Dry Creek in South Australia, flooded after a heavy rain
Dry Creek in South Australia, flooded after a heavy rain. In a new study, researchers test hydrological models to better predict events like this. Credit: Peripitus, CC BY-SA 3.0
Source: Water Resources Research

Hydrologists use mathematical models to predict the movement of water through river channels, floodplains, and aquifers, with the goals of helping water managers decide how to use water efficiently, protect against floods, and deal with a range of other problems. However, hydrological modelers face a common dilemma: The most robust algorithms for optimizing hydrological models are often the slowest and most computationally costly. A new study could help modelers find optimization algorithms that strike the best balance between robustness and cost.

Hydrological models attempt to predict various aspects of the water cycle, including snowmelt, evaporation, and surface runoff, among other phenomena. In assessing and improving the accuracy of their models, hydrologists use a process called calibration, in which optimization algorithms are used to find the best match between a model and observed reality. Some of these optimization algorithms are fast but not very robust, often failing to find the best model solutions. Other algorithms find good solutions much more consistently but are slow and costly to run.

To create a method for helping researchers balance robustness and cost, Kavetski et al. defined an algorithm’s efficiency based on the cost to achieve a certain level of confidence in its results. They then compared two optimization algorithms widely used in environmental applications, the “fast but nonrobust” Levenberg-Marquardt (LM) algorithm and the “robust but slow” shuffled complex evolution (SCE) algorithm. They pitted the two against each another, judging how they optimized four hydrological models in three different Australian river basins: Tambo River, Bass River, and Cooper Creek.

The team found that although the SCE algorithm had a much higher chance of success, the LM algorithm could achieve a comparable level of confidence by being invoked multiple times and would often outperform SCE by virtue of a lower total computational cost. The authors argue that a similar comparison process can be used to benchmark other optimization algorithms, helping environmental modelers calibrate their models more quickly. (Water Resources Research, https://doi.org/10.1029/2017WR022051, 2018)

—Emily Underwood, Freelance Writer


Underwood, E. (2019), Balancing robustness and cost in hydrological model optimization, Eos, 100, https://doi.org/10.1029/2019EO114619. Published on 06 February 2019.

Text © 2019. The authors. CC BY-NC-ND 3.0
Except where otherwise noted, images are subject to copyright. Any reuse without express permission from the copyright owner is prohibited.