Earth Planet. In our model, we specifically computed this parameterized function for each individual glacier larger than 0.5km2, representing 80% of the total glacierized area in 2015, using two DEMs covering the whole French Alps: a photogrammetric one in 1979 and a SPOT-5 one in 2011. All values correspond to ensemble means under RCP 4.5. Res. CPDD, winter snowfall or summer snowfall) was modified for all glaciers and years. The two recent iterations of the Glacier Model Intercomparison Project (GlacierMIP7,8) have proved a remarkable effort to aggregate, compare and understand global glacier evolution estimates and their associated uncertainties. J. Hosp. Such ice caps cannot retreat to higher elevations in a warming climate, which inhibits this positive impact on MB40 (Fig. We argue that such models can be suitable for steep mountain glaciers. Nature 577, 364369 (2020). A consensus estimate for the ice thickness distribution of all glaciers on Earth. Verfaillie, D., Dqu, M., Morin, S. & Lafaysse, M. The method ADAMONT v1.0 for statistical adjustment of climate projections applicable to energy balance land surface models. Multiple copies of this dataset were created, and for each individual copy a single predictor (i.e. 58, 267288 (1996). Strong Alpine glacier melt in the 1940s due to enhanced solar radiation. Nonlinear sensitivity of glacier mass balance to future climate change unveiled by deep learning. Then in 1884, Allen Mason photographed the glacier for the first time . 282, 104115 (2003). If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate. Another source of discrepancy between both models comes from the different MB data used to calibrate or train the MB models. 799904) and from the Fonds de la Recherche Scientifique FNRS (postdoctoral grant charg de recherches). Glacier Research on Mt. Rainier - Portland State University Nisqually Glacier is the lengthiest of any made in North America. 2013). Both models agree around the average values seen during training (i.e. 12, 1959 (2020). S1a). Nature 568, 382386 (2019). 60, 11401154 (2014). GLAMOS. This work was funded by the Labex OSUG@2020 (Investissements davenir, ANR10 LABX56) and the Auvergne-Rhne-Alpes region through the BERGER project. The linear Lasso MB model suggests a stabilization of glacier evolution, reaching neutral MB rates by the end of the century. The ice thickness data for two of the largest glaciers in the French Alps were modified in order to improve data quality. Six, D. & Vincent, C. Sensitivity of mass balance and equilibrium-line altitude to climate change in the French Alps. ICCV (2015) https://doi.org/10.1109/iccv.2015.123. Therefore, their sensitivities to the projected 21st century increase in PDDs are linear. ALPGM uses a feed-forward fully connected multilayer perceptron, with an architecture (40-20-10-5-1) with Leaky-ReLu44 activation functions and a single linear function at the output. 5). The rest of the story appears to lie primarily in the unique dynamic response of the region's glaciers to climate change. The projections without glacier geometry adjustment explore the behaviour of glaciers which cannot retreat to higher elevations (i.e. 0.78m.w.e. Mer de Glace, 29km2 in 2015), which did show important differences under RCP 8.5 (up to 75%), due to their longer response time. (a) Topographical predictors were computed based on the glaciers annually updated digital elevation model (DEM). 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. Average cumulative MB projections of French Alpine glaciers with a nonlinear deep learning vs. a linear Lasso model for 29 climate scenarios; a with topographical feedback (allowing for glacier retreat) and e without topographical feedback (synthetic experiment with constant mean glacier altitude). A recent study he did found that 80 percent of the glaciers in Alberta and British Columbia could melt in the next 50 years. This is well in agreement with the known uncertainties of glacier evolution models, with glacier ice thickness being the second largest uncertainty after the future GCM-RCM-RCP climate members used to force the model29. For these 32 glaciers, a total of 1048 annual glacier-wide MB values are available, covering the 19672015 period with gaps. ISSN 2041-1723 (online). The cumulative positive degree days (CPDD), snowfall and rainfall dl, are at the glaciers annually evolving centroids. In the meantime, to ensure continued support, we are displaying the site without styles Z. et al. Rveillet, M. et al. 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. 4). For such cases, we assumed that ice dynamics no longer play an important role, and the mass changes were applied equally throughout the glacier. The same was done with winter snowfall anomalies, ranging between 1500mm and +1500mm in steps of 100mm, and summer snowfall anomalies, ranging between 1000mm and +1000mm in steps of 100mm. Ten . Glacier topography is a crucial driver of future glacier projections and is expected to play an important role in determining the magnitude that nonlinearities will have on the mass balance. With this cross-validation we determined a deep learning MB model spatiotemporal (LSYGO) RMSE of 0.59m.w.e. However, the use of ANNs remains largely unexplored in glaciology for regression problems, with only a few studies using shallow ANNs for predicting the ice thickness14 or mass balance13 of a single glacier. The initial glacier ice thickness data for the year 2003 also differs slightly between both models. The Cryosphere 12, 13671386 (2018). Appl. The machine learning models used in this study are useful to highlight and quantify how nonlinearities in MB affect climate-glacier interactions, but are limited in terms of process understanding. This oversensitivity directly results from the fact that temperature-index models rely on linear relationships between PDDs and melt and that these models are calibrated with past MB and climate data. This suggests that linear MB models are adequate tools for simulating MB of mountain glaciers with important topographical adjustment, with the only exception being the most optimistic climate scenarios and glaciers with long response times. The effect of glaciers shrinking to smaller extents is not captured by these synthetic experiments, but this effect is less important for flat glaciers that are dominated by thinning (Fig. Moreover, these differences between nonlinear and linear models appear to come from an over-sensitivity of linear models to increasing ablation season air temperatures, when ice is exposed in a large fraction of glaciers. In this study, we demonstrated the advantages of using deep learning to model glacier MB at regional scales, both in terms of variance and bias. Geomorphology 350, 106913 (2020). Ecol. https://doi.org/10.1016/B978-0-12-821575-3.00009-8. 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. Melting Glaciers: Effects on the Environment, Humans, and Biodiversity Between 1857 and 1979, Nisqually Glacier receded a total of 1,945 meters and advanced a total of 294 meters. Geophys. H.Z. To obtain a1 and an r2 of 0.3531. Grenoble Alpes, Universit de Toulouse, Mto-France, CNRS, CNRM, Centre dtudes de la Neige, Grenoble, France, Department of Geoscience and Remote Sensing, Delft University of Technology, Delft, Netherlands, Laboratoire de Glaciologie, Universit Libre de Bruxelles, Brussels, Belgium, Univ. Graphics inspired by Hock and Huss40. GlaciersUnderstanding Climate Drivers | U.S. Geological Survey 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. Since 2005, study finds that surface melt off glaciers in the North has risen by 900%. S5 and S6). 3). Rev. 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. We further assessed the effect of MB nonlinearities by comparing our simulated glacier changes with those obtained from other glacier evolution studies from the literature, which rely on temperature-index models for MB modelling. Simulating these processes at a large geographical scale is challenging, with models requiring several parametrizations and simplifications to operate. Changes in DDFs with respect to air temperature also strongly depend on albedo, with ice presenting a substantially more nonlinear response than snow. Climate Change Indicators: Glaciers | US EPA 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. Since in ALPGM the climate forcing of glaciers is extracted at the mean glacier altitude, we do not expect these altitude differences to drive important MB differences between models. 44, 13761383 (2017). Here, we perform the first-ever glacier evolution projections based on deep learning by modelling the 21st century glacier evolution in the French Alps. IPCC. Nat. In order to overcome these differences, some adaptations were performed to the GloGEMflow output, accompanied with some hypotheses to ensure a realistic comparison. 2015 IEEE Int. Therefore, we were capable of isolating the different behaviours of the nonlinear deep learning model and a linear machine learning model based on the Lasso30. Nature Communications thanks Mohd Anul Haq, Lauren Vargo, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. In the United States, glaciers can be found in the Rocky Mountains, the Sierra Nevada, the Cascades, and throughout Alaska. Atmospheres 121, 77107728 (2016). Nisqually Glacier | glacier, Washington, United States 6 (2018). Our results confirm an over-sensitivity of temperature-index models, often used by large-scale studies, to future warming. 12, 909931 (2019). "The Patagonia Icefields are dominated by so-called 'calving' glaciers," Rignot said. Since the neural network used here virtually behaves like a black box, an alternative way is needed to understand the models behaviour. In Climate Change 157176 (Elsevier, 2021). The main uncertainties in future glacier estimates stem from future climate projections and levels of greenhouse gas emissions (differences between RCPs, GCMs, and RCMs), whose relative importance progressively increases throughout the 21st century. Zemp, M. et al. Future projections of glacier-wide MB evolution were performed using climate projections from ADAMONT25. Gaining a better understanding of how warming ocean water affects these glaciers will help improve predictions of their fate. The 29 RCP-GCM-RCM combinations available, hereafter named climate members, are representative of future climate trajectories with different concentration levels of greenhouse gases (TableS1). Clarke, G. K. C., Berthier, E., Schoof, C. G. & Jarosch, A. H. Neural networks applied to estimating subglacial topography and glacier volume. The images or other third party material in this article are included in the articles Creative Commons license, unless indicated otherwise in a credit line to the material. Sci. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (2018). and JavaScript. A sensitivity analysis of both MB models revealed nonlinear relationships between PDDs, snowfall (in winter and summer) and glacier-wide MB, which the linear model was only able to approximate (r2=0.41 for the Lasso vs. r2=0.76 for deep learning in cross-validation31; Fig. All these glacier models, independently from their approach, need to resolve the two main processes that determine glacier evolution: (1) glacier mass balance, as the difference between the mass gained via accumulation (e.g. Res. Recent efforts have been made to improve the representation of ice flow dynamics in these models, replacing empirical parametrizations with simplified physical models9,10. Contrasting glacier responses to recent climate change in high-mountain This has the strongest impact under RCP 2.6, where positive MB rates are more frequent (Fig. ADAMONT provides climate data at 300m altitudinal bands and different slope aspects, thus having a significantly higher spatial resolution than the 0.11 from EURO-CORDEX. "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. A.R. This means that these differences linked to MB nonlinearities observed in this experiment could be even greater for such ice caps. This modelling approach was described in detail in a previous publication dedicated to the methods, where the ALpine Parameterized Glacier Model (ALPGM43) was presented31. Model Dev. 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. J. Glaciol. on various mass balance and radiation components) are opening the door for updated and better constrained projections. Ioffe, S. & Szegedy, C. Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift (2015). Simulations for projections in this study were made by generating an ensemble of 60 cross-validated models based on LSYGO. Glacier variations in response to climate change from 1972 to 2007 in contributed to the climate analyses. An accurate prediction of future glacier evolution will be crucial to successfully adapt socioeconomic models and preserve biodiversity. This behaviour is not observed with the nonlinear model, hinting at a positive bias of linear MB models under RCP 2.6. The two models with linear MB responses to PDDs and accumulation simulate more positive MB rates under RCP 2.6, highlighting their over-sensitivity to negative air temperature anomalies and positive snowfall anomalies (Fig. Sci. At the Edge: Monitoring Glaciers to Watch Global Warming - NASA The mountain has three major peaks: Liberty Cap, Point Success, and Columbia Crest (the latter is the summit, located on the rim of the caldera). Annual glacier-wide mass balance (MB) is estimated to remain stable at around 1.2m.w.e. 1). Deep learning applied to glacier evolution modelling. 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. Ser. The Open Global Glacier Model (OGGM) v1.1. MATH Here, we compare our results with those from a recent study that focused on the European Alps10. Climate variations change a glacier's mass balance by affecting ablation and accumulation amounts. Preliminary results suggest winter accumulation in 2018 was slightly above the 2003-2017 average for the Emmons & Nisqually. Nonetheless, since the main GCM-RCM climate signal is the same, the main large-scale long-term trends are quite similar. Bolibar, J., Rabatel, A., Gouttevin, I. et al. Grenoble Alpes, CNRS, G-INP, Laboratoire Jean Kuntzmann, Grenoble, France, You can also search for this author in GloGEMflow relies on EURO-CORDEX ensembles26, whereas ALPGM uses ADAMONT25, an adjusted version of EURO-CORDEX specifically designed for mountain regions. Roe, G. H. Orographic precipitation. Canada's glaciers and ice caps are now a major contributor to sea level change, a new UCI study shows. 12, 168173 (2019). PubMedGoogle Scholar. The Elements of Statistical Learning. However, glacier projections under low-emission scenarios and the behaviour of flatter glaciers and ice caps are likely to be biased by mass balance models with linear sensitivities, introducing long-term biases in sea-level rise and water resources projections. (Zenodo, 2020). A glacier is a large mass of snow and ice that has accumulated over many years and is present year-round. Ecography 40, 913929 (2017). The smallest best performing architecture was used, in order to find a good balance between predictive power, speed, and extrapolation outside the training data. Lett. (b) Climate predictors are based on climatic anomalies computed at the glaciers mean altitude with respect to the 19672015 reference period mean values. Nature 575, 341344 (2019). 0.5) than lower values typical from ice34. Share sensitive information only on official, secure websites.. Earths Future 5, 418435 (2017). Several aquatic and terrestrial ecosystems depend on these water resources as well, which ensure a base runoff during the warmest or driest months of the year6. The advantage of this method is that by only changing the MB model, we can keep the rest of the model components (glacier dynamics and climate forcing) and parameters the same in order to have a controlled environment for our experiment. 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. Our results suggest that, except for the lowest emissions climate scenarios and for large glaciers with long response times, MB models with linear relationships for PDDs and precipitation are suitable for mountain glaciers with a marked topographical feedback. Data 12, 19731983 (2020). 4 vs.S5). Hock, R. Temperature index melt modelling in mountain areas. For intermediate and pessimistic climate scenarios, no significant differences were found (Fig. Huss, M. & Hock, R. A new model for global glacier change and sea-level rise. We ran glacier evolution projections for both the deep learning and Lasso MB models, but we kept the glacier geometry constant, thus preserving the glacier centroid where the climate data is computed constant through time. Gabbi, J., Carenzo, M., Pellicciotti, F., Bauder, A. 51, 573587 (2005). Evol. Glacier Length Variations and Climate Change: Comparative Glacier S5b). See how Mount Rainier glaciers have vanished over time, with this eye Cite this article. Tom R. Andersson, J. Scott Hosking, Emily Shuckburgh, Shfaqat A. Khan, Anders A. Bjrk, Toni Schenk, Romain Hugonnet, Robert McNabb, Andreas Kb, Atanu Bhattacharya, Tobias Bolch, Tandong Yao, Christian Sommer, Philipp Malz, Matthias H. Braun, Romain Millan, Jrmie Mouginot, Mathieu Morlighem, Matthias H. Braun, Philipp Malz, Thorsten C. Seehaus, Nature Communications Nisqually Glacier is well known for its kinematic waves ( Meier, 1962 ), but its mass balance has never been measured due to the difficulty of the glacier terrain. 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. CoRR abs/1505.00853 (2015). 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. Despite the existence of slightly different trends during the first half of the century, both the Lasso and the temperature-index model react similarly under RCP 4.5 and 8.5 during the second half of the century, compared to the deep learning model. Differences for individual glaciers can be much more pronounced, as large and flat glaciers will have topoclimatic configurations that produce more extreme MB rates than small and steep glaciers with a short response time. 15 - Response of glaciers to climate change - Cambridge Core Future high-mountain hydrology: a new parameterization of glacier retreat. The lower fraction of variance explained by linear models is present under all climate scenarios. Average ice velocities on the Nisqually Glacier were previously measured at approximately 200 mm/day (8 in) (Hodge 1974). Earth Sci. NASA finds Asian glaciers slowed by ice loss - Climate Change: Vital S10). Then, we ran multiple simulations for this same period by altering the initial ice thickness by 30% and the glacier geometry update parametrizations by 10%, according to the estimated uncertainties of each of the two methods31. 1960). Park, and S. Beason. In this study, we investigate the future evolution of glaciers in the French Alps and their nonlinear response to multiple climate scenarios. 47 (2020). South American Glaciers Melting Faster, Changing Sea Level Large glaciers and glaciers flowing slowly down shallow slopes respond more sluggishly to short-term climate changes, as might be expected.