Fabien Maussion

This paper is motivated by growing concern about tipping dynamics and processes that can cause abrupt, self-amplifying, and irreversible changes across the Earth system and societies. It assesses current scientific understanding of the most severe climate outcomes, and proposes ways to address the deep uncertainty surrounding them. We find substantial, though difficult-to-quantify, probabilities of major disruptions to large-scale ocean circulation, including the Atlantic Meridional Overturning Circulation and the Southern Ocean Abyssal Circulation, with the potential to cause global climate shifts. The probability of ice sheet instability in vulnerable sectors, particularly Antarctica's marine-based sectors, is already high and could accelerate sea level rise over centennial to multi-centennial timescales. Permafrost thaw is underway, with the potential for accelerated greenhouse gas release. Tropical forests face increasing risks of widespread tree mortality, with major consequences for biodiversity, regional climate, and water availability. Coral reef dieback and mountain glacier loss are also underway, while severe compound marine events may trigger irreversible ecosystem changes. Even gradual climate change can produce nonlinear, abrupt impacts when hazards breach physiological or socioecological thresholds, disrupting health, food, or economic systems. It is essential to determine how the probability of such outcomes depends on the level and rate of warming. Current modelling systems challenge our ability to reliably quantify these risks. Sustained investment in observations, process understanding and next-generation earth system modelling is critical to identify precursors and improve early warning. Decision frameworks are needed to support proactive, flexible, anticipatory adaptation. Effective mitigation reduces risks of these highly disruptive outcomes.

Building on Bayesian calibration techniques and leveraging high-resolution climate simulations, we test the current capabilities of calibrating a surface energy balance model for unmonitored glaciers using only globally available satellite observations at Hintereisferner. We developed a multi-objective Bayesian framework for the Coupled Snowpack and Ice surface energy and mass balance model in Python (COSIPY), calibrated with satellite-derived geodetic mass balance, transient snowline altitudes, and mean glacier albedo. The framework is evaluated with in-situ observations and weather station measurements. A Latin Hypercube Sampling ensemble was used to investigate the model's parameter behaviour, establish priors and enable a Markov Chain Monte Carlo calibration helped by a novel and computationally efficient COSIPY emulator. The multi-objective calibration successfully constrains parameter distributions, addressing the equifinality between accumulation and albedo parameters. However, tighter parameter constraints, combined with imperfect climate simulations as forcing, create a model solution that requires a compromise between the three observational targets. Model evaluation shows that the calibrated ensemble reproduces glacier-mean albedos and inter-annual mass-balance variability well, but exhibits a negative bias in mean annual mass balance and a delayed snowline rise. This apparent contradiction arises from both forcing and model limitations. Incoming longwave radiation is overestimated throughout the year, while a warm and humid bias in the meteorological input during the later melt season enhances the positive turbulent fluxes. Early season melt is delayed by underestimated incoming shortwave radiation, and by an overly slow albedo decay caused by too high albedo aging and firn albedo parameters. Such biases in the meteorological forcing data remain a major obstacle to applying surface energy balance models to unmonitored glaciers. The calibration framework presented in this study provides diagnostic tools that help identify shortcomings and compensation effects within the modelling chain, paving the way for correcting forcing biases within a Bayesian framework as more observations become available. The open-source tools developed here have the potential to lower the barrier to studying the atmospheric drivers of glacier change at unmonitored sites with explicit treatment of uncertainty.

Building on Bayesian calibration techniques and leveraging high-resolution climate simulations, we test the current capabilities of calibrating a surface energy balance model for unmonitored glaciers using only globally available satellite observations at Hintereisferner. We developed a multi-objective Bayesian framework for the Coupled Snowpack and Ice surface energy and mass balance model in Python (COSIPY), calibrated with satellite-derived geodetic mass balance, transient snowline altitudes, and mean glacier albedo. The framework is evaluated with in-situ observations and weather station measurements. A Latin Hypercube Sampling ensemble was used to investigate the model's parameter behaviour, establish priors and enable a Markov Chain Monte Carlo calibration helped by a novel and computationally efficient COSIPY emulator. The multi-objective calibration successfully constrains parameter distributions, addressing the equifinality between accumulation and albedo parameters. However, tighter parameter constraints, combined with imperfect climate simulations as forcing, create a model solution that requires a compromise between the three observational targets. Model evaluation shows that the calibrated ensemble reproduces glacier-mean albedos and inter-annual mass-balance variability well, but exhibits a negative bias in mean annual mass balance and a delayed snowline rise. This apparent contradiction arises from both forcing and model limitations. Incoming longwave radiation is overestimated throughout the year, while a warm and humid bias in the meteorological input during the later melt season enhances the positive turbulent fluxes. Early season melt is delayed by underestimated incoming shortwave radiation, and by an overly slow albedo decay caused by too high albedo aging and firn albedo parameters. Such biases in the meteorological forcing data remain a major obstacle to applying surface energy balance models to unmonitored glaciers. The calibration framework presented in this study provides diagnostic tools that help identify shortcomings and compensation effects within the modelling chain, paving the way for correcting forcing biases within a Bayesian framework as more observations become available. The open-source tools developed here have the potential to lower the barrier to studying the atmospheric drivers of glacier change at unmonitored sites with explicit treatment of uncertainty.

The representation of glacier ice flow dynamics in glacier models in response to climate change, such as basal sliding or rheology, remains a critical challenge, particularly in integrating mechanistic models based on differential equations with the growing availability of observational data. Here, we present ODINN.jl, an open-source, modular framework for hybrid glacier modelling that combines scientific machine learning (SciML) with physical process-based approaches based on partial differential equations (PDEs). The framework is designed to facilitate both forward and inverse simulations of glacier evolution, enabling the assimilation of diverse datasets—such as ice thickness, surface velocities, and climate reanalyses—into a unified modelling ecosystem.We have recently released ODINN.jl v1.0 after almost 5 years of work, providing a new architecture and stable API structured as an interconnected suite of Julia packages, each addressing specific tasks: Sleipnir.jl for data management, Muninn.jl for surface mass balance, Huginn.jl for ice flow dynamics, and ODINN.jl itself as the SciML interface for differentiation, optimisation, and hybrid modelling. This architecture allows users to easily customise model components, swap physical parametrisations, and integrate data-driven models (e.g., neural networks) to represent sub-grid processes or empirical laws. A key feature of this framework is its capacity to leverage automatic differentiation and adjoint methods to optimise model parameters, initial conditions, and statistical regressors. Parallelization is available for both forward simulations and advanced inverse methods, such as universal differential equations (UDEs), to explore poorly understood processes like basal sliding or calving. An early prototype of the model showed its potential to learn hidden laws in a noisy synthetic dataset, and with this new stable release we are now moving to large-scale applications using regional remote sensing and field observations such as high-resolution ice surface velocities and ice thickness. ODINN.jl is compatible with the Open Global Glacier Model (OGGM) ecosystem, enabling simulations for virtually any glacier worldwide using preprocessed datasets (e.g., RGI outlines, DEMs, climate reanalyses). This new modelling framework offers a reproducible, open-source solution to bridge the gap between physical understanding and data-driven discovery. Through modularity, scalability, and open-source collaborative approaches, ODINN.jl aims to explore both methodological advancements and large-scale applied modelling in glaciology.

The Arctic has been warming four times faster than the global mean over the last forty years. In response, the mass loss of glaciers has been accelerating and contributing to global sea-level rise. However, many of the mechanisms of this mass loss process are not well understood, especially the calving dynamics of marine-terminating glaciers, in part due to a lack of high-resolution calving front observations. To address this limitation, we developed a novel automated deep learning framework to automatically detect calving fronts in multimodal satellite imagery, including optical and SAR satellite images from Landsat, Terra-ASTER, Sentinel-2, and Sentinel-1 missions. In total, this provides an approximately four-decade time series of calving dynamics across Arctic glaciers and ice caps. The method was tested and developed initially on the Svalbard archipelago, where we identified widespread calving front retreats during the past four decades, including attribution to climate forcing mechanisms. Here, we have extended the analysis across the entire Arctic to produce a time series that starts in the 1980s with quasi-annual temporal resolution, but achieves sub-monthly sampling from 2014 onward, following the launch of Sentinel-1. This provides both seasonal and inter-annual calving dynamics for almost all marine-terminating glaciers in the Arctic, which have been used to explore and understand the drivers of calving processes.

The accelerating loss of glacier mass is disrupting local ecosystems, reshaping hydrological regimes, increasing the likelihood of glacier-related hazards, and undermining the resilience of dependent communities. Quantifying annual glacier mass balance remains challenging because existing estimates rely on either sparse in situ and remote sensing observations or process-based models, each with inherent limitations. Observational approaches, while consistent at global scales, exhibit large regional variations. In contrast, modelling frameworks provide complete spatiotemporal fields but are typically deterministic, sensitive to calibration data, and constrained by assumptions that may overlook key energy balance drivers. Through our approach, we combine the strengths of models and various observational methods in a Bayesian neural network framework. Specifically, we use a Bayesian Neural Field architecture that is first pre-trained on fixed-geometry annual mass balance outputs from the Open Global Glacier Model (OGGM) and then finetuned using multimodal observations available at different spatial and temporal scales. The model uses static glacier characteristics and near-surface climate variables as predictors. We demonstrate through a blocked testing strategy that our framework can fill gaps for glaciers with missing or highly uncertain mass balance records as well as reconstruct meaningful long-term time series. In summary, this approach provides a scalable, uncertainty-aware method for generating spatially and temporally complete annual glacier mass balance estimates.

Mountain glaciers are critical elements of the Earth’s hydrological and climate systems. The rapid changes in glaciers due to climate change hinder our ability to monitor and address associated risks effectively. To address these challenges, we present the Digital Twin Component for Glaciers (DTC Glaciers), part of ESA’s Digital Twin Earth (DTE) programme. In this presentation, we will demonstrate the early prototype of the DTC Glaciers system, developed through close co-design with our stakeholders in the hydropower and water sectors. The demonstration will highlight current capabilities, including regional glacier mass-balance assessment, runoff estimation, and user-informed scenarios. We will also share key lessons learned from Phase 1, focussing on data assimilation of heterogeneous datasets, the practicalities of building adaptive architectures, and the challenges of meeting diverse user needs within a unified framework. Despite these challenges, DTC Glaciers offers a well defined test case to assess the transformative potential of digital twins in climate risk assessments.

Recognising the global importance and vulnerability of mountain glaciers, the United Nations General Assembly declared 2025 the International Year of Glaciers’ Preservation, highlighting the critical role of glaciers in the hydrological cycle and the growing societal risks associated with their rapid decline. These concerns are well founded: glaciers worldwide are retreating rapidly, yet projections have traditionally focused on changes in mass and area rather than on the fate of individual glaciers.Here, we quantify the future evolution of the global number of glaciers and introduce the concept of peak glacier extinction: the year in which the largest number of individual glaciers is projected to disappear. Using three global glacier models and the Randolph Glacier Inventory v6.0, we project the fate of more than 200,000 glaciers worldwide under four policy-relevant global warming scenarios by 2100 (+1.5 °C, +2.0 °C, +2.7 °C and +4.0 °C relative to pre-industrial levels). A glacier is classified as extinct when its area falls below 0.01 km² or its remaining volume declines to less than 1% of its initial value.Across all scenarios, we identify a pronounced mid-century peak in glacier extinction. Under a +1.5 °C pathway, this peak occurs around 2041, with approximately 2,000 glaciers disappearing per year, whereas under a +4.0 °C scenario it shifts to the mid-2050s and intensifies to nearly 4,000 glacier extinctions annually. Regional differences reflect contrasts in glacier size distributions and climatic conditions. Our results highlight the urgency of ambitious climate action. Whether the world faces the loss of 2,000 or 4,000 glaciers per year by mid-century, and whether roughly 20,000 or 100,000 glaciers remain by the end of the century, will be determined by near-term policy and societal choices made today.

Building on Bayesian calibration techniques and leveraging high-resolution climate simulations, we test the current capabilities of calibrating a surface energy balance model for unmonitored glaciers using only globally available satellite observations at Hintereisferner. We developed a multi-objective Bayesian framework for the Coupled Snowpack and Ice surface energy and mass balance model in Python (COSIPY), calibrated with satellite-derived geodetic mass balance, transient snowline altitudes, and mean glacier albedo. The framework is evaluated with in-situ observations and weather station measurements. A Latin Hypercube Sampling ensemble was used to investigate the model's parameter behaviour, establish priors and enable a Markov Chain Monte Carlo calibration helped by a novel and computationally efficient COSIPY emulator. The multi-objective calibration successfully constrains parameter distributions, addressing the equifinality between accumulation and albedo parameters. However, tighter parameter constraints, combined with imperfect climate simulations as forcing, create a model solution that requires a compromise between the three observational targets. Model evaluation shows that the calibrated ensemble reproduces glacier-mean albedos and inter-annual mass-balance variability well, but exhibits a negative bias in mean annual mass balance and a delayed snowline rise. This apparent contradiction arises from both forcing and model limitations. Incoming longwave radiation is overestimated throughout the year, while a warm and humid bias in the meteorological input during the later melt season enhances the positive turbulent fluxes. Early season melt is delayed by underestimated incoming shortwave radiation, and by an overly slow albedo decay caused by too high albedo aging and firn albedo parameters. Such biases in the meteorological forcing data remain a major obstacle to applying surface energy balance models to unmonitored glaciers. The calibration framework presented in this study provides diagnostic tools that help identify shortcomings and compensation effects within the modelling chain, paving the way for correcting forcing biases within a Bayesian framework as more observations become available. The open-source tools developed here have the potential to lower the barrier to studying the atmospheric drivers of glacier change at unmonitored sites with explicit treatment of uncertainty.