EGU22-5252
https://doi.org/10.5194/egusphere-egu22-5252
EGU General Assembly 2022
© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

Improving interpretation of sea-level projections through a machine-learning-based local explanation approach

Jeremy Rohmer1, Remi Thieblemont1, Goneri Le Cozannet1, Heiko Goelzer2, and Gael Durand3
Jeremy Rohmer et al.
  • 1BRGM, Natural Risks, Orléans, France (j.rohmer@brgm.fr)
  • 2NORCE, Norwegian Research Centre, Bjerknes Centre for Climate Research, Bergen, Norway
  • 3IGE, University Grenoble Alpes, Grenoble, France

Sea-level projections are usually calculated from numerical simulations using complex long-term numerical models (or a chain of models) as part of multi-model ensemble studies. Because of their importance in supporting the decision-making process for coastal risk assessment and adaptation, improving the interpretability of these projections is of great interest. Specifically, it is assumed that clear and transparent explanations of projected sea-level changes can increase the trust of the end-users, and ultimately favor their engagement in coastal adaptation. To this end, we adopt the local attribution approach developed in the machine learning community, and we combine the game-theoretic approach known as ‘SHAP’ (SHapley Additive exPlanation, Lundberg & Lee, 2017) with tree-based regression models. We applied our methodology to sea-level projections for the Greenland ice sheet computed by the ISMIP6 initiative (Goelzer et al., 2020) with a particular attention paid to the validation of the procedure. This allows us to quantify the influence of particular modelling decisions and to express the influence directly in terms of sea level change contribution. For Greenland, we show that the largest predicted sea level change, 19cm in 2100, is primarily attributable to >4.5cm (i.e. nearly 25%) to the choice of the model parameter that controls the retreat of marine-terminating outlet glaciers, i.e. to the modelling of the retreat rate of tidewater glaciers; other modelling decisions (choice of global climate model, formulation of the ice sheet model ISM, model grid size, etc.) have only a low-to-moderate influence for this case (with contribution of 1-2cm). This type of diagnosis can be performed on any member of the ensemble, and we show how the aggregation of all local attribution analyses can help guide future model development as well as scientific interpretation, particularly with regard to model spatial resolution or the selection of a specific model formulation.

This study was supported by the PROTECT project, which received funding from the European Union's Horizon 2020 research and innovation program under grant agreement No 869304.

References

Goelzer, H., et al. (2020). The future sea-level contribution of the Greenland ice sheet: a multi-model ensemble study of ISMIP6. The Cryosphere 14, 3071-3096.

Lundberg, S.M., & Lee, S.I. (2017). A unified approach to interpreting model predictions. In Proceedings of the 31st international conference on neural information processing systems (pp. 4768-4777).

How to cite: Rohmer, J., Thieblemont, R., Le Cozannet, G., Goelzer, H., and Durand, G.: Improving interpretation of sea-level projections through a machine-learning-based local explanation approach, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5252, https://doi.org/10.5194/egusphere-egu22-5252, 2022.

Displays

Display file