Validation Practices in Mountain Palaeoglacier Modelling

Numerical palaeoglacier modelling is being applied to mountain glacier systems worldwide to investigate past ice dynamics, climatic controls, and glacier–climate interactions.  At present, there is no overview or consensus framework for evaluating the performance of these models, nor agreed standards for determining whether simulated glacier distribution and geometry are plausibly reconstructed through time.

To assess how palaeoglacier models in mountains are being validated and evaluated, we conducted a systematic review of 94 coupled mass-balance–ice-dynamics palaeoglacier models worldwide. For each study, we recorded validation approaches (visual, quantitative, and/or statistical), the glaciological attributes evaluated (extent, area, ice thickness, mass balance, ice flow, glacier distribution), the validation datasets used (direct geomorphological evidence versus previous model outputs), and the spatial structure of validation (point-, line-, polygon-, or grid-based). We also synthesised validation workflows to document the datasets, metrics, and decision criteria employed.

Our results show that the methodological design used to validate model outputs vary considerably between studies. We found that 80% of reviewed studies incorporate visual evaluation of model outputs to some extent, and 43% rely exclusively on subjective visual interpretation. A further 4% of studies do not perform any form of model validation or evaluation. Most studies validate glacier length (89%), whereas 48% of cases validate only one individual model output. Geomorphic reconstructions are used as validation datasets in only 27% of studies, indicating that most validation workflows rely on mapped glacial landforms. Because glacier length and glacial landforms dominate validation strategies, point- and line-based features constitute the majority of spatial validation data, with polygon- or grid-based approaches remaining comparatively rare.

We observe that the interpretation of model performance often remains subjective and reliant on the judgement of a limited number of research authors, which hinders reproducibility and intercomparison across studies. This reliance on subjective visual interpretation reflects persistent challenges in palaeoglacier model validation that are not being solved by existing tools and workflows designed to measure model-data fit. Our review indicates that uneven geomorphic evidence coverage, positional uncertainty between mapped landforms and simulated ice margins, resolution mismatches between geomorphic data and model outputs, chronological dating uncertainties, parameter uncertainty, and ‘equifinality’ (i.e., where multiple parameter combinations can yield similarly plausible glacier geometries) within glacier models collectively hinder robust quantitative validation.

Therefore, we propose a probabilistic, equifinality-aware validation framework that integrates geomorphically based reconstructions, multiple model outputs (e.g., ice extent, ice thickness), temporal steps (e.g., LGM and present-day), and performance metrics (e.g., overestimation, underestimation). Our approach evaluates ice cover as a probability field derived from an ensemble of acceptable simulations, explicitly acknowledging parameter non-uniqueness of equifinal modelling outputs. This approach identifies spatial patterns of robust agreement and persistent uncertainty, avoids subjective selection of a best-fit simulation, and enables domain-wide validation that captures spatial and temporal heterogeneity in glacier behaviour, thereby providing a more transparent and reproducible basis for evaluating palaeoglacier model–data fit.