Fig. 3: Glaciological, topoclimatic and geological controls on glacial erosion rates.

a–c, Predicted versus measured glacial erosion rate in surge-type (a), land-terminating (b) and marine-terminating (c) environments based on equations presented in this study. These data are compared with previously proposed velocity–erosion rules, based only on previously published datasets^5,7,11,15. MAE values are the mean absolute error of predictions relative to observations, while R² is the coefficient of determination for the natural logarithm of erosion rate measurements and predictions. d, Variable importance in surge-type, marine- and land-terminating environments. Cross-hatched variables are excluded owing to insufficient data. Scores are the scaled coefficient of each explanatory variable, such that the most important (that is, largest coefficient) is 100 for each bootstrap resample (n = 1,000); bars represent the mean variable importance across all resamples. Thick and thin error bars show the approximate ±1 standard deviation (15.87th–84.13th percentile) and ±2 standard deviation (2.28th–97.72nd percentile) ranges of the bootstrapped distributions, respectively. X markers indicate the median importance from the resampling, while hollow circles indicate variable importances from the final trained model. Each continuous variable is presented as both ‘linear’ and ‘log-transformed’.