Nature Communications thanks Mohd Anul Haq, Lauren Vargo, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Ice-surface altitude changes of as much as 25 meters occurred between 1944 and 1955. Glacier landscapes are expected to see important changes throughout the French Alps, with the average glacier altitude becoming 300m (RCP 4.5) and 400m (RCP 8.5) higher than nowadays (Fig. Farinotti, D. et al. MB rates only begin to approach equilibrium towards the end of the century under RCP 2.6, for which glaciers could potentially stabilize with the climate in the first decades of the 22nd century depending on their response time (Fig. Overall, this results in linear MB models overestimating both extreme positive (Fig. a1 throughout the whole century under RCP 4.5, with glacier retreat to higher elevations (positive effect on MB) compensating for the warmer climate (negative effect on MB). Common climatic signal from glaciers in the European Alps over the last 50 years: Common Climatic Signal in the Alps. 65, 453467 (2019). Universal Differential Equations for Scientific Machine Learning. For these 32 glaciers, a total of 1048 annual glacier-wide MB values are available, covering the 19672015 period with gaps. Through his research in that area, he's seen firsthand the impact of climate change and has been studying the long-term effects of a warming planet. Models were trained using the SAFRAN reanalysis dataset47, including observations of mountain regions in France for the 19582015 period. "Such glaciers spawn icebergs into the ocean or lakes and have different dynamics from glaciers that end on land and melt at their front ends. Ice caps in the Canadian Arctic, the Russian Arctic, Svalbard, and parts of the periphery of Greenland are major reservoirs of ice, as well as some of the biggest expected contributors to sea level rise outside the two polar ice sheets7. Glaciers smaller than 0.5km2 often display a high climate imbalance, with their equilibrium line being higher than the glaciers maximum altitude. 4). The high spatial resolution enables a detailed representation of mountain weather patterns, which are often undermined by coarser resolution climate datasets. acknowledges the funding received from a EU Horizon 2020 Marie Skodowska-Curie Individual Fellowship (grant no. Summer melt was also above average. Relative performance of empirical and physical models in assessing the seasonal and annual glacier surface mass balance of Saint-Sorlin Glacier (French Alps). S4). The vertical blue and red lines indicate the distribution of extreme (top 5%) values for all 21st century projected climate scenarios, with the mean value in the center and 1 indicated by dashed lines. DDFs are known to vary much less with increasing temperatures for intermediate values of albedo (i.e. However, to further investigate these findings, experiments designed more towards ice caps, and including crucial mechanisms such as ice-ocean interactions and thermodynamics, should be used for this purpose. Smiatek, G., Kunstmann, H. & Senatore, A. EURO-CORDEX regional climate model analysis for the Greater Alpine Region: performance and expected future change: climate change in the gar area. is central to a glacier's response: Fig.2ashows 1L.t/for a warming trend of 1 C per century, for three glaciers with dierent (and fixed ). Bolibar, J. ALPGM (ALpine Parameterized Glacier Model) v1.1. Temperature-index models are known to be over-sensitive to temperature changes, mainly due to important differences in the processes contributing to future warming. At present, using complex surface energy balance models for large-scale glacier projections is not feasible yet, mainly due to the lack of input data. Taking into account that for several regions in the world about half of the glacierized volume will be lost during this first half of the 21st century, glacier models play a major role in the correct assessment of future glacier evolution. Interestingly, this matches the nonlinear, less sensitive response to summer snowfall in the ablation season of our deep learning model (Fig. Publishers note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Each one of these cross-validations served to evaluate the model performance for the spatial, temporal and both dimensions, respectively. Overall, the evolving glaciers are expected to undergo rather stable climate conditions under RCP 4.5, but increasingly higher temperatures and rainfall under RCP 8.5 (Fig. This means that these flatter ice bodies, under a warming climate, will be subject to higher temperatures than their steeper counterparts. Secure .gov websites use HTTPS A lock ( ) or https:// means you've safely connected to the .gov website. These predictors are composed of: the mean glacier altitude, maximum glacier altitude, slope of the lowermost 20% altitudinal range of the glacier, glacier surface area, latitude, longitude and aspect. Xu, B., Wang, N., Chen, T. & Li, M. Empirical Evaluation of Rectified Activations in Convolutional Network. GloGEMflow has been previously applied in a study over the whole European Alps, and its temperature-index model was mainly calibrated with MB data from the Swiss Alps. 0.5) than lower values typical from ice34. Hydrol. Bolibar, J., Rabatel, A., Gouttevin, I. 4e). The smallest best performing architecture was used, in order to find a good balance between predictive power, speed, and extrapolation outside the training data. 4e and 5). Three different types of cross validation were performed: a Leave-One-Glacier-Out (LOGO), a Leave-One-Year-Out (LOYO) and a Leave-Some-Years-and-Glaciers-Out (LSYGO). Geophys. Many studies have investigated the effects of climate change on glacier runoff using observations or modelling, with a recent focus on High Mountain Asia 14,16,17 and the Andes 18,19,20.The degree . Google Scholar. In order to do so, we applied a deterministic sampling process as a sensitivity analysis to both the deep learning and the Lasso MB models. Despite these differences, the average altitude difference of the glaciers between both models is never greater than 50m (Fig. A comparison between the two MB models shows that a nonlinear response to climate forcings is captured by the deep learning MB model, allowing a better representation of glacier mass changes, including significantly reduced biases for extreme values (see Methods). Geophys. ADS Each one of these models was created by training a deep learning model with the full dataset except all data from a random glacier and year, and evaluating the performance on these hidden values. Our results show that the mean elevation is far more variable than the kinematic ELA ( Fig. the Open Global Glacier Model - OGGM9) is likely to be less affected by an over-sensitivity to future warming than a more complex model with dedicated DDFs for ice, snow, and firn. A global synthesis of biodiversity responses to glacier retreat. snowfall, avalanches and refreezing) and the mass lost via different processes of ablation (e.g. Provided by the Springer Nature SharedIt content-sharing initiative. Zekollari, H., Huss, M. & Farinotti, D. On the Imbalance and Response Time of Glaciers in the European Alps. By monitoring the change in size of glaciers around the world, scientists can learn about global climate change. New research suggests that climate change-induced melting of the Nisqually Glacier near Seattle, Wash., and other high-elevation glaciers will offset seasonal declines in streamflow until. a1), but when conditions deviate from this mean training data centroid, the Lasso can only linearly approximate the extremes based on the linear trend set on the main cluster of average values (Fig. Under warmer conditions (RCP 8.5), the differences between the linear and nonlinear MB model become smaller, as the topographical feedback from glacier retreat compensates for an important fraction of the losses induced by the late century warmer climate (Fig. The Cryosphere 13, 11251146 (2019). Scand. By performing glacier projections both with mountain glaciers in the French Alps and a synthetic experiment reproducing ice cap-like behaviour, we argue that the limitations identified here for linear models will also have implications for many other glacierized regions in the world. A globally complete, spatially and temporally resolved estimate of glacier mass change: 2000 to 2019. https://meetingorganizer.copernicus.org/EGU2020/EGU2020-20908.html (2020) https://doi.org/10.5194/egusphere-egu2020-20908. B Methodol. Sign up for the Nature Briefing newsletter what matters in science, free to your inbox daily. A well-established parametrization based on empirical functions50 was used in order to redistribute the annually simulated glacier-wide mass changes over each glacier. Together with recent findings by another study41 highlighting the increased uncertainties in ice thickness distribution estimates of ice caps compared to mountain glaciers, our results raise further awareness on the important uncertainties in glacier projections for ice caps. J. Glaciol. 4). Both models agree around the average values seen during training (i.e. 3c). This synthetic experiment is an approximation of what might occur in other glacierized regions with ice caps. Huss, M. & Hock, R. A new model for global glacier change and sea-level rise. ISSN 2041-1723 (online). Immerzeel, W. W. et al. contributed to the extraction of nonlinear mass balance responses and to the statistical analysis. Nonlinear sensitivity of glacier mass balance to future climate change unveiled by deep learning. For that, a dataset of input predictors covering all the glaciers in the French Alps for the 19672015 period was generated from a past MB reconstruction study15. These differences in the received climate signal are explained by the retreat of glaciers to higher altitudes, which keep up with the warming climate in RCP 4.5 but are outpaced by it under RCP 8.5. a Glacier-wide annual MB, b Ice volume, c Glacier area. Analyses were made of the annual photographs . Additionally, glacier surface area was found to be a minor predictor in our MB models31. Alternatively, the Lasso MB model displayed an RMSE of 0.85m.w.e. Large glaciers and glaciers flowing slowly down shallow slopes respond more sluggishly to short-term climate changes, as might be expected. Google Scholar. The Cryosphere 14, 565584 (2020). Average ice velocities on the Nisqually Glacier were previously measured at approximately 200 mm/day (8 in) (Hodge 1974). Internet Explorer). Google Scholar. The climatic forcing comes from high-resolution climate ensemble projections from 29 combinations of global climate models (GCMs) and regional climate models (RCMs) adjusted for mountain regions for three Representative Concentration Pathway (RCP) scenarios: 2.6, 4.5, and 8.525. Years in white in c-e indicate the disappearance of all glaciers in a given massif. Strong Alpine glacier melt in the 1940s due to enhanced solar radiation. Carlson, B. (b) Climate predictors are based on climatic anomalies computed at the glaciers mean altitude with respect to the 19672015 reference period mean values. Grenoble Alpes, CNRS, G-INP, Laboratoire Jean Kuntzmann, Grenoble, France, You can also search for this author in 3b). Roberts, D. R. et al. Both DEMs were resampled and aligned at a common spatial resolution of 25m. For each glacier, an individual parameterized function was computed representing the differences in glacier surface elevation with respect to the glaciers altitude within the 19792011 period. The temperature-index model includes up to three different DDFs, for ice, firn and snow, resulting in three parameters. Since the climate and glacier systems are known to be nonlinear13, we investigate the benefits of using a model treating, among others, PDDs in a nonlinear way in order to simulate annual glacier-wide MB at a regional scale. Nature 577, 364369 (2020). The maximum advance of Nisqually Glacier in the last thousand years was located, and retreat from this point is believed to have started about 1840. Mer de Glace, 29km2 in 2015), which did show important differences under RCP 8.5 (up to 75%), due to their longer response time. Clarke, G. K. C., Berthier, E., Schoof, C. G. & Jarosch, A. H. Neural networks applied to estimating subglacial topography and glacier volume. https://zenodo.org/record/5549758. Earths Future https://doi.org/10.1029/2019EF001470 (2020). The scheme simulates the mass balance as well as changes of the areal . Salim, E., Ravanel, L., Deline, P. & Gauchon, C. A review of melting ice adaptation strategies in the glacier tourism context. A physically-based method for mapping glacial debris-cover thickness from ASTER satellite imagery: development and testing at Miage Glacier, Italian Alps Discovery - the University of Dundee Research Portal Here, with our newly presented approach, we were able for the first time to quantify the effect that stationary parameters in temperature-index mass balance models have on transient glacier evolution. Nonetheless, a better understanding of the underlying processes guiding these nonlinear behaviours at large geographical scales is needed. However, both the climate and glacier systems are known to react non-linearly, even to pre-processed forcings like PDDs13, implying that these models can only offer a linearized approximation of climate-glacier relationships. S5h, j, l). pay council tax north tyneside,