ICEBOOST is a boosted-decision tree model. It averages equally the predictions from XGBoost and CatBoost. Both are trained independently for regression, using a l2 loss, and optimized globally.

MODEL DOMAIN

The model runs on all glaciers defined in the Randolph Glacier Inventory [Pfeffer and The Randolph Consortium, 2017]. The model is trained on version v.62, which includes ice bodies with direct connection to the ice sheets. The model can be deployed on v. 62 or the most recent v. 70.

MODEL TARGET

• Ice thickness measurements from GlaThiDa v. 3.1.0 [Welty et al. 2020]

MODEL INPUTS

• TanDEM-X - Edited Digital Elevation Model (EDEM) - Global, 30m from DLR [González et al. 2021], Website
• Surface ice velocity:
  • Glaciers: optical and SAR-derived map from [Millan et al 2022]
  • Greenland: multi-year InSAR, SAR and Landsat8-derived, from [Joughin et al 2016]
  • Antarctica: SAR phase-derived from [Mouginot et al 2019]
• Temperature above 2 meter: combined ERA5 0.25° [Hersbach et al 2020] and ERA5-Land 0.1°, [Muñoz-Sabater et al 2021]
• Glacier-integrated mass balance from [Hugonnet et al. 2021]
• Glacier spatially-distributed mass balance
  • Glaciers outside the ice sheets: method described in the paper [Maffezzoli et al. 2025]
  • Glaciers inside the ice sheets: from RACMO2 (Greenland, 1km from [Noël and van Kampenhout, 2019]; Antarctica, 2km from [Noël et al, 2023])
• Coastlines from the Global Self-consistent Hierarchical, High-resolution Geography Database (GSHHG), from Wessel P. Walter H. F. Smith, [Website]

Paper accepted for publication in Geoscientific Model Development. https://egusphere.copernicus.org/preprints/2024/egusphere-2024-2455/