Quantifying the Errors Involved in Transferring Suspended Sediment Rating Equations Across Ablation Seasons

Abstract

Bias-corrected suspended sediment rating equations are developed for the data from six separate ablation seasons in the Glacier de Tsidjiore Nouve basin. These equations differ significantly from one another. Suspended loads as predicted by equations used beyond their year of origin fall in the range of 34-278% of the true load. Mean absolute-percentage errors (MAPE) based on forecasts computed from each year’s model applied to the other 5 years of data range from 35 to 81%, with a mean of 52%. The equivalent MAPE for a lumped multi-year model is 38%, for an autocorrelation-correcting generalized least-squares (GLS) regression model, is 15%, and for a transfer function (TF) developed from a single 25 d period, only 5%. Simple structure-transfer functions appear to be more robust to temporal transfer than the usual type of rating equation, and offer the possibility of greatly improving upon the poor performance of OLS rating equations in providing estimates of suspended loads for periods beyond frames of origin.