reference Marti2015; Vidaller and others, Reference Vidaller2021). The cross-validation resulted in a mean ice thickness error of ±1.2 m compared to the validated GPR transects. In the Monte Perdido survey (2016), there is a risk of signal attenuation due to liquid water in the fall surveys, but a residual uncertainty of ±5 m is assumed in water-rich sectors (López-Moreno and others, Reference López-Moreno2019). In the Aneto survey (2020), the dense 6.8 km transects provided high-resolution data and the interpolation uncertainty (±0.3 m) was validated using a randomly selected 30% (test) of the GPR data (Vidaller and others, Reference Vidaller2023).

To estimate the ice thickness distribution in intermediate years (2011, 2020 and 2024), the published surface elevation changes for the periods 2011–20 (Vidaller and others, Reference Vidaller2021) and 2020–23 (Izagirre and others, Reference Izagirre2024a) were then applied to the base maps. The ice thickness reconstructions were updated to 2024 using unmanned aerial vehicles surveys in September/October 2024, using the same procedure as in the previously published studies. An average thinning of −0.7 ma⁻¹ for 2023–24 was calculated for the three glaciers studied.

Future air temperature and precipitation data are based on CMIP5 scenarios from the CNRM-CM5 model, downscaled to 2.5 km using the SAFRAN atmospheric model (available at https://digital.csic.es/handle/10261/271116, accessed on 26/10/2024). The historical climate period, from 1960 to 2006, was data-assimilated with meteorological observations from the mountain range. Climate projections were bias-corrected using the analog method (Quintana-Seguí and others, Reference Quintana-Seguí, Turco, Herrera and Miguez-Macho2017). To estimate committed future ice loss due following the current trend (2010–20), we used the SAFRAN reanalysis dataset. This dataset is based on ARPEGE (2002–23) models corrected with in situ meteorological observations for the mountain range. A moderate greenhouse gas emission scenario (RCP4.5) was also considered to assess the response of Pyrenean glaciers to a slightly warmer climate (+0.2°C) beyond the committed ice loss (FS3). The 2022 extreme ablation year was simulated by identifying the 2022 temperature anomalies during the 2006–23 period.

The IGM simulates the evolution of ice thickness, extent and velocity based on ice mass conservation, surface mass balance (SMB) and ice-flow physics (Jouvet and others, Reference Jouvet, Cordonnier, Kim, Lüthi, Vieli and Aschwanden2022). Ice flow is modeled using a convolutional neural network (CNN) trained on land-terminating alpine glaciers (Jouvet and Cordonnier, Reference Jouvet and Cordonnier2023). This CNN emulator replicates the full-Stokes equations at a reduced computational cost (Jouvet and others, Reference Jouvet, Cordonnier, Kim, Lüthi, Vieli and Aschwanden2022). The IGM is driven by air temperature and precipitation data. SMB is estimated using a monthly positive degree-day (PDD) model (Hock and others, Reference Hock2019). The PDD melt rate factor is calibrated to match with in-situ observed SMB data from 2015 to 2022 (FS 4) and the extreme temperature year (FS5). Three simulations were performed to estimate (i) the committed future loss, regardless of future climate scenarios, based on the 2010–20 climate period (committed ice loss), (ii) future loss for an extreme temperature year (2022) in a loop (extreme) and (iii) future loss for a moderate emissions scenario (RCP4.5). RCP4.5 represents slightly warmer (0.20°C) conditions than those of the committed period and is used as an upper bound for temperature scenarios (FS3). We conducted a calibration process to tune the PDD melt rate factor using literature values ranging from 4 to 7 mm °C⁻¹ d⁻¹, consistent with similar values reported in the literature (Hock, Reference Hock2005). For the future and committed future scenarios, the melt rate was set to 4  mm °C⁻¹ d⁻¹ to reduce the mismatch between observations and IGM simulations. In the extreme temperature year, melt rates were set to 6 mm °C⁻¹ d⁻¹ for Monte Perdido and Aneto, and 4.5 mm °C⁻¹ d⁻¹ for Ossoue, (FS5), based on the SMB and melt rate sensitivity during extreme temperature conditions. This sensitivity analysis is also compared with OGGM’s calibration included in the IGM SMB module, as described in the OGGM documentation (https://docs.oggm.org/en/v1.1/mass-balance.html#calibration), which showed that a PDD melt rate factor of 5 aligns the SMB with the geodetic mass balance for each glacier (RGI6.0 level, Hugonnet and others, Reference Hugonnet2021).

The IGM is forced using a high-resolution DEM acquired from at a resolution of 3 m. Temperature downscaled using a lapse rate of −0.6°C per 100 m, and precipitation downscaled using a vertical gradient of 35 mm per 100 m. The SMB simulations are subtracted from the original ice thickness distribution to obtain a yearly estimate of future ice loss.

3. Results and discussion

The estimated average thickness of the glaciers Aneto, Monte Perdido and Ossoue in October 2024 is 11.6, 15.0 and 15.1 m, respectively. Only below 24.1% of the total area of these glaciers is the ice >20 m thick, while below 36.4% of the area has an ice thickness of <10 m (Fig. 1). Since 2011, the area with a thickness of >20 m was reduced in average by 33% on average. The largest decrease was reconstructed for the Monte Perdido glacier (from 70% to 19%). Estimated thicknesses must be considered together with the thinning rates for these glaciers from 2011 to 2024 (Aneto: 16.4 m, 1.26 m a⁻¹; Monte Perdido: 15.8 m, 1.22 m a⁻¹; Ossoue: 21 m, 1.61 m a⁻¹). Ice thickness losses are accompanied by other concerning signs, such as the collapse and fragmentation of ice bodies and the rapid increase in debris cover (Izagirre and others, Reference Izagirre2024b), that indicate transformation to glacial stagnation.

Figure 1. (a) Current spatial distribution of glaciers in central and southern Europe. (b) Reconstructed ice thickness of the Aneto, Monte Perdido, and Ossoue glaciers during 2024. (c) Areas of these glaciers that had thickness <10 m, 10–20 m and > 20 m during 2011, 2020 and 2024. 3D orthoimages are derived from point clouds obtained with unmanned aerial vehicles in 2024.

Using the IGM, the expected lifetimes of these Pyrenean glaciers were quite similar when using a replica of the observed climate data of the 2010–20 period (committed ice loss) and using climate projections under the RCP4.5 scenario. By 2030, 94% of the ice of Monte Perdido, 91% of the ice of Ossoue and 79% of the ice of Aneto will have melted under the RCP4.5 scenario; these numbers are 83%, 72% and 57% under the committed ice loss scenario (Fig. 2a). For the later scenario and this time period (2024–30), the Aneto Glacier is the only glacier that will have any glacier area thicker than 20 m (Fig. 3), and the probability of glaciers to show distinct ice motion (a defining characteristic of glaciers) is minimal based on Copland (Reference Copland2022). By 2030, the total ice area of Aneto will decline to 0.06 km², Ossoue to 0.05 km² and Monte Perdido to 0.01 km² under the RCP4.5 scenario. By 2034 (the end of the UN Decade of Action for Cryospheric Sciences), there will be little or no remaining ice in Monte Perdido and Ossoue; and Aneto will fragment into small patches with a total glacier area <2 ha, and no single ice body will exceed 1 ha (Figs 2b and 3 and FS1 and FS2).

Figure 2. (a) Reconstructed glacier area and ice thickness during 2024 and predictions for 2030 under two scenarios (committed ice loss, RCP4.5) for the Aneto, Monte Perdido, and Ossoue glaciers, with details in Figures S1–S3. (b) Present and predicted glacier areas under two scenarios (committed ice loss, RCP4.5). (c) Present and predicted distributions of ice thickness under the RCP4.5 scenario. (d) Predicted ice thickness changes (%) relative to 2024 under three scenarios (committed ice loss, RCP4.5 and extreme temperature year).

Figure 3. Spatio-temporal distribution of ice thickness in 2024 (reconstructed) and predictions for 2030 and 2034 under the committed ice loss scenario for the Aneto, Monte Perdido and Ossoue glaciers.

The glacier areas under the committed ice loss scenario are only slightly larger. Interestingly, simulations that replicate the climatic conditions of 2022 show that the RCP 4.5 scenario for 2034 will be reached within only 3–4 years from 2024 (Fig. 2). Even assuming inherent uncertainties of the modeling approach and data used, results of this study confirm that the current climate in the Pyrenees is incompatible with the existence of glaciers, and that this region will be one the next mountain ranges to be declassified as ‘glacierized’. Only a significant and unlikely shift in the current climatic conditions could prolong the duration of the glaciers over time. In the past 14 years, there have been instances of neutral or even slight snow accumulation on the glaciers, as observed in the geodetic mass balance of the Monte Perdido glacier (López-Moreno and others, Reference López-Moreno2019). However, these years represent an outlier compared to those with a negative mass balance, and the positive effects have not been sufficient to counteract the substantial ice losses documented above. Only an improbable sequence of years with favorable conditions for the glaciers could potentially delay the hopeless projections presented in this study.

Pyrenean glaciers have survived the Roman and Medieval Warm Periods (Moreno and others, Reference Moreno2021), but the current warming seems to be too much for them. The disappearance of Pyrenean glaciers means there will be no mountain ranges in southern Europe (below 43° N latitude) with active glaciers, making this event a ‘canary in the coal mine’ warning that similar outcomes will occur in many other mountain ranges around the world (Rounce and others, Reference Rounce2023). The loss of the small glaciers of the Pyrenees will generally have no significant hydrological or environmental consequences compared to other mountain ranges in terms of water availability, energy and food production, natural hazards and mountain erosion (Biemans and others, Reference Biemans2019; Herman and others, Reference Herman, De Doncker, Delaney, Prasicek and Koppes2021; Clason and others, Reference Clason2022; Taylor and others, Reference Taylor, Robinson, Dunning, Carr and Westoby2023). However, the loss of these small glaciers means the disappearance of a unique landscape feature, and this will prevent studies of important paleoenvironmental and microbiological information that is unique to these glaciers (Huber and others, Reference Huber2024). There is also evidence that the disappearance of small glaciers can alter the cultural identity and landscape perception for inhabitants and tourists. It may be hoped that the disappearance of the Pyrenean glaciers will send a clear message to the immediately affected societies and to the world at large, which should then take urgent actions to mitigate the effects of climate change.