Glacio‐hydrological model calibration and evaluation

Abstract

Glaciers are essential for downstream water resources. Hydrological modeling is necessary for a better understanding and for future projections of the water resources in these rapidly changing systems, but modeling glacierized catchments is especially challenging. Here we review a wealth of glacio-hydrological modeling studies (145 publications) in catchments around the world. Major model challenges include a high uncertainty in the input data, mainly precipitation, due to scarce observations. Consequently, the risk of wrongly compensating input with model errors in competing snow and ice accumulation and melt process parameterization is particularly high. Modelers have used a range of calibration and validation approaches to address this issue. The review revealed that while a large part ($35%) of the reviewed studies used only streamflow data to evaluate model performances, most studies ($50%) have used additional data related to snow and glaciers to constrain model parameters. These data were employed in a variety of calibration strategies, including stepwise and multi-signal calibration. Although the primary aim of glaciohydrological modeling studies is to assess future climate change impacts, longterm changes have rarely been taken into account in model performance evaluations. Overall, a more precise description of which data are used how for model evaluation would facilitate the interpretation of the simulation results and their uncertainty, which in turn would support water resources management. Moreover, there is a need for systematic analyses of calibration approaches to disentangle what works best and why. Addressing this need will improve our system understanding and model simulations of glacierized catchments.

Improving the Pareto Frontier in multi-dataset calibration of hydrological models using metaheuristics

“…The presented results fortify the conclusion that multi-data calibration is essential to constrain the parameter uncertainty, reduce the equifinality problem and improve the overall consistency performance of hydrological models for different runoff components being modelled. This is in agreement with previous studies (Finger et al, 2015;Sikorska et al, 2015a;Tarasova et al, 2016;van Tiel et al, 2020). While it remains important to maximise the model performance of a primary target variable, which is in most cases the total water streamflow, only the simultaneous calibration with relevant components of the water cycle can improve the overall robustness of the model calibration and can ensure the correct calibration for the right reason (Kirchner, 2006).…”

“…In terms of high-altitude catchments with significant snow processes, a logical inclusion is to utilize observations on snow cover (Chen et al, 2017;Duethmann et al, 2014;Finger, 2018), and in glaciated catchments also on glacier mass balance (Konz and Seibert, 2010;Finger et al, 2011Finger et al, , 2015Etter et al, 2017;He et al, 2019), in addition to streamflow. van Tiel et al (2020) provide an extensive review on multi-objective calibration of hydrological models in glaciated catchments. Indeed multi-output calibration has proven to be very beneficial for realistically simulating hydrological processes in high-altitude catchments with significant snowmelt or glacier-melt contributions (Finger et al, 2011), while a single-output calibration tends to a large underestimation of these processes (Finger et al, 2015;He et al, 2018).…”

Comment on hess-2021-44

“…Engelhardt et al, 2014;Jobst et al, 2018;Stahl et al, 2017). Nonetheless, we need to be careful using such models because there is quite a risk of internal model compensation and on the event scale, this may result in wrong conclusions van Tiel et al (2020b). Also, modelling one or a few catchments will not provide the insights that we have shown here, that there is a dependence on relative glacier cover but that there are regional differences and that catchment and climate characteristics play a role too.…”

On constraining a lumped hydrological model with both piezometry and streamflow: results of a large sample evaluation

“…In catchments affected by snow and/or glaciers, related data -i.e. snow depth or glacier state -can be used in model calibration (Riboust et al, 2018;Tiel et al, 2020). Beyond calibration, extra data can be assimilated into the model to correct its trajectory during runtime; several studies showed an improved performance of hydrological models with assimilation of soil moisture (Aubert et al, 2003a, b;Oudin et al, 2003) or snowpack (Thirel et al, 2013).…”