Expectile-based hydrological modelling for uncertainty estimation: Life after mean

TL;DR

This paper introduces a novel expectile-based method for probabilistic hydrological modelling, providing a new way to estimate uncertainty that complements existing quantile-based approaches and generalizes mean-based methods.

Contribution

It proposes calibrating hydrological models using expectile loss functions, enabling direct uncertainty estimation and extending the modeling framework beyond mean predictions.

Findings

GR6J model outperforms others at all expectile levels

Expectile-based calibration improves uncertainty quantification in hydrology

Method applicable to multiple hydrological models and regions

Abstract

Predictions of hydrological models should be probabilistic in nature. Our aim is to introduce a method that estimates directly the uncertainty of hydrological simulations using expectiles, thus complementing previous quantile-based direct approaches as well as generalizing mean-based approaches. Expectiles are new risk measures in hydrology. Compared to quantiles that use information of the frequency of process realizations over a specified value, expectiles use additional information of the magnitude of the exceedances over the specified value. Expectiles are least square analogues of quantiles and can characterize the probability distribution in much the same way as quantiles do. Moreover, the mean of the probability distribution is the special case of the expectile at level 0.5. To this end, we propose calibrating hydrological models using the expectile loss function, which is strictly consistent for expectiles. We apply our method to 511 basins in contiguous US and deliver predictive expectiles of hydrological simulations with the GR4J, GR5J and GR6J hydrological models at expectile levels 0.5, 0.9, 0.95 and 0.975. An honest assessment empirically proves that the GR6J model outperforms the other two models at all expectile levels. Great opportunities are offered for moving beyond the mean in hydrological modelling by simply adjusting the objective function.