A Gaussian process subglacial drainage model emulator

For this project, we constructed a fast Gaussian process emulator of the GlaDS subglacial drainage model. We explore in-depth the various architectural and experimental choices and how these impact the accuracy of the emulator. Using the emulator, which makes predictions about 1000x faster than the physics-based model, we evaluate the model’s sensitivity to eight parameters, highlighting that ~90% of the variance in model outputs is described by three of the parameters. Read the preprint for more!

Preprint

Hill, T., Bingham, D., Flowers, G. E., Hoffman, M. J. (2024). Computationally Efficient Subglacial Drainage Modeling Using Gaussian Process Emulators. ESS Open Archive Preprint https://doi.org/10.22541/essoar.172736254.41350153/v2