Searcharxiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.23748

Neural Operators Reveal Scale-Dependent Limits on Recovering Greenland Ice Loss from Sea Level Fingerprints

Xiyuan Bao·Sophie Coulson·Natasha Valencic·S. Mostafa Mousavi·Sönke Dangendorf·Andrew Jason Lloyd·Jerry X. Mitrovica
View original source ↗Read source document ↗Export metadata ↓

Abstract

Sea level fingerprints connect the spatial distribution of rapid ice mass change to sea surface height (SSH) change, but the spatial resolving power of this inverse problem remains poorly quantified. Here we provide the first systematic, scale-resolved quantification of Greenland Ice Sheet mass change structure recoverable from its SSH fingerprint when ice loss geometry varies. We develop a physics-informed neural operator, constrained by an accurate neural-operator emulator of the forward sea level equation, to probe the information content of this ill-conditioned inverse problem. Across synthetic ice change ensembles, broad ice loss patterns can be recovered, but distinct ice fields can produce nearly identical fingerprints, so fine-scale structure depends on the assumed a-priori ice change distribution. Inversion fidelity varies with the wavelength and amplitude of ice mass change. An empirical error law separates these effects and defines a scale-dependent accuracy ceiling. For a satellite-altimetry-derived reference field, the estimated fingerprint-only ceiling is $R^2\sim0.79$ under the present SSH observation geometry and matched-prior assumption. The leading three principal components explain 99.76% of fingerprint variance, yet higher-order components retain information needed to distinguish finer ice change structure. SSH observations near Greenland carry most information on ice loss locations. Applying the leading fingerprint principal components to the observed 1993--2019 SSH rate recovers the large-scale Greenland fingerprint, while the residual field is correlated with basin-scale bathymetry, consistent with unresolved sterodynamic variability. These results establish a quantitative empirical bound on fingerprint-only ice-loss resolving power and identify how independent observations and ice sheet constraints can improve inference.

Explore related subjects

physics.geo-ph

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Xiyuan Bao, Sophie Coulson, Natasha Valencic, S. Mostafa Mousavi, Sönke Dangendorf, Andrew Jason Lloyd, Jerry X. Mitrovica. 2026-10-16. Neural Operators Reveal Scale-Dependent Limits on Recovering Greenland Ice Loss from Sea Level Fingerprints. https://arxiv.org/abs/2610.23748

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Bayesian Geophysical Inversion using Stein Mixture Variational Inference for Multimodal or Complex Solutions

Bayesian geophysical inversion provides a probabilistic framework to estimate subsurface properties, in which uncertainties are represented by a so-called posterior probability density function (pdf). Estimating this pdf is computationally challenging when it is high dimensional or complex, particularly if it may be multimodal. We introduce Stein mixture variational inference (SMVI) for efficient Bayesian geophysical inversion with multimodal or complex posterior solutions. The method combines parametric variational inference with Stein variational gradient descent (SVGD). In SMVI, the posterior pdf is approximated by a mixture (sum) of distributions, in which component distributions are optimised jointly using the SVGD update rule. This construction combines computational efficiency with the flexibility to represent a range of plausible Earth models and allows components to be updated in parallel. Here, SMVI is first tested on two synthetic seismic inverse problems, demonstrating that SMVI can recover non-Gaussian, complex and multimodal posterior solutions accurately and efficiently. Then, we apply SMVI to real seismic data by performing Love wave tomography of the British Isles, and obtain inversion results that are consistent with previous Monte Carlo solutions. Around the East Irish Sea sedimentary basins, SMVI captures bimodal uncertainty associated with ambiguity in basin boundary locations, revealing alternative subsurface structures compatible with the observations. These results indicate that SMVI generalises various previous variational methods, and provides a practical and effective framework for quantifying complex uncertainties and supporting geological interpretation of seismic images.

physics.geo-ph↗
arXiv2026

An exact finite-state representation of viscoelastic memory for accelerated sea-level modeling: SLEvolve v1.0

Glacial Isostatic Adjustment (GIA) models solve the gravitationally self-consistent sea-level equation for the response of the solid Earth and the oceans to changing ice loads. As these models are increasingly embedded in coupled ice-sheet simulations, glacial-cycle calculations, and large ensembles, standard pseudo-spectral 1-D solvers face a steep computational barrier: they evaluate the viscoelastic response by explicitly re-summing every past load increment, scaling quadratically with time steps as $O(N^2)$. Here we show that this computational penalty is an artifact of algorithmic formulation rather than Earth physics. By exploiting the exponential form of the viscoelastic relaxation kernel, we reformulate the viscous response as an exact recursive update of a finite set of modal states, without re-summing the load history. This exact reformulation alone reduces the cost of a run to $O(N)$, yielding a 20.6-fold speed-up relative to a standard full-history 1-D GIA solver for a case study involving a 550-step Antarctic future sea-level projection. Implemented in our open-source solver SLEvolve v1.0, this formulation---combined with fewer, table-free and multithreaded spherical-harmonic transforms, improved file handling, and physics-inspired iteration starts---achieves an end-to-end speed-up of over three orders of magnitude (up to 4,000-fold) against the same 1-D GIA solver, reducing day-long simulations to under thirty seconds on a workstation. Beyond this solid-Earth application, SLEvolve illustrates how human-derived analytical reformulations can be combined with benchmark-harnessed AI-assisted engineering to modernize scientific software. In particular, Numerical regression tests and performance profiling guided the AI agents' implementation choices and helped reject ineffective optimizations.

physics.geo-ph↗
arXiv2026

Gabor-Enhanced Physics-Informed Neural Networks for Fast Simulations of Acoustic Wavefields

Physics-Informed Neural Networks (PINNs) have gained attention for solving partial differential equations, including the scattered Helmholtz equation, due to their flexibility and mesh-free formulation. However, their performance suffers from low-frequency bias, particularly in high-frequency wavefield simulations, limiting convergence speed and accuracy. To address this, we propose a novel and simplified PINN framework that incorporates explicit, trainable Gabor basis functions to efficiently capture the localized and oscillatory nature of wavefields. Unlike previous Gabor-based PINNs that rely on multiplicative filters or auxiliary networks to learn Gabor parameters, our approach redefines the network's task as learning a nonlinear mapping from input coordinates to a custom Gabor coordinate system, where a Gabor function captures the dominant oscillatory behavior of the wavefield. This formulation absorbs the effect of two Gabor parameters into the learned mapping, reducing computational complexity and eliminating the need for manual tuning of hyperparameters. We also present an efficient formulation for incorporating a Perfectly Matched Layer (PML) into the training by deriving real-valued loss components and introducing an analytical expression for the background wavefield. Numerical experiments on various velocity models show that our Gabor-PINN achieves faster convergence, higher accuracy, and greater robustness to architectural design and initialization compared to both traditional PINNs and prior Gabor-based methods. The improvement lies not in adding architectural complexity --as is common in enhanced PINNs -- but in absorbing this complexity into the learned coordinate transformation, making the method both simpler and more effective. Our implementation is publicly available to support reproducibility and future research.

physics.geo-ph↗