Extended Data Fig. 3: Error assessment of the hierarchical V-A model.
From: Evolving resource potential of glacial lakes with ongoing deglaciation
We evaluate the median difference between observed and predicted lake volumes on either side of the suspected breakpoints at A = 0.5 km2 (light red and light blue lines) or A = 5 km2 (dark red and dark blue lines). Legend in the top right applies to all panels. a, Absolute errors for log10-transformed data (that is, the direct output of the V-A model). These errors represent differences in exponents between observation and prediction, effectively showing errors in terms of orders of magnitude. b, Absolute errors after back-transforming the median predicted volume to the original scale, measured in 106 m³. c, Relative errors, expressed as the ratio between predicted and observed lake volume. Positive values indicate an n-fold overestimation, while negative values indicate an n-fold underestimation.
