Ice thickness of the world's glaciers (IceBoost model)

The repository contains: All the world's glaciers: regional zip files (Randolph Glacier Inventory RGI v.62, n=216,502 and RGI v.70, n=274,531) containing individual glacier tif files. IceBoost model (XGBoost mudules: .json files, CatBoost modules: .cbm file). Total 6 files. Training datasets. Users can find both the raw datasets, and the final glacier-encoded dataset used for training. 1. Modeled world's glaciers: Every tif file contains 5 layers: the ice thickness, the ice thickness error, the surface elevation, the geoid elevation, and the Jensen Gap. Ice thickness measurements collected within the glacier, if any, can be found in the attributes. The projection is UTM, except for: Greenland (rgi5, EPSG:3413), and Antarctica (rgi19, south of 60° S, EPSG:3031). Glaciers in rgi19 north of 60° S are released in UTM projection. The horizontal resolution is 100 m, unless specified otherwise. 2. IceBoost models: there are 3 pairs of modules. Each pairs is trained with a different set of features. For XGBoost (the same applies for CatBoost): iceboost_xgb_20251009_with_v.json: trained with all features, with velocity and without lmax. iceboost_xgb_20251009_without_v.json: trained without velocity and without lmax. iceboost_xgb_20251010_without_v_with_lmax.json: trained without velocity, and with lmax. A human-readable brief explanation of the features: 'elevation': surface elevation. 'dist_from_border_km_geom': distance to any ice free region. 'slope50', 'slope75', 'slope100', 'slope125', 'slope150', 'slope300', 'slope450', 'slopegfa': surface slopes after smoothing with gaussian kernels. 'curv_50', 'curv_100', 'curv_150', 'curv_300', 'curv_450', 'curv_gfa': surface curvature. 'smb': surface mass balance. 't2m': temperature above 2 m. 'dist_from_ocean': distance to the ocean. 'v50', 'v100', 'v150', 'v300', 'v450', 'vgfa': ice velocity. 'lmax': length of the glacier (convex hull). For each point inside the glacier, collect the feature set X, and run the model iceboost(X) to get the ice thickness. 3. Training datasets: iceboost_train_20250927_hmineq1.0_tmin20050000_mean_grid_100.csv: training data encoded per glacier. This file is the one used for training. It is created by combining the following raw datasets together: glathida44.csv alaska_ice_thick_train_iceboost.csv asia_ice_thick_train_iceboost.csv jostedalsbreen_ice_thick_train_iceboost.csv patagonia_ice_thick_train_iceboost.csv ruth_glacier_ice_thick_train_iceboost.csv polar_ice_thick_train_iceboost4.parquet =============== IceBoost Web Visualizer at: https://nmaffe.github.io/iceboost_webapp/ IceBoost code on Github at: https://github.com/nmaffe/iceboost ===============