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. This synthetic setup allowed us to reproduce the climatic conditions to be undergone by most ice caps, with their mean surface altitude hardly evolving through time. https://doi.org/10.1016/B978-0-12-821575-3.00009-8. This behaviour has already been observed for the European Alps, with a reduction in DDFs for snow during the ablation season of 7% per decade34. Nonlinear sensitivity of glacier mass balance to future climate change The training was performed with an RMSprop optimizer, batch normalization46, and we used both dropout and Gaussian noise in order to regularize it. Huss, M. et al. Finally, there are differences as well in the glacier dynamics of both models, with ALPGM using a glacier-specific parameterized approach and GloGEMflow explicitly reproducing the ice flow dynamics. S8 and Fig. "The Patagonia Icefields are dominated by so-called 'calving' glaciers," Rignot said. 4). When working with spatiotemporal data, it is imperative to respect spatial and temporal data structures during cross-validation in order to correctly assess an accurate model performance48. This is particularly important for the ablation season and for ice DDFs, which need to accommodate the progressively decreasing role that shortwave radiation will play in the future glacier surface energy budget under warmer conditions. The original ice thickness estimates of the methods used by both models are different10,32, and for ALPGM we performed some additional modifications to the two largest glaciers in the French Alps (see Glacier geometry evolution for details). Envelopes indicate based on results for all 660 glaciers in the French Alps for the 19672015 period. Nisqually Glacier | glacier, Washington, United States The Karakoram and the Himalayan mountain range accommodate a large number of glaciers and are the major source of several perennial rivers downstream. This modelling approach was described in detail in a previous publication dedicated to the methods, where the ALpine Parameterized Glacier Model (ALPGM43) was presented31. Nature 577, 364369 (2020). Climate Change 2013: The Physical Science Basis. Glacier-wide MB is simulated annually for individual glaciers using deep learning (i.e. Glacier response to climate perturbations: an accurate linear geometric He, K., Zhang, X., Ren, S. & Sun, J. Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification. Our analyses suggest that these limitations can also be translated to temperature-index MB models, as they share linear relationships between PDDs and melt, as well as precipitation and accumulation. This parametrization reproduces in an empirical manner the changes in glacier geometry due to the combined effects of ice dynamics and MB. This translates into more frequent extreme negative MB rates, and therefore greater differences due to nonlinearities for the vast majority of future climate scenarios (Fig. 5). 3 (2015). 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. The processing chain for extracting glacier outlines from images is composed of four steps: (1) calculation of band ratio, (2) selection of threshold value, (3) creation of binary image and (4) manual digitization. Bolibar, J. et al. Interestingly, this matches the nonlinear, less sensitive response to summer snowfall in the ablation season of our deep learning model (Fig. These results revealed that the main uncertainties on glacier simulations arise from the initial ice thickness used to initialize the model. Clim. This behaviour is not observed with the nonlinear model, hinting at a positive bias of linear MB models under RCP 2.6. The lower fraction of variance explained by linear models is present under all climate scenarios. Earths Future https://doi.org/10.1029/2019EF001470 (2020). Botanical Evidence of the Modern History of Nisqually Glacier - USGS Ser. This dataset applies a statistical adjustment specific to French mountain regions based on the SAFRAN dataset, to EURO-CORDEX26 GCM-RCM-RCP members, covering a total of 29 different future climate scenarios for the 20052100 period. Nat Commun 13, 409 (2022). CAS Jordi Bolibar. Spandre, P. et al. Thus, glacier sensitivity to a step change in climate , glacier response to climate trends , and glacier variance driven by stochastic climate fluctuations are all proportional to , making an important number to constrain. Deep learning captures a nonlinear response of glaciers to air temperature and precipitation, improving the representation of extreme mass balance rates compared to linear statistical and temperature-index models. Article This implies that current global glacier mass loss projections are too low for the lowest emissions climate scenarios and too high for the highest emissions ones, which has direct consequences for related sea-level rise and water resources projections. 10, 42574283 (2017). Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (2018). South American Glaciers Melting Faster, Changing Sea Level Nature 575, 341344 (2019). Nonlinear sensitivity of glacier mass balance to future climate change unveiled by deep learning, https://doi.org/10.1038/s41467-022-28033-0. Such ice caps cannot retreat to higher elevations in a warming climate, which inhibits this positive impact on MB40 (Fig. Nonlinear sensitivity of glacier mass balance to future climate change unveiled by deep learning. 'When the Glaciers Disappear, Those Species Will Go Extinct' Here, we perform the first-ever glacier evolution projections based on deep learning by modelling the 21st century glacier evolution in the French Alps. Our results serve as a strong reminder that the outcomes of existing large-scale glacier simulations should be interpreted with care, and that newly available techniques (such as the nonlinear deep learning approach presented here) and observations (e.g. Carlson, B. Bartk, B. et al. We acknowledge the more than 50 years of glaciological monitoring performed by the GLACIOCLIM French National Observatory (https://glacioclim.osug.fr), which provided essential observations for our modelling study. a Projected mean glacier altitude evolution between 2015 and 2100. Alluvial landscape response to climate change in glacial rivers and the implications to transportation infrastructure. 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. J.B. was supported by a NWO VIDI grant 016.Vidi.171.063. This reduced sensitivity is captured through the response to summer snowfall anomalies, since the sensitivity to positive CPDD anomalies is quite similar for the linear and nonlinear models, as it encompasses both the accumulation and ablation seasons (Fig.